Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the...

257
Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey and Air Pollution Modelling By SHARIF Rayhan SIDDIQUE A thesis presented for the degree of Doctor of Philosophy Business School The University of Western Australia Perth WA, Australia December 2006

Transcript of Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the...

Page 1: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City,

Using the Results of a Stated Preference Survey and Air Pollution Modelling

By

SHARIF Rayhan SIDDIQUE

A thesis presented for the degree of Doctor of Philosophy

Business School The University of Western Australia

Perth WA, Australia

December 2006

Page 2: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Abstract

i

AAbbssttrraacctt

Air pollution is increasingly perceived to be a serious intangible threat to humanity,

with air quality continuing to deteriorate in most urban areas. The main sources of inner

city pollution are motor vehicles, which generate emissions from the tail pipe as well as

by evaporation. These contain toxic gaseous components which have adverse health

effects. The major components are carbon monoxide (CO), nitrogen dioxide (NO2),

nitric oxide (NO), sulphur dioxide (SO2), particulates (PM10), and volatile organic

compounds (VOC). CO and oxides of nitrogen (NOx) are major emissions from cars.

This study focuses on pollutant concentration in Perth city and has sought to develop

measures to improve air quality. To estimate concentrations, the study develops air

pollution models for CO and NOx; on the basis of the model estimates, effective policy

is devised to improve the air quality by managing travel to the city.

Two peaks, due to traffic, are observed in hourly CO and NOx concentrations. Unlike

traffic, however, the morning peak does not reach the level of the afternoon peak. The

reasons for this divergence are assessed and quantified.

Separate causal models of hourly concentrations of CO and NOx explain their

fluctuations accurately. They take account of the complex effects of the urban street

canyon and winds in the city. The angle of incidence of the wind has significant impact

on pollution level; a wind flow from the south-west increases pollution and wind from

the north-east decreases it. The models have been shown to be equivalent to

engineering and scientific models in estimating emission rate in the context of street

canyons. However the study models are much more precise in the Perth context.

A number of existing transport and environmental policies and regulations applied in

various cities could be applicable in Perth. Four broad policy categories are fixed

charge, variable charges, parking fee, and lane restriction measures. Initial analysis

indicated that these measures could appreciably improve air quality in Perth.

Perth city is unlike Sydney and Melbourne in that about 70% of sampled travellers use

private transport to the city and only 30% use public transport, which indicates the

potential for car suppression policies. The study identified the factors which influence

travellers’ mode choice decisions by developing a nested logit model using discrete

choice analysis with existing survey (RP) data.

Page 3: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Abstract

ii

Whereas other studies have used the stated preference method to assess travel mode

selection, trip destination or parking preference, the next step in this study was to use

penalising attributes to determine the impact on car use. The responses from a stated

preference (SP) survey showed about 72% of the sample taking the car to the city for

purposes other than work. Among these about 25% go to shop and 33% for personal

business or recreation. Another significant observation is that about 71% of the sample

are willing to switch to public transport if taking a car to the city is not convenient. This

indicates the potential for imposing measures to reduce car use.

Binary logit models with and without socio-demographic variables explain the

respondents’ reaction to potential policies. Separate work and non-work models were

developed. The models are used to calculate the marginal effects for all attributes and

elasticity for fuel price. In almost all attributes the non-work group is more responsive

than the work group.

Finally, the SP model results are integrated into an econometric model for the purpose

of prediction. The travel behaviour prediction is used to estimate the policy impact on

air quality. The benefit from the air quality improvement is reported in terms of life

saved. The estimated relationships between probability of death and air pollution

determines the number of lives that could be saved under various policy scenarios. A

ratio of benefits to the financial and perceived sacrifices by drivers is calculated to

compare the effectiveness of the suggested policies. A car size charge policy was

found to be the most cost effective measure to ameliorate the environmental impact of

cars in Perth, with a morning peak entry time charge being almost as cost effective.

The study demonstrates the need for appropriate modelling of air pollution and travel

behaviour. It brings together analytical methods at three levels of causality, vehicle to

air pollution, charge to travel response, and air pollution to health.

Page 4: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Acknowledgement

iii

AAcckknnoowwlleeddggeemmeennttss

My deep gratitude and sincere appreciation to my supervisor, Professor John H. E.

Taplin, for his invaluable advice, enormous support, important suggestions,

accessibility, and continuous encouragements in conducting this research work. I would

also like to thank him for taking the burden of correcting my writing and helping me

improves my English. I am very fortunate to have him as my supervisor. I believe his

direction provides the positive turning point of my life.

I would also express my thanks to my co-supervisor Dr. Min Qiu for his help in

understanding analytical aspects of this research work. Mr. Brett Smith and Dr. Doina

Olaru have extended their patient help in specifying models and providing

interpretations.

I gratefully acknowledge the Planning and Transport Research Centre (PATREC) for

awarding me a scholarship to complete the degree. The PATREC scholarship helped

me concentrate on my research and avoid effort of earning other financial support.

I would also thank the Department of Environment, Main Roads Western Australia,

Bureau of Meteorology, and Department for Planning and Infrastructure for providing

data for this research.

Finally, I would thank my friends and family for their support in completing this

research. My wife Sayma and daughters Saba and Samarah always support me by

sacrificing their time with me. My mother, father-in-law and mother-in-law are the

main source of encouragement to complete this research. I am really grateful to them

for their emotional support.

Page 5: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Table of content

iv

TTaabbllee ooff CCoonntteenntt

Abstract ......................................................................................................................... i

Acknowledgements ..................................................................................................... iii

List of Tables .............................................................................................................. ix

List of Figures ........................................................................................................... xiii

Acronyms .................................................................................................................. xvi

CHAPTER ONE ....................................................................................................... 1

Introduction to the Research .......................................................................... 1 1.1 CARS AND AIR QUALITY........................................................................ 1

1.2 PROBLEM STATEMENT IN RELATION TO PREVIOUS WORK......... 2

1.3 PUBLIC POLICY MEASURES AND INSTITUTIONS............................. 3

1.4 AIR COMPOSITION AND POLLUTION .................................................. 5

1.4.1 Health and environmental impact of pollutants .................................... 6

1.4.2 Sources of pollutants in Perth ............................................................... 7

1.5 CLIMATE AND OTHER CHARACTERISTICS OF PERTH.................. 10

1.6 AIM OF THE RESEARCH ........................................................................ 13

1.7 ORGANISATION OF THE THESIS ......................................................... 15

CHAPTER TWO .................................................................................................... 17

Previous analysis and policies on air pollution and travel behaviour ...... 17 2.1 APPROACHES TO AIR POLLUTION AND TRANSPORT .................. 17

2.2 HEALTH IMPACT OF AIR POLLUTION ............................................... 20

2.3 AIR POLLUTION MODELS..................................................................... 22

2.3.1 The STREET Model .................................................................... 24

2.3.2 The OSPM Model ........................................................................ 25

2.3.3 The CALINE4 Model .................................................................. 26

2.3.4 Factors influencing vehicle emission rate .................................... 26

2.4 AIR POLLUTION CONTROL POLICY ................................................... 28

2.4.1 Technology Measures .................................................................. 30

2.4.2 Pricing Measures.......................................................................... 30

2.4.3 Car Manufacturing Industry Measures ........................................ 31

2.4.4 Vehicle Maintenance Measures ................................................... 31

2.4.5 Awareness Measures.................................................................... 32

Page 6: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Table of content

v

2.5 AIR POLLUTION CONTROL IN PERTH, WESTERN AUSTRALIA ... 33

2.6 DISCRETE CHOICE ANALYSIS............................................................. 36

Appendix 2A: The Gaussian Plume Model ............................................................ 38

Appendix 2B: The STREET Model........................................................................ 40

Appendix 2C: The OSPM Model............................................................................ 42

Appendix 2D: The MOBILE6 Emission Model and the AusVeh 1.0 emission

Model .............................................................................................. 44

CHAPTER THREE ................................................................................................. 47

Causal relationships between traffic and air pollution in a Perth city canyon 3.1 INTRODUCTION ...................................................................................... 47

3.2 AIR POLLUTION DEVELOPMENT IN PERTH CITY........................... 48

3.2.1 Factors in the formation of pollutants ................................................. 48

3.2.1.1 Meteorology ...................................................................... 48

3.2.1.2 Street geometry ................................................................ 51

3.2.1.3 Perth city pollution monitoring station............................ 53

3.2.1.4 Traffic volume and emission factor.................................. 54

3.3 DATA STRUCTURE ................................................................................. 55

3.3.1 Air quality data............................................................................. 55

3.3.2 Meteorological data...................................................................... 58

3.3.3 Traffic data ................................................................................... 60

3.4 AIR POLLUTION MODEL ....................................................................... 61

3.4.1 ARIMA Model ............................................................................. 62

3.4.2 Causal Model ............................................................................... 65

3.5 COMPARISON WITH PREVIOUSLY DEVELOPED MODELS ............. 76

3.6 CONCLUSION........................................................................................... 78

Appendix 3A: ARIMA (111) Model – CO.......................................................... 80

Appendix 3B: ARIMA (111) Model – NOx......................................................... 81

Appendix 3C: CM1CO and CM3CO Models...................................................... 82

Appendix 3D: CM1NOx and CM3NOx Models .................................................. 83

CHAPTER FOUR ................................................................................................... 84

Traffic Control Policies to Reduce Pollution in Perth City: a first assessment based on previously estimated elasticities.................................................... 84

4.1 INTRODUCTION ...................................................................................... 84

4.2 MEASURES TARGETED TO THE CITY................................................ 86

Page 7: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Table of content

vi

4.3 ELASTICITY ESTIMATION .................................................................... 86

4.3.1 Response to a Fixed Charge......................................................... 87

4.3.2 Responses to Variable Charges.................................................... 89

4.3.3 Responses to Parking Measures ................................................... 91

4.3.4 Responses to Lane Restriction ..................................................... 93

4.3.5 Average Elasticities...................................................................... 93

4.4 IMPACT ON AIR QUALITY .................................................................... 95

4.4.1 Impact of a Fixed Charge............................................................. 95

4.4.2 Impact of Variable Charges ......................................................... 97

4.4.3 Impact of Parking Measures......................................................... 99

4.4.4 Impact of Lane Restriction......................................................... 100

4.4.5 Combined Impact ....................................................................... 102

4.5 CONCLUSION......................................................................................... 104

CHAPTER FIVE ................................................................................................... 105

Factors Influencing Car Use: A Revealed Preference Analysis .............. 105 5.1 A TWO-PHASE MODELLING APPROACH......................................... 106

5.2 METHODOLOGY.................................................................................... 108

5.2.1 The Revealed Preference (RP) Model........................................ 108

5.2.2 Elasticities .................................................................................. 112

5.2.3 Value of travel time savings (VTTS) ......................................... 113

5.3 DATA STRUCTURE ............................................................................... 113

5.4 ASSUMPTIONS FOR DATA IMPUTATION ........................................ 115

5.5 MODEL ESTIMATION ........................................................................... 117

5.5.1 Multinomial Logit (MNL) Model .............................................. 118

5.5.2 Nested Logit (NL) Model .......................................................... 119

5.5.3 Nested Logit Model with different time parameters (NLDT).... 120

5.5.4 Elasticity Estimation .................................................................. 123

5.5.5 Value of Travel Time Savings (VTTS) Estimation ................... 124

5.6 DISCUSSION AND CONCLUSION....................................................... 124

Appendix 5A: Monthly average unleaded petrol price ...................................... 126

Appendix 5B: Public Transport Fares................................................................ 127

Appendix 5C: Example of Data set.................................................................... 128

Appendix 5D: MNL model without socio-demographic variable ..................... 129

Appendix 5E: MNL model with socio-demographic variable........................... 130

Appendix 5F: NL model with socio-demographic variable............................... 131

Page 8: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Table of content

vii

Appendix 5G: NL model with socio-demographic variable and different time parameter.................................................................................... 132

CHAPTER SIX ..................................................................................................... 133

Stated Preference Survey of car travel to Perth city .......................................... 133 6.1 INTRODUCTION .................................................................................... 133

6.2 POLICY REACTION MODEL................................................................ 134

6.2.1 The Model .................................................................................. 135

6.2.2 Inapplicability of combined SP-RP............................................ 137

6.3 DESIGNING THE SP MODEL ............................................................... 138

6.3.1 Experimental design................................................................... 138

6.3.2 Data collection instrument ......................................................... 140

6.3.3 Sampling frame and sample size................................................ 142

6.3.4 Data collection ........................................................................... 143

6.4 SURVEY OUTCOMES............................................................................ 143

6.5 SUMMARY .............................................................................................. 148

Appendix 6A: Car Trip Response Survey 2005 Questionnaire ......................... 150

CHAPTER SEVEN ............................................................................................... 160

Modelling the reactions of car travellers to Perth city............................. 160 7.1 INTRODUCTION .................................................................................... 160

7.2 REACTIONS TO ATTRIBUTE LEVELS............................................... 161

7.3 MODEL ESTIMATION ........................................................................... 162

7.3.1 Binary logit model for Q1: Whether to take the car .................. 163

7.3.2 Panel Data Model....................................................................... 166

7.3.3 Latent Class Model .................................................................... 167

7.3.4 Binary logit model for work and non-work groups ................... 174

7.4 MARGINAL EFFECTS ANALYSIS ...................................................... 176

7.5 FUEL PRICE ELASTICITY .................................................................... 177

7.6 BINARY LOGIT MODEL FOR Q2 ........................................................ 178

7.7 SUMMARY AND CONCLUSION.......................................................... 179

Appendix 7A: Q1SP Model ............................................................................... 181

Appendix 7B: Q1SPSD Model .......................................................................... 183

Appendix 7C: Q1SPP Model ............................................................................. 185

Appendix 7D: Q1LC Model .............................................................................. 187

Appendix 7E: Q1SPSDW and Q1SPSDNW Models ........................................ 189

Page 9: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Table of content

viii

Appendix 7F: Q2SP Model................................................................................ 192

Appendix 7G: Q2SPSD Model .......................................................................... 194

CHAPTER EIGHT ................................................................................................ 196

Modelling the impact of policy measures on air quality in Perth...................... 196 8.1 INTRODUCTION .................................................................................... 196

8.2 MODEL IMPLEMENTATION................................................................ 197

8.2.1 Estimation of policy impacts...................................................... 197

8.3 MODELLED IMPACT ON TRAFFIC .................................................... 198

8.3.1 Combined SP and RP................................................................. 199

8.3.2 Application of the SP model alone............................................. 203

8.4 ESTIMATED TRAFFIC IMPACT ON AIR QUALITY ......................... 204

8.5 HEALTH IMPACT OF AIR QUALITY IMPROVEMENT ................... 207

8.5.1 Mortality and morbidity from air pollution................................ 207

8.5.2 Health impact of policy implementation.................................... 210

8.6 BENEFIT-SACRIFICE RATIO ............................................................... 212

8.7 CONCLUSION......................................................................................... 215

Appendix 8A: Base case model ........................................................................ 217

Appendix 8B: Fuel price change simulation model........................................... 219

Appendix 8C: Parking charge simulation model ............................................... 220

CHAPTER NINE .................................................................................................. 221

Conclusions ............................................................................................ 221 9.1 RESEARCH FINDINGS ......................................................................... 222

9.1.1 The pollution models ................................................................. 223

9.1.2 The behavioural models ............................................................. 224

9.2 IMPLICATIONS AND INFERENCES.................................................... 226

9.2.1 Financial assessment .................................................................. 226

9.2.2 Other implications and inferences.............................................. 227

REFERENCES......................................................................................................... 229

Page 10: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

List of tables

ix

LLiisstt ooff TTaabblleess

Table 1.1: Sea level air composition

Table 2.1: Comparison of key environmental and transport policy approaches

Table 3.1: Comparative results of ARIMA models for CO level

Table 3.2: Comparative results of ARIMA models for NOx level

Table 3.3: Correlations between CO & explanatory variables

Table 3.4: Correlations between NOx & explanatory variables

Table 3.5: Comparative results of regression models for CO

Table 3.6: Coefficients of the explanatory variables for first difference of hourly CO level

Table 3.7: Comparative results of regression models for NOx

Table 3.8: Coefficients of the explanatory variables for first difference of hourly NOx level

Table 3.9: A comparison of CO/NOx ratio between present study and other studies

Table 3.10: A comparison of emission factors between this study and other studies

Table 3C1: Coefficients of the explanatory variables for hourly CO level for CM1CO model

Table 3C2: Coefficients of the explanatory variables for hourly Ln CO level for CM3CO model

Table 3D1: Coefficients of the explanatory variables for hourly NOx level for CM1NOx model

Table 3D2: Coefficients of the explanatory variables for hourly Ln NOx level for CM3NOx model

Table 4.1: Summary of potential measures for ameliorating air pollution

Table 4.2: Policies applicable to Perth City

Table 4.3: Comparative performance of cordon pricing for Norwegian cities

Table 4.4: Estimates of elasticity of fuel consumption with respect to fuel price

Table 4.5: Travel demand elasticity

Table 4.6: Peak and off-peak travel demand elasticity

Page 11: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

List of tables

x

Table 4.7: Demand elasticities of car and public transport for work trips

Table 4.8: Parking price elasticity

Table 4.9: Average elasticities for four suggested measures

Table 4.10: Impact of fixed charge

Table 4.11: Impact of variable charge by time of days (peak/between peaks/rest of day)

Table 4.12: Impact of parking measure (peak/off-peak)

Table 4.13: Impact of lane restriction

Table 4.14: Combined impact of policy measures

Table 4.15: Annual reduction of pollution in tonnes in the Perth airshed

Table 5.1: Characteristics of sampled travellers to Perth city (from PARTS survey 2002-03)

Table 5.2: Estimated results: MNL model with socio-demographic variables

Table 5.3: Estimated results with NL model

Table 5.4: Success table for NL model

Table 5.5: Estimated results with NL model with different travel time parameter (NLDT)

Table 5.6: Success table for NL model with different travel times

Table 5.7: Comparative results from different models

Table 5.8: Direct and Cross-elasticities with respect to the trip cost

Table 5A1: Monthly average unleaded petrol price

Table 5B1: Public Transport Fares

Table 5C1: First 16 observations of the data set

Table 6.1: Attributes and their levels used for the decision process

Table 6.2: Orthogonal main effect design profiles

Table 6.3: Respondents profile from their last car trip to Perth city

Table 6.4: Summery of metric variables

Table 6.5: Classification by purpose

Table 6.6: Relation between trip purpose and selected alternative if not taking a car

Table 6.7: Car size and trip purpose

Page 12: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

List of tables

xi

Table 6.8: Responses to SP profiles: Proportion of respondents who would take their car to the city

Table 6.9: Responses to SP profiles for work and non-work groups

Table 7.1: Odds ratios for the attributes

Table 7.2: Odds ratios of attributes for two groups

Table 7.3: Binary logit model for Q1 with only SP data (Q1SP)

Table 7.4: Binary logit model for Q1 with SP and socio-demographic data (Q1SPSD)

Table 7.5: Panel data model for Q1 with only SP data (Q1SPP)

Table 7.6: Latent class model with SP and socio-demographic data (Q1LC)

Table 7.7: Cross tabulation of class members and choice alternatives

Table 7.8: Mean values of selected variables for class 1 and class 2

Table 7.9: Latent class and purpose group membership

Table 7.10: Latent class and selected variables

Table 7.11: Responses to SP profiles for work and non-work groups

Table 7.12: Binary logit model for the work (Q1SPSDW) and the non-work (Q1SPSDNW) purpose group with SP and socio-demographic data

Table 7.13: Marginal effects of attributes for the work and non-work groups

Table 7.14: Fuel price elasticities for various models

Table 7.15: Binary logit model for Q2 with only SP data (Q2SP)

Table 7.16: Binary logit model for Q2 with SP and work purpose (Q2SPSD)

Table 8.1: Potential policies and charges assessed

Table 8.2: Model coefficients of various charges

Table 8.3: Conversion of policy measures to parking fee equivalents

Table 8.4: Estimated impact on car use to the city

Table 8.5: Policy responses on car use using SP models

Table 8.6: Relative Risks of death from individual studies

Table 8.7: Relative Risks for respiratory and cardiovascular deaths from individual studies

Table 8.8: Annual value of statistical life saved from different policies or charges

Page 13: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

List of tables

xii

Table 8.9: Benefit-Sacrifice Ratio calculation for various policies

Table 9.1: Relative contributions of variables explaining CO and NOx levels

Table 9.2: Coefficients of the RP model for trips by all modes and the binary SP models for car only: trips to Perth city

Page 14: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

List of figures

xiii

LLiisstt ooff FFiigguurreess

Figure 1.1: Pollutants emitted in Perth and Sydney airsheds in 2003-04

Figure 1.2: Sources of pollutants in Perth in 2003-2004

Figure 1.3: Monthly average temperature in Perth

Figure 1.4: Monthly average rainfall in Perth

Figure 1.5: Monthly average wind speed in Perth

Figure 1.6: Average wind direction in Perth at 9am and at 3pm

Figure 1.7: Research framework

Figure 2.1: Two cross-sections of a Gaussian plume

Figure 2A1: The Gaussian Plume Model

Figure 2B.1: Schematic of cross-street circulation between buildings

Figure 3.1: Morning emissions lost out at sea

Figure 3.2: Morning emissions are trapped in the city

Figure 3.3: Inland event showing emission flows

Figure 3.4: Kwinana event showing emission flows

Figure 3.5: Schematic diagram of flow and dispersion condition in street canyon

Figure 3.6: Dimensions of a street canyon

Figure 3.7: The flow regimes associated with air flow over building and aspect ratio

Figure 3.8: Air quality monitoring stations in Perth Metropolitan area

Figure 3.9: Queens Building monitoring station

Figure 3.10: Hourly CO level in Perth City from October 2003 to June 2005

Figure 3.11: Hourly NO2 level in Perth City from October 2003 to June 2005

Figure 3.12: Hourly NOx level in Perth City from October 2003 to June 2005

Figure 3.13: Hourly variation in an average day, (a) for CO, and (b) for NOx in Perth city

Figure 3.14: Monthly average Temperature, wind speed and pollution level in Perth city over 2003 to 2005

Page 15: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

List of figures

xiv

Figure 3.15: Wind directions in Perth

Figure 3.16: Average hourly traffic in the city for the period between October 2003 and March 2004

Figure 3.17: Pollutant levels for a) a week in late October ’03, and b) a week in late December ‘03

Figure 3.18: ACF and PACF for CO levels for 16 hours lag periods

Figure 3.19: ACF and PACF after first differencing of CO levels for 16 hours lag periods

Figure 3.20: Residual plots for a) ‘seasonal’ model and b) ‘non-seasonal’ model

Figure 3.21: ACF and PACF after first differencing of NOx levels for 16 hours lag periods

Figure 3.22: Residual plots for a) ‘seasonal’ and b) ‘non-seasonal’ models of NOx

Figure 3.23: Traffic and a) average hourly CO level, b) average hourly NOx level in Perth city during Oct 2003 to Mar 2004

Figure 3.24: Residual plots for models a) CM1CO, b) CM2CO, and c) CM3CO

Figure 3.25: Residual plots for models a) CM1NOx, b) CM2NOx, and c) CM3NOx

Figure 3.26: Comparison between actual and model CO levels for (a) a week in late October ‘03, (b) a week in late December ‘03

Figure 3.27: Comparison between actual and model NOx levels for (a) a week in late October ‘03, (b) a week in late December ‘03

Figure 3.28: William Street canyon with wind directions

Figure 3.29: Assumed dimensions of the Perth airshed

Figure 4.1: Price elasticity for small and large change

Figure 4.2a: Impact of fixed charge of $1.0 on CO level for 6-month average data

Figure 4.2b: Impact of fixed charge of $1.0 on NOx level for 6-month average data

Figure 4.3a: Impact of variable charges of $0.4/$0.35/$0.3 for three periods of a day on hourly CO for 6-month average data

Figure 4.3b: Impact of variable charges of $0.4/$0.35/$0.3 for three periods of a day on hourly NOx for 6-month average data

Figure 4.4a: Impact of parking measure of $0.4.0/$3.0 on hourly CO for 6-month average data

Figure 4.4b: Impact of parking measure of $0.4.0/$3.0 on hourly NOx for 6-month average data

Page 16: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

List of figures

xv

Figure 4.5: Example of lane restriction at William Street and Wellington Street intersection

Figure 4.6a: Impact of 25% lane restriction on hourly CO for 6-month average data

Figure 4.6b: Impact of 25% lane restriction on hourly NOx for 6-month average data

Figure 4.7a: Short and long term impact of combined policy on hourly CO

Figure 4.7b: Short and long term impact of combined policy on hourly NOx

Figure 5.1: Hierarchical structure of mode distribution for travellers to Perth City

Figure 6.1: Response pattern of the survey

Figure 7.1: Latent class model structure

Figure 8.1: Hourly CO level under alternative policy measures: a new or added $1.00 charges in each case

Figure 8.2: Hourly NOx level under alternative policy measures: a new or added $1.00 charges in each case

Figure 8.3: Average daily CO and NOx reduction under alternative policy measures

Figure 8.4: Impact of parking fee on car use in the city

Figure 8.5: Lives saved under different policy measures in 2004

Figure 8.6: Benefit-Sacrifice Ratios and financial sacrifices for different measures

Figure 9.1: Initial estimates of annual reduction of air pollution in Perth city under four suggested policies

Figure 9.2: Benefit-sacrifice ratios for four potential policy measures

Page 17: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Acronyms

xvi

AAccrroonnyymmss ABS Australian Bureau of Statistics

ACF Autocorrelation Function

ADR Australian Design Rules

AGO Australian Greenhouse Office

AIC Akaike’s Information Criterion

AQCP Air Quality Control Policy

AQMP Air Quality Management Plan

ARIMA Auto Regressive Integrated Moving Average

ASC Alternative Specific Constant

BTRE Bureau of Transport and Regional Economics

BTCE Bureau of Transport and Communications Economics

CBD Central Business District

CNG Concentrated Natural Gas

CO Carbon Monoxide

CO2 Carbon Dioxide

COPD Chronic Obstructive Pulmonary Disease

CPI Consumer Price Index

CSIRO Commonwealth Scientific and Industrial Research Organisation

DEP Department of Environmental Protection

DOTARS Department of Transport and Regional Service

EA Environment Australia

EEA European Environment Agency

EPA Environmental Protection Agency

ERP Electronic Road Pricing

HC Hydro Carbon

IIA Independence of Irrelevant Alternatives

IID Independently and Identically Distributed

IV Inclusive Value

LPG Liquefied Petroleum Gas

MNL Multinomial Logit

MSE Mean Square Error

NEPM National Environmental Protection Measure

NL Nested Logit

Page 18: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Acronyms

xvii

NO Nitric Oxide

NO2 Nitrogen Dioxide

NOx Oxides of Nitrogen

NPI National Pollutant Inventory

OSPM Operational Street Pollution Model

PACF Partial Autocorrelation Function

PARTS Perth and Regions Travel Survey

PCU Passenger Car Unit

PPSS Perth Photochemical Smog Study

RP Revealed Preference

RR Relative Risk

SBC Schwarz Bayesian Criterion

SP Stated Preference

UNEP United Nations Environment Program

VKT Vehicle Kilometres Travelled

VMT Vehicle Miles Travelled

VOC Volatile Organic Compound

VOSL Value of Statistical Life

VTTS Value of Travel Time Savings

WA Western Australia

WHO World Health Organisation

Page 19: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 1: Introduction to the research

1

CHAPTER ONE

Introduction to the Research

1.1 CARS AND AIR QUALITY

A serious reason for government reluctance to undertake unpalatable measures to

combat air pollution is concern about actual outcomes. If charges or restrictions are to

be imposed on motorists in order to improve conditions then the government taking this

step needs to be reasonably assured that a certain impost will produce a fairly certain

result. Then the achievement can be publicised to demonstrate that the sacrifice made by

those bearing the burden produces a quantifiable benefit.

This study addresses the problem by developing precise measures of car contributions to

pollution and, even more importantly, measures of driver responses to a variety of

potential steps to limit car use in critical areas at critical times. A city with a relatively

modest pollution problem but good data provides the basis for a model of how the

benefits can be measured reliably, even though air quality is considerably better than in

many other cities. Perth is an important case because of its high car ownership and

extremely high car share of commuter trips to the central business district.

Thus the study has two main areas of contribution: one in modelling air pollution and

the other in modelling relevant travel behaviour. These have been brought together to

determine pollution control measures. The study has dealt with three levels of causality:

vehicle to air pollution, charges to responses, and pollution to health. The final step is

to measure the value of lives saved in relation to the burdens that would be imposed on

motorists.

Air pollution modelling is a major part of the study. The estimated causal models are

shown to be equivalent to engineering and scientific models in estimating the emission

rates. They will be shown to be not only equivalent but also precise in providing

unbiased and accurate vehicle coefficients with tight confidence limits. The estimated

pollution models have also been interpreted in terms of a street canyon model in a

precise way.

Page 20: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 1: Introduction to the research

2

1.2 PROBLEM STATEMENT IN RELATION TO PREVIOUS WORK

A review of previous studies in this area indicates the growing concern about

sustainable transport and its environmental impact, but these questions are addressed

from different points of view – socio-economic, transport, and management. According

to Camagni et al. (1998) ‘environment’ can be categorised as physical, economic, and

social and interactions between them, but many previous studies are limited by the

traditional borders of the discipline in which they are rooted and do not address all

components of the problem.

Research conducted by microeconomists has considered social costs, pricing, marginal

costs, who should be liable for the environmental pollution, and what would be the

efficient way to charge travellers. This research has been less strong on equity issues,

social barriers and implementation. There has been a tendency to examine air and noise

pollution, and congestion from a short term point of view.

Transport specialists have tended to investigate environmental issues such as air

pollution, congestion, demand management, and public transport from an engineering

point of view. Studies of the sustainable development of urban areas have covered the

short, medium, and long term. However, the studies have not answered the question,

how could travel behaviour be managed so as to do more to alleviate pollution?

Public health researchers have studied the impact of atmospheric pollution on the

human body. Other social scientists have looked at environmental issues in terms of

equity, ‘livability’, urban development, use of land, and economic growth.

This study goes beyond the boundaries of previous studies. A model to analyse and

predict atmospheric pollution in Perth, as an example of a medium sized city, using

traffic and meteorological variables is developed. Then, environmental control policies

are formulated to improve air quality. In turn, the predicted result is used in stated

preference (SP) analysis to reveal the probable responses of Perth travellers. Better

understanding of traveller sensitivity to environmental impact and control measures will

facilitate improved pollution control management (Louviere et al. 2000) and sustainable

development. Finally, suitable environmental control policies are identified by

observing costs and benefits of the policy implementation. Thus, three major questions

are addressed:

Page 21: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 1: Introduction to the research

3

• What is the nature of the variation in air pollution in Perth and what are

the causal factors that influence this variation?

• How would urban people respond to policies specifically designed to

reduce air pollution?

• Which policies would be most effective in reducing air pollution in the

urban area?

• What would be the health benefits of the reduced pollution?

1.3 PUBLIC POLICY MEASURES AND INSTITUTIONS

Emissions from road, air, rail, and water transport have been partly responsible for acid

deposition, stratospheric ozone depletion and climate change. Road traffic exhaust

emissions have been the cause of much concern about the effects of urban air quality on

human health and troposphere ozone production (Colvile et al. 2001). Transport is

recognised as a major and growing source of air pollution worldwide. The transport

sector causes increasingly serious pollution and health problems since it is closely

associated with heavy urbanisation and high population densities, especially in

developing countries (Chaaban et al. 2001). As a consequence, throughout the world,

many policies, programs and acts regarding emissions and vehicle standards have

evolved over time. These include the 1970 US Clean Air Act, which in 1975 introduced

the first generation of catalytic converter to reduce vehicle emissions and in 1985

imposed stringent emission standards.

City residents are becoming increasingly aware of the deteriorating quality of life. The

pressure on space, the growing problem of urban pollution and growing differences in

living standards are all making cities increasingly difficult places for planners to

manage and for residents to enjoy. Most developed countries like the USA, Australia,

the UK, and Germany are conducting research programs to achieve sustainable

development. In order to address the sustainability of cities, we need to understand the

forces that shape them. Sustainability planning may require changing the way people

think about and solve transport problems.

Effective implementation of policies related to transport and urbanisation requires a

thorough understanding of the behaviour of people living in urban areas. Responsible

Page 22: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 1: Introduction to the research

4

authorities would like to optimise policies related to environmental impact and to know

how responsive people are to policy changes.

The United Nations Environment Program (UNEP), established in 1972, works to

encourage sustainable development through sound environmental practices everywhere.

The World Bank states that in some countries the annual losses of productivity due to

environmental degradation have been estimated at 4% to 8% of GDP (The World Bank

2001).

The U. S. Environmental Protection Agency (EPA), one of the leading environment

monitoring authorities in the world, protects human health and the environment through

regulation and voluntary programs such as Energy Star and Commuter Choice. Under

the Clean Air Act, the EPA sets limits on how much of a pollutant is allowed in the air

anywhere in the United States. Although national air quality has improved over the last

20 years, many challenges remain in protecting public health and the environment. The

EPA’s goal is to have clean air to breathe for this generation and those to follow.

Other environmental protection organisations such as the European Environment

Agency (EEA) and the Protection of Human Environment, a sector of the World Health

Organization (WHO), are involved in monitoring environmental impacts and policies

related to the environment.

Since 1975, Environment Australia (EA) (also known as The Department of

Environment and Heritage) has developed a series of measures to protect the

environment. One of the EA national strategies deals with efficient transport and

sustainable urban planning. It states that the actions to be taken by Australia should

include: i) integrated land use-transport planning; ii) travel demand and traffic

management; iii) encouraging greater use of public transport, walking and cycling; iv)

improving vehicle fuel efficiency and fuel technology; and v) sustainable freight and

logistic systems (http://www.deh.gov.au/atmosphere/transport/index.html).

There are also non-government organisations active in developing environmental

policies, making people aware of environmental issues and encouraging them to use

more public transport, cycling, and walking rather than private motor transport.

However, the world is still at risk of a heavily polluted environment in the future.

Unfortunately, when the environmental effects are not obvious, people do not take them

very seriously. However, successful implementation of any policy is subject to human

Page 23: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 1: Introduction to the research

5

behaviour: “urban planning which reduces the need for motorised travel and encourages

public transport use, and action to influence the behaviour of transport users”1.

Most countries are nowadays concerned about the environmental impact of transport

and are trying to prevent the situation becoming worse. Countries like the USA,

Australia, the UK, Germany, and others, are engaged in formulating policies regarding

the use of public transport and other non-motorised modes in order to reduce emissions

from private transport.

1.4 AIR COMPOSITION AND POLLUTION

According to the Royal Commission on Environmental Pollution (The Committee of

Royal College 1970), pollution is defined as “The introduction by man into the

environment of substances or energy liable to cause hazard to human health, harm to

living resources and ecological systems, damage to structure or amenity or interference

with legitimate use of the environment”. This definition covers a wide range of

pollution including air and water pollution. The pollution created from human activities

is called anthropogenic, while that from animals or plants is called biogenic. We may

have more control over anthropogenic pollution than biogenic.

There is no such thing as “pure air”. Air is a composite of various gaseous components.

Lide (2005) gives the sea-level composition of air (in percent by volume at the

temperature of 15°C and the pressure of 101325 Pa) as presented in Table 1.1. The

main components which form air are nitrogen (78%) and oxygen (21%). Some other

gaseous components exist in the air at very low volumes.

The composition of air varies with meteorological factors and human activities.

Anthropogenic emissions mainly include six common pollutants: carbon monoxide

(CO), nitrogen oxides (NOx), ozone (O3), lead, particles smaller than 10μm (PM10), and

sulphur dioxide (SO2), and also the combined gases called greenhouse gas. Greenhouse

gases include (but are not limited to) carbon dioxide (CO2), ozone (O3), methane (CH4),

nitric oxide (N2O), sulphur hexafluoride (SF6), and chlorofluorocarbons (CFC). These

gases affect human life by contributing to global warming. All of the pollutants have

direct or indirect effects on plants and animals (including humans). A brief description

of health and environmental impacts is given in the following section.

1 Module 5, The National Greenhouse Strategy, The Department of Environment and Heritage, Australia

Page 24: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 1: Introduction to the research

6

Table 1.1: Sea level air composition

Name Symbol Percent by Volume

Nitrogen N2 78.080000%

Oxygen O2 20.950000%

Argon Ar 0.930000%

Carbon Dioxide CO2 0.031400%

Neon Ne 0.001818%

Methane CH4 0.000200%

Helium He 0.000524%

Krypton Kr 0.000114%

Hydrogen H2 0.000050%

Xenon Xe 0.000009%

Source: Lide (2005)

1.4.1 Health and environmental impact of pollutants

Carbon monoxide (CO) is a colourless and odourless gas that is formed when carbon in

fuel is not burnt completely. It can cause harmful health effects by reducing oxygen

delivery to the body’s organs and tissues. Even a low level of CO can be a major threat

to a heart disease sufferer; a healthy person can be affected in their central nervous

system and respiratory system.

Nitrogen oxides (NOx) is a generic term for a group of highly reactive gases containing

nitrogen and oxygen in varying proportions. They cause a wide variety of health and

environmental impacts, including respiratory problems, damage to lung tissues, and can

even be a cause of premature death. NOx is also one cause of acid rain, water quality

deterioration, global warming, and visibility impairment. NOx is also an ingredient in

ozone formation as discussed below.

Ozone (O3) is a gas composed of three atoms of oxygen. It is not usually emitted

directly into the air, but is created by a chemical reaction between oxides of nitrogen

(NOx) and volatile organic compounds (VOC) in the presence of sunlight and heat. It

can irritate lung airways and cause inflammation much like sunburn. Other symptoms

include wheezing, coughing, and breathing difficulty during outdoor activities.

Repeated exposure to ozone for several months may cause permanent lung damage.

Even a low level of this gas may trigger a variety of health problems, such as asthma,

Page 25: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 1: Introduction to the research

7

pneumonia, and bronchitis. Other than health problems, this gas has the ability to

damage the leaves of trees and other plants and to reduce crop and forest yields.

Lead is a cumulative poison that can damage human organs, such as kidneys, liver,

brain and nerves. It can also cause mental retardation and behavioural disorders. High

blood pressure and increased heart disease can be caused by excessive exposure to lead.

Lead can enter into the water system through sewage and industrial waste and can cause

reproductive damage to aquatic life.

Particulate Matter smaller than 10µm (PM10) is found in the air and includes dust, dirt,

soot, smoke, and liquid droplets. It causes a wide variety of health and environmental

impacts. Scientific studies have found links between breathing PM10 and a series of

health problems, such as asthma, respiratory symptoms, chronic bronchitis, and even

premature death. PM10 can also create problems for aquatic life and plants. The effect

of particles on health is becoming a matter of serious concern but is outside the scope of

this study.

Finally sulphur dioxide (SO2) is a gas which can dissolve easily in water. It is found to

have similar effects to other pollutants on humans and on plants; in particular,

respiratory problems can be initiated by exposure to SO2. Sulphate particles are the

major cause of reduced visibility in many parts of cities.

All air pollutants discussed have the ability to cause deterioration of human and plant

life. Certainly an increasing amount of these pollutants will have impacts on our regular

activities and on future generations. To deal with these pollutants we need to know

their sources.

1.4.2 Sources of pollutants in Perth

Motor vehicle exhaust contributes about 66%2 of total CO emissions nationwide in the

USA, whereas in Perth, about 82%3 of all CO emissions are from motor vehicles.

Another important consideration in generation of CO is the age of vehicle; the older the

vehicle the higher the contribution to exhaust emission. In the 2004 Australian vehicle

fleet 31% of vehicles were pre-1990 but Western Australia has 34% in this category

(ABS 2004). However Perth is better than Sydney and Melbourne in terms of air

2 2001 figure according to US Department of Transportation 3 2003-2004 figure according to National Pollutant Inventory, Australia

Page 26: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 1: Introduction to the research

8

quality. All six pollutant levels in Perth were less than in Sydney in 2003-2004 (Figure

1.1). More carbon monoxide was produced in both cities than other pollutants. Air

quality in Perth is far better than some polluted cities such as Mexico, Bangkok and

Mumbai, but Perth was selected for study for several reasons. First, air is polluted

mainly through industrial activities, residential activities, and transport; and since there

are no industrial activities and very few residences in central Perth city, the impact of

transport on air pollution can be measured separately. Secondly, Perth is one of the

rapidly growing cities in the world. The increasing number of cars in use is one of the

main concerns for air quality deterioration. As already noted, cars account for a very

large share of trips.

Carbon monoxide and Volatile Organic Compound (VOC) are major pollutants in Perth

as we can see in Figure 1.1. The nitrogen oxides also contribute to air quality

deterioration. A chemical reaction between VOC and NOx forms ozone; however ozone

is not reported by the National Pollutant Inventory (NPI) of the Department of the

Environment and Heritage, Australia. An indication of ozone formation is given by the

amounts of NOx and VOC.

CO NOx VOC PM10 SO2

0

100

200

300

400

500

600

700

800

900

mill

ion

kg PerthSydney

Figure 1.1: Pollutants emitted in Perth and Sydney airsheds in 2003-04 Source: National Pollution Inventory (NPI) website

Page 27: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 1: Introduction to the research

9

CO

Solid fuel burning

(domestic)9%

Others3%

Motor Vehicles82%

Burning/ Wildfires

6%

NOx

Biogenics20%

Shipping & Boating

7%

Others6%

Motor Vehicles

67%

VOC

Others27%

Dom/Comm solvents & aerosols

10%Solid fuel burning

(domestic)19%

Motor Vehicles44%

SO2

Motor Vehicles

29%

Others8%

Shipping & Boating

63%

PM10

Solid fuel burning

(domestic)37%

Others5%

Burning/ Wildfires

31%

Motor Vehicles

27%

Figure 1.2: Sources of pollutants in Perth in 2003-2004 Source: constructed from National Pollution Inventory (NPI) data

The mixture of sources of pollutants is illustrated in Figure 1.2 which shows the

proportions of different sources in Perth. Pollutants are generated mainly from human

activities and most are concentrated in urban areas. Figure 1.2 shows that motor

vehicles are the main source of pollutant emissions in Perth. About 82% of CO is

emitted from motor vehicles. Motor vehicles also contribute about 67% of the total

NOx emitted in Perth, the remainder being biogenic 20%, and shipping & boating 7%.

The main sources of PM10 and SO2 are not motor vehicles, though motor vehicles

contribute about 27% and 29% of these pollutants respectively. Solid domestic fuel

Page 28: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 1: Introduction to the research

10

burn (37%) and wildfires (31%) are two major sources of PM10 in Perth, whereas

shipping (63%) is the major source of SO2 pollution. From 1st January 2002 leaded

petrol was phased out, so that lead is no longer considered one of the major pollutants in

Australia. In the case of VOC, motor vehicles (44%) are the main source followed by

domestic fuel burn (19%).

It is clear that most pollutants are caused by motor vehicles. Conventional cars use

fossil fuel, which produces pollutants during the combustion process. These pollutants

come through the exhaust. Moreover cars also generate emissions through evaporation

from the engine and refuelling losses. The combustion process is discussed in a later

chapter.

The increasing number of vehicles in Perth is a major concern, as well as the increasing

number of vehicle-kilometres-travelled (VKT). Other factors, such as life of vehicles,

cold and hot start of vehicles, fuel efficiency, congestion, and meteorological factors

also influence the concentration of air pollution.

1.5 CLIMATE AND OTHER CHARACTERISTICS OF PERTH

Western Australia (WA) covers one-third of Australia, spreading over 2.5 million

square kilometres. The eastern border is mostly deserted and the west is bounded by

12,500 kilometres of coastline. Perth is located at the south-western corner of WA at

Latitude 31° 57' south and Longitude 115° 51' east and is spread along the coast.

Perth was established in 1829 as the capital city in Western Australia. It is considered

to be the most remote capital city in the world, the closest city Adelaide being 2,200

kilometres away to the east.

The average hours of sunlight each day and rain-free days each year in Perth are more

than in Brisbane, Melbourne, Sydney, Hobart or Adelaide. Perth enjoys pleasant

weather most of the year. However because of the geography of the western coast of

Australia and the effects of Indian Ocean air and water currents, Perth experiences some

more extreme weather than Sydney and Adelaide (which share approximately the same

latitude).

The global climate has changed over the last few decades, in Perth as elsewhere. Figure

1.3 shows monthly variation of temperature averaged over the last 12 years. The

Page 29: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 1: Introduction to the research

11

highest temperature is observed during January-February and the lowest in July-August.

The bars represent the high-low range of temperature within each month.

Monthly Temperature

0

5

10

15

20

25

30

35

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

deg

CAverage

Perth has its highest rainfall during winter and very little rain in summer. However the

last 12 years observation (Figure 1.4) show that rainfall varies significantly during June

and July, with large gaps between high and low levels.

Monthly Rainfall

0

50

100

150

200

250

300

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

mm

Mean

Another significant meteorological observation in Perth is wind speed. Figure 1.5

shows that afternoon wind speed is higher than morning wind speed in all seasons.

Figure 1.3: Monthly average temperature in Perth Source: constructed from data from the Bureau of Meteorology, WA

Figure 1.4: Monthly average rainfall in Perth Source: constructed from data from the Bureau of Meteorology, WA

Page 30: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 1: Introduction to the research

12

Wind Speed

9

11

13

15

17

19

21

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

km/h

9am

3pm

Wind direction is another important meteorological observation. Perth has a striking

wind direction feature compared to other cities in Australia. Figure 1.6 shows average

wind direction for the last 2 years at 9 am and 3 pm. In the morning wind flows mainly

from the north-east, which is from inland to the ocean and in the afternoon wind flows

mainly from south-west, which is from the ocean to inland. A somewhat similar feature

is observed in Adelaide, but the opposite is observed in Sydney and Brisbane (Bureau of

Meteorology, http://www.bom.gov.au/climate/averages/wind/selection_map.shtml).

Wind also flows from other directions in the morning and afternoon, but with lesser

frequency.

NE

SESW

NWNE

SESW

NW

Figure 1.5: Monthly average wind speed in Perth Source: constructed from data from the Bureau of Meteorology, WA

Figure 1.6: Average wind direction in Perth at 9am and at 3pm Source: constructed from data from the Bureau of Meteorology, WA

(9 am) (3 pm)

N

Page 31: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 1: Introduction to the research

13

Geographical location and climate have a significant impact on air quality in Perth.

During winter when wind speed and temperature are low the air pollutants become

trapped near the ground beneath a layer of warm air.

Other than geographical and meteorological characteristics, Perth has its own social

character. The Perth Metropolitan area covers 5,386 square kilometres, comprising only

0.2% of the area of Western Australia, while 72% of the population of WA (1.34

million out of 1.85 million according to the 2001 Census) live in Perth. The 2001

Census reports that 49% of the total population are in the work force and of these 64%

are full time and 36% are part time workers. The 2006-07 State Budget reported that

the unemployment rate was 4.25% in 2005-06. Other social characteristics in 2001

included median age of 34, median weekly individual income of $300-$399, median

weekly household income of $800-$999 and mean household size of 2.6.

One important feature of Perth city is that it has the highest car ownership among

Australian cities. Car ownership in Perth was at least 587 per 1000 inhabitants in 2001,

whereas the figures were 472 in Sydney, 546 in Melbourne, and 533 in Brisbane.

People in Perth generally prefer to use a car to travel to work rather than using other

modes of transport. About 90% of the work force uses car (as a driver or a passenger)

to travel to work, 4% use trains and 6% use buses. Thus Perth is a car dominated city.

1.6 AIM OF THE RESEARCH

To fill the gap in the area of human responses, the study addresses the behaviour of

travellers in order to determine suitable environmental policies. The objectives of the

study are as follows:

• To estimate the air pollution in Perth due to road transport under various

atmospheric conditions;

• To identify potential policies related to pollution and transport;

• To assess the mode choice behaviour of travellers to Perth city, particularly the

factors underlying car choice;

• To estimate the responses of travellers (both commuters and discretionary

travellers) with respect to potential environmental policies using stated

preference methods;

Page 32: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 1: Introduction to the research

14

• To formulate a composite environment and transport policy by analysing costs

and benefits of implementing the policy.

It has been recognised that effective implementation of environmental control policies is

not easy because the characteristics of environmental systems are uncertain and

externally dynamic (Papakyriazis and Papakyriazis 1998). A ‘suitable’ control policy

must be based on human behaviour. Commuter use of private cars is a central concern

of environmental policy makers. “Among the socio-economic characteristics, age,

income, and education level, seem to be influencing the choice of transit” (Abdel-Aty

2001). This study emphasises the combined use of stated preferences and revealed

preferences methods to derive travellers’ responses in relation to air quality control

policies.

Formulating appropriate solutions to the air quality problem in the Perth airshed is a

major objective. The aim is to identify suitable policies for improving air quality in the

Perth Central Business District (CBD). The research framework is shown in Figure 1.7.

The research is an integrated study conducted through several stages. At Stage 1 the car

pollution model is developed with meteorological and traffic data. Stage 2 involves

assessing likely responses to the proposed measures to limit car pollution using

previously estimated elasticities. The RP model for transport mode selection is

developed at Stage 3 on the basis of the Perth and Regional Travel Survey (PARTS) and

at Stage 4 the SP model is developed from the Car Trip Response Survey 2005. The

questions in this survey were based on the measures to control car pollution subjected to

preliminary testing at Stage 2.

The SP model results are integrated into the RP model and the combined results are

used to predict responses to the proposed measures at Stage 5. Finally at Stage 6 the

study identifies the most appropriate solution to the problem of improving air quality by

calculating health benefits and relating them to the burden that would be imposed on

motorists.

Page 33: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 1: Introduction to the research

15

1.7 ORGANISATION OF THE THESIS

Chapter 2 deals with previous work on air pollution and travel behaviour. Various

factors influencing the concentration of air pollution in any urban area are identified.

This chapter also reviews the environmental policies implemented in a number of cities

around the world. Chapter 3 describes the air pollution model building process for

Perth city. This pollution model is used later in Chapter 4 and in Chapter 8 to estimate

Figure 1.7: Research framework

Stage 1 Stage 2

Stage 3

Stage 4

Stage 5

Stage 6

Meteorological and traffic data

Car pollution model

Initial response assessment

Previous elasticity estimates

Proposed initial measures to limit car pollution

Integrate SP results into RP model

RP model based on PARTS data

SP model based on Car Trip Response Survey 2005

Full response assessment

Calculation and assessment of health benefits

Proposed final measures to limit car pollution

Page 34: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 1: Introduction to the research

16

the policy impacts. Chapter 4 identifies the potential measures to reduce air pollution in

Perth city and also presents calculations of the impacts of these potential measures in

terms of air pollution. Chapter 5 then develops a transport mode choice model with RP

data for travellers to Perth city. Chapters 6 and 7 discuss the Stated Preference survey.

Chapter 6 reports the descriptive survey outcomes and the following chapter develops a

binary logit model using discrete choice methods. Chapter 8 integrates the SP model

results into the RP model and then assesses the impact of the policies through benefit-

sacrifice analysis. In this chapter mode choice is simulated using the parameters of the

SP model for the purpose of prediction. Finally Chapter 9 summarises the research and

outlines the findings and their implications.

Page 35: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

17

CHAPTER TWO

Previous analysis and policies on air pollution and travel

behaviour

Chapter 1 introduced the research topic on air pollution and transport. The problems

were identified and the direction of research formulated. This chapter gives a global

view of air pollution and transport in Section 2.1. Section 2.2 reviews previous studies

of the health impacts. Then in Section 2.3 air pollution modelling is discussed. Air

pollution control policies implemented in various cities in the world are discussed in

Section 2.4. Relevant studies of travel behaviour are reviewed in Section 2.5.

2.1 APPROACHES TO AIR POLLUTION AND TRANSPORT

Air pollution is a universally recognised challenge faced in the past, the present and the

future. In the bronze and iron ages, settlements were polluted by dust and fumes created

from many sources. Gold and copper were fabricated and clay was kilned and glazed to

produce pottery and bricks (Boubel et al. 1994). People used charcoal as the main

source of fuel at that time; later coal was used. Then at the beginning of the industrial

revolution in the early eighteenth century steam was used to provide the power to move

machinery. During most of the nineteenth century coal was the primary fuel, even

though some oil was used for steam generation late in the century. The leading air

pollution problem of the nineteenth century was smoke and ash generated from coal and

oil burning in power plants, locomotives, marine vessels and in heating houses.

Although smoke and ash were recognised as a problem in the fifteenth century, impacts

on human health and on plants received little attention until very recently. The air

pollution problem can also be projected into the future as more and more fuel is used to

meet the demands of growing population.

Since pollution is recognised as one of the challenges faced by urban populations, a

number of studies have been conducted to address this problem. Anthropogenic

pollution, which is created by people, can be reduced by controlling human activities

and behaviour. According to Weatherley and Timmis (2001) pollutants can be assigned

to seven types: i) nuisance (e.g. noise, odour), ii) toxic, iii) acidifying/atrophying, iv)

photochemical oxidant precursors, v) radio-nuclides, vi) stratospheric ozone depleting

Page 36: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

18

substances, and vii) greenhouse gases. Of these, toxic and nuisance pollutants are

familiar in urban areas. The Weatherley and Timmis (2001) study among many others

(Newman and Kenworthy 1999, Nijkamp et al. 2002, Camagni et al. 1998,

Gudmundsson and Hojer 1996, Jones and Lucas 2000) have discussed the urban air

pollution problem and suggested measures to control it. These studies are used to

identify the potential challenges and measures to control air pollution in the Perth case.

The study by Weatherley and Timmis (2001) used an atmospheric management cycle

framework, which explains the factors influencing atmospheric degradation and the

probable means to resolve the problems. Because of the anthropogenic nature of the

pollution, the suggested solutions are mainly regulations and policies which can change

the activities and behaviour of people to reduce pollution generating elements. The

study concluded that developing an accurate quantitative measure of air pollution and

understanding the human health impact of pollutants can lead to the solution of the

problems.

In their book “Sustainability and Cities: overcoming automobile dependence”, Newman

and Kenworthy (1999) discussed sustainability and automobile dependence. After using

a number of cities as examples to show the pattern of auto-dependence, they visualised

the environmental situation with less auto-dependence, and finally sought to promote

urban changes in order to reach sustainability. This was a development of the Newman

and Kenworthy (1996) study on land use and transport. The aims of the 1996 paper

were to analyse the pattern of various cities in relation to land use and transport systems,

identifying problems linked to particular cities, and to evaluate awareness of the

connection between urban land use and transport. The paper compared cities in the

USA, Australia, Europe, and Asia in relation to transport and land use, using Perth as

one of the cases, and concluded that people can use their cars less, even in an

automobile-dependent city. In both of their studies Newman and Kenworthy concluded

that achievement of sustainability is possible from an economic point of view but

difficult from a political point of view. The book (Newman and Kenworthy 1999) is a

good source of comparative descriptions among cities in terms of demography and

transport, but it did not analyse the situation using mathematical models. Relational

models for sustainability and transport would give specific direction in solving

sustainability problems. Nevertheless, this book helps the present study in providing

background on sustainable transport systems from an air pollution point of view.

Page 37: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

19

Camagni et al. (1998) analysed sustainable city policies in terms of the economy and

the environment. They identified three ‘environments’: the natural, the artifact, and the

social, and sought to integrate them for effective implementation of policies to achieve

sustainability in cities. A categorisation of positive and negative external effects due to

the interaction of these three ‘environments’ was summarised in their paper. The paper

aimed to identify environmental policies and how to put them into effect but it

essentially reviewed and analysed existing policies without trying to develop an optimal

composite policy.

Gudmundsson & Hojer (1996) dealt with sustainable development principles and their

implications for transport. They analysed ‘sustainability’ and ‘development’ using a

multi-directional concept. Four principles of sustainable development were evaluated –

i) safeguard natural resources, ii) maintain the option value of productive capital, iii)

improve the quality of life, and iv) equitably distribute quality of life. The authors found

that all principles are applicable in relation to transport. The study identified a challenge

in the way of sustainable development by arguing that the different sectors involved in

sustainability are highly integrated with each other. The criteria for sustainable

development for one sector may link to criteria for other sectors and making the entire

process complex. The paper considered only the broader principles of sustainable

development and did not quantitatively validate the effectiveness of the principles.

In another study Jones & Lucas (2000) sought to employ a more integrated approach to

policy appraisal. They identified 28 sustainability indicators related to transport and

suggested the integration of various different government units and departments in order

to achieve successful implementation of environmental policies. However, the paper

did not address the quantification of costs and benefits in appraising the transport

policies. Addressing many objectives to achieve the overall goal may become complex

and would not be practically feasible to implement.

In general a sustainable environment can be ensured by controlling human activities that

lead to air pollution or any other pollution. Many policies or regulations are being tried

in various cities in order to reduce air pollution; however increasing population is

making the implementation of such policies very difficult. Perth is one of the rapidly

growing cities. The human activities in this city, especially motor vehicle use, are

increasing as in other big cities. An accurate estimation of motor vehicle impact on air

Page 38: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

20

pollution is required to control air pollution in Perth city and lead to measures to control

motor vehicle use.

2.2 HEALTH IMPACT OF AIR POLLUTION

Increasing recognition of economic and social costs resulting from transport

externalities directs researchers to measure the actual impact of air pollution and other

effects. Taylor and Taplin (1998) reported a BTCE (1996a and 1996b) estimate of

congestion impact in capital cities in Australia of about $2.16 billion per annum, which

does not include the costs associated with air pollution generated as a consequence of

the congestion. Many studies estimated the economic costs of air pollution by

calculating the cost in lives affected by air pollution. Studies like Amoako et al. (2003),

Scoggins at al. (2004), Kunzli et al. (2000), Finkelstein et al. (2003) are recent ones

which discussed the relationships between air pollution and mortality in various urban

areas.

Relative risk is the usual measure to estimate the impact of a harmful event, especially

in medical studies. The relative risk of death is defined as the probability of death due

to the occurrence of an event relative to the absence of that event. Many air pollution

and mortality related studies have reported the relative risk of mortality due to one unit

increase in a specific pollutant. The present study adopts this approach and quantifies

the number of deaths that could have been saved if the pollution levels were reduced by

a certain amount. To achieve this, estimates of relative risk of death due to one unit

increase in CO and NOx are required to quantify the benefit of air quality improvement.

The previous estimates of relative risk of death that are reviewed in this section have

been taken into account in the present study in the assessment of relative risks of death

for one unit increase in CO and NOx in Perth city.

The general health impacts of six major pollutants generated in urban areas were

discussed in Chapter 1. According to the Australian Bureau of Statistics (ABS) 45% of

death in Australia in 2004 was caused by diseases of the circulatory and respiratory

systems, which are to a considerable extent caused by air pollution. The figure was

41% for Western Australia but it rose to 44% for the period between 1997 and 2004.

The Melbourne Mortality Study (2000) was the first conducted in Melbourne to

estimate the association between air pollution (mainly particles, ozone, nitrogen

dioxide, and carbon monoxide) and mortality rate. The study used two methodological

Page 39: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

21

approaches in estimating the relative risk (RR) of death associated with air pollution.

These two approaches are trigonometric analysis and a generalised additive model

(GAM). It is argued that the effects of individual pollutants are difficult to separate

from combined effects because of the high correlation between pollutants; the study

addresses this challenge by controlling for other pollutants by fitting the potential

confounding pollutants to the model then fitting the pollutant of interest to the residuals

of that model. This was a rigorous approach to the collinearity problem. The study also

considered seasonal effects on mortality. The results are in terms of estimates of

relative risk of mortality associated with PM10, O3, NO2, and CO. The relative risks of

mortality for CO and NO2 as calculated in the Melbourne Mortality Study (2000) have

been adopted to measure the number of lives that could have been saved if air pollution

were reduced in Perth city. The study is comprehensive in estimating RRs of mortality

due to all causes of death, and also deaths caused by respiratory and cardiovascular

diseases for both the 65+ and below 65 age groups.

Amoako et al. (2003) is another recent study conducted to estimate the economic

consequences of the health effects of transport emissions in Australian capital cities.

The study basically reports the economic impact of health effects in Australian cities

using the Kunzli et al. (2000) estimation of relative risks for a 10μg/m3 increase in

PM10. The health impact of transport emissions was assessed in terms of number of

deaths in all capital cities in Australia. Again the impacts were measured in monetary

terms by using the human capital method, although the study recognised the alternative

willingness to pay method. It finally compares the economic costs of air pollution for

all capital cities in Australia in terms of both mortality and morbidity. Perth ranked fifth

among Australian cities for total cost of health effects of air pollution. This study is a

good source, except that ‘willingness to pay’ is the more appropriate method of

determining the economic impact of air pollution (Deng 2006, Ortuzar et al. 2000,

Johannesson 1996).

Morgan et al. (1998a and 1998b), and Petroeschevsky et al. (2001) reported daily

mortality in Sydney, hospital admissions in Sydney, and hospital admissions in

Brisbane associated with an increase in pollution concentration from the 10th to the 90th

percentile of the average observation. These studies have estimated the relative risks of

mortality or morbidity (hospital admission) with the increase in pollution, using

generalised linear models. The studies also estimated the impacts in terms of

Page 40: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

22

cardiovascular and respiratory mortalities. All these studies provide a good

representation of Australian health impacts of increasing air pollution. The estimated

relative risks are used to calculate the average relative risk of mortality for one unit

increase in CO and NOx for the present study.

A few other studies reported relative risks of mortality due to increase in pollution

levels in various cities. Income and air pollution levels are correlated with mortality

rate in southern Ontario (Finkelstein et al. 2003). A similar scenario was analysed in

another study (Finkelstein et al. 2004), where a strong negative relationship was found

between relative risk of mortality and the distance of residence from major roads and

highways. This study also established a link between risk of death and income levels.

A study in China (Chen et al. 2004) found increased risk of mortality and hospital

admission for COPD (chronic obstructive pulmonary disease) with increased air

pollution. The study estimates the relative risk of non-accidental deaths with an

increase in 10μg/m3 of NO2, SO2 and PM10. The present study does not consider the

income effect of mortality simply because of unavailability of data.

2.3 AIR POLLUTION MODELS

Although air quality is also affected by plants and animals, air is polluted mainly by

human activities in urban areas. The main sources are motor vehicles, industrial

emissions, and area-based emissions. Details of these sources in Perth are discussed in

Chapter 3. As was mentioned above, accurate measurement of quantitative

relationships between air pollution and the sources assists in formulating policies to

control air pollution.

For a large urban centre, pollution sources include individual point sources (industry

emissions) and mobile sources (motor vehicles). The combined pollution affects air

masses which spread over hundreds of kilometres. Such an air mass is described as an

urban plume. If the urban plume comes from a point source (emitted from a

smokestack) then estimation follows a Gaussian plume model, whereas if it is from

mobile sources, the estimation of the concentration follows a street canyon model. For

both models meteorological factors, especially wind, influence the concentration of

pollutants. More specifically, the concentration of a plume from a point source depends

on i) the physical and chemical nature of the pollutants, ii) meteorological factors, iii)

location of source relative to physical obstructions, and iv) topological factors that

Page 41: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

23

affect air movement. The Gaussian plume model takes all these factors into account to

estimate the pollution concentration coming from point sources. The model is based on

a set of equations describing three-dimensional concentration (Figure 2.1). The three

Gaussian equations are provided in Appendix 2A.

The model assumes that concentrations from a continuously emitting source are

proportional to the emission rate and inversely proportional to the wind speed, and that

the time-averaged pollutant concentrations, vertically and horizontally, are well

described by Gaussian or normal (bell-shaped) distributions (Boubel et al. 1994). The

Gaussian plume equations strictly rely on the ratio of horizontal and vertical standard

deviations of the distributed plume. Apart from industrial application, Gaussian plume

models can be used to estimate pollution concentration in the vicinity of a highway.

However, these models are not directly applicable for small-scale dispersion within an

urban canyon (Vardoulakis et al. 2003).

Vardoulakis et al. (2003) reviewed various air quality models applicable to street

canyons. Air pollution concentrations in an urban street depend on the characteristics of

the street canyon: the canyon geometry, wind flow, traffic volume, and emission factor.

The term street canyon refers to a relatively narrow street with buildings along both

sides. The dimensions of a canyon play a vital role on pollution concentration in the

canyon. The dimensions are width (W), height (H), and length (L) of the canyon.

Depending on the dimensions of the canyon, the wind flow may be described in terms

Figure 2.1: Two cross-sections of a Gaussian plume Source: http://www.rpi.edu/dept/chem-eng/Biotech-Environ/SYSTEMS/plume/gaussian.html

Page 42: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

24

of three regimes (Oke 1988). These are isolated roughness, wake interface, and

skimming flows. The wind speed and wind direction relative to the canyon determine

the pollution concentration in the canyon. The street pollution includes the direct

emission from vehicles and the contribution from recirculating air. Details of street

geometry and wind flows are discussed in Chapter 3.

While wind flow determines the dispersion of pollution into the street canyon, traffic

volume and the emission factor are variables determining the pollution concentration.

There are various influences associated with motor vehicles which affect the emission

factor. These are discussed in Section 2.3.4.

In the 1970s air quality modelling started with limited computer resources. With the

increase in computer power the models have improved significantly by including all

types of factors that affect air quality. Such air pollution modelling, especially street

canyon modelling, is relevant to this study in developing an air pollution model for

Perth city in which pollution concentrations follow the canyon effect. The modelling

approach identifies the factors and their impacts on plume concentration.

Vardoulakis et al. (2003) reviewed several air pollution models and tried to categorise

them according to their physical or mathematical principles and their level of

complexity. The study grouped them into statistical, receptor, screening, box, street

canyon, Gaussian, microscale, and urban scale types. The commonly used models

include STREET, OSPM, CAR, CALINE4, and ADMS-Urban. A brief discussion of

the three most relevant models is provided in the following section.

2.3.1 The STREET Model

The STREET model is an empirical model which calculates a series of hourly

concentration at various receptor locations within a street canyon (Vardoulakis et al.

2002). This model, developed by Johnson et al. (1973), assumes that the wind flow at

roof level is perpendicular to the street axis. The pollution concentration (C) on the

roadside consists two components, the background concentration (Cb) and direct

emission from motor vehicles (Cd). The direct emission component was derived from a

simple box model, where pollutants are assumed to be uniformly mixed from ground to

the depth of the boundary layer; source release rates and winds are assumed constant

over the model domain (Kallos 1998). In this model the emission concentration is a

function of emission rate, distance between the source and receptor, canyon geometry,

Page 43: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

25

and wind speed, along with some empirical constants. Model details for direct emission

concentration on leeward and windward sides are provided in Appendix-2B.

According to the model, direct emission concentration on the leeward side is higher than

on the windward side. The model shows that if emission rate and wind speed vary,

other factors being fixed, the direct emission concentration on the leeward side is

approximately 2.4 times higher than on the windward side. It also shows that pollution

concentration is directly related to the emission rate and inversely related to wind speed.

In case of a wind parallel or near parallel to the street axis, an average of leeward and

windward concentrations should be calculated to estimate direct pollution concentration.

This model does not take the angle of wind to the street axis into account. It is simple to

estimate and has been used in recent studies like Vardoulakis et al. (2002) and Mensink

et al. (2006).

2.3.2 The OSPM Model

The Operational Street Pollution Model (OSPM) was developed by Berkowicz (1998).

This model estimates the pollution concentration using a combination of the Gaussian

plume model for direct contribution and the box model for recirculating part of the

pollutants in the street (see Figure 3.5 in Chapter 3). The total pollution concentration

(C) consists of direct contribution (Cd), recirculation of air (Cr), and background air

(Cb).

This model is similar to the STREET model. The pollution concentration is estimated

from the number of motor vehicles, average speed of vehicles, dimensions of vehicles,

aerodynamic drag coefficient, dimensions of recirculation zone, angle of wind to the

street axis, vertical fluctuation due to mechanical turbulences along with the factors

mentioned in the STREET model. It is more comprehensive than the STREET model.

In this model it is assumed that a vortex is formed inside the canyon if the wind is not

parallel to the street axis, and the length of the vortex is twice the up-wind building

height. The direct contribution follows the Gaussian model as a function of vertical

fluctuation of emission, emission rate, canyon geometry, and wind speed. On the other

hand the recirculation part of emission is determined by an equation which is practically

identical to the STREET model for the windward side. A significant improvement of

the OSPM over the STREET model is that it takes into consideration the angle of the

Page 44: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

26

wind direction to the street axis. Details of the OSPM model are provided in

Appendix-2C.

The pollution concentration estimated on the leeward side is the sum of the direct and

recirculation contributions, but on the windward side only the direct contribution of

emissions generated outside the recirculation zone are taken into account. The OSPM

model has been used in many studies, including Berkowicz (2000), Vardoulakis et al.

(2002), and Mensink et al. (2006), to measure air quality in the urban street canyon. All

of these studies estimated values which were very close to the observed values. The

concept of the OSPM model can be related to the model developed for Perth city. The

present study model parameters can easily be interpreted in terms of the concepts of the

OSPM model as discussed in Chapter 3.

2.3.3 The CALINE4 Model

The CALINE4 model is the latest version of the CALINE series of pollution dispersion

models. Although CALINE4 is able to handle canyon effects, it has been used in

relatively few urban street air quality studies. The model mainly uses the Gaussian

plume theory to simulate the dispersion of pollutants emitted from a line source

(vicinity of a highway). The region directly above the road is called the mixing zone,

which is considered as a zone of uniform emission and turbulence. This model has been

widely used in scientific and engineering applications (Vardoulakis et al. 2003). The

details of this model are not discussed further because of it is less applicable in the street

canyon situation.

While the pollution models emphasise meteorological factors and street geometry,

vehicle emission rate is one of the important inputs to the models. However the vehicle

emission rate depends on a range of factors associated with vehicles. These factors are

discussed in the following section.

2.3.4 Factors influencing vehicle emission rate

Types of vehicles: Different categories of vehicle emit different amounts of pollutants.

Vehicles are categorised differently in various countries depending on size, weight, and

shape. According to Australian Bureau of Statistics (ABS) vehicles are classified as

passenger vehicles, motor cycles, light commercial vehicles, rigid trucks, articulated

trucks, non-freight carrying trucks, and buses. Usually smaller cars produce less

Page 45: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

27

emission. In most CBDs, passenger cars generally predominate over other types of

vehicle. Therefore many emission models convert all vehicles to equivalent passenger

car units (PCU).

Types of fuel used: Different types of fuel produce different amounts of emissions. Un-

leaded petrol, diesel, liquefied petroleum gas (LPG), and compressed natural gas (CNG)

are the major types of fuel used to run cars. The different fuel types produce pollutants

in differing quantities. Formerly leaded petrol emitted lead into the air, but from 1st

January 2002 leaded petrol has been eliminated from Australia. Diesel produces more

SO2 than petrol does, whereas petrol produces more CO. On the other hand LPG and

CNG are considered cleaner fuels.

Types of road used by the cars: Road type may influence pollution generation. Four

categories of road are used in urban areas; these are freeway, highway, arterial road, and

local road. The smoothness of road surface and the average speed limit make a

difference to the production of emissions. Rough surface and lower speed generally

produce more pollutants.

Age of the car: Older car generally create more pollution than new cars. New model

cars are usually fitted with improved pollution control equipment. The catalytic

converter was introduced in 1986 to convert harmful pollutants like HC, CO, and NOx

into harmless carbon dioxide (CO2), water (H2O), and nitrogen (N2). At the same time

new cars have better fuel economy which ensures less emission. Very recently

alternative technologies are gaining popularity. One of them is the hybrid car, which

uses gasoline and an electric battery. In urban running, a hybrid car may use as little as

half the petrol used by a comparable conventional car (Australian Government 2003

quoted in Taplin 2004) and the hybrid is a very clean car in terms of pollution creation.

Therefore in emission model building older cars are considered to be greater polluters

than new cars.

Travel pattern: Other than the above mentioned physical factors, behavioural factors

also influence the level of emission. The factors include driving in a congested period,

speed of car, duration of car running, use of air conditioner, and cold-hot start. These

factors result in different rates of pollution production. For example, in a congested

period a car produces more pollution because of low speed and consuming more fuel; a

cold start produces more pollution than a hot start.

Page 46: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

28

Vehicle emission models are used by various agencies. The US Environmental

Protection Agency (US EPA) has used a new version of MOBILE6 to estimate emission

rate; the UK Department for Transport uses a version of the EURO model. In Australia,

departments and agencies use various models similar to the MOBILE model. The

Australian Greenhouse Office has developed a methodology for estimating greenhouse

gas. The EPA Victoria uses the Australian Motor Vehicle Emission Factors System

(AusVeh 1.0). Other recognised emission models used in Australia include aaSIDRA

(Akçelik & Besley 2003) emission module and CSIRO power-based (Leung and

Williams 2000) emission model. The present study develops an air pollution model

with traffic as one of the explanatory variables. The coefficient of traffic is converted to

an equivalent emission factor which can be compared with emission factors estimated

by various emission models. The MOBILE6 and AusVeh1.0 emission models are

briefly presented in Appendix-2D.

The air pollution model estimated in this study takes an entirely different approach but a

comparative analysis with models such as MOBILE and AusVeh is discussed in

Chapter 3. The air pollution model quantifies the impact of vehicles on air pollution

concentration in Perth city.

2.4 AIR POLLUTION CONTROL POLICY

A number of policies and regulations are implemented in various cities around the

world to control and manage the use of cars. Different cities use different approaches to

achieve the goal of improving air quality. The following discussion shows that the USA

is more focused on using improved technology vehicles, whereas the UK approach is

more on reducing vehicle kilometres travelled (VKT). In general the air pollution can

be reduced in any city centre, such as the City of Perth, by implementing the following

strategies:

• Reduce private car use, particularly during periods of congestion.

• Discourage private car use in the city.

• Encourage use of public transport and non-motorised modes.

• Encourage people to shift to improved technology vehicles.

The US Environmental Protection Agency (EPA) is an organisation to protect human

health and the environment and, since 1970, it has been working for a cleaner and

healthier environment for the citizen of the USA. The UK is a leader in Europe in

Page 47: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

29

implementing policies and regulations to improve the atmospheric environment, with its

policy on transport emissions being to minimise vehicle kilometre travelled (VKT). In

Australia the Department of Transport and Regional Services (DOTARS) is responsible

for managing development of policy and standards on vehicle emissions. It has specific

responsibility for the Australian Design Rules (ADR) related to vehicle emissions. The

first emissions related ADRs were introduced in the early 1970s and have been

gradually made more rigid since then. Moreover, the Australian Greenhouse Office

(AGO 2005), a part of the Department of the Environment and Heritage, develops

programs under the Australian Government’s climate change strategy. One of the aims

of these programs is to ensure transport sustainability in Australia.

Table 2.1: Comparison of key environmental and transport policy approaches

Measures US, EPAa UK, DfTb Australia

Technology Promote cleaner vehicle technology by improving fuel efficiency, alternative fuel vehicles.

Encouraging use of fuel-efficient vehicles, alternative fuel vehicles.

Encouraging use of natural gas (NG) and LPG as alternative fuels.

Pricing Parking cash-out program.

High Occupancy Toll Lanes.

Cordon pricing policy in central London.

Car manufacturing Industry

Introduce the low emission vehicle (LEV) program.

Setting a voluntary target for car manufacturers

Vehicle maintenance

Introduce inspection and maintenance (I/M) program

Fuel consumption labelling.

Green Vehicle Guide program.

Awareness Pollution awareness program

Encouraging public transport.

Walk or cycle to school program.

Emphasising cycling activities.

In-town-without-my-car program

Travel demand management

TravelSmart

Encourage cycling

a Environmental Protection Agency b Department for Transport

Page 48: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

30

Table 2.1 gives a comparison of environmental and transport policy approaches. The

policies are categorised into six groups on the basis of the focus of the policy. The

measures deal with technology, pricing, car manufacturing, vehicle maintenance, and

awareness. The comparison of policies adopted by the US EPA, the UK DfT, and

Australia (Table 2.1) mainly shows similarities, yet some policies are different. Details

of the individual policies are discussed in the following sections.

2.4.1 Technology Measures

The US Environmental Protection Agency promotes the use of clean technology rather

than reducing the use of vehicles, engines, or equipment. People need to use vehicles

for mobility and this is seen as an inevitable part of human activity. But the EPA

encourages people to use more alternative fuel vehicles, hybrid vehicles, and vehicles

with good fuel economy. Similarly the UK DfT encourages people to use fuel-efficient

vehicles, hybrid vehicles, and fuel-cell vehicles, which produce less emission than

traditional cars. A program has also been introduced in Australia to increase the use of

alternative fuels, especially natural gas (NG) and liquefied petroleum gas (LPG), in

medium to heavy road vehicles. The technological aspect of environmental policy is

well recognised by all countries.

2.4.2 Pricing Measures

A number of cities have imposed pricing (or taxation) policies to manage travel

demand. Common methods are to impose a charge on car users to enter a city centre or

for travelling on a tolled road. The charge is usually designed to cover the social costs

of vehicle operation. The pricing policy may also impose added charges for workplace

parking. The policy basically discourages people from taking cars to the city. London

has implemented cordon pricing as have some European cities and Singapore.

However, cordon pricing is not applied in American cities, or in Australia, though tolled

roads are common in American cities and in Melbourne and Sydney in Australia.

Another form of pricing measure, used in California is the parking cash-out program,

under which employees are encouraged to earn extra cash by sacrificing their parking

space, so that they leave their car at home. In the 1990s the USA established a Value

Pricing Pilot Program to fund innovative road pricing measures for congestion relief.

One type of measure implemented is High Occupancy Toll (HOT) lanes. The HOT

lanes are an alternative to High Occupancy Vehicle (HOV) lanes that allow vehicles

Page 49: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

31

that do not meet occupancy requirements to use the lanes for a toll. This type of

measure is an effective policy to control travel demand, however it is difficult to

implement from a political point of view.

2.4.3 Car Manufacturing Industry Measures

The US EPA has implemented the National Low Emission Vehicle (LEV) program.

Under this program 23 major car manufacturers were required to follow certain

standards on emissions in producing new cars by the model year 1999. The program

also indicated that the proportion of LEV sales should be at least 25% in 2001, 50% in

2002, 85% in 2003, and 100% in 2004 and in later years. This reflects the EPA

emphasis on low emission technology rather than reducing car use. This type of

measure is not adopted explicitly in the UK, nor in Australia. In Australia the National

Average CO2 Emissions (NACE) target is set by the Australian Greenhouse Office

(AGO). According to the NACE target the government arranged with the automotive

industry a voluntary target for fuel efficiency of 6.8L/100km for petrol passenger cars

by 2010. This represents an 18% improvement in the fuel efficiency of new vehicles

between 2002 and 2010.

2.4.4 Vehicle Maintenance Measures

The EPA issued its first guidance for the Inspection/Maintenance (I/M) program in

1978. This guidance addressed the State Implementation Plan (SIP) elements such as

minimum emission reduction requirements, administrative requirements, and

implementation schedules. Under this program vehicle tail pipe exhaust should be

checked in an authorised test stations on a regular basis, and should satisfy the standard

for exhaust emissions. The program also includes an evaporative system test,

centralised annual test, and visual inspection. This original I/M guidance was quite

broad and difficult to implement. Then in 1990 amendments to the Clean Air Act

(CAAA) set more I/M guidance including minimum performance standards for basic

and enhanced I/M programs. It also addressed a range of program implementation

issues, such as network design, test procedures, oversight and enforcement

requirements, waivers, funding, etc. In contrast the UK does not have this form of

regulation to control air pollution.

In Australia mandatory Fuel Consumption Labelling on all new cars sold promotes

consumer demand for fuel efficient vehicles by making comparative model specific

Page 50: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

32

information available to buyers. In addition, mandatory CO2 labelling will raise

consumer awareness on reducing greenhouse gas emissions.

2.4.5 Awareness Measures

In Australia the Department of Transport and Regional Services (DOTARS) has

developed measures to raise awareness of the impact of vehicles on the environment by

launching the Green Vehicle Guide website and the requirement for a Fuel

Consumption Label on all new light vehicles. The Green Vehicle Guide provides

ratings on the environmental performance of new vehicles sold in Australia. This

Internet site helps to compare vehicles performance in terms of greenhouse gas and air

pollution emissions. A nationwide environmental awareness program is implemented in

Australia. The Australian Greenhouse Office (AGO) is joining with States, Territories,

local governments and communities to support various approaches to reduce greenhouse

gas emissions from passenger transport in urban areas. This initiative adds to

TravelSmart activities, which are already in operation across Australia. TravelSmart is

an approach which tries to make people aware of the environmental impact of motor

vehicles and of the opportunities available to them personally to use public transport.

They provide information about air pollution associated with the cars and its impact on

health, and encourage people to use alternatives to the private car, especially when

travelling to work or on business. The program encourages people to walk or cycle or

use public transport.

Public awareness has probably been given more attention in the UK than in any other

country. The program encourages people to use public transport or a non-motorised

mode. One of the program activities is the National Cycling Strategy (NCS). The aim

is to establish a culture favourable to the increased use of bicycles for all age groups; to

develop sound policies and good practice; and to seek out effective and innovative

means of fostering accessibility by bike. Furthermore, the cycling activities program

ensures workplace cycle parking facilities priority for cyclists. Apart from this people

are encouraged to follow the “In Town Without My Car” program. This suggests that

people not take a car to the city on a particular day of the year. Its intention is to stop

taking the car to the city every day of the year. In another program students are

encouraged to walk or cycle to school.

Page 51: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

33

Most cities in the USA follow a similar environmental awareness program. People are

encouraged to use public transport, cycle, or use another non-motorised mode of

transport, though the main approach to controlling air pollution is to achieve improved

technology.

2.5 AIR POLLUTION CONTROL IN PERTH, WESTERN AUSTRALIA

In Western Australia the Department of Environment introduced the Perth Air Quality

Management Plan (AQMP) in December 2000 to ensure clean air throughout the Perth

metropolitan area. The AQMP is designed to manage air quality in Perth until the year

2029. The key areas addressed by the plan are:

• Health effect research.

• Monitoring, modelling, and research.

• Land use and transport planning.

• Vehicle emissions management.

• Domestic activities emissions.

• Burning emissions management.

• Industry emission management.

• Community information and education.

In the Perth AQMP context, this study indicated measures which would reduce air

pollution in Perth city airshed. Details of these measures are discussed in Chapter 4.

They are price and control measures which can be categorised into four types – fixed

charge, variable charges, parking measure, and lane restriction measure. Since none of

the suggested measures has been implemented in Perth before, a first assessment

(Chapter 4) is based on demand elasticities from other parts of the world. Later chapters

deal with a stated choice survey and the application of the resulting estimates.

A fixed charge would be imposed on a car each time it enters the city centre. This

measure is similar to cordon pricing which has already been implemented in cities such

as Singapore, London, Bergen, Oslo, Trondheim, Stockholm, and tried in Hong Kong.

The fixed charge measure has been assessed on the basis of studies by Luk (1999),

Lettice (2004), Matas & Raymond (2003), and Arentze et al. (2004) which estimated

Page 52: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

34

the travel demand elasticities with respect to road tolls. Details of these studies are

discussed in Chapter 4; however some are briefly reviewed here.

Arentze et al. (2004) examined individual behaviour under a congestion pricing

scenario. They conducted an Internet based stated preference experiment to examine

individual adjustments in activity-travel patterns. The study analysed individual choice

data using the multinomial logit function and developed models for work activities and

non-work activities. The study also estimated the price elasticities of travel demand

under a congestion pricing policy, essentially a cordon pricing policy. The estimated

elasticities vary within a range of -0.13 to -0.19 for the entire network, and -0.35 to

-0.39 for congested roads and times. These estimates are used in the present study to

calculate aggregate responses to a fixed charge policy.

Variable charges are imposed on car users according to size of car and time of entering

the city. Records of responses to this type of charge were not available but the effects

are assumed to be similar to variable costs; fuel price elasticities are used in this study.

Some of the previous studies (Taplin et al. 1999, Mayeres 2000, Luk & Hepburn 1993,

Hensher & Young 1990, Button 1993) have been reviewed and their estimated

elasticities are used to calculate average demand elasticity with respect to variable costs.

Among these studies Mayeres (2000) used an applied general equilibrium model to

analyse transport pricing policy, such as fuel tax, to solve transport related problems.

The study formulated a utility model which contains four components: i) the direct

utility from the consumption of the commodities and time use, ii) the utility derived

from the public goods provided by the government, iii) the disutility from emissions,

and iv) the disutility from accidents. The study estimated the marginal external costs of

transport use in terms of air pollution and accidents. These marginal external costs are

used to determine the welfare impact of transport policies suggested in the study.

Finally the study estimated the impact of the policy in terms of percentage change in

traffic flow in peak and off-peak situations. These estimates have been adopted to

calculate average elasticity of travel demand with respect to variable charges.

Many studies (Hensher & King 2001, Hess 2001, Willson & Shoup 1990, Calthrop

2002) have estimated travel demand elasticity with respect to parking charges. Hensher

& King (2001) conducted a study on parking demand and pricing in the Sydney CBD.

The study used a Stated Preference (SP) experiment to explain the behaviour of the

travellers to Sydney CBD. They defined three different parking zones and compared

Page 53: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

35

them in terms of travellers’ responses. The results were presented in terms of

elasticities of parking demand with respect to parking fee per hour. Since the study area

was in Australia, it seems appropriate to use its results in assessing travel behaviour in

Perth. Therefore the estimated travel demand elasticity is used to estimate aggregate

response to a proposed Perth parking measure.

Another potential policy considered in this study is lane restriction. A number of

studies (Noland 2001, Noland & Cowart 2000, Fulton et al. 2000, Hensen & Huang

1997) have developed relationships between increasing road capacity and induced

vehicle travel. All of these studies have estimated the elasticities of travel demand with

respect to increased road supply. The present study deals with the inverse of this; it is

tentatively assumed that the elasticities are reversible and can be applied to the

suggested lane restriction measure which would reduce road capacity. Noland (2001)

found that increased road capacity influences traveller’s behaviour in terms of mode

shifts, route shifts, redistribution of trips, and generations of new trips. The study

ultimately estimated elasticity of travel demand in vehicle miles travelled (VMT) with

respect to increased road capacity. A logarithmic transform of VMT was related to a

logarithmic transformation of lane miles (a proxy for cost of travel time) and to

demographic variables. The model is basically a simple form of regression model. The

study estimated the elasticity of vehicle miles travelled with respect to lane capacity as

0.3 to 0.6 in the short run and 0.7 to 1.0 in the long run. These values are used initially

in calculating the average elasticity of travel demand with respect to lane closure.

After estimating the average impact of suggested policies on travel demand, a study

conducted by Taylor and Taplin (1998) is used as a basis to set the actual values for

fixed and variable charges. The study was conducted in the Australian context;

therefore its estimations seem to be appropriate as the base of the charges. The present

study uses Taylor & Taplin (1998) costs figures indexed up to 2004 with the consumer

price index (CPI).

All of these previous studies provide reasonable bases for aggregate measures of the

impact of air quality control policies for Perth city and are used in the first assessment

reported in Chapter 4. However, to know the actual travel behaviour for the travellers

to the Perth city, we need to analyse actual data for Perth city. Consequently in the

present study a stated preference (SP) discrete choice survey was conducted to assess

the actual reaction to the suggested policies (Chapters 6 and 7).

Page 54: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

36

2.6 DISCRETE CHOICE ANALYSIS

Discrete choice analysis is an integral part of this study. It is the key to assessing driver

responses to pollution reducing measures. Travel choice has many different aspects

including but not limited to transport mode choice, responses to price changes and

policies, and responses to travel time changes, parking fees and availability of parking

spaces. Travellers’ choices may be observed in their past activities or approached

through their intended activities in the future. The information collected from past

activities is called revealed preference (RP) information. The revealed preference data

is used to analyse travel demand and travel behaviour in the transport research field

(Caldas & Black 1997). On the other hand, when travellers have not encountered a

particular situation their responses to it cannot be observed. However their likely

responses can be assessed through stated preference (SP) information. This involves

setting up some hypothetical situation and finding how they would react in that event.

The approach is appropriate to investigate a new or a hypothetical event (Louviere et al.

2000, Hensher et al. 2005, Ben-Akiva et al. 1991). Limitations to both approaches are

discussed in Chapter 6 in Section 6.2.2. Morikawa (1994) and others (Louviere et al.

2000, Ben-Akiva and Morikawa 1990, Cherchi and Ortuzar 2002) suggested that SP and

RP data should be combined to exploit their advantages and overcome their limitations.

Travellers’ choice or preference data are analysed using discrete choice analysis. The

Multinomial Logit (MNL) function based on utility concepts is used to estimate the

probability of choosing a certain option. The MNL function has been modified and

extended to Nested Logit (NL), Generalised Extreme Value (GEV), Mixed Logit,

Non-normalised Nested Logit (NNNL), and the Heteroscedastic Extreme Value (HEV)

model. As well as the multinomial logit (MNL) and nested logit (NL) models used in

this study to analyse the behaviour of travellers’ to Perth city, a latent class model has

also been used. The nested logit and multinomial logit models are discussed in Chapter

5 where a transport mode choice model is developed for Perth city travellers on the

basis of RP survey data. Chapters 6 and 7 explain the binary logit, panel data, and latent

class models. The following paragraphs give an initial introduction.

Among the case studies discussed by Louviere et al. (2000) one is on the valuation of

travel time savings and urban route choice using SP information for tolled and free

routes. The attributes of the choices were toll and travel time, each attribute having

three levels which gave an orthogonal design of nine choice sets. Five different models

Page 55: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

37

were developed with travel time, toll, and their interaction as explanatory variables

along with personal income as a demographic variable. Using these models the study

estimated the value of travel time savings (VTTS) for five groups based on responses to

a $1.00 toll. The study reported the VTTSs for five groups as $4.35 per hour for private

commute, $7.07 for business commute, $4.59 for travel as a part of work, $5.68 for

non-work related travel, and $8.33 for other personal business travel. The present study

has also developed a mode choice model using discrete choice analysis. That model is

used to estimate VTTS for the travellers to Perth city. The study is unable to estimate

VTTS for different groups due to small sample size. However, the estimated VTTS for

all travellers is very close to Louviere et al. (2000) estimations. Details are discussed in

Chapter 5.

Morikawa (1994), an early proponent of the combined SP-RP approach, suggested a

method of correcting correlation effects in SP-RP estimation which was a major

problem in some studies. The theory related to this correction is discussed in Section

6.2.2 in Chapter 6.

A latent class modelling approach to choice problems was introduced by Swait (1994).

The latent class model is essentially another form of nested logit. It is useful for

segmenting respondents by using their explicit choices and unobserved preferences.

The results from the latent class model provide improved classification of respondents

which can be used to target for marketing or other purposes. The present study

developed a latent class model to segment responses to air pollution control policies

(Chapter 7).

Page 56: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

38

Appendix 2A

The Gaussian Plume Model

For stable conditions or unlimited vertical mixing the plume concentration is (Boubel et

al. 1994):

]})2/[(]}{)2/[(){/1( 5.02

5.01 zy gguQ σπσπχ = .................................... (2A.1)

Where,

)/5.0exp( 221 yyg σ−=

]/)(5.0exp[]/)(5.0exp[ 22222 zz zHzHg σσ +−+−−=

χ is plume concentration Q is emission rate

u is wind speed

σy is standard deviation of horizontal distribution of plume concentration

σz is standard deviation of vertical distribution of plume concentration

Figure 2A1: The Gaussian Plume Model Source: Boubel et al. 1994

Page 57: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

39

L is mixing height

h is physical stack height

H is effective height of emission

x is downwind distance

y is crosswind distance

z is receptor height above ground

For unstable or neutral condition, where σz is greater than 1.6L, the plume concentration is:

)/1]}()2/[(){/1( 5.01 LguQ yσπχ = ..................................................... (2A.2)

For unstable or neutral condition, where σz is less than 1.6L, the plume concentration is:

]})2/[(]}{)2/[(){/1( 5.03

5.01 zy gguQ σπσπχ = .................................... (2A.3)

Where,

∑∞

−∞=

++−++−−=N

zz NLzHNLzHg ]}/)2(5.0exp[]/)2(5.0{exp[ 22223 σσ

Page 58: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

40

Appendix 2B

The STREET Model

The following model is taken from Dabberdt et al. (1973).

Pollution concentration (C) on the roadside is shown in equation (2B.1).

db CCC += ............................................................ (2B.1)

Where, Cb is background concentration, and Cd is direct emission from motor vehicles.

The flow of wind in a street canyon is shown in Figure 2B.1.

For the leeward side of the street the direct emission concentration is expressed in

equation (2B.2).

)]()[( 05.022

s

Ld UUhzx

QKC+++

= ................................... (2B.2)

Figure 2B.1: Schematic of cross-street circulation between buildings. Source: Dabberdt et al. 1973

Page 59: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

41

Where, K is an empirical constant (K=7) Q is emission rate on the street (gm/s) x is horizontal distance between the receptor and centre of the nearest traffic lane z is height of the receptor h0 is a constant that accounts for the height of initial pollutant dispersion (h0 = 2m) U is roof level wind speed Us is a constant for additional air movement (empirical value is 0.5 m/s)

For the windward side this concentration can be expressed in equation (2B.3).

H

zHUUW

QKCs

Wd

−+

=)(

.................................................. (2B.3)

Where H is the height of the canyon W is the width of the canyon

Page 60: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

42

Appendix 2C

The OSPM Model

The following models are taken from Berkowicz (1998).

The total pollution concentration (C) is expressed in equation (2C.1) by adding direct

contribution (Cd), recirculation of air (Cr), and background air (Cb).

brd CCCC ++= ............................................................... (2C.1)

According to Gaussian plume model the direct plume contribution can be estimated by

using equation (2C.2).

FW

QCw

d σπ2

= .................................................................. (2C.2)

Where Q is emission rate on the street (gm/s) W is the width of street canyon F is a factor depending on synoptic wind

σw is the vertical velocity fluctuation due to mechanical turbulence generally by wind and traffic on the street

This vertical velocity fluctuation can be expressed as:

22)( wow u σασ +=

Where α is a constant (empirical value is 0.1) u is the street level wind speed σwo is traffic created turbulence

Again traffic created turbulence can be defined as:

⎟⎟⎠

⎞⎜⎜⎝

⎛=

WNVSbwo

2

σ

Where b is aerodynamic drag coefficient (empirical value is 0.3) N is the number of vehicles on the street per time unit V is the average vehicle speed S2 is the road surface occupied by a single vehicle W is the width of the canyon

Page 61: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

43

The contribution from the recirculation zone is estimated using a simple box model,

assuming that the pollutants are well mixed inside the box. The calculation of this

component can be expressed in equation (2C.3).

21 SSttwt

rr uLLUL

LWQC

++=

σ .............................................. (2C.3)

Where Lr, Lt, LS1, and LS2 are the dimensions of the recirculation zone, assuming that the canyon vortex has a trapeze shape.

σwt is the ventilation velocity of the canyon, which can be expressed as:

22)( woroofwt FU σλσ +=

Where U roof level wind speed λ and Froof are proportionality constants given the values of 0.1 and 0.4

respectively

The length of the recirculation zone (Lr) can be defined as:

)sin,min( ψvortexr LWL =

Where Lvortex is the length the vortex ψ is the angle between roof level wind and street axis

Page 62: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

44

Appendix 2D

The MOBILE6 Emission Model

The MOBILE6 model was developed by the US Environmental Protection Agency

(EPA). This comprehensive model incorporates many factors which influence emission

rate. It mainly focuses on vehicle performance, fuel performance, and travel pattern.

The algorithm used to estimate running emission rates for different classes of vehicle is

summarised in equations (2D.1) and (2D.2) (EPA 2001).

[Fleet-Ave Emission Rate]veh class = ∑ [Travel Fraction] × {[Linear Emission Rate + Tampering Offset + Aggressive Driving + Air Conditioning] × [Temperature Adjustment] × [Speed Adjustment] × [Fuel Adjustment]} ................... (2D.1)

[Fleet –Ave Emission Rate] = ∑ [VMT Mix]veh class

× [Fleet-Ave Emission Rate]veh class ....... (2D.2)

Where

Travel fraction based on VMT distributions, registration distribution (25 years), and mileage accumulation.

Linear emission rate is historical linear rate of emission.

Tampering offset based on production features of the vehicle.

Aggressive driving considers acceleration rate of the car.

Temperature adjustment is the variation of temperature at testing cycle. Low temperatures range from -7° C to 24° C and high temperatures range from 28° C to 35° C.

Speed adjustment considers different road speed conditions.

Fuel adjustment considers fuel performance, e.g. oxidised, non-oxidised, sulphur level etc.

Age=1

25

veh=1

n

Page 63: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

45

The AusVeh 1.0 emission model

The Australian Motor Vehicle Emission Factors System (AusVeh 1.0) has been

developed by EPA Victoria to estimate motor vehicle emissions for states/territories or

nationally in Australia. This model provides the methodology for estimating running

emissions and evaporative emissions for different pollutants and for different types of

vehicles. It is similar to the MOBILE6 model. The methodology (EPA Victoria 2003)

used to estimate running emissions is shown in equation (2D.3).

( ){ }∑ ×××+×=y

ypscfvyfvgfvgpfvo

gpcfvppscfv scrvcvdrefcfef ,,,,,,,,,,,,,,,,,,,, ,min 150 ........ (2D.3)

where ef is average emission factor (g/km),

cf is conversion factor,

efo is base emission factor or zero-kilometre emission (g/km),

dr is deterioration rate (g/1000km²),

cv is average cumulative vehicle kilometre travelled (VKT) (in 103km),

min(cv,150) is minimum of cv and 150,

rv is relative VKT,

sc is speed correction factor,

v is index for motor vehicle type,

f is index for fuel type,

c is index for process,

s is index for speed,

p is index for pollutant,

y is index for year of manufacture, and

g is index for year group.

The conversion factor (cf) is used to convert the emission factor of one pollutant to that

of another. Equation 2D.3 assumes that there is no further deterioration after 150,000

km of travel due to engine replacement (Carnovale et al. 1996). The deterioration rate

(dr) is only applicable for exhaust emission and if it is not entered for a vehicle type,

fuel type and pollutant then it is assumed to be zero. In this case, the base emission

factor is an average emission factor for the vehicle type, fuel type, pollutant, and year

group concerned. If a speed correction factor is not entered for a vehicle type, fuel type,

and pollutant, it is assumed to be 1.

Page 64: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 2: Previous work on air pollution and travel behaviour

46

The average cumulative VKT and relative VKT are calculated according to equations

(2D.4) and (2D.5).

∑∑==

×=2

1

2

1

y

yyyfv

y

yyyfvyfvgfv nvcvnvcv ,,,,,,,, ....................................... (2D.4)

∑=y

yfvyfvyfv tvtvrv ,,,,,, ................................................................ (2D.5)

Where

nv is number of vehicles

y1 and y2 are the start and end years of the year group between 1965 and 2021

tv is total VKT (106 km/yr)

aiy −=

i is year in that the vehicle commence in use a is vehicle age

Page 65: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

47

CHAPTER THREE Causal relationships between traffic and air pollution in a

Perth city canyon

Transportation gives people access to goods, services, and activities. On the other hand

it contaminates the environment by producing emissions. Chapter 2 reviewed and

identified the factors that may affect the level of pollution concentration in any urban

area. This chapter first discusses air pollution formation in Perth city and the factors

influencing the concentration and dispersion of pollution. Section 3.3 discusses the

structure of data used in developing an air pollution model. Air quality data,

meteorological data, and traffic data are analysed. Section 3.4 presents the process of

development of pollution models for CO and NOx levels in Perth city. Both ARIMA

and Causal relationship models are developed. Section 3.5 presents a comparative

analysis of the results of the causal model and previously developed models.

3.1 INTRODUCTION

The air around us can be polluted from sources such as on-road vehicles, non-road

engines and equipment, and aircraft. As noted in Chapter 1 surface transport is the

major source of air pollution in most urban areas. The atmosphere is polluted by

various gases and particles, mainly carbon monoxide (CO), nitrogen oxides (NOx),

ozone (O3), sulphur dioxide (SO2), lead, and inhalable particles (PM10 and PM2.5).

Among these, CO and NOx come directly from motor vehicles. These are by-products

of the combustion process in any petrol or diesel vehicle. The chemical reaction of the

combustion process can be generalised in the following expression.

Typical Engine Combustion:

FUEL (C9H20 or C14H30) + AIR (N2 & O2) ⇒ UNBURNED HYDROCARBONS (HC) + NITROGEN OXIDE (NO2) + NITRIC OXIDE (NO) + CARBON MONOXIDE (CO) + CARBON DIOXIDE (CO2) + water (H2O)

All of the by-products other than water can adversely affect the human body directly

and indirectly. A small portion of carbon monoxide reduces the flow of oxygen in the

Page 66: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

48

bloodstream and is particularly dangerous to a person with heart disease and NOx also

exacerbates respiratory symptoms and cardiovascular diseases. Hydrocarbons, carbon

dioxide and nitrogen oxides are the major contributors to ozone and “greenhouse gas”

formation.

This section of the study explores air pollution in Perth city by developing mathematical

models for CO and NOx concentrations. Before investigating the variability of CO and

NOx concentrations and the causes of variability, we need to consider air pollution

development in Perth city.

3.2 AIR POLLUTION DEVELOPMENT IN PERTH CITY

The air pollution development process in an urban area is not simple and is complicated

by the nature of air flow within city street ‘canyons’. The canyon refers here to a street

channel bounded by high rise buildings on both sides. The factors influencing pollution

concentration in such a canyon are discussed in the following sections.

3.2.1 Factors in the formation of pollutants

The pollution level at any particular location depends on the dispersion process. This

process is complex in a street canyon. The dispersion process is a function of

meteorology, street geometry, receptor location, traffic volume, and emission factor.

All of these have significant impacts on air quality in Perth city.

3.2.1.1 Meteorology

Wind is probably the most obvious factor influencing dispersion of air pollution from

one place to another. Light winds allow emissions to remain close to the source and

accumulate high concentrations. The summer season in Perth is characterised by

prevailing offshore winds in the morning with long period of fine warm weather (Perth

Photochemical Smog Study 1996). A temperature difference between inland and the

sea creates an onshore pressure difference, which leads to a regular sea breeze. This sea

breeze is usually south-westerly. The morning wind flow takes the city’s concentrated

pollution offshore, while the afternoon sea breeze brings it back and adds to the mid-day

emissions.

Page 67: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

49

In the winter, at night, there are high concentrations of emissions from wood fires and

motor vehicles; these accumulate near ground level. Few variations can be observed in

these processes.

Generally the sea breeze forms at some distance offshore. When the offshore flow is

strong and/or the sea breeze forms close to the shoreline the city’s morning emissions

may be removed from the region. This is illustrated in Figure 3.1.

In contrast when the morning easterly wind is light and/or the sea breeze forms at a

distance offshore, the morning emissions may be trapped in the sea breeze and returned

to the city. The effect is shown in Figure 3.2.

Furthermore, the Perth Photochemical Smog Study (PPSS 1996) reported on other

significant events. The “inland event” and “Kwinana event” are the two most

Morning

Afternoon

Sea breeze

Figure 3.1: Morning emissions lost out at sea

Morning

Afternoon

Figure 3.2: Morning emissions trapped in the city

Page 68: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

50

significant in relation to pollution concentration in Perth city. The “inland event” has

been identified as morning wind direction between northeast and east and wind speed

inland of 11 km/h or less. Usually (but not always) a low pressure occurs just offshore

and the sea breeze arrives slightly earlier than usual. Because of the northerly

component of morning winds, the sea breeze has more westerly direction than usual,

which ensures that the morning emissions return to the city but are not carried to the

northern suburbs. Figure 3.3 shows the directions for city emission flow and Kwinana

emission flow.

The “Kwinana event” was also observed during the PPSS (1996) study. It is generally

similar to the “inland event” except that there are more southerly morning winds. City

emissions return to the northern suburbs and Kwinana industrial emissions return to the

city region. In this event the afternoon city emission level is augmented by the Kwinana

emission. The afternoon NOx level may be higher than the morning NOx level in the

city, because Kwinana emissions have high levels of NOx. This event is illustrated in

Figure 3.4.

Figure 3.3: Inland event showing emission flows Source: Developed from the study PPSS (1996)

Figure 3.4: Kwinana event showing emission flows Source: Developed from the study PPSS (1996)

Page 69: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

51

3.2.1.2 Street geometry

A higher level of pollution is observed in a street canyon (Vardoulakis et al. 2003). The

unique feature of street canyon wind flow is the formation of a wind vortex that makes

the direction of wind at street level opposite to the direction of flow above the roof level

(Berkowicz 2000). This is illustrated in Figure 3.5.

Estimation of air pollution from a motor vehicle is not just simple measurement of the

emission factor. Wind flows and street geometry influence the concentration and

dispersion of pollution in street canyons. Total pollution concentration is the

summation of street contribution and background contribution, and again street

contribution includes the direct plume and recirculating air. The pollutants emitted

from traffic in the street primarily move toward the upwind building (leeward side)

while the downwind side (windward) is exposed to background pollution and the

pollution that has recirculated in the street (Berkowicz 2000). Therefore, the pollution

concentration also depends on the dimensions of the street canyon, height (H), width

(W) and the ratio of height to width (H/W), called the aspect ratio. The canyon is

called regular if the aspect ratio is approximately 1; if the aspect ratio is 2 or more the

canyon is called a deep canyon. Other than the height and width, length (L), usually

expressing the road distance between two major intersections, can also influence

pollution dispersion. The street canyon can also be characterised in terms of ratio

Figure 3.5: Schematic diagram of flow and dispersion condition in street canyon Source: Berkowicz 2000

Page 70: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

52

between length and height (L/H). The canyon can be cubic (L/H = 1), short (L/H ≈ 3),

medium (L/H ≈ 5), or long (L/H ≈ 7). The dimensions of a street canyon are shown in

Figure 3.6.

Hunter et al. (1991) investigated relationships between street geometry and anticipated

types of wind flow. The wind flow regimes in an urban street canyon, classified by Oke

(1988) as isolated roughness, wake interface, and skimming flow, are shown in Figure

3.7.

For a wide canyon (H/W<0.3), the buildings are well apart and act essentially as

isolated roughness elements. In this situation the air flows a considerable distance

Figure 3.6: Dimensions of a street canyon Source: Hunter et al. 1992

Figure 3.7: The flow regimes associated with air flow over building and aspect ratio Source: Oke 1988

Page 71: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

53

downward from the first building before facing the next one. When the buildings are

more closely spaced (H/W ≈ 0.5) the air flow has inadequate distance to readjust the

flow before facing the next obstacle; this flow is called wake interface. In the case of a

regular canyon (H/W ≈ 1), the major flow skims over the buildings producing skimming

flow. Hunter et al. (1991) reported that for a long (L/H = 7) and somewhat deep (H/W

≥ 2) canyon the anticipated flow would be skimming flow. These approximate to the

relative dimensions of William street canyon in Perth city, where the pollution

monitoring station is located.

3.2.1.3 Perth city pollution monitoring station

The Department of Environmental Protection (DEP) in Western Australia currently

monitors air quality at 10 monitoring stations in Perth. These 10 stations are spread

across the entire metropolitan area. The locations are shown in Figure 3.8.

The station called “Perth” is located at the Queen’s Building in William Street at the

centre of Perth Central Business District (CBD). The exact view of the Queens

Building monitoring station is shown in Figure 3.9. The station has a sampling inlet at

4.4 metres above the kerb (‘4’ in Figure 3.9). The main section of William Street is

about 400 metres (measured from Online Mapping System, City of Perth) between the

intersections at Wellington Street and St. Georges Terrace. The height of the buildings

on both sides of the street averages about 25 metres (although the buildings are

Figure 3.8: Air quality monitoring stations in Perth Metropolitan area Source: The Department of Environmental Protection, WA website

Page 72: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

54

asymmetric) and the width of the street is about 15 metres (measured from Online

Mapping System, City of Perth). Therefore the approximate dimensions of the William

street canyon are L=400m, H=25m and W=15m. Hunter et al. (1992) reported that for a

deep canyon, like this, air flow would be skimming flow, giving a very low movement

of air at the bottom of the canyon, with a vortex created in the upper part of the canyon.

3.2.1.4 Traffic volume and emission factor

The final factors which affect the pollution concentration and dispersion are traffic

volume and emission rates. In addition to the number of motor vehicles operating in the

city, vehicle speed, density of moving cars, and aerodynamic drag have impacts on

pollution. Details of traffic volume and emissions in Perth city are discussed in section

3.3.3.

All of these factors have been used to develop air pollution models for the urban

canyon. There is a range of mathematical models which are modified to suit the street

geometry. The models include STREET, CAR, OSPM, CFD, UK-ADMS, ADMS-

Figure 3.9: Queens Building monitoring station: [1] monitoring room, [2] nephelometer inlet, [3] PM10 sampler, [4] inlet sample tube

Source: Monitoring Plan for WA 2001

Page 73: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

55

Urban, and CALINE4 among many others (Vardoulakis et al. 2003). Some of these

were discussed in Chapter 2.

The present study depends entirely on secondary data as there was no opportunity to

experiment in collecting actual wind flow information at the source of emissions.

Therefore the study develops a pollution model for Perth city using available data from

various sources. It establishes relationships between air pollution, especially CO and

NOx, and motor vehicles travelling in Perth city as well as wind speed and direction.

The main objective of this part of the study is to develop and calibrate an appropriate

model for Perth air quality. Effectively, this part of the study is an extension of the

study reported by Siddique (2004). Air quality data were collected from the Department

of Environmental Protection, WA, meteorological data from the Bureau of

Meteorology, and traffic data from Main Roads, WA.

3.3 DATA STRUCTURE

3.3.1 Air quality data

Although Perth has 10 air quality monitoring stations, not all stations record every

pollutant. The only pollution data used in this study was from the Perth monitoring

station in Queen’s Building which monitors CO and NOx every ten minutes.

The National Environmental Protection Measure (NEPM) air quality standard sets 9

ppm (parts per million air volume) for an 8-hour average as the standard for CO

nationwide. Figure 3.10 shows hourly CO level in Perth city for the period from

October 2003 to June 2005. It was found that the NEPM limit was exceeded once when

Perth CBD level of CO reached 12.8 ppm at 2 AM on 17th February 2005.

Page 74: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

56

0

1

23

4

5

67

8

9

10

1112

13

14

Oct

Nov

Dec Ja

n

Feb

Mar

Apr

May Ju

n

Jul

Aug

Sep Oct

Nov

Dec Ja

n

Feb

Mar

Apr

May Ju

n

hourly (Oct 2003 to June 2005)

ppm

Although average hourly CO level in Perth is currently within the NEPM limit most of

the time, increasing car use will cause the level to rise. According to the Perth

Metropolitan Transport Strategy 1995-2029 (State Government of WA 1995), average

personal car travel to work or business was 8.4 kilometres in 1991 and this increased to

12.6 km in 2004 (according to the Perth and Regional Transport Survey data).

The other pollutant considered in this study, nitrogen dioxide (NO2), is a brown gas,

most being transformed from nitric oxide (NO) contained in emissions. These two

nitrogen oxides are recorded in the Perth monitoring station. In an urban area, motor

vehicle emissions along with industrial boilers and furnaces are the major source of

NO2. There are two primary standards for ambient NO2 in Australia. One is a 1-hour

average of 0.12 ppm (12.0 pphm) (parts per hundred million air volume) and another is

a 1-year average of 0.03 ppm (3.0 pphm). Figure 3.11 shows the hourly average NO2

level in Perth from October 2003 to June 2005; during that period, the level exceeded

the standard once to 0.16 ppm at 1 PM on 8th March 2004. The yearly averages for

three years were 0.02, 0.01, 0.02 ppm in 2003, 2004, and 2005 respectively. Data are

not available for the month of January 2005.

Figure 3.10: Hourly CO level in Perth City from October 2003 to June 2005

NEPM standard

Page 75: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

57

0

2

4

6

8

10

12

14

16

18

Oct

Nov

Dec Ja

nFe

bM

arA

prM

ay

Jun

Jul

Aug

Sep Oct

Nov

Dec Ja

nFe

b

Mar

Apr

May

Jun

hourly (Oct 2003 to June 2005)

pphm

Nitric oxide (NO) gas is also generated from combustion. It has an open shell

configuration, which makes it very reactive and unstable. In air this gas reacts quickly

with oxygen to form nitrogen dioxide (NO2). During the vehicle combustion process a

greater volume of NO is produced than NO2 but it disappears very soon. The study

considers NOx as a summation of NO2 and NO for the purpose of model development.

Transportation contributed 47% of NOx in USA in 2001 (US Department of

Transportations 2003), whereas it contributed 68% (in year 2003-2004) in Perth (NPI

website). The level of NOx in Perth city for the period between October 2003 and June

2005 is shown in Figure 3.12.

0

10

20

30

40

50

60

Oct

Nov

Dec Ja

n

Feb

Mar

Apr

May

Jun

Jul

Aug

Sep

Oct

Nov

Dec Ja

n

Feb

Mar

Apr

May

Jun

hourly (Oct 2003 to June 2005)

pphm

Figure 3.11: Hourly NO2 level in Perth City from October 2003 to June 2005

Figure 3.12: Hourly NOx level in Perth City from October 2003 to June 2005

NEPM standard

Page 76: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

58

The maximum level of NOx was 53.4 pphm at 11 PM on 17th June 2005. There is no

national standard set for NOx in Australia.

Figures 3.10, 3.11, and 3.12 show overall variation of hourly CO and NOx levels in

Perth city for the period between October 2003 and June 2005. Hourly variation of CO

and NOx in an average day is shown in Figures 3.13 (a) and (b).

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

100

300

500

700

900

1100

1300

1500

1700

1900

2100

2300

hour

ppm

0

2

4

6

8

10

12

14

16

100

300

500

700

900

1100

1300

1500

1700

1900

2100

2300

hourpp

hm

Maximum levels of CO and NOx are reached in the afternoon at around 6 PM in Perth.

Although there is a peak in the morning at around 9 AM, it is not to the level of the

afternoon peak. This disparity in peaks can be due to a range of reasons, which are

discussed later.

3.3.2 Meteorological data

Although motor vehicles are the major cause of air quality deterioration, temperature,

wind speed, wind direction, rainfall, humidity, and hours of sunshine can modify the

concentration of any pollutant. To build a relationship between air pollution and

meteorological factors, the data should be collected from the same geographical

location. However, there was no opportunity to set up an experiment which could

ensure data collection from the same location. Therefore the study had to rely on the

data collected from the closest weather station.

Air temperature is an important factor in the chemical process leading to the formation

of ozone from nitrogen oxides in the presence of sunlight. The levels of CO and NOx in

the outdoor air are typically higher during the colder months of the year when inversion

(a) CO (b) NOx

Figure 3.13: Hourly variation in an average day in Perth city (a) for CO, and (b) for NOx Source: Constructed from the data provided by Department of Environmental Protection, WA

Page 77: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

59

conditions are more frequent. Air pollutants become trapped near the ground beneath a

layer of warm air. Though CO and NOx levels are low during summer (December-

February) the high temperatures may lead to high ozone concentrations. Figure 3.14

shows monthly variation of pollution levels with air temperature and wind speed.

0.0

5.0

10.0

15.0

20.0

25.0

30.0

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

NO

x (p

phm

), Te

mp

(C),

Spee

d (k

m/h

)

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

CO

(ppm

)

Wind speed

Temperature

NOx

CO

Wind speed and direction influence the distribution of CO and NOx in Perth. The city is

located close to the sea; therefore a strong wind from the sea blows away most of the

pollutants. It was mentioned before that the morning wind blows the morning

emissions of the city offshore, and the afternoon sea breeze brings it back and adds to

the midday emissions. Thus direction and speed of the wind are factors determining

pollution concentration and dispersion. In this study, all wind directions were

categorised as North-East, South-East, South-West, or North-West. Figure 3.15 shows

a Perth map with wind directions.

Figure 3.14: Monthly average Temperature, wind speed and pollution level in Perth city over 2003 to 2005

Source: Constructed from the data provided by Department of Environmental Protection, WA

Page 78: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

60

3.3.3 Traffic data

Main Roads WA records traffic count data from various roads throughout the Perth

metropolitan area. Many of the traffic counts are intermittent but some are done on a

continuous basis. Main Roads records traffic counts every 15 minutes at most of the

traffic lights in the city area. The study combines the traffic counts (passenger car units)

of 10 intersections close to the Perth air quality monitoring station and uses the data on

an hourly basis. Traffic data was collected for the period between October 2003 and

March 2004. According to Main Roads WA this huge data set is stored on a six-month

basis. They could only provide the data for this period.

0

5000

10000

15000

20000

25000

30000

100

200

300

400

500

600

700

800

900

1000

1100

1200

1300

1400

1500

1600

1700

1800

1900

2000

2100

2200

2300

2400

hour

num

ber

weekdayweekend

Figure 3.15: Wind directions in Perth

Figure 3.16: Average hourly traffic in the city for the period between October 2003 and March 2004

Source: constructed from the data provided by Main Roads WA

Page 79: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

61

Hourly average traffic volumes in the city differ between weekdays and weekends as

shown in Figure 3.16. On weekdays there are two peaks, one at around 9 AM and

another at about 6 PM. People come to the city in the morning for work and leave at

around 6 PM during weekdays. Between the peaks people may come for other

purposes, such as shopping and personal business. In contrast, the weekend days have

only one peak in the middle of the day, and this peak is far less than the weekday peaks,

indicating that fewer people go to the city during weekends.

3.4 AIR POLLUTION MODEL

This study builds models of air pollution for CO and NOx levels based on data for the

period between October 2003 and March 2004 in which traffic data were available. The

fluctuations of CO and NOx levels for a two-week period in late October ‘03 and

another in late December ’03 are shown in Figure 3.17. Both CO and NOx vary during

a day and it is evident that weekend (last two peaks in panel (a) and last 4 peaks for

Christmas holidays in panel (b)) levels are lower than weekday levels.

00.20.40.60.8

11.21.41.61.8

2

100

1000

1900 40

013

0022

00 700

1600 10

010

0019

00 400

1300

2200 70

016

00 100

1000

1900

hour (20-26 Oct, 2003)

CO (ppm)

0

5

10

15

20

25

30NOx (pphm)

CONOx

(a)

0

0.5

1

1.5

2

2.5

3

100

1000

1900 40

013

0022

00 700

1600 10

010

0019

00 400

1300

2200 70

016

00 100

1000

1900

hour (22-28 Dec, 2003)

CO (ppm)

0

5

10

15

20

25

30

35NOx (pphm)

CONOx

(b)

The variation of the CO and NOx levels can be explained through mathematical models.

This study attempts to explain the variations of CO and NOx levels in Perth city within

an atmospheric pollution modelling framework using multivariate statistics. Two types

of models are developed and compared: ARIMA (autoregressive null relationships) and

multiple regressions (causal relationships) models.

Figure 3.17: Pollutant levels for a) a week in late October ’03, and b) a week in late December ‘03

Page 80: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

62

3.4.1 ARIMA Model

Autoregressive Integrated Moving Average (ARIMA) models are used to establish

relationships between predicted variables and the same variables in previous periods.

This modelling approach is especially suitable when little or no information on causal

relationships is available. They are used in this case to estimate the extent to which

pollution can be explained in terms of lagged values, before traffic is taken into account.

The present study considers both hourly CO and NOx levels related to the levels for the

previous 16 hours. Two separate daily models were developed for CO and NOx levels.

After preparing the datasets, an estimation of ARIMA model parameters (p=order of the

autoregressive part, d=degree first differencing involved, and q=order of the moving

average part) is required. To estimate these parameters we need to determine an

autocorrelation function (ACF) and partial autocorrelation function (PACF). Mainly

the shapes of ACF and PACF are used to determine the parameters. The study

examines these functions for 16 lagged periods. Figure 3.18 shows ACF and PACF for

CO with lag periods.

ACF - CO

Lag Number

16151413121110987654321

ACF

1.0

.8

.6

.4

.2

.0

-.2

-.4

-.6

PACF - CO

Lag Number

16151413121110987654321

Parti

al A

CF

1.0

.8

.6

.4

.2

.0

-.2

-.4

-.6

-.8

-1.0

ACF and PACF cannot be used to determine the parameters at this stage as the shapes

do not convey any decisive information. The ACF moves in a sine-wave manner,

therefore it is not possible to determine the values of the parameters. The next step is

differencing data at the first order level. The shapes of ACF and PACF after first order

differencing are shown in Figure 3.19.

Figure 3.18: ACF and PACF for CO levels for 16 hour lag periods

Page 81: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

63

ACF - 1st differencing CO

Lag Number

16151413121110987654321

ACF

1.0

.8

.6

.4

.2

.0

-.2

-.4

-.6

-.8

-1.0

PACF - 1st differencing of CO

Lag Number

16151413121110987654321

Parti

al A

CF

1.0

.8

.6

.4

.2

.0

-.2

-.4

-.6

-.8

-1.0

At this stage it is possible to determine the parameters of the model. Autoregressive

order (p) will be 1 as the correlation of the first lag period is significantly higher than

other correlations, moving average order (q) will also be 1 because of a similar shape of

PACF and the differencing parameter is 1 as we use first difference data. However,

Figure 3.19 shows a diurnal (‘seasonal’) effect as the 12th period lag has a spike on both

ACF and PACF. Therefore another model with ‘seasonal’ difference was developed as

well. Two models of ARIMA (0,0,0)(1,1,1) for ‘seasonal’ difference and ARIMA

(1,1,1) with first order difference are compared and the results are shown in Table 3.1.

Table 3.1: Comparative results of ARIMA models for CO level

Model Log likelihood

AICa SBCb R2 MSEc

ARIMA (000) (111), ‘seasonal’ -2088.15 4182.31 4201.39 0.42 0.19

ARIMA (111), ‘non-seasonal’ -898.01 1802.03 1821.13 0.64 0.16a AIC is Akaike’s Information Criterion. AIC=-2 log L + 2m, where m=(p+q+P+Q) where p and q are

as above; P=seasonal part of autoregressive order and Q=seasonal part of moving average order (Makridakis et al. 1998)

b SBC is Schwarz Bayesian Criterion. SBC=-2Log L + k Log(n), where k is number of parameters and n is number of observations (Brockwell and Davis 1996) .

c MSE is Mean Square Error, the average of squared differences between actual and predicted values. Lower absolute values of Log likelihood, AIC, SBC, and MSE and higher R2 indicate a

better model. Therefore the ARIMA (111) model (‘non-seasonal’ difference) is a better

predictor of CO level in Perth than the ARIMA (000)(111) model with ‘seasonal’

difference. The detailed ARIMA (111) model is provided in Appendix-3A. Another

way of assessing the model’s fit is to examine residual plots. The ‘non-seasonal’ model

Figure 3.19: ACF and PACF after first differencing of CO levels for 16 hour lag periods

Page 82: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

64

(panel b) data series looks more stationary than ‘seasonal’ model (panel a) in Figure

3.20.

-6.000

-4.000

-2.000

0.000

2.000

4.000

6.000

resi

dual

for

CO

from

AR

IMA

(000

)(11

1)l

(a)

-6.000

-4.000

-2.000

0.000

2.000

4.000

6.000

resi

dual

for C

O fr

om A

RIM

A (1

11)

(b)

A similar investigation was conducted for NOx model building. ACF and PACF for the

NOx dataset after first differencing are shown in Figure 3.21.

ACF - 1st differencing of NOx

Lag Number

16151413121110987654321

ACF

1.0

.8

.6

.4

.2

.0

-.2

-.4

-.6

-.8

-1.0

PACF - 1st differencing of NOx

Lag Number

16151413121110987654321

Parti

al A

CF

1.0

.8

.6

.4

.2

.0

-.2

-.4

-.6

-.8

-1.0

The shapes of ACF and PACF indicate that the values of both autoregressive and

moving average parameters are one, although it may be argued that the series needs

‘seasonal’ differences as the 12th period lag has the highest spikes. Therefore, the study

again compared two models with ‘seasonal’ difference and ‘non-seasonal’ difference.

Table 3.2 shows the comparative results for the models and Figure 3.22 shows residual

plots of these models. Most of the numerical values for the NOx model are higher than

the CO model because the data set used is in pphm compared to ppm for CO.

Figure 3.21: ACF and PACF after first differencing of NOx levels for 16 hour lag periods

Figure 3.20: Residual plots for a) ‘seasonal’ model and b) ‘non-seasonal’ model

Page 83: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

65

Table 3.2: Comparative results of ARIMA models for NOx level

Model Log likelihood

AIC SBC R2 MSE

ARIMA (000)(111), ‘seasonal’ -12361.49 24728.99 24748.06 0.48 23.9

ARIMA (111), ‘non-seasonal’ -10665.67 21337.35 21356.44 0.68 17.2

-30.000

-20.000

-10.000

0.000

10.000

20.000

30.000

resi

dual

for N

Ox

from

AR

IMA(

000)

(111

)

(a)

-30.000

-20.000

-10.000

0.000

10.000

20.000

30.000

resi

dual

for N

Ox

from

AR

IMA

(111

)

(b)

Table 3.2 and Figure 3.22 indicate that the ‘non-seasonal’ model is a better model to

estimate the level of NOx. The detailed ARIMA (111) model for NOx is provided in

Appendix-3B.

In summary, the ARIMA models established the extent to which CO and NOx pollution

can be explained as functions of their own previous values. Although it is possible to

achieve some prediction of CO and NOx levels in Perth city using ARIMA, these are not

causal models. However the estimates indicate the strength of the lagged relationships

which must be taken into account in the causal models which incorporate traffic.

3.4.2 Causal Model

The development a functional relationship between hourly levels of CO and NOx and

traffic volume in Perth city is based on the fact that the major contributors to air

pollution in the urban area are motor vehicles. Figure 3.23 shows the fluctuation of

traffic and CO level in (a) and NOx level in (b). The morning peaks for CO and NOx do

not match the level of the traffic peak, though afternoon peaks follow the traffic level.

The probable reasons are:

Figure 3.22: Residual plots for a) ‘seasonal’ and b) ‘non-seasonal’ models of NOx

Page 84: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

66

i) Accumulation of air pollution, emission in the morning hours being

accumulated throughout the day and dropping after traffic falls back in the

afternoon,

ii) The cold start effect, which means that in the afternoon, despite warm

ambient temperature, the engines of the parked cars are started before

leaving the city, thus generating more emissions at that time (Joumard and

Serie 1999).

iii) Finally, there could be some effect of wind flows; as discussed before, the

afternoon sea breeze brings back morning emissions which were taken

offshore by the morning wind.

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

100

300

500

700

900

1100

1300

1500

1700

1900

2100

2300

hour

ppm

0

5000

10000

15000

20000

25000number

CO Traffic

0.0

2.0

4.0

6.0

8.0

10.0

12.0

14.0

16.0

18.0

100

300

500

700

900

1100

1300

1500

1700

1900

2100

2300

hour

pphm

0

5000

10000

15000

20000

25000

number

NOx Traffic

This study identified numbers of variables which could explain the variation of hourly

CO and NOx level in Perth. It identified hourly traffic, wind speed, previous period’s

wind speed, cross product of wind direction (as dummy variables) and previous

period’s wind speed, and cross product of previous period’s wind speed and previous

period’s pollution level as explanatory variables. A series of regression models was

developed with different combinations of explanatory variables. Wind speed and

direction for the same period and previous period were the main causal variables for

accumulation of pollution. Wind speed and direction are measured at roof top level, not

at street level. Traffic is the main causal variable for the generation of pollution. As was

discussed in Chapter 2, in a deep and medium length canyon, like William Street where

Figure 3.23: Traffic and a) average hourly CO level, b) average hourly NOx level in Perth city during Oct 2003 to Mar 2004

Source: Constructed from the data provided by Department of Environmental Protection, WA and Main Roads WA

(a) CO (b) NOx

Page 85: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

67

the receptor is located, a skimming wind flow could be anticipated with a very low air

movement at the bottom of the canyon. Therefore, wind speed experienced at the

downwind building would be relatively higher than roof top wind speed (Hunter et al.

1991).

Tables 3.3 and 3.4 show the matrices of correlations between pollutants and

independent variables. Traffic and south-west wind are highly correlated with pollution

levels.

Table 3.3: Correlations between CO & explanatory variables (bold shows higher correlation)

Traffic Wind Speed

North-East wind

direction (dummy1)

South-East wind

direction (dummy2)

South-West wind direction (dummy3) CO

Traffic 1

Wind Speed 0.438 1 North-east wind direction (dummy1) -0.139 -0.277 1 South-east wind direction (dummy2) -0.221 -0.171 -0.358 1 South-west wind direction (dummy3) 0.319 0.412 -0.376 -0.620 1

CO 0.664 0.220 -0.168 -0.440 0.507 1

Table 3.4: Correlations between NOx & explanatory variables (bold shows higher correlation)

Traffic Wind Speed

North-East wind

direction (dummy1)

South-East wind

direction (dummy2)

South-West wind

direction (dummy3) NOx

Traffic 1

Wind Speed 0.438 1 North-east wind direction (dummy1) -0.139 -0.277 1 South-east wind direction (dummy2) -0.221 -0.171 -0.358 1 South-west wind direction (dummy3) 0.319 0.412 -0.376 -0.620 1

NOx 0.724 0.276 -0.179 -0.418 0.520 1

The correlation matrices show individual variables, not cross-products. The correlation

matrices for variables in cross-product form are very similar (not shown here). The

regression model uses cross-products of wind speed and the dummy variables for wind

directions. The direction of wind provides better information when taken in conjunction

Page 86: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

68

with wind speed in relation to the estimation of pollution level. In the ARIMA model

section it was recognised that the previous period’s pollution level influences the

present period’s pollution level. The causal model also uses the cross-product of

previous period wind speed and previous period pollution level. This cross-product

would explain the combined effect of static pollution level and the diffusion process.

Most of the explanatory variables used in the regression modelling are cross-products

for the previous period. One of the four wind directions (North-West) was omitted. A

North-West wind is not very common in Perth city.

Different models were developed using the same sets of variables and goodness of fit

was compared. The initial model (CM1CO) was developed with a linear relationship

between hourly CO level as the predicted variable and the explanatory variables. It was

found that the data set needs to be made stationary around the mean and variance.

Hence the model was further modified in two stages. The second model (CM2CO) was

developed with first difference of CO as the predicted variable, keeping the explanatory

variables unchanged. Later the study examines non-linearity in the relationship. A

further model (CM3CO) was developed with logarithmic transformation of CO level

and keeping explanatory variables unchanged except that traffic was logarithmically

transformed. Details of CM1CO and CM3CO models are provided in the Appendix-3C

and CM2CO model is discussed below. The R2 and MSE are calculated using actual

and predicted values for the three models so that the models are directly comparable. A

comparative analysis of model fit is shown in Table 3.5.

Table 3.5: Comparative results of regression models for CO calculated with actual and predicted values

Model R2 MSE DPa NPb

Linear model (CM1CO) 0.65 0.11 4215 7

Linear model with first difference of CO (CM2CO) 0.68 0.11 4208 7

Non-linear model (CM3CO) 0.63 0.13 3924 7 a DP = Data point b NP = Number of parameters

The CM2CO model gives a high R2 and low MSE. Furthermore, as noted by Harvey

(1980), a first difference model yields a better estimator of β than a model using

aggregates. This model is also justified by observing the residual plots which shows

Page 87: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

69

that CM2CO has a more stationary data series than other models (Figure 3.24). The

residual plot for CM2CO has higher proportion of cases lies within ±0.5 ppm.

-6.0

-4.0

-2.0

0.0

2.0

4.0

6.0re

sidu

al fo

r CO

from

CM

1CO

-6.0

-4.0

-2.0

0.0

2.0

4.0

6.0

resi

dual

for C

O fr

om C

M2C

O

-6.0

-4.0

-2.0

0.0

2.0

4.0

6.0

resi

dual

for C

O fr

om C

M3C

O

Figure 3.24: Residual plots for models (a) CM1CO, (b) CM2CO, and (c) CM3CO

(a) Model: CM1CO

(b) Model: CM2CO

(c) Model: CM3CO

Proportion of cases within ±0.5 ppm = 95%

Proportion of cases within ±0.5 ppm = 95%

Proportion of cases within ±0.5 ppm = 94%

Page 88: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

70

The CM2CO model results are shown in Table 3.6.

Table 3.6: Coefficients of the explanatory variables for first difference of hourly CO

level

Un-standardised

Coefficients Standardised Coefficients t-ratio Sig.

B Std. Error Beta (Constant) -0.1061 0.0146129 -7.33 0.000Wind speed -0.0040 0.0015764 -0.109 -5.05 0.000Previous period’s wind speed 0.0210 0.0025470 0.406 12.02 0.000Previous period’s north-east wind speed -0.0163 0.0022169 -0.200 -8.44 0.000Previous period’s south-east wind speed -0.0159 0.0020899 -0.317 -10.61 0.000Previous period’s south-west wind speed 0.0059 0.0020615 0.059 1.56 0.119Cross product of previous period’s wind speed and previous period’s CO level -0.0326 0.0008909 -0.979 -41.32 0.000

Traffic (in ‘000) 0.0263 0.0000008 0.508 29.55 0.000Dependent Variable: First difference of CO level in parts per million F-statistic = 1999.37, df=4201

The un-standardised coefficient for traffic is very small in value which is due to the

large actual value of traffic as compared to the very small value of CO. The Beta

coefficient gives a relative measure of the traffic effect.

A similar process was followed for NOx (in parts per hundred million) model

development. It was found that the first difference of NOx level can be explained by the

same explanatory variables as those used in CO modelling. As in the CO modelling,

two other models were developed with actual NOx level (CM1NOx) and Logarithmic

transformation of NOx (CM3NOx), however CM2NOx is the best of these models.

Model fit results are shown in Table 3.7 and residual plots are in Figure 3.25 for the

three models.

Table 3.7: Comparative results of regression models for NOx calculated with actual and predicted values

Model R2 MSE DPa NPb

Linear model (CM1NOx) 0.72 11.4 4210 7

Linear model with first difference of NOx (CM2NOx) 0.75 11.5 4210 7

Non-linear model (CM3NOx) 0.71 11.7 4159 7 a DP = Data point b NP = Number of parameters

Page 89: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

71

The CM2NOx model is best among these three with a high R2 and reasonably low mean

square error.

-30.0

-20.0

-10.0

0.0

10.0

20.0

30.0re

sidu

al fo

r NO

x fro

m C

M1N

Ox

-30.0

-20.0

-10.0

0.0

10.0

20.0

30.0

resi

dual

for N

Ox

from

CM

2NO

x

-30.0

-20.0

-10.0

0.0

10.0

20.0

30.0

resi

dual

for N

Ox

from

CM

3NO

x

Figure 3.25: Residual plots for models (a) CM1NOx, (b) CM2NOx, and (c) CM3NOx

(a) Model: CM1NOx

(b) Model: CM2NOx

(c) Model: CM3NOx

Proportion of cases within ±5 pphm = 94.5%

Proportion of cases within ±5 pphm = 94.14%

Proportion of cases within ±5 pphm = 94.55%

Page 90: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

72

Detailed model results are shown in Table 3.8 (other models’ details are shown in

Appendix 3D).

Table 3.8: Coefficients of the explanatory variables for first difference of hourly NOx level

Unstandardised

Coefficients Standardised Coefficients t-ratio Sig.

B Std. Error Beta (Constant) -1.1006 0.152269 -7.23 0.000

Wind speed -0.0236 0.015612 -0.034 -1.51 0.130

Previous period’s wind speed 0.1365 0.023726 0.199 5.75 0.000

Previous period’s north-east wind speed -0.1152 0.020788 -0.135 -5.54 0.000

Previous period’s south-east wind speed -0.1280 0.019065 -0.206 -6.72 0.000

Previous period’s south-west wind speed 0.0481 0.018504 0.104 2.60 0.009

Cross product of previous period’s wind speed and previous period’s NOx level -0.0285 0.000830 -0.853 -34.30 0.000

Traffic (in ‘000) 0.2411 0.000009 0.490 26.78 0.000

Dependent Variable: First difference of NOx level in parts per hundred million F-statistic = 3216.41 df=4203

As for CO, in this model traffic and cross product of previous period’s wind speed and

previous period’s NOx level have major influences on NOx level prediction (high Beta

values). The signs of the coefficients for both CM2CO and CM2NOx are the same,

which shows model consistency.

To achieve comparability, the R2 and MSE values for all models have been calculated

from actual and predicted values. The higher values of R2 and lower values of MSE in

regression models for both CO and NOx than those in the ARIMA models indicate the

improvement that has been achieved, which demonstrate the superiority of the

regression model over the ARIMA model. In the causal models the traffic is a major

explanatory variable so that these models can be used to assess pollution levels for a

specific traffic volume. To illustrate the model fit Figure 3.26 (a) and (b) show the

comparison between observed and model CO levels for two weeks (same weeks shown

in Figure 3.17) and Figure 3.27 (a) and (b) show the comparison between observed and

model NOx levels.

Page 91: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

73

-0.5

0

0.5

1

1.5

2

100

1100

2100 70

017

00 300

1300

2300 90

019

00 500

1500 10

011

0021

00 700

1700

hour (20-26 Oct, 2003)

ppm

observed

CM2CO

-0.5

0

0.5

1

1.5

2

2.5

3

100

1000

1900 40

013

0022

00 700

1600 10

010

0019

00 400

1300

2200 70

016

00 100

1000

1900

hour (22-28 Dec, 2003)

observed

CM2CO

-5

0

5

10

15

20

25

30

100

1100

2100 70

017

00 300

1300

2300 90

019

00 500

1500 10

011

0021

00 700

1700

hour (20-26 Oct, 2003)

pphm

observed

CM2NOx

Figure 3.26: Comparison between actual and model CO levels for (a) a week in late October ‘03, (b) a week in late December ‘03

(a)

(b)

(a)

Page 92: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

74

-5

0

5

10

15

20

25

30

35

100

1000

1900 40

013

0022

00 700

1600 10

010

0019

00 400

1300

2200 70

016

00 100

1000

1900

hour (22-28 Dec, 2003)

pphm

observedCM2NOx

In order to fully interpret the model estimates, it is necessary to consider the actual

situation where the pollution was measured. This is illustrated in Figure 3.28.

(b)

Figure 3.27: Comparison between actual and model NOx levels for (a) a week in late October ‘03, (b) a week in late December ‘03

Figure 3.28: William Street canyon with wind directions

William Street Canyon

Receptor

S-W wind

N-E wind

S-E wind

φ

North

θ

30o

δ

Page 93: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

75

The William Street canyon is aligned at approximately 30o from due north. Although

the wind directions are assumed to be from the middle of North-East, South-East,

South-West, and South-East, the actual average directions were 47o, 131o, 213o and 300o

from North. Therefore in the canyon the north-east wind is coming at approximately a

13o (φ) angle with respect to the street axis and south-west wind is at 3o (θ) (almost

parallel to the street axis) and the south-east wind is coming at 79o (δ) angle to the street

axis. If the recirculation zone extends through the whole canyon, no direct contribution

will be recorded at the receptor on the windward side (Berkowicz et al. 1997). The

emission contribution on the leeward side would be more than on the windward side, as

was illustrated in Figure 3.5. When the angle between wind direction and street axis is

small then the effect at the receptor would be less than when the angle is big (Berkowicz

et al. 1997).

In this context, the results from both CM2CO and CM2NOx models (shown in Table 3.6

and 3.8) support the arguments reported in the previous studies. The coefficients for

north-east and south-east wind directions in both models have negative signs, which

indicate decreasing concentration at the windward side receptor (see Figure 3.28).

Again the north-east wind has a smaller angle (φ=13o) with the street axis than the

south-east wind (δ=79o); therefore the effect of the north-east wind on pollution

concentration at the receptor would be less. The Beta coefficients are -0.200 and -0.317

for north-east wind and south-east wind respectively in the CM2CO model, indicating

that the south-east wind has a higher negative impact on pollution concentration at the

windward side receptor. A similar situation is observed in the CM2NOx model, where

-0.135 and -0.206 are the Beta coefficients for north-east wind and south-east wind

respectively.

For the south-west wind the receptor is located at the leeward side; therefore pollution

concentration would be increased for this wind direction. The Beta coefficient for

south-west wind was 0.059 in the CM2CO model and 0.104 in the CM2NOx model. In

both the cases the coefficients are positive, which indicates an increase in pollution

concentration. This confirms the expectation that on the leeward side the direct plume

from the vehicles would be added to the recirculated pollution. But since the angle

(θ=3o) between south-west wind and street axis is very small, the impact at the receptor

is relatively low compared to the south-east wind. In addition, the south-west wind

often brings back polluted air, as discussed earlier in this chapter.

Page 94: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

76

Traffic and the cross-product of previous period wind speed and previous period

pollution have relatively large Beta values and thus make the major contributions to the

explanation of the dependent variable. The standard error of the traffic coefficient is

extremely small, so that this coefficient is very reliable for forecasting purposes.

3.5 COMPARISON WITH PREVIOUSLY DEVELOPED MODELS

The model reported in Section 3.4 can be compared with previous studies in terms of

the ratio of CO to NOx level. Because of the different units used in reporting the results

of various models, the ratio CO/NOx enables comparison of results as the ratio is

independent of the units employed. The estimated coefficients for traffic in the CM2CO

and CM2NOx models are used to calculate the ratio CO/NOx. These coefficients are

expressed in volumetric units (i.e. ppm and pphm per 1000 cars). Since most of the

previous estimations were expressed in gravimetric units (i.e. grams per cubic meter)

the traffic coefficients are converted to gravimetric units using the conversion factors of

Colls (1997). Depending on molecular masses for these gases, the conversion factors

are not the same for CO and NOx. The conversion factors for CO and NOx from ppb to

μg/m3 are 1.16 and 1.58 respectively at 20o C. These factors are used to convert the

traffic coefficients to g/m3/vehicle. The ratios of CO to NOx from the present models

and some previous studies are shown in Table 3.9. The ratio for this study is 7.99, and

the ratios from other studies are between 7.95 and 14.00.

Table 3.9: A comparison of CO/NOx ratio between present study and other studies

Studies CO NOx CO/NOx

0.0263 (ppm/1000 cars) 0.00241 (ppm/1000 cars) Present study

3.06E-08a (g/m3/vehicle) 3.82E-09a (g/m3/vehicle) 7.99

Palmgren et al. 1999 25.2b (g/vehicle km) 1.8b (g/vehicle km) 14.00

Eerens et al. 1993 12 (g/vehicle km) 1.5 (g/vehicle km) 8.00

Pokharel et al. 2002 53 (g/kg of fuel) 6.3 (g/kg of fuel) 8.41

EPA Victoria 1999 11.3-12.7 (g/km) 1.42-1.6 (g/km) 7.95

Ketzel et al. 2003 16 (g/vehicle km) 1.6 (g/vehicle km) 10.00

a using conversion factors according to Colls (1997) b estimations are for year 1994 as a base year

Page 95: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

77

Comparison in terms of actual emissions is more difficult. Emission models have been

developed in various studies for use by environmental protection agencies. They

generally estimate the emission rate, in grams per kilometre travelled by a vehicle

(g/km) or grams per mile (g/m). The factors discussed in Chapter 2 (Section 2.3.4) play

significant roles in estimating emission rate. Many studies have developed emission

models from empirical data.

The coefficients of traffic in this study are converted from ppm and pphm to grams per

cubic metre (g/m3) with factors due to Colls (1997). That still leaves the problem of

relating these measures to emission model outputs which are in terms of grams per

kilometre. The CM2CO and CM2NOx models are not directly comparable with

previously developed models because of the different variables used. As an

approximation the coefficient of traffic in g/m3 is converted to g/km by assuming that

the volume of the Perth airshed is 0.219 km3. This is based on the area of the city of

Perth of 8.75 km2 and an assumed depth of the relevant air mass of 25 m, the height of

the William Street canyon. As a very rough approximation, the average car is assumed

to travel one kilometre after entering the central area (Figure 3.29).

On this basis, the contribution to air pollution per car estimated in this study can be

converted to emission factors for comparison with some previous studies (Table 3.10).

2.96 km

2.96 km

0.25 km

Figure 3.29: Assumed dimensions of the Perth airshed

Page 96: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

78

Table 3.10: A comparison of emission factors between this study and other studies Models CO NOx

0.0263 (ppm/1000 cars) 0.00241 (ppm/1000 cars)

3.05E-08a (g/m3/car) 3.82E-09a (g/m3/car) Present study model

6.67 (g/km)b 0.84 (g/km)b

Other studies MOBILE6 (Pokharel et al. 2002) 6.36 (g/km)c 0.75 (g/km)c

MOBILE5 (Robinson et al. 1996) 11.09 (g/km)d 2.1 (g/km)d

MOBILE4.1 (Robinson et al. 1996) 5.86 (g/km)d 1.0 (g/km)d

AusVeh 1.0 (EPA Victoria 1999) 11.3-12.7 (g/km) 1.42-1.6 (g/km)

AGO method (EPA NSW 1995) 11.57 (g/km) pre 1986 cars 3.2 (g/km) post 1986 cars

1.17 (g/km) pre 1986 cars 0.73 (g/km) post 1986 cars

aaSIDRA (Dia et al. 2005) 39.9 (g/km) 0.9 (g/km)

CSIRO (Dia et al. 2005) 17.9 (g/km) 4.7 (g/km)

BTRE 2003a 6.9-12.4 (g/km) 1.37-1.75 (g/km) a using conversion factors according to Colls (1997) b using Perth airshed assumption: the area of Perth city is 8.75 km2 and an assumed depth of the relevant air mass of 25 m produces Perth airshed of 0.219 km3, and average car assumed travel 1 km after entering the city.

c using conversion factor from McGaughey et al. 2004 d results for Fort McHenry site

It is clear that the estimates vary within wide ranges. This study produces results which

are well within those ranges.

While the correspondence between models is reassuring, it is clear that those estimated

in this study are superior from a cause and effect point of view. It would be quite

arbitrary to select one of the previous models as the basis for estimating the impact on

pollution of a car entering the city. The model parameters vary so widely that there

could be little confidence in any such inference. In contrast, the vehicle coefficients in

the CO and NOx models estimated in this study are reliable within narrow confidence

limits.

3.6 CONCLUSION

The topic of this chapter was the development of models of air pollution in Perth. To

express pollution levels mathematically we need to know the factors which influence

the level of pollution concentration. The development of air pollution in Perth is a

complex process combining wind flow, vehicle emission rate and canyon geometry.

Page 97: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

79

Causal models demonstrated that wind speed and wind directions have significant

influences on pollution concentration in the city. For the purpose of this study, the main

result was to obtain accurate measures of the contributions to CO and NOx pollution

made by each vehicle entering Perth city. A comparative analysis of the ratio CO/NOx

and the emission rates for the present model and previously developed models indicated

that the results for the present models are well within the ranges of previous estimates.

The next chapter identifies potential measures to control traffic volume in the city. The

impact of those measures is estimated using the air pollution model developed in this

chapter. Chapters 5, 6 and 7 discuss the travel behaviour of travellers to Perth city. The

outcomes of Chapters 5, 6 and 7 and the air pollution model are combined in Chapter 8.

Page 98: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

80

Appendix-3A

MODEL: ARIMA (111) – CO Model Description: Variable: CO_ppm Regressors: NONE ‘Non-seasonal’ differencing: 1 No ‘seasonal’ component in model. 95.00 percent confidence intervals will be generated. Split group number: 1 Series length: 4368 Number of cases containing missing values: 75 Termination criteria: Parameter epsilon: .001 Maximum Marquardt constant: 1.00E+09 SSQ Percentage: .001 Maximum number of iterations: 10 Initial values: AR1 -.25684 MA1 -.15330 CONSTANT -.00002 Marquardt constant = .001 Adjusted sum of squares = 381.87509 FINAL PARAMETERS: Number of residuals 4292 Standard error .29797092 Log likelihood -898.01901 AIC 1802.038 SBC 1821.1315 Analysis of Variance: DF Adj. Sum of Squares Residual Variance Residuals 4289 381.87399 .08878667 Variables in the Model: B SEB T-RATIO APPROX. PROB. AR1 -.26243707 .13828459 -1.8978042 .05778897 MA1 -.16049131 .14144148 -1.1346835 .25657137 CONSTANT -.00002433 .00414497 -.0058690 .99531755

Page 99: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

81

Appendix-3B

MODEL: ARIMA (111) –NOx Model Description: Variable: NOx_pphm Regressors: NONE ‘Non-seasonal’ differencing: 1 No ‘seasonal’ component in model. 95.00 percent confidence intervals will be generated. Split group number: 1 Series length: 4368 Number of cases containing missing values: 82 Termination criteria: Parameter epsilon: .001 Maximum Marquardt constant: 1.00E+09 SSQ Percentage: .001 Maximum number of iterations: 10 Initial values: AR1 -.21065 MA1 -.39169 CONSTANT -.00043 Marquardt constant = .001 Adjusted sum of squares = 36485.955 FINAL PARAMETERS: Number of residuals 4285 Standard error 2.9105344 Log likelihood -10665.677 AIC 21337.354 SBC 21356.443 Analysis of Variance: DF Adj. Sum of Squares Residual Variance Residuals 4282 36429.641 8.4712103 Variables in the Model: B SEB T-RATIO APPROX. PROB. AR1 -.05275905 .08352462 -.6316587 .52764361 MA1 -.23312078 .08136505 -2.8651219 .00418879 CONSTANT -.00044728 .05158744 -.0086702 .99308264

Page 100: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

82

Appendix – 3C

Table 3C1: Coefficients of the explanatory variables for hourly CO level for CM1CO model

Unstandardised Coefficients

Standardised Coefficients t Sig.

B Std. Error Beta

(Constant) 0.465 0.015 30.94 0.000

Wind speed -0.009 0.002 -0.086 -5.62 0.000

Previous period’s wind speed -0.009 0.002 -0.088 -3.71 0.000

Previous period’s north-east wind speed -0.022 0.002 -0.176 -10.57 0.000

Previous period’s south-east wind speed -0.027 0.002 -0.290 -13.75 0.000

Previous period’s south-west wind speed -0.002 0.002 -0.035 -1.32 0.187

Cross product of previous period’s wind speed and previous period’s CO level 0.020 0.001 0.342 20.61 0.000

Traffic (in ‘000) 0.035 0.000 0.488 40.28 0.000

Dependent Variable: hourly CO level

Table 3C2: Coefficients of the explanatory variables for hourly Ln CO level for

CM3CO model

Unstandardised

Coefficients Standardised Coefficients t Sig.

B Std. Error Beta (Constant) -5.280 0.090 -58.55 0.000

Wind speed -0.023 0.002 -0.144 -9.75 0.000

Previous period’s wind speed -0.009 0.004 -0.055 -2.43 0.015

Previous period’s north-east wind speed -0.040 0.003 -0.197 -12.86 0.000

Previous period’s south-east wind speed -0.056 0.003 -0.372 -19.56 0.000

Previous period’s south-west wind speed -0.004 0.003 -0.038 -1.52 0.130

Cross product of previous period’s wind speed and previous period’s CO level 0.028 0.001 0.305 20.57 0.000

Ln of Traffic 0.552 0.011 0.562 51.44 0.000

Dependent Variable: Ln of CO

Page 101: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 3: Causal relationships between traffic and air pollution

83

Appendix – 3D Table 3D1: Coefficients of the explanatory variables for hourly NOx level for CM1NOx

model

Unstandardised

Coefficients Standardised Coefficients t Sig.

B Std. Error Beta

(Constant) 3.876 0.151 25.62 0.000

Wind speed -0.032 0.016 -0.028 -2.05 0.041

Previous period’s wind speed -0.212 0.024 -0.186 -8.98 0.000

Previous period’s north-east wind speed -0.170 0.021 -0.120 -8.21 0.000

Previous period’s south-east wind speed -0.196 0.019 -0.190 -10.33 0.000

Previous period’s south-west wind speed 0.024 0.018 0.031 1.30 0.193

Cross product of previous period’s wind speed and previous period’s NOx level

0.023 0.001 0.423 28.43 0.000

Traffic (in ‘000) 0.398 0.000 0.488 43.68 0.000Dependent Variable: hourly NOx level Table 3D2: Coefficients of the explanatory variables for hourly Ln NOx level for

CM3NOx model

Unstandardised

Coefficients Standardised Coefficients t Sig.

B Std. Error Beta

(Constant) -5.957 0.102 -58.13 0.000

Wind speed -0.032 0.003 -0.139 -11.43 0.000

Previous period’s wind speed -0.012 0.004 -0.053 -2.82 0.005

Previous period’s north-east wind speed -0.049 0.004 -0.174 -13.22 0.000

Previous period’s south-east wind speed -0.070 0.003 -0.336 -20.43 0.000

Previous period’s south-west wind speed -0.001 0.003 -0.008 -0.36 0.717

Cross product of previous period’s wind speed and previous period’s NOx level

0.002 0.000 0.212 17.24 0.000

Ln of Traffic 0.885 0.012 0.672 72.98 0.000Dependent Variable: Ln of NOx level

Page 102: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 4: Traffic Control Policies to Reduce Pollution

84

CHAPTER FOUR Traffic Control Policies to Reduce Pollution in Perth City: a first assessment based on previously estimated elasticities A causal approach to air pollution has been developed in Chapter 3, with separate

models for CO and NOx levels in Perth City. This work has established the significant

effect of traffic in increasing air pollution. Consequently, measures to control traffic

volume are required in order to improve air quality, and the focus of this chapter is to

identify potential means of improvement in Perth city. The first section identifies

possible strategies from a global perspective. Section 4.2 suggests specific measures to

limit traffic and thus reduce pollution. Section 4.3 reviews previously estimated car trip

demand elasticities with respect to various potential measures. Specific responses to

these measures are discussed. Section 4.4 has quantified the impacts on air quality of

the suggested measures individually and collectively.

4.1 INTRODUCTION

Because transportation problems are experienced in most cities in the world, many are

trying to implement a range of policies in order to moderate traffic as well as reducing

air pollution. To assess the behavioural response of travellers to the “green” issue the

study identifies the potential policies to influence travel behaviour leading to improve

air quality in Perth city. The study follows three stages to quantify the effectiveness of

the potential policies. Initially previous travel behaviour estimations have been used to

measure the impact of suggested policies by applying the pollution coefficients reported

in Chapter 3. At this stage the study uses generalised estimates to the travellers’

responses. The following stage (Chapter 5) is to assess private car drivers’ responses in

the specific context of Perth, where transport mode choice behaviour is modelled. The

third stage (Chapters 6 and 7) estimates the actual responses to various charging policies

with the information collected through a Stated Preference survey. This chapter

discusses the first stage in measuring the effectiveness of proposed measures.

The fact that Perth is one of the cities that have experienced high demand for private

cars and has one of the highest car ownership levels in the world is serious from both

Page 103: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 4: Traffic Control Policies to Reduce Pollution

85

sustainability and air pollution viewpoints. This chapter identifies measures which can

control air pollution by discouraging private car use, especially for commuting.

Policies used in various countries have been introduced in Chapter 2. Table 4.1

summarises the measures that can be implemented generally in Australia, and

specifically in Perth.

Table 4.1: Summary of potential measures for ameliorating air pollution

Policies How VKT measure Full-cost pricing

o Road pricing o Pollution tax & fees o High charge during peak periods

Discriminatory policies

Charging more for potentially higher pollutant vehicles, like big engine, SUV, diesel engine.

Allow selected licence plate numbers to enter the city on alternate days.

Cars in the city Parking control o Increasing parking fee o Reducing long time parking space o Parking cash-out program

Not allowing any vehicles to enter the inner city Public-transport measure

Offering season ticket High capacity of mass transit system, train, trams, etc.

Cycling and Walking measures

Improve pavements More space given to cyclists and pedestrians Workplace parking for cycles Buffer zone, where cars are not allowed to enter

Vehicle monitoring Regular vehicle testing program Setting high emission standard and random emission test

Fuel measure Improve the quality of fuel Use alternative fuel vehicles CNG, Electric, fuel cell, bio-fuel, etc Use technologically advanced vehicles o Hybrid cars, hydrogen cars

Traffic demand management

Land use plan Lane restriction

Car scrapping program

Offering incentives for scrapping older vehicles

Other measures Working at home 1 or 2 days a week

These are the measures that are used or proposed in various cities in the world in order

to reduce congestion and improve the traffic system.

Page 104: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 4: Traffic Control Policies to Reduce Pollution

86

4.2 MEASURES TARGETED TO THE CITY

“If a tax were imposed on only one of two roads, both the speed and travel time would

be better on the ‘toll’ road than on the free road, and the result is increased efficiency”

(Pigou 1920). In a recent study (Sapkota 1999) investigated traffic flows on tolled roads

by income groups. Increased car ownership, increased job opportunities in the city and

shopping in the city area should be taken into account in formulating policies. Some of

the measures in Table 4.1 may be applicable to Perth city.

This study identified seven different measures to control travel behaviour in order to

reduce air pollution (Table 4.2). They can be categorised into 4 types, i) fixed charge,

ii) variable charge, iii) parking measures, and iv) lane restriction. Impacts of these

measures on travel behaviour would vary with the circumstances. Studies reported in

the literature found diverse sensitivity of travel demand with respect to various pricing

and control policies. This sensitivity is measured as elasticity. The literature on this

topic was introduced in Chapter 2 and is reviewed in more detail in this chapter.

Table 4.2: Policies applicable to Perth City

Measure Type A fixed pollution charge for bringing a vehicle into the city at any

time during weekdays Fixed charge

A differential charge imposed with respect to the size of the cars A differential charge imposed with respect to time of a day to

enter the city A variable charge imposed for distance travelled within the city

Variable charge

Increase parking fee for long term and peak period Reduce parking space in city Parking measure

Reduce lanes for the vehicles and provide that space to cyclists, pedestrians, and public transport use Lane restriction

4.3 ELASTICITY ESTIMATION

A wide range of research has been conducted on own (direct) price and cross elasticity

of transport demand. Some studies have estimated both short and long run elasticities,

whereas other estimates are only for the short run. The BTRE database on elasticities

has been consulted but the original source references are given here. Not all previous

estimation can be exactly categorised as being based on i) fixed charge, ii) variable

charge, iii) parking measures, or iv) lane restriction. However the following discussion

tries to relate previously estimated elasticities to these categories. Elasticities are with

Page 105: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 4: Traffic Control Policies to Reduce Pollution

87

respect to generalised cost, to take account of cases where there was previously no

charge. The charge is treated as an increase in running cost and time cost in money

terms.

4.3.1 Response to a Fixed Charge

A number of studies have been carried out on road or congestion pricing. The main

objective of this type of pricing is to reduce traffic in the congested area, especially in

the peak period, and thus increase average speed. The standard approach to optimal

congestion pricing depends on three prime elements; those are, the speed-flow

relationship, the demand function, and the generalized cost (Yang et al. 2004).

Although the primary objective is to reduce congestion in a specific area at a specific

time of a day, the supplementary objective is to reduce air pollution. This latter purpose

is getting more attention recently and hence this type of pricing is sometimes called an

“emission fee”. A wide range of studies have attempted to determine the optimal

charge under this approach. Most of them used marginal cost pricing; some even used a

specific trial-and-error approach to optimise pricing. The impact may not be the same

in all cases. Before measuring the impact of pricing we look at some of the successes of

this approach around the world.

First Singapore and then several cities in Norway experimented with congestion charges

for central cities. Recently London has introduced charging for entry to the central area,

in order to reduce congestion levels (Stopher 2004). Taking Singapore first, the Area

Licensing Scheme (ALS) was implemented in 1975 and significantly modified in1994.

The total impact was a 71.1% decrease in number of private vehicles entering the

restricted zone during the morning peak period just after introduction of the scheme.

Tolls ranging from A$0.36 to A$2.15 are collected through 34 non-stop overhead

gantries. The Straits Times reported that within two months of the implementation of

this policy traffic dropped by 20-24% in the restricted zone. Luk (1999) estimates that

toll elasticities in Singapore are -0.19 to -0.58, with an average of -0.34. Cost of

implementing this Electronic Road Pricing (ERP) system was S$192 million (A$160

million), however, it smoothed traffic flow and even increased speed to about 60 km per

hour during the peak (McNulty 2000).

Norway also implemented congestion pricing in three major cities starting in 1986. A

comparison of road pricing in Norwegian cities is shown in Table 4.3.

Page 106: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 4: Traffic Control Policies to Reduce Pollution

88

Table 4.3: Comparative performance of cordon pricing for Norwegian cities

Characteristics Bergen Oslo Trondheim

City population 213,000 456,000 138,000

Commencement in 1986 1990 1991

Number of toll stations 7 19 11 (22 in 1998)

Charging period Mon-Fri 6 am to 10 pm

All days All hours

Mon – Fri 6 am to 5 pm

Entry charge for a small vehicle (NOKa) 5 12 10

Traffic impact (traffic reduction) 6-7% 3-4% 10%

a NOK is Norway Kroner equivalent to A$0.201 on 8th March 2006 Source: Tretvik (2003)

London Transport reports that their charging scheme has reduced traffic by 18% inside

the charging zone, and has reduced delay by 30% (Lettice 2004). However, many

people in London are reported to have negative reactions toward this scheme (Lettice

2004).

An electronic road pricing project (1983-1985) in Hong Kong was undertaken

experimentally but not subsequently implemented, although the project was

economically and technically feasible. According to the findings the pricing would

reduce travel time by 20-24%, average speed at the central area during peak period

would be increased by 16%, and traffic flow would be reduced significantly. The

project was not continued for a range of reasons, though the benefits of road pricing

were recognised.

The objective of the Melbourne City Link Project, with road pricing, is to provide the

city link to toll paying travellers. Though the purpose is not to restrict travellers to

Melbourne CBD, the project has at least one toll station at the point of entering the city.

The project has successfully diverted traffic from some other links, thus improving

traffic flow with reduced travel time. At the same time the system is generating revenue

to repay the project cost.

A recent study conducted by Matas and Raymond (2003) summarises previous

estimates of toll elasticities, and develops a model for toll road demand in Spain. They

Page 107: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 4: Traffic Control Policies to Reduce Pollution

89

have calculated that short term price elasticities for road tolls range from -0.21 to -0.83,

which is rather higher than previous estimates. Again Arentze et al. (2004) developed

models to estimate toll price elasticity of travel demand. Their estimates are within a

range from -0.13 to -0.19 for the entire system and from -0.35 to -0.39 for congested

areas and times.

From the above discussion it is clear that congestion pricing or emission pricing would

certainly reduce travel demand in a specific area at a specific time. This study evaluates

the application of this pricing approach along with other policy measures in order to

reduce air pollution in the city centre.

4.3.2 Responses to Variable Charges

Three different types of variable charge could be imposed on private vehicle users in

Perth city depending on use of their vehicles. The three categories are: i) variable

charges imposed on peak and off-peak entrance; ii) variable charges imposed with

respect to distance travelled within the city; and iii) variable charges imposed with

respect to the size of the vehicle and/or type of fuel used.

None of the studies reviewed has estimated the impact of these forms of variable charge

on travel demand. However due to the nature of the suggested charges they may be

viewed as additional costs of running a car. Such a charge can easily be matched with

the cost of fuel to run a car. There have been many studies of elasticity of travel

demand and it is still an important area of transport study. Many of these studies tried

to establish both long term and short term relationships with travel behaviour.

Johansson and Schipper (1997) measured long term fuel price elasticity of -0.55 to -0.05

for car travel demand and -0.35 to -0.05 for mean driving distance (per car per year).

The authors’ “best guess” values are -0.3 and -0.2 respectively. Goodwin (1992)

reviewed studies of travel demand elasticity and summarised the elasticities of traffic

levels with respect to fuel price from 11 studies of -0.16 and -0.33 for short-run and

long-run respectively.

Luk and Hepburn (1993) reviewed travel demand elasticities in Australia and

summarised studies relating to different states in Australia. Table 4.4 shows various

estimates of fuel consumption elasticities with respect to fuel price in Western

Australia.

Page 108: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 4: Traffic Control Policies to Reduce Pollution

90

Table 4.4: Estimates of elasticity of fuel consumption with respect to fuel price

Author Coverage Data Short run Long run

Schou & Johnson

(1979)

Australia 1955-1976 -0.05 Not available

Donnelly (1981) WA 1966-1980 -0.10 -0.30

Donnelly (1981) WA 1958-1981 -0.19 -0.78

Donnelly (1984) WA 1958-1981 -0.16 -1.03

Australiaa -0.10 -0.42

Australiab -0.12 -0.67 a Estimated using a weighted average of State results. b Estimated using a single national equation. Source: Luk & Hepburn (1993)

A study by Mayeres (2000) indicated the fuel price elasticities of car mileage of -0.16

and -0.43 for peak and off-peak transport respectively for essential trips and -0.43 and

-0.36 for optional trips.

A few other similar studies tried to establish the relationship between out-of-pocket

expenses and travel demand. Button (1993) estimated the elasticity of road travel with

respect to out of pocket expenses of -0.3 to -2.9 for urban commuting. Oum et al.

(1990) reviewed a number of studies on transport demand elasticity. They summarised

elasticities as -0.12 to -0.49 for peak period and -0.06 to -0.88 for off-peak period and

-0.001 to -0.52 for the whole day.

A review by the Industry Commission (1993) summarises travel demand elasticity with

respect to variable car cost and petrol price. Table 4.5 shows the result.

Table 4.5: Travel demand elasticity

Car travel demand w.r.t. Short run Long run

Variable car costs -0.09 to -0.24 -0.22 to -0.31

Petrol price -0.04 to -0.20 -0.30

Source: Industry Commission (1993) Hensher and Young (1990) examined fuel price elasticity of energy demand in Australia

with data for the years from 1961 to 1988. They have estimated this elasticity at -0.17

using ordinary least squares and -0.21 using two-stage least squares methods for

Page 109: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 4: Traffic Control Policies to Reduce Pollution

91

passenger cars. A study by De Borger and Wouters (1998) estimated cross elasticities

of travel demand for peak and off-peak period car use (in Table 4.6).

Table 4.6: Peak and Off-peak travel demand elasticity

Elasticity w r t cost of car trip Price

Peak Off-peak

Peak car -0.3 0.048

Off-peak car 0.05 -0.6

Source: De Borger and Wouters (1998)

Taplin et al. (1999) estimated commuter elasticities for public transport and car as

shown in Table 4.7.

Table 4.7: Demand elasticities of car and public transport for work trips

Elasticity w r t fare or cost of trips by Travel mode

Public transport Car

Public transport -0.15 0.17

Car 0.08 -0.09

Source: Taplin et al. (1999)

This section has briefly reviewed the relationship between variable costs of transport

and the demand for it. A major part of variable operating costs is the cost of petrol;

other variable costs are in-vehicle time and parking charges. The analysis in this study

is based on the assumption that the elasticities for variable operating costs (principally

fuel costs) would be the same as those for variable charges imposed as a policy

measure. Some may argue that private car users already pay ample fuel tax, so why

should an additional charge be imposed. The answer is that the targets of the proposed

charges are private car users who enter into the city centre, not all car users. The

argument is that those who create more pollution in a polluted area should pay more.

4.3.3 Responses to Parking Measures

Another set of measures this study considers as means of reducing air pollution would

affect parking. There are two types of measure, i) increased parking fee, and ii) reduced

parking space in the city. Both would discourage private car users from bringing their

cars to the city. The study uses previously estimated elasticities to calculate the impact

of this measure on air pollution.

Page 110: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 4: Traffic Control Policies to Reduce Pollution

92

A study by Hensher and King (2001) on parking policy in Sydney CBD suggested that

1% increase in hourly parking rate would result in a 0.54% reduction in CBD parking

and 0.29% increase in public transport use. Hess (2001) estimated price elasticity of

demand for parking at work in Portland CBD with varying daily parking fees. The

elasticity was -0.44 for $6 or more parking fee and -0.07 for $1. The results of a review

by Willson & Shoup (1990) on parking price elasticities is summarised in Table 4.8.

Table 4.8: Parking price elasticity

Study Type City Change Elasticity

Surber et al. 1984 Before/after Los Angeles - near CBD

Ended employer-paid parking for solo drivers

-0.68

Soper 1989 Before/after Los Angeles – suburban

Price of solo driver parking raised to two thirds market rate

-0.32

Transport Canada 1974

Before/after Ottawa Stopped providing free parking

-0.11

Francis & Groninga 1969

With/without Los Angeles – CBD

Comparison employees with & without free parking

-0.29

Shoup & Pickrell 1980

With/without Los Angeles – suburban

Comparison employees with & without free parking

-0.10

Note With/without studies compare commuting behaviour of matched samples of employees with and without employer paid parking. Before/after employer paid parking was eliminated. Source: Willson & Shoup (1990)

ICF Consulting Group (1997) indicated that 1% increase in parking fees would reduce

vehicle miles travelled (VMT) to work by 12% to 39% and would even reduce solo

driving by 66% to 81%. Another study found a price elasticity of on-street parking of

-0.37 (Calthrop 2002). Wilson (1992) commented that “best performing models predict

that between 25 and 34 percent fewer automobiles are driven to work when workers

have to pay to park, as compared to when they park free”.

In summary, parking price policy has a significant impact on travel behaviour.

Increasing price would encourage people not to drive solo to the city centre and also to

take transit to work. Although studies have been done on price elasticity of parking, no

study has been found which addresses the relationship between commuter behaviour

and increasing or decreasing parking space. Therefore, for simplicity of analysis this

Page 111: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 4: Traffic Control Policies to Reduce Pollution

93

study considers parking fee and parking space measures as a composite parking price

measure.

4.3.4 Responses to Lane Restriction

To ease congestion and improve traffic speed, transport planners are concerned with

increased road supply. Hensen & Huang (1997) estimate a long term elasticity of lane

capacity and vehicle miles travelled as 0.9 in the metropolitan area and 0.6 to 0.7 at the

county level. They also summarise elasticities from previous studies varying from 0.1

to as high as 0.7 depending on the methodology of the studies. Another study (Noland

2001) suggested that the relationship between lane capacity and vehicle miles travelled

is 0.3 to 0.6 in the short run and 0.7 to 1.0 in the long run. Noland and Cowart (2000)

also estimated a corresponding elasticity of 0.655 for freeways and arterials. Fulton et

al. (2000) measured induced traffic in the US mid-Atlantic region using cross-section

and time series data. The estimated average elasticities of vehicle miles travelled with

respect to lane-miles range from 0.2 to 0.6.

All of these studies focus on increased supply, meaning increased road capacity or more

lanes. However, in the present study one of the suggested pollution control policies is

to reduce lanes for private vehicles in the urban area. For the purpose of estimating the

response to lane restriction the study assumes that estimated elasticities are reversible.

This measure may increase congestion but that effect may be partly offset by

implementing congestion charges, discussed before.

4.3.5 Average Elasticities

Table 4.9 summarises the elasticity estimates reviewed for i) fixed charge, ii) variable

charge, iii) parking measures, and iv) lane restriction. There are short and long term

elasticities of demand for trips with respect to variable charges and lane restriction but

only short term elasticities for fixed charges and parking measures. Long term

elasticities for fixed charge and parking measures are assumed to be double the short

run elasticities, as a similar doubling is observed (approximately) in the case of variable

charges and lane restriction.

There are also cross elasticity effects but these are beyond the scope of this study, which

considers only the own price effect of each measure. In summarising the elasticities,

extreme estimates are ignored (Table 4.9).

Page 112: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 4: Traffic Control Policies to Reduce Pollution

94

Table 4.9: Average elasticities for four suggested measures

Type of measure

Average Previously estimated travel demand elasticity with respect to own price

Fixed charge Short run -0.15 -0.06, -0.07, -0.03, -0.04, -0.1, -0.2, -0.24, -0.18, -0.34, -0.21, -0.13, -0.19

Long run -0.30 (assumed to be double the short run elasticity)

Variable charge Short run -0.18 -0.3, -0.16, -0.05, -0.1, -0.19, -0.16, -0.12, -0.16, -0.3, -0.12, -0.49, -0.09, -0.24, -0.17, -0.169, -0.212, -0.3, -0.094

Long run -0.39 -0.33, -0.29, -0.3, -0.78, -0.22, -0.31, -0.3, -0.55, -1.03

Parking

measure

Short run -0.30 -0.54, -0.44, -0.07, -0.12, -0.68, -0.32, -0.11, -0.29, -0.1, -0.39, -0.37, -0.25, -0.34

Long run -0.59 (assumed to be double the short run elasticity)

Lane reduction

measure

Short run 0.43 0.1, 0.7, 0.3, 0.6, 0.655, 0.2, 0.6

Long run 0.83 0.9, 0.7, 1.0, 0.6, 0.9, 0.904 Note: Italicised values in the last column are extreme ones, which are ignored for the calculation of

averages For small changes, elasticity can be expressed in equation (4.1).

e = VP

PV

PPVV

×∂∂

=∂∂

// ............................................... (4.1)

However, for larger changes, the applicable elasticity expression is:

( )( )PP

VVLnLnLnLn

e21

21

−=

This can be rewritten as,

( ){ }PPVV LnLneLnExp 2112 −−= ………………….. (4.2)

Where, e is own price elasticity

V1 is initial traffic

V2 is reduced traffic

P1 is initial price/cost

P2 is final price/cost

Both equations (4.1) and (4.2) can be represented graphically as in Figure 4.1. For a

small change in price, the elasticity expression follows the solid line whereas for a

larger change it follows the broken line.

Page 113: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 4: Traffic Control Policies to Reduce Pollution

95

The average elasticities (from Table 4.9) for different measures are used in equation

(4.2) to estimate the reduced traffic for different measures.

4.4 IMPACT ON AIR QUALITY

The following section uses the elasticity estimates to quantify reductions in car use in

Perth city leading to reduced CO and NOx levels. Traffic levels are estimated for the

various measures using average elasticity, then CO and NOx levels are determined by

using the air pollution models developed in Chapter 3.

4.4.1 Impact of a Fixed Charge

Long term price elasticity of travel demand with respect to a fixed charge in any area at

any time is estimated as -0.30 (Table 4.9). To estimate the traffic response to a charge

using equation (4.2), we need the base cost (P1) to which the fixed charge is added.

Taylor and Taplin (1998) estimated equilibrium cost per km for both social cost and

private cost. This study uses private cost for the analysis, as it is considering out-of-

pocket expenses for commuters. Taylor and Taplin (1998) consider only fuel cost and

time value for calculating private cost, while Bray and Tisato (1997) identified various

approaches to private costs, such as i) time costs and fuel costs, ii) time costs, fuel costs

and parking costs, iii) time costs, fuel costs and vehicle maintenance costs. Estimated

private cost per km is $0.264 which has been updated to 2004 from the figure indicated

in Taylor and Taplin (1998) using the consumer price index. Average car kms travelled

Figure 4.1: Price elasticity for small and large change

V

P

Page 114: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 4: Traffic Control Policies to Reduce Pollution

96

per trip is 12.6 in 2004 (Perth and Regional Transport Survey). Therefore, average cost

per trip for the commuter in Perth is $3.33. This figure is used to calculate the impact of

a fixed charge.

The proposed fixed charge would be levied on private vehicles each time they enter the

city regardless of the time of a day. This fixed charge is subject to further research;

however social cost may be about $ 2.29 (in 2004 value) more per trip than private costs

(based on Taylor and Taplin 1998). The present study investigated the long term impact

on air pollution of a $1 charge per trip. Table 4.10 shows impacts on CO and NOx

levels for four different levels of charge.

Table 4.10: Impact of fixed charge: percentage reduction in pollutants and traffic

Amount charged Pollutants & traffic $0.5 $1.0 $1.5 $2.0

CO 2 4 6 7

NOx 2 3 5 6

Traffic 4 8 11 13

Table 4.10 shows that $1.0 per trip extra charge would reduce average daily traffic by

8%, which would lead to reduction of 4% and 3% in daily average of CO and NOx

levels respectively.

0.000

0.200

0.400

0.600

0.800

1.000

1.200

1.400

1.600

100

300

500

700

900

1100

1300

1500

1700

1900

2100

2300

hour

ppm

0

5000

10000

15000

20000

25000

30000tr

affic

CO-no charge CO-fixed charge traffic-no charge traffic-fixed charge

Figure 4.2a: Impact of fixed charge of $1.0 on hourly CO level (6-month average data)

Page 115: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 4: Traffic Control Policies to Reduce Pollution

97

0.00

2.00

4.006.00

8.00

10.00

12.0014.00

16.00

18.00

100

300

500

700

900

1100

1300

1500

1700

1900

2100

2300

hour

pphm

0

5000

10000

15000

20000

25000

30000

traffi

c

NOx-no charge NOx-fixed charge traffic-no charge traffic-fixed charge

The reductions in pollution levels are not uniform. Figures 4.2a and 4.2b show that

traffic in the morning and afternoon peaks are at about the same level, whereas both CO

and NOx in the afternoon peak are higher than in the morning peak. The reasons have

already been discussed in Chapter 3. In each hour the vehicle emissions combine with

lingering emissions from previous hours that eventually build up to the highest point at

around 6 PM even though the traffic level is falling.

4.4.2 Impact of Variable Charges

Long term elasticity of travel demand with respect to variable costs is estimated from

previous studies as -0.39 (Table 4.9). This study considers charges which vary by time

of day. Three time periods are identified for this analysis: i) peak period from 6 am to

10 am, ii) during the day from 10 am to 5 pm, and iii) other times of the day. The peak

starts with the build-up of traffic and ends when it declines. Emissions accumulate

rapidly in this period. The proposed rate would be $0.45 per km in the peaks, $0.40 per

km between peaks, and $0.35 per km for the rest of the day. This charge is again

arbitrary, however it is not baseless as total social cost per trip is about $5.60 (indexed

up to 2004) (Taylor and Taplin 1998). For the average trip, the proposed variable

charges together with the fixed charge come to a total of $8.74, $8.25, and $7.76 per trip

for the three periods during a day. Table 4.11 also shows two other alternative charging

scenarios.

Figure 4.2b: Impact of fixed charge of $1.0 on hourly NOx level (6-month average data)

Page 116: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 4: Traffic Control Policies to Reduce Pollution

98

Table 4.11: Impact of variable charge by time of day (peak/between peaks/rest of day): percentage reduction in pollutants and traffic

Alternative charging scenarios ($ per km) Pollutants & traffic $0.4/$0.35/$0.3 $0.45/$0.4/$0.35 $0.5/$0.45/$0.4

CO 5 8 10

NOx 4 6 8

Traffic 10 14 18

The suggested variable charges would reduce traffic by 14% which leads to 8% and 6%

reductions in daily average of CO and NOx levels respectively. The effects by time of

day are shown in Figures 4.3a and 4.3b for CO and NOx respectively. They are similar

to Figures 4.2a and 4.2b, the only difference being a scale effect.

0.000

0.200

0.400

0.600

0.800

1.000

1.200

1.400

1.600

100

300

500

700

900

1100

1300

1500

1700

1900

2100

2300

hour

ppm

0

5000

10000

15000

20000

25000

30000

traf

fic

CO-no charge CO-variable charge traffic-no charge traffic-variable charge

0.002.004.006.008.00

10.0012.0014.0016.0018.00

100

300

500

700

900

1100

1300

1500

1700

1900

2100

2300

hour

pphm

0

5000

10000

15000

20000

25000

30000

traf

fic

NOx-no charge NOx-variable charge traffic-no charge traffic-variable charge

Figure 4.3b: Impact of variable charges of $0.4/$0.35/$0.3 for three periods of a day on hourly NOx (6-month average data)

Figure 4.3a: Impact of variable charges of $0.4/$0.35/$0.3 for three periods of a day on hourly CO (6-month average data)

Page 117: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 4: Traffic Control Policies to Reduce Pollution

99

Impacts of variable charges on pollution levels are generally similar to those of a fixed

charge. However the proposed charges would reduce pollution levels more than the

fixed charge.

4.4.3 Impact of the Parking Measures

Long term elasticity of travel demand with respect to parking price is estimated from the

previous studies as -0.59 (Table 4.9). The proposed measure is to increase parking fees

per hour in the city to $4 in the peak (6 am to 10 am) and $3 off-peak (rest of the day).

Current parking fees in the city vary from $0.8 to $3.5 (City of Perth 2004). Traffic as

well as pollution levels would be reduced in the city centre after imposing additional

parking fees. Air pollution levels are shown with varying parking fees for peak and off-

peak periods in Table 4.12 with three other scenarios.

Table 4.12: Impact of parking measure (peak/off-peak): percentage reduction in pollutants and traffic

Alternative charging scenarios Pollutants & traffic $ 3.5/$3.0 $ 4.0/$2.5 $ 4.0/$3.0 $ 4.5/$3.5

CO 9 5 9 13

NOx 7 5 8 11

Traffic 16 10 18 25

The suggested parking fees would reduce daily traffic by 18% leading to 9% and 8%

reductions of average daily CO and NOx levels respectively. The hourly result is shown

in Figures 4.4a and 4.4b for CO and NOx respectively.

0.000

0.200

0.400

0.600

0.800

1.000

1.200

1.400

1.600

100

300

500

700

900

1100

1300

1500

1700

1900

2100

2300

hour

ppm

0

5000

10000

15000

20000

25000

30000

traf

fic

CO-no charge CO-park measure traffic-no charge traffic-park measure

Figure 4.4a: Impact of parking measure of $0.4.0/$3.0 on hourly CO (6-month average data)

Page 118: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 4: Traffic Control Policies to Reduce Pollution

100

0.00

2.00

4.00

6.00

8.00

10.00

12.00

14.00

16.00

18.00

100

300

500

700

900

1100

1300

1500

1700

1900

2100

2300

hour

pphm

0

5000

10000

15000

20000

25000

30000

traffi

c

NOx-no charge NOx-park measure traffic-no charge traffic-park measure

The parking fee measure would reduce pollution levels, but not as much as it would

reduce traffic.

4.4.4 Impact of Lane Restriction

Long term elasticity of travel demand with respect to lane capacity is estimated from

previous studies to be 0.83 (Table 4.9). This value indicates that expanded road supply

increases travel demand by this proportion in the long run. On the assumption that the

relationship is reversible, this elasticity is used to calculate the effect of reducing lane

availability on private vehicle traffic in the city. At this point, the study considers a

25% reduction in lanes open to private vehicles. This means that the left lane of any

4-laned road in the city would be closed to private vehicle users and restricted to public

transport, pedestrians and cyclists. Figure 4.5 gives an example.

Figure 4.4b: Impact of parking measure of $0.4.0/$3.0 on hourly NOx (6-month average data)

Restricted lane

Figure 4.5: Example of lane restriction at William Street and Wellington Street intersection Source: based on Main Roads Western Australia

William St Horseshoe Bridge

Page 119: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 4: Traffic Control Policies to Reduce Pollution

101

A sensitivity analysis of lane reduction is shown in Table 4.13. There would be 21%

reduction of private traffic leading to 11% and 10% reductions in daily average of CO

and NOx respectively from a 25% lane restriction policy. Of total traffic in Perth in

2002, 82% was private vehicles, 17% commercial vehicles, and 1% buses. Figures 4.6a

and 4.6b show the reduction of traffic and pollution levels for CO and NOx respectively.

Table 4.13: Impact of lane restriction: percentage reduction in pollutants and traffic

Proportion of lane reduction Pollutants & traffic 25% 50% 75%

CO 11 23 36

NOx 10 20 31

Traffic 21 44 69

0.000

0.200

0.400

0.600

0.800

1.000

1.200

1.400

1.600

100

300

500

700

900

1100

1300

1500

1700

1900

2100

2300

hour

ppm

0

5000

10000

15000

20000

25000

30000

traf

fic

CO-no charge CO-lane restriction traffic-no charge traffic-lane restriction

Figure 4.6a: Impact of 25% lane restriction on hourly CO (6-month average data)

Page 120: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 4: Traffic Control Policies to Reduce Pollution

102

0.00

2.00

4.00

6.00

8.00

10.00

12.00

14.00

16.00

18.00

100

300

500

700

900

1100

1300

1500

1700

1900

2100

2300

hour

pphm

0

5000

10000

15000

20000

25000

30000

traf

fic

NOx-no charge NOx-lane restriction traffic-no charge traffic-lane restriction

4.4.5 Combined Impact

If we assume that the effects of all the policies can be combined linearly then the overall

impact on air pollution would be substantial. There may even be a synergistic effect

when a variety of different measures are implemented simultaneously. The impact of

combined effects can be either stronger or weaker than the total of individual effects

(Shiftan and Suhrbier 2002). Table 4.14 shows the long term combined impacts of

pollution and traffic levels. The sum of the individual impacts on traffic, CO and NOx

is more than the combined policy effect which is not estimated by the simple addition of

individual effects. The combined effect is calculated with equation (4.3) which is the

same as equation (4.2) except that the second part of the right hand side of the equation

is in terms of a summation of four policies.

The reduced traffic is applied to the air pollution models developed in Chapter 3. Thus

the impact of the combined policy is a 49% reduction in average daily traffic and 26%

reduction in CO and 22% reduction in NOx.

( )⎥⎦

⎤⎢⎣

⎡−−= ∑

=

4

12112

iiii PPV LnLneLnVExp ……………………………. (4.3)

Where: V2 = combined reduced traffic

V1 = existing traffic P = price level i = 1,2,3,4 policy measures

Figure 4.6b: Impact of 25% lane restriction on hourly NOx (6-month average data)

Page 121: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 4: Traffic Control Policies to Reduce Pollution

103

Table 4.14: Combined impact of policy measures: percentage reduction of pollutants and traffic

Impacts Fixed charge

Variable charge

Parking measure

Lane restriction Combined

CO 4 8 9 11 26

NOx 3 6 8 10 22

Traffic 8 14 18 21 49

The reductions can be converted to totals for the Perth airshed on an annual basis. The

area of the City of Perth is 8.75 Square km. The dimension of the Perth city airshed was

considered in Chapter 3 as an 8.75 square kilometres area by a 25 metre elevation. The

annual reduction of CO and NOx (in tonnes) for the Perth airshed is shown in Table 4.15

(unit conversion is taken from Colls 1997).

Table 4.15: Annual reduction of pollution in tonnes in the Perth airshed

Pollution level

Fixed charge

Variable charge

Parking measure

Lane restriction Combined

CO 2 4 5 7 15

NOx 0.3 0.3 0.3 0.4 0.9

The short and long term combined impacts of pollution are compared in Figures 4.7a

and 4.7b for CO and NOx respectively. In the short run both CO and NOx levels are

moderately reduced but in the long run there is a much greater impact. For any

transport policy one would expect the impact to reach equilibrium in the long run.

CO Level

0.0000.2000.4000.6000.8001.0001.2001.4001.600

100

300

500

700

900

1100

1300

1500

1700

1900

2100

2300

hour

ppm

no charge long term combined policy short term combined policy

Figure 4.7a: Short and long term impact of combined policy on hourly CO

Page 122: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 4: Traffic Control Policies to Reduce Pollution

104

NOx Level

0.002.004.006.008.00

10.0012.0014.0016.0018.00

100

200

300

400

500

600

700

800

900

1000

1100

1200

1300

1400

1500

1600

1700

1800

1900

2000

2100

2200

2300

2400

hour

pphm

no charge long term combined policy short term combined policy

4.5 CONCLUSION

This chapter formulated four measures to reduce pollution due to traffic in Perth city.

Average price elasticities of travel demand have been used in calculating the resulting

traffic flows. The air pollution models (developed in Chapter 3) for hourly CO and NOx

levels in relation to traffic and meteorological factors are used to estimate the pollution

levels. The measures would have significant impacts in reducing CO and NOx levels in

Perth city. This chapter also estimated the combined effect of the policies and found

significant improvement of air pollution in both short and long run.

In identifying feasible measures to reduce traffic in Perth city and thus improve air

quality, this chapter has also provided a preliminary assessment of policy impacts. The

next stage of this study assesses actual travel behaviour in the Perth context. The

policies suggested in this chapter would target private transport users. In order to

estimate the effects of implementing those policies in Perth it is necessary to estimate

transport mode choice behaviour. Chapter 5 analyses transport mode selection

behaviour for travellers’ to the Perth city.

Figure 4.7b: Short and long term impact of combined policy on hourly NOx

Page 123: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

105

CHAPTER FIVE

Factors Influencing Car Use: A Revealed Preference Analysis

Perth city is a prime candidate for car suppression. A rule of thumb is that about 80%

of travellers use public transport to go to the major city centres in the world and 20%

use cars. However the Perth and Regional Travel Survey (PARTS) found the converse.

About 70% use car and 30% use public transport to go to Perth city.

Policies for suppressing car use were identified in Chapter 4 and initial estimates of

their potential for controlling air pollution were made. They showed that pricing and

control measures could reduce CO and NOx levels in Perth city by reducing traffic.

Generalised estimates indicated the likely effectiveness of the proposed policies. This

chapter deals with the next step, estimating private car drivers’ responses in the specific

context of Perth.

Individual responses and mode selection are based mainly on the level of service

provided. Level of service includes travel time and trip cost. Other attributes

associated with each mode include reliability, comfort, and safety, but these are difficult

to quantify. Consequently, the problem of ridership attraction has been discussed

largely in qualitative rather than quantitative terms (Ben-Akiva and Morikawa 2002).

To overcome the quantification problem, researchers may also include socio-

demographic characteristics of individuals in the model to estimate travel demand and

assess travel behaviour.

A mode choice model is developed using the Perth and Regional Travel Survey

(PARTS) data. Section 5.2 discusses the methods used to analyse the choice models.

Section 5.3 refers to the data structure and its description. After that, section 5.4

describes the few assumptions needed to make the database ready for use in choice

modelling. Section 5.5 reports the empirical models using discrete choice analysis and

interprets the models through elasticities and value of travel time savings (VTTS). The

whole chapter is designed to provide insight into traveller mode selection and

behaviour.

Page 124: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

106

5.1 A TWO-PHASE MODELLING APPROACH

Discrete choice analysis is used to model the choice of one among a set of mutually

exclusive alternatives. The estimation of discrete choice models is often based on only

one type of data, usually revealed preference (RP) data based on actual trip records

(Ortuzar and Iacobelli 1998). Another source of information is stated preference (SP)

which captures respondents’ preferences in hypothetical situations. The revealed

preference method has been used widely to analyse travel demand and behaviour in the

transport research field (Caldas and Black 1997). The advantage of RP modelling is

that the data set is based on the individual’s actual behaviour.

In cases where responses to hypothetical stated preference (SP) scenarios are being

tested, there is always an element of doubt about reliabilty and the degree of ‘contextual

realism’ (Swait et al. 1994, Louviere et al. 2000). The now conventional way of

overcoming the difficulties is to combine SP and RP data. The SP data set can provide

measures of the impact of potential air pollution control policies but may not give

unbiased predictions of car travel reactions if respondents misjudge how they would

behave in a real situation. Being based on what has already happened, RP models in

contrast are expected to give unbiased estimates of travel parameters, subject to accurate

model specification and quality of data.

In general, the SP model should not be used to estimate future impacts unless the

parameters are adjusted by a scale factor but for a single data set it is not possible to

estimate a scale factor and it is usually normalised to the constant 1 (Adamowicz et al.

1997, Boxall et al. 2003). In addition to the scale factor problem, SP data have not been

used widely for estimation of predictive models due to unreliability of information from

hypothetical scenarios (Ben-Akiva et al. 1991, Huang et al. 1997, Ortuzar and Iacobelli

1998). Ben-Akiva et al. (1991) refer to the reliability in terms of ‘validity’ and

‘stability’. Lack of validity involves the discrepancy between stated and actual

behaviour of respondents, whereas lack of stability refers to the magnitude of the

random error in SP data.

There is a well established method of combining SP and RP in a special type of nested

logit joint estimation; the two data sets are combined and a relative scale parameter

estimated (Louviere et al. 2000). Morikawa (1994) and others (Ben-Akiva and

Morikawa 1990, Hensher and Bradley 1993, Louviere et al. 2000, Cherchi and Ortuzar

Page 125: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

107

2002) have recommended using SP and RP data jointly to exploit their advantages and

overcome their limitations. The process of pooling RP and SP data and estimating a

model from the pooled data has been called data enrichment (Louviere et al. 2000) and

the purpose of data enrichment is to produce a model which can be used to forecast real

future market scenarios. A number of studies have reported that a combined SP-RP

model produces more efficient estimates. These studies include Adamowicz et al.

(1997), Boxall et al. (2003), Ben-Akiva et al. (1991), Ortuzar and Iacobelli (1998),

Hensher et al. (1999) and Hensher et al. (2005) among many others.

Despite the general acceptance of the SP-RP joint estimation procedure, ‘sequential

estimation’ as presented in Swait et al. (1994) and discussed in Louviere et al. (2000) is

an alternative in which the SP and RP estimations are done separately but the results are

combined. In their discussion of this procedure, Swait et al. (1994) commented: “In our

presentation and empirical application, we used RP and SP data collected from the same

respondents. This need not be the case, of course. The RP data might come from

existing data sets, collected in a form completely independent of the SP data.”

In the present study, combined SP-RP model estimation was not feasible because of the

inherent limitations of a car user survey. An SP survey testing car driver responses to

hypothetical charges could have no RP counterparts to these responses. The only

meaningful answer to an SP scenario was whether the respondent would take their car to

the city or not. The factual section of the study could seek socio-demographic

information about the respondent but there could be no mode choice in the RP data

except taking a car to the city. The consequences are discussed in Chapter 6.

An alternative approach was adopted, dividing choice modelling into two phases. The

first was to develop an RP model of mode selection by travellers to Perth city using a

subset of the Perth and Regional Transport Survey (PARTS) data. However the

questions in the survey limited the estimation of mode choice parameters to trip costs

and parking charges – as well as travel time. Parking charge alone is too generic for the

purposes of this study.

Consequently a second phase was needed to differentiate between potential responses to

hypothetical charges as well as response to a possible lane restriction. The hypothetical

charges would be for entering the city, for large car size and for entry time, as well as a

possible increase in parking charge. The Car Trip Response Survey 2005 was

Page 126: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

108

conducted as the basis for an SP model assessing reactions to the suggested policy

measures. As already noted, the SP data collection could not generate matching RP data

because of different choice sets. The SP data covered the behaviour of car travellers

only.

The problem of the parameters estimated for the detailed measures being based entirely

on the SP responses is addressed by integrating them into the estimated RP model. The

relevant SP model coefficients are converted to the equivalent of the parking fee

coefficient in the RP model. The conversion is made because the SP scenarios are

hypothetical whereas the RP response to the parking fee is an actual response. In reality

any pricing policy imposed on the motorist may be viewed as the cost of taking a car to

a specific location, equivalent to a parking fee. Once the SP estimates are converted to

parking fee equivalents then this value is used through an RP model simulation process.

Details of this method are discussed in Section 8.3. In effect the simulated model

rescales the SP coefficients through the RP parking fee coefficient so that the probable

impacts of the potential policy measures can be realistically assessed. This means that

the impact of a charge for entering the city or a charge for a large vehicles or a charge

according to entry time can be calculated, as well as the impact of an increased parking

fee.

5.2 METHODOLOGY

5.2.1 The Revealed Preference (RP) Model

In the marketing discipline or any other field where consumers make choices, several

steps in reaching a decision are recognised. The first is that consumers become aware

of the needs and/or problems, then they search for information about the products or

services that could satisfy their needs. In searching for information, consumers identify

the features or attributes associated with the products or services. Once consumers have

assessed the attributes of the alternatives, they develop a preference ordering about the

products or services by trading off attributes to maximise satisfaction (utility). Train

(2003) reports that to fit within the discrete choice framework, the set of alternatives

needs to have three characteristics. First, the alternatives must be mutually exclusive

from a decision maker’s perspective. Second, the choice set must be exhaustive, in that

all possible alternatives are included. Third, the number of alternatives must be finite.

Page 127: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

109

This process is also applicable to choosing a transport mode from alternatives to travel

to any destination. Travellers make a choice of either taking a car or riding a bus or

train (depending on availability) to the destination. In making a selection, travellers

usually assess the attributes of modes in terms of in-vehicle time, out of vehicle time

and trip fare or cost. Then they are assumed to maximise satisfaction (utility) in terms

of the attributes associated with the modes and their socio-demographic characteristics.

According to the random utility approach, the probability of choosing a particular

alternative can be expressed as in equation (5.1).

Piq = Pr (Uiq > Ujq) ................................................... (5.1)

= Pr (Viq + εiq > Vjq + εjq)

= Pr (Viq - Vjq > εjq - εiq)

Where Vi is the observed component of the utility function and εi is the unobserved

(random) component. An expanded form of the utility function is shown in equation

(5.2).

iqikikii XU εγβα +++= ……………………….. (5.2)

Where αi is the alternative-specific constant for choice alternative i

βik is the parameter associated with attribute k of choice alternative i

Xik is attribute k of choice alternative i

γq is the systematic (observed) component associated with the individual response

εi is the random (unobserved) component of the individual response

The multinomial logit (MNL) choice model is based on the principles of utility

maximisation and has the advantage of simple mathematical structure (Koppelman and

Wen 1998, Train 2003). However, a basic assumption made in using the MNL model in

discrete choice modelling is the IID assumption, that the variances associated with the

unobserved component of the utility function for each alternative are identical, and

these effects are not correlated between any pair of alternatives (Louviere et al. 2000).

Researchers normalise this unobserved variance term to any number. The simplest and

easiest way of doing that is to attach a scale parameter (τ) to this unobserved term and

normalise the scale parameter by making τ = 1. Multiplying the scale parameter on both

sides of the utility function does not make any difference in choosing an alternative to

maximise utility.

Page 128: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

110

The mathematical expression of the MNL model to estimate the probability of choosing

alternative i from choice set C is shown in equation (5.3).

∑∈

++

++

∑=

Cj

X

X

ik

jjkjkj

kiikiki

e

epεβα

εβα

…...…...………........……… (5.3)

This model implies that when the attributes of one alternative improve (e.g. decreasing

travel time), the probability of choosing that alternative rises and there is a shift to this

alternative from other alternatives. The logit model implies a certain pattern of

substitution across alternatives (Train 2003). For any two alternatives i and m from j

alternatives, the ratio of the logit probabilities is:

mqiq

mq

iq

jq

mq

jq

iq

VVV

V

j

V

Vj

V

V

mq

iq eee

ee

ee

PP −===

∑ ....................................... (5.4)

This ratio does not depend on any alternatives other than i and m. Since, the ratio is

independent of other alternatives, it is said to be independent from irrelevant

alternatives. The logit model is based on this assumption of Independence of Irrelevant

Alternatives (IIA). While the IIA property is realistic in some choice situations, it is

clearly inappropriate in others (Train 2003, Chipman 1960, Debreu 1960). Often the

researcher is unable to capture all sources of correlation explicitly, so that the

unobserved components of utility are correlated and IIA does not hold. In this situation

a more general model than standard logit is needed (Train 2003). The most widely used

general MNL model is Nested Logit (NL) or Hierarchical Logit (HL) which allows

interdependence between the pairs of alternatives in a common group. This assumption

can also be relaxed when the model uses a ‘variable choice set’ for the individual.

Nested logit reflects a more generalised form of multidimensional structure of the

choice set (Ben-Akiva and Lerman 1985).

In this study a nested logit choice model has been developed because the structure of the

choice set can be arranged hierarchically. The modes chosen by the travellers are

mainly car, bus and train. A basic classification of the modes is private and public.

Bus and train are public modes and car is private. About 30% of those who used cars

Page 129: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

111

were passengers and 70% drivers. Hence, the private mode is again divided into car use

as a driver and car use as a passenger categories. The convenient way to picture the

substitution patterns is with a tree diagram. The tree diagram in this case is shown in

Figure 5.1.

The diagram shows two nests with two alternatives in each nest. It is assumed that

unobserved components are more correlated within the nest and less correlated between

the nests. This correlation or justification of using Nested Logit can be measured with

the estimation of the inclusive value (log sum). This term will be discussed in the

subsequent section.

The probability of choosing the alternative car as a driver, for example, using the above

tree diagram can be estimated by using a Bayes probability. This probability is the

product of the conditional probability of choosing car as a driver given that the private

mode is chosen and the marginal probability of choosing the private mode. This

probability is expressed mathematically in equation (5.5).

cqciqiq PPP |= ........................................... (5.5)

Where Piq|c is the conditional probability of individual q choosing alternative i given

that the person chooses nest c, and Pcq is the marginal probability of individual q

choosing nest c. This decomposes Piq in such a way that both the components can take

the logit form. The marginal and conditional probabilities can be expressed as:

∑ +

+

=lqlcl

cqccq

IZ

IZ

cq eeP τ

τ

....................................... (5.6)

Private Public

Car as driver

Car as passenger

Bus Train

Figure 5.1: Hierarchical structure of mode distribution for travellers to Perth City

Page 130: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

112

=cij

ciq

Y

Y

ciq eeP τ

τ

/

/

| ........................................ (5.7)

Zcq is the observed component of utility which is variable over the nests, but not over

the alternatives within each nest, and Yiq is another observed component which is

variable over alternatives within a nest. ∑= ciqYcq eI τ/ln , is called the inclusive value.

The value of Icq links the upper and lower models by bringing information from the

lower model into the upper model. A higher value of τc means greater independence

and less correlation, which leads to a lower inclusive value. This model is applicable

where the dataset can be arranged as in Figure 5.1.

5.2.2 Elasticities

Although the probabilities estimated from the dataset are important in an appropriate

model, it is often useful to have a generalised measure of response to a change in some

observed factor. This is done by computing the elasticity. Estimating own and cross

elasticities using the logit model helps to assess people’s responsiveness to any attribute

in choosing any alternative.

The probabilities in equations (5.5, 5.6, and 5.7) can be used to calculate the own or

direct elasticity as well as the cross-elasticities of demand for the mode. The first step

in calculating the weighted average is the probability weighted direct elasticity for

person q is expressed in equation (5.8).

iqiq

ikq

ikq

iqPx P

PX

XP

E iq

ikq..

∂∂

= …………………….....……. (5.8)

Therefore the probability weighted direct point elasticity for person q from the logit

model is:

iqiqikqikPx PPXE iq

ikq)1( −= β ................................................ (5.9)

And the probability weighted cross elasticity of person q’s choice mode i with respect to

attribute Xjkq of mode j is:

iqjqjkqjkPx PPXE iq

jkqβ−= ........................................................... (5.10)

These weighted individual elasticities are summed and divided by the sum of the

weights Piq to obtain a weighted average elasticity, which is shown in equation (5.11).

Page 131: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

113

∑∑=

iq

PX

w PE

Eiq

ikq

……………………………………. (5.11)

In this case the elasticities measure how an increase in trip cost on one mode decreases

the demand for that mode and increases demand for other modes.

5.2.3 Value of travel time saving (VTTS)

A by-product of this analysis is an estimate of the value of travel time saving (VTTS).

If the utility model contains a travel-time (tt) attribute and trip-cost (tc) attribute then

VTTS can be estimated by the simple calculation shown in equation (5.12) (Hess et al.

2005).

tcVttVVTTS

∂∂∂∂

=//

........................................................................ (5.12)

Equation (5.12) reduces to tctt ββ / , where βtt and βtc are the time and the cost

coefficients in the choice model. Equation (5.12) is based on the assumption that the

derivative of the unobserved part of utility with respect to travel-time and trip-cost is

zero.

The choice elasticities of transport modes and the value of travel time saving have been

estimated using a nested logit model with a dataset extracted from the Perth and

Regions Travel Survey (PARTS). This is an on-going project started in 2002. The time

frame of the data set used in this study is between 2002 and 2003.

5.3 DATA STRUCTURE

The PARTS database was designed for other purposes than this study. The data were

collected in two different forms. One is information about the respondents’ households

and the other is about daily trips. The entire database contains around 15000 trip

records. In this analysis, the records selected were for those who stopped first at Perth

city on their travel day. The number of those records is 406; however 30 who cycled or

walked to the city and the one who used taxi were excluded from the analysis. The

study considers those trips where only car or bus or train were used as a mode of

transport; this number of records is 375. Table 5.1 summarises the characteristics of the

Page 132: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

114

travellers. Some descriptive statistics of this dataset are provided in the following

paragraphs.

Car use dominates other modes. 70% of the travellers used car, 14% used bus and 16%

train. Of those who used cars about 30% rode as passengers. This passenger group can

be identified by licence status or being in the minor age group.

Table 5.1: Characteristics of sampled travellers to Perth city (from PARTS survey 2002-03)

Frequency Percent Main Mode

Car as driver 188 50.1 Car as passenger 75 20.0 Bus 54 14.4 Train 58 15.5

Gender

Female 190 50.7 Male 185 49.3

Income

Not reported 65 17.3 No income 31 8.3 $1 - $199 per week 35 9.3 $200 - $399 per week 34 9.1 $400 - $599 per week 35 9.3 $600 - $799 per week 48 12.8 $800 - $999 per week 41 10.9 $1000 - $1499 per week 50 13.3 $1500+ per week 36 9.6

Purpose of trip Work 143 38.1 Non-work 232 61.9 No. of vehicles in household

No car 10 2.7 1 car 94 25.1 2 cars 180 48.0 3 cars 58 15.5 4 cars 22 5.9 5 cars 10 2.7 6 cars 1 0.3

Licence status

Full licence 318 84.8 P-plates 8 2.1 Learners permit 6 1.6 No licence 43 11.5

Mean SD Age 39.10 15.80 Travel time (min) 23.83 12.12 Distance travelled (km) 13.52 9.12

Page 133: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

115

An important characteristic is income which was reported in a range per week.

However, about 17% of the respondents did not report their income.

Many travellers made their trips to Perth city by public transport even though they had

car(s) in their household. About 72% of the reduced sample of households had 2 or

more cars and 3% had no car. Among the respondents about 11% had no driver’s

licence. This group would not have the choice alternative of car as driver.

Travel time is one of the major considerations in making a mode selection decision.

Mean travel time is 39.1 min with standard deviation of 12.12 min. Mean distance

travelled is 13.52 km with standard deviation of 9.12 km.

The descriptive statistics give an overview of those who travel to the city on a regular or

occasional basis. As the survey was done for different purposes from the objectives of

this study, data needed to be prepared to be useful for a mode choice model by making

some reasonable assumptions.

5.4 ASSUMPTIONS FOR DATA IMPUTATION

The PARTS database can be used to develop a mode choice model after making some

logical assumptions about each individual’s choice set. The main attributes of the

modes are travel-time and trip-cost. The database contains only actual travel time spent

on the mode used by the traveller. There is no estimated travel time for other

alternatives. A logical approach is used to estimate travel time and trip cost for each

alternative mode in each case. Assumptions used for imputing data are:

a) In-vehicle travel times for car use as a driver and as a passenger are assumed to

be the same.

b) In-vehicle travel time for car is the actual time spent by the travellers using car.

It is assumed that the same time from the same suburb would be the option for

those who haven’t used their car.

c) In-vehicle travel time for bus is estimated using the Transperth web site

http://www.transperth.wa.gov.au/DesktopDefault.aspx. The “Journey Planner”

link provides alternative routes with estimated duration of trip if someone used

only bus from a particular suburb to ‘William Street’ in Perth city. The quickest

time is taken to be the travel-time by bus.

Page 134: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

116

d) The same procedure is used to estimate travel-time for train only. However,

some of the suburbs do not have train service. For them the train mode is not an

alternative choice.

e) After estimating in-vehicle time, access and egress time are added for different

modes. Five minutes egress time is added to in-vehicle time for the car as

driver mode, whereas 2 minutes are added for car as passenger mode. On the

other hand a total of 15 minutes are added as access and egress time for both bus

and train modes.

f) Trip-costs for car as driver and car as passenger are considered to be the same.

For the car mode, trip-cost is considered to be only fuel cost. Fuel consumption

is estimated by assuming average performance by the car used in this case.

According to the Bureau of Transport and Regional Economics

(www.btre.gov.au/docs/r107/r107.pdf) the average on-road fuel performance of

cars in Australian cities is 11-12 litres per 100 km. A performance of 11.5 litres

per 100 km is used as the estimate of car fuel consumption.

g) It is assumed that only unleaded petrol is used. Monthly average unleaded petrol

price per litre is used to calculate fuel cost. Average monthly petrol price for the

years 2002 and 2003 (Appendix-5A) has been collected from the Australian

Automobile Association’s web site (http://www.aaa.asn.au/petrol.htm).

h) The product of fuel consumption and monthly average petrol price gives trip-

cost for the car mode.

i) In the case of bus and train, a standard ‘MultiRider 40’ rate is used to estimate

the trips fare. This rate varies among Transperth zones. The fare is estimated

according to each traveller’s origin zone (Appendix-5B).

j) Hourly parking fee is imputed on the basis of individual situations. First the

specific location of the destination (according to Main Roads WA zone number)

is identified for each individual and then the closest parking place located from

the parking map of the City of Perth. Then the hourly parking fee is recorded

for that parking space assuming that the individual would park at that location.

In the choice model travel-time and trip-cost are used as attributes associated with the

modes.

Page 135: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

117

5.5 MODEL ESTIMATION

To estimate the parameters of the utility functions the LIMDEP statistical package was

used. Every choice alternative for a respondent required one row of data. That means

for one respondent there will be 4 rows of information for 4 mode choice alternatives

with their corresponding attributes. A sample of data presentation is shown in

Appendix-5C.

Before formulating the model the number of choices available for each individual was

identified. The maximum number of choice alternatives was 4, car as driver, car as

passenger, bus, and train, but not all of them were available to all travellers, and the set

of choices varies between 1 and 4. This variable choice set is determined on the basis of

physical and logical availability of the alternatives.

• Those who do not have a drivers licence will not have the car as driver choice.

• Those who live in suburbs where train service is not offered will not have the

train choice.

• Those who do not have a car will not have car as driver or car as passenger

choices.

On the basis of the above conditions the choice sets are as follows:

i) CSET = 1, means only Bus

ii) CSET = 2, means EITHER CarP and Bus OR Bus and Train

iii) CSET = 3, means EITHER CarD, CarP, & Bus OR CarP, Bus, & Train

iv) CSET = 4, means CarD, CarP, Bus, & Train

Where CSET is variable choice set

CarD is car as driver

CarP is car as passenger

The independent variables used in developing the logit model are:

Travel Time (TRAVELTIME): Total time spent on the mode including access

and egress time (in minutes).

Trip Cost (TRIPCOST): Fuel cost for CarD and CarP, and fare for Bus and

Train ($ per trip).

Page 136: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

118

Parking fee (PARKHOUR): Parking fee per hour ($).

Age (AGE): Age in years of the respondent.

Gender (GENDER): A dummy variable, 1 for males and 0 for females.

The study has gone through several steps. First the MNL model was developed with

only attributes associated with the modes, and then another MNL model including some

socio-demographic factors was developed. After that, the NL model was formulated

and finally a NL model with different travel time parameters for different modes was

developed.

5.5.1 Multinomial Logit (MNL) Model

The first model, a preliminary model using only the attributes associated with the

modes, travel time and trip cost, without socio-demographic variables, gave results

which were not good. The log likelihood was -350.99 with pseudo R2 of 0.24 and

t-ratios for travel time and trip cost were poor (Appendix-5D).

The second model, an MNL with socio-demographic variables had these utility

functions:

UcarD = αcarD + βtt * TRAVELTIME + βtc* TRIPCOST + βpark* PARKHOUR

UcarP = αcarP + βtt * TRAVELTIME + βtc* TRIPCOST + βage* AGE + βgender* GENDER

UBus = αBus + βtt * TRAVELTIME + βtc* TRIPCOST

UTrain = βtt * TRAVELTIME + βtc* TRIPCOST

Where, βtt is the coefficient for travel time, βtc trip cost, βpark for hourly parking fee, βage

for age, βgender for gender (a dummy variable), and α is the alternative specific constant

(ASC). There is no ASC for train so that the other ASCs are estimated relative to the

train alternative. Because the socio-demographic variables have the same values across

all alternatives, they have been included in only some utility functions. The minor age

group can be logically included in the utility function for car as a passenger. Clearly

parking cost is associated with the person driving the car. The expected signs of the

coefficients for travel time, trip cost, and parking fee are negative as people maximise

utility by minimising costs. Again, if the sign of the ASC for car as a passenger mode

is positive, then the expected signs for gender and age are negative, implying that males

are more likely to drive and minors are likely to be a passenger.

Page 137: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

119

The estimated MNL model coefficients using the above model specification are shown

in Table 5.2 and the model output from LIMDEP is provided in Appendix-5E.

Table 5.2: Estimated results: MNL model with socio-demographic variables

Parameters Coefficient Std. Error t-ratio

βtt -0.0073 0.008 -0.89

βtc -0.3499 0.259 -1.35

βpark -0.2388 0.317 -0.75

βage -0.0301 0.009 -3.17

βgender -0.8233 0.295 -2.78

αcarD 0.132 0.469 0.28

αcarP 0.104 0.452 0.23

αBus -1.353 0.258 -5.23

In the results the coefficients for travel-time, trip-cost, and parking fee are negative, as

expected. Travel time, trip cost, and parking fee are inversely related to mode selection,

meaning that travellers try to maximise utility by reducing travel time and trip cost. The

t-ratios indicate generally low levels of confidence in the estimated parameters, which

may reflect misspecification of the model or poor quality of data. The log likelihood of

the model, used for comparing with other models, is -340.80 with pseudo R2 of 0.26.

The closer the log likelihood to zero the better the model.

5.5.2 Nested Logit (NL) Model

A more appropriate model was developed using the NL model with a tree structure as in

Figure 5.1. Two nests were specified as private and public, with two alternatives under

each of these nests. The nested logit model is more appropriate here as people using

private and public modes may be influenced by different factors. To capture this

heterogeneity of individuals, the NL model is more realistic. The NL model produces

comparatively better results with the log likelihood of -339.88. Adjusted Pseudo R2 is

0.34, which again does not mean much except to compare with other models. The

model is shown in Table 5.3 (for detail see Appendix-5F).

Page 138: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

120

Table 5.3: Estimated results with NL model

Parameters Coefficient Std. Error t-ratio

βtt -0.0135 0.01 -1.33

βtc -0.398 0.20 -1.90

βpark -0.295 0.32 -0.93

βage -0.031 0.01 -3.24

βgender -0.804 0.29 -2.74

αcarD 0.444 0.51 0.87

αcarP 0.357 0.46 0.76

αBus -1.493 0.31 -4.77

IV parameters, tau(j|i,l),sigma(i|l),phi(l)

PRIVATE 1.000 0.00 Fixed

PUBLIC 0.780 0.144 5.41

This model produces better results as t-ratios are appreciably larger in absolute value

than in the MNL models. The inclusive value (IV) parameter for the private mode was

fixed to estimate IV for the public mode. The estimated IV parameter for public of 0.78

implied that the two nests are not highly correlated, indicating the superiority of the NL

model over the MNL model. An IV parameter approaching zero would imply

independent nests. The success table (crosstab) for the NL model is shown in Table 5.4.

Table 5.4: Success table for NL model (Successes are in bold on the diagonal)

Car as a driver

Car as a passenger

Bus Train Total

Car as a driver 114 29 24 22 188

Car as a passenger 30 27 12 5 75

Bus 25 10 14 6 54

Train 20 9 4 25 58

5.5.3 Nested Logit Model with different time parameters (NLDT)

Other associated attributes such as ability to read during the trip mean that a given time

on car, bus and train may not be viewed as the same. Therefore another NL model was

Page 139: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

121

developed using different travel time parameters for car, bus, and train. The results are

shown in Table 5.5 (details in Appendix-5G).

Table 5.5: Estimated results with NL model with different travel time parameter (NLDT)

Coefficient Std. Error t-ratio

βtt_car -0.010 0.01 -0.65

βtt_bus -0.066 0.03 -2.27

βtt_train 0.034 0.02 1.44

βtc -0.706 0.31 -2.27

βpark -0.285 0.32 -0.89

βage -0.031 0.01 -3.18

βgender -0.834 0.29 -2.82

αcarD 1.591 0.72 2.19

αcarP 1.525 0.68 2.23

αBus 2.388 1.24 1.91

IV parameters, tau(j|i,l),sigma(i|l),phi(l)

PRIVATE 1.000 0.00 Fixed PUBLIC 0.621 0.14 4.35

Although the t-ratios for the time parameters raise difficulties, the log likelihood of

-329.07 is better than for the simpler NL model. The Pseudo R2 (0.36) also indicates a

better model. One of the main issues here is the positive sign of the travel time

parameter for train. Hess et al. (2005) report that the use of an unbounded distribution

(Normal) may lead to a non-zero probability of a positive travel-time coefficient. The

sign could be an artefact of the model specification or the poor quality of the data used

in the model (Hess et al. 2005). Few cases have been found where the disutility of

travel time by train is estimated to be less than for car travel time but it is just possible

that this estimate partly reflects the opportunity to read in comfort on the train.

The estimate of train travel time was based on the Transperth web site from ‘suburb

train station’ to Perth city station which may be too short. In some cases, car travel time

from the same suburb to the Perth city is more than the estimated train travel time. The

problem is due to not knowing the exact location of the traveller’s origin point. Precise

location of the origin would help to estimate travel time accurately. The success table

(Table 5.6) gives better predictions than before.

Page 140: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

122

Table 5.6: Success table for NL model with different travel times (Successes are in bold on the diagonal)

Car as a driver

Car as a passenger

Bus Train Total

Car as a driver 116 29 24 19 188

Car as a passenger 30 28 13 4 75

Bus 25 10 15 5 54

Train 18 8 3 28 58

Table 5.7 shows comparative results for the different models. Although the NLDT

model produces better log likelihood and pseudo R2 and even better crosstabs, the

unreliable coefficient estimates for car and train travel-time weaken the model.

Consequently the simpler NL model is considered the best estimator for mode choice

decisions from this dataset.

Table 5.7: Comparative results from different models, coefficient (t-ratio) MNL without

socio-demo MNL with socio-demo

NL NLDT

βtt -0.007 (-0.91) -0.007 (-0.89) -0.014 (-0.87)

βtt_car -0.010 (-0.65)

βtt_bus -0.066 (-2.27)

βtt_train 0.034 (1.44)

βtc -0.358 (-1.38) -0.350 (-1.35) -0.398 (-1.90) -0.706 (-2.27)

βpark -0.238 (-0.75) -0.295 (-0.93) -0.285 (-0.89)

βage -0.030 (-3.17) -0.031 (-3.24) -0.031 (3.18)

βgender -0.823 (-2.78) -0.804 (-2.74) -0.834 (-2.82)

αcarD -0.131 (-0.48) 0.132 (0.28) 0.444 (0.87) 1.591 (2.19)

αcarP -1.365 (-4.65) 0.104 (0.23) 0.357 (0.76) 1.525 (2.23)

αBus -1.404 (-5.40) -1.353 (-5.23) -1.493 (-4.77) 2.388 (1.91)IV parameters

PRIVATE 1.000 (fixed) 1.000 (fixed)PUBLIC 0.780 (5.41) 0.621 (4.35)

Log likelihood for function -350.99 -340.80 -339.88 -337.59

Pseudo R2 0.34 0.36

Further exploratory study identified two groups of respondents, work and non-work.

Two separate nested logit models were estimated with the same model specification as

Page 141: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

123

the first NL model. Neither provided a better model, mainly because of small sample

sizes. Only 143 respondents entered the city for work purposes and 232 for non-work

purposes (Table 5.1). These two groups are small in size for discrete choice modelling.

5.5.4 Elasticity Estimation

The NL model is used to estimate direct and cross elasticities of mode choice with

respect to trip cost. The NL model output has been used to estimate a set of probability

weighted direct and cross elasticities (top part of Table 5.8). The SUMPRODUCTs of

each column and the number of trips for the top part of Table 5.8 are not zero, which

conflicts with the requirement that the total number of trips before and after switching

modes should be the same. This means that there is no trip generation effect (Taplin

1982, Taplin et al. 1999). The bottom part of Table 5.8 has been adjusted with Solver

to satisfy the following conditions.

• The Sumproduct of each column and the number of trips is zero, i.e. a traveller

changing from one mode must travel by another.

• The cross-elasticities below the diagonal are symmetric with those above.

Expressed in equation (5.13):

jii

jij e

ExpExp

e = ....................................................... (5.13)

Where, eij is choice elasticity for mode i with respect to cost of mode j

eji is choice elasticity for mode j with respect to cost of mode i

Expj is expenditure on mode j

Expi is expenditure on mode i

• The cross-elasticities are assumed to be non-negative as these modes are

substitutes for one another.

Page 142: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

124

Table 5.8: Direct and Cross-elasticities with respect to the trip cost

Original

Car as driver cost

Car as passenger

cost

Bus fare Train fare

Car as driver -0.229 0.092 0.086 0.070

Car as passenger 0.230 -0.399 0.113 0.069

Bus 0.270 0.150 -0.532 0.080

Train 0.192 0.079 0.076 -0.395

Adjusted Trips

Car as driver -0.229 0.159 0.066 0.035 188

Car as passenger 0.399 -0.399 0.000 0.000 75

Bus 0.159 0.000 -0.532 0.301 54

Train 0.079 0.000 0.281 -0.395 58

Column sumproduct 0.000 0.000 0.000 0.000

On the no-generation assumption, if the bus fare is increased by 1% then the demand for

bus would be reduced by 0.53%, and at the same time the demand for car as driver and

train would be increased by 0.07%, and 0.28% respectively, but there would be no shift

from car as a passenger mode. These elasticities are comparable to the estimates made

by Louviere et al. (2000).

5.5.5 Value of Travel Time Savings (VTTS) Estimation

Although it is a by-product of the study, the estimate of the value of travel time saving

(VTTS) provides a means of comparison and validation in relation to previous studies.

On the average, travellers will trade off travel time against trip cost on the different

modes. The VTTS is estimated according to equation (5.11). The NLDT model has the

opportunity to estimate VTTS for different modes, but a model indicating a non-zero

probability of positive travel-time coefficient should not be used in VTTS calculation

(Hess et al. 2005). The VTTS of $2.04 per hour is estimated using the simpler NL

model. This value is comparable to the estimate by Louviere et al. (2000).

5.6 DISCUSSION AND CONCLUSION

The modelling work reported in this chapter has provided a set of estimates relating

specifically to travel to Perth city. The most important is the estimate of response to

travel cost which can be used to predict the effects of car suppression measures.

Page 143: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

125

Unfortunately the estimated coefficient with respect to parking fees has a small

t-statistics so that it is not as reliable as had been hoped.

Thus the model provided estimates indicating that further refinement was needed. As

the available data source had been exhausted, this could only be achieved by a targeted

survey using stated choice.

Two alternative multinomial logit models and two nested logit models had been

developed for travellers’ mode choice. Travellers to Perth city used mainly four modes:

car as driver, car as passenger, bus, and train. This part of the study investigated

reasons for using different modes to travel to Perth city. Travel-time and trip-cost are

the main attributes travellers consider at the time of mode selection. However, some

other socio-demographic factors also influence their decisions. Two MNL models were

developed, one with only travel-time and trip-cost and another including some socio-

demographic factors. Both models produced poorer fit as compared to NL models.

Two nested logit models provided better results from this dataset.

This part of the study provides insight into the behaviour of people who travel to Perth

city regularly, especially with regard to the costs of their own and other modes. As the

main objectives of this project is to assess the behaviour of car users when subjected to

charges and/or other measures, the stated preference survey reported in the next chapter

was needed to give a more complete assessment of travellers’ reactions to the various

measures indicated in Chapter 4.

Page 144: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

126

Appendix-5A

Table 5A1: Monthly average unleaded petrol price

Month Perth Metro Average

(cents per litre) October 2002 92.5 November 2002 89.8 December 2002 90.6 January 2003 95.1 February 2003 98.7 March 2003 102.0

Source: www.aaa.asn.au/petrol.htm (Australian Automobile Association)

Page 145: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

127

Appendix-5B

Table 5B1: Public Transport Fares

Cash MultiRider 10 MultiRider 40 Standard Concession Standard Concession Standard Concession 2 Section $ 1.30 $ 0.50 $ 11.05 $ 4.25 $ 39.00 $ 15.00 1 Zone $ 2.00 $ 0.80 $ 17.00 $ 6.80 $ 60.00 $ 24.00 2 Zone $ 3.00 $ 1.30 $ 25.50 $ 11.05 $ 90.00 $ 39.00 3 Zones $ 3.80 $ 1.60 $ 32.30 $ 13.60 $ 114.00 $ 48.00 4 Zones $ 4.50 $ 1.90 $ 38.25 $ 16.15 $ 135.00 $ 57.00 5 Zones $ 5.50 $ 2.10 $ 46.75 $ 17.85 $ 165.00 $ 63.00 6 Zones $ 6.40 $ 2.40 $ 54.40 $ 20.40 $ 192.00 $ 72.00 7 Zones $ 7.30 $ 2.80 $ 62.05 $ 23.80 $ 219.00 $ 84.00 8 Zones $ 8.00 $ 3.10 $ 68.00 $ 26.35 $ 240.00 $ 93.00 DayRider $ 7.50 $ 3.00 NA $ 25.50 NA $ 90.00 FamilyRider $ 7.50 NA NA NA NA NA

Source: http://www.transperth.wa.gov.au/DesktopDefault.aspx?tabid=26

Page 146: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

128

Appendix-5C

Table 5C1: First 16 observations of the data set

ID choice cset mode travel time trip cost park fee gender age

1 0 3 2 32 2.39 0.00 1 161 0 3 3 85 2.85 0.00 1 161 1 3 4 36 2.85 0.00 1 162 0 3 1 35 2.44 1.50 1 302 1 3 2 32 2.44 0.00 1 302 0 3 3 85 2.85 0.00 1 303 1 3 1 35 2.55 1.20 0 273 0 3 2 32 2.55 0.00 0 273 0 3 3 85 2.85 0.00 0 274 1 1 3 50 2.25 0.00 1 435 1 1 3 50 2.25 0.00 0 726 0 3 1 25 1.13 1.30 0 346 0 3 2 22 1.13 0.00 0 346 1 3 3 50 2.25 0.00 0 347 0 3 1 25 1.04 1.50 0 307 0 3 2 22 1.04 0.00 0 307 1 3 3 50 2.25 0.00 0 308 1 4 1 30 1.39 0.70 0 238 0 4 2 27 1.39 0.00 0 238 0 4 3 45 2.25 0.00 0 238 0 4 4 26 2.25 0.00 0 239 0 3 1 30 1.16 0.70 0 299 0 3 2 27 1.16 0.00 0 299 1 3 3 60 2.25 0.00 0 29

10 1 3 1 25 1.19 1.30 1 4810 0 3 2 22 1.19 0.00 1 4810 0 3 3 55 2.25 0.00 1 4811 0 3 1 30 1.22 1.20 1 7511 1 3 2 27 1.22 0.00 1 7511 0 3 3 55 2.25 0.00 1 7512 0 3 2 17 0.67 0.00 0 1512 0 3 3 30 1.50 0.00 0 1512 1 3 4 26 1.50 0.00 0 1513 1 4 1 20 1.09 1.20 0 3513 0 4 2 17 1.09 0.00 0 3513 0 4 3 42 2.25 0.00 0 3513 0 4 4 31 2.25 0.00 0 3514 1 3 1 20 1.13 1.40 0 4714 0 3 2 17 1.13 0.00 0 4714 0 3 3 35 2.25 0.00 0 4715 0 3 1 20 1.09 1.50 0 1915 1 3 2 17 1.09 0.00 0 1915 0 3 3 35 2.25 0.00 0 1916 1 4 1 35 1.14 1.20 1 3116 0 4 2 32 1.14 0.00 1 3116 0 4 3 47 2.25 0.00 1 3116 0 4 4 115 2.25 0.00 1 31

Page 147: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

129

Appendix – 5D

Multinomial Logit Model (MNL) without socio-demographic variable --> sample;all$ --> NLOGIT; lhs=CHOICE,CSET,MODE; CHOICES=carD,carP,bus,train; MODEL: U(carD)=A_carD+time*TRAVELTI+cost*TRIPCOST/ U(carP)=A_carP+time*TRAVELTI+cost*TRIPCOST/ U(bus)=A_bus +time*TRAVELTI+cost*TRIPCOST/ U(train)= time*TRAVELTI+cost*TRIPCOST; crosstab; $ Normal exit from iterations. Exit status=0. +---------------------------------------------+ | Discrete choice (multinomial logit) model | | Maximum Likelihood Estimates | | Model estimated: Dec 21, 2005 at 11:21:27AM.| | Dependent variable Choice | | Weighting variable None | | Number of observations 375 | | Iterations completed 5 | | Log likelihood function -350.9941 | | Number of parameters 5 | | Info. Criterion: AIC = 1.89864 | | Finite Sample: AIC = 1.89907 | | Info. Criterion: BIC = 1.95099 | | Info. Criterion:HQIC = 1.91942 | | R2=1-LogL/LogL* Log-L fncn R-sqrd RsqAdj | | Constants only. Must be computed directly. | | Use NLOGIT ;...; RHS=ONE $ | | Chi-squared[ 2] = 224.85927 | | Prob [ chi squared > value ] = .00000 | | Response data are given as ind. choice. | | Number of obs.= 375, skipped 0 bad obs. | +---------------------------------------------+ +---------+--------------+----------------+--------+---------+ |Variable | Coefficient | Standard Error |b/St.Er.|P[|Z|>z] | +---------+--------------+----------------+--------+---------+ A_CARD -.13108203 .26901136 -.487 .6261 TIME -.00753565 .00826683 -.912 .3620 COST -.35783243 .25957887 -1.379 .1680 A_CARP -1.36572973 .29347376 -4.654 .0000 A_BUS -1.40384979 .25974305 -5.405 .0000 +------------------------------------------------------+ | Cross tabulation of actual vs. predicted choices. | | Row indicator is actual, column is predicted. | | Predicted total is F(k,j,i)=Sum(i=1,...,N) P(k,j,i). | | Column totals may be subject to rounding error. | +------------------------------------------------------+ Matrix Crosstab has 5 rows and 5 columns. CARD CARP BUS TRAIN Total +----------------------------------------------------------- CARD | 112.00000 33.00000 21.00000 22.00000 188.00000 CARP | 31.00000 23.00000 14.00000 6.00000 75.00000 BUS | 24.00000 11.00000 14.00000 6.00000 54.00000 TRAIN | 21.00000 8.00000 5.00000 24.00000 58.00000 Total | 188.00000 75.00000 54.00000 58.00000 375.00000

Page 148: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

130

Appendix – 5E

Multinomial Logit Model (MNL) with socio-demographic variable --> sample;all$ --> NLOGIT; lhs=CHOICE,CSET,MODE; CHOICES=carD,carP,bus,train; MODEL: U(carD)=A_carD+time*TRAVELTI+cost*TRIPCOST+park*PARKHOUR/ U(carP)=A_carP+time*TRAVELTI+cost*TRIPCOST+age*AGE+sex*SEX/ U(bus)=A_bus +time*TRAVELTI+cost*TRIPCOST/ U(train)= time*TRAVELTI+cost*TRIPCOST; crosstab; $ Normal exit from iterations. Exit status=0. +---------------------------------------------+ | Discrete choice (multinomial logit) model | | Maximum Likelihood Estimates | | Model estimated: Dec 21, 2005 at 11:20:54AM.| | Dependent variable Choice | | Weighting variable None | | Number of observations 375 | | Iterations completed 5 | | Log likelihood function -340.8038 | | Number of parameters 8 | | Info. Criterion: AIC = 1.86029 | | Finite Sample: AIC = 1.86134 | | Info. Criterion: BIC = 1.94406 | | Info. Criterion:HQIC = 1.89355 | | R2=1-LogL/LogL* Log-L fncn R-sqrd RsqAdj | | Constants only. Must be computed directly. | | Use NLOGIT ;...; RHS=ONE $ | | Chi-squared[ 5] = 245.23980 | | Prob [ chi squared > value ] = .00000 | | Response data are given as ind. choice. | | Number of obs.= 375, skipped 0 bad obs. | +---------------------------------------------+ +---------+--------------+----------------+--------+---------+ |Variable | Coefficient | Standard Error |b/St.Er.|P[|Z|>z] | +---------+--------------+----------------+--------+---------+ A_CARD .13207640 .46911504 .282 .7783 TIME -.00728143 .00819382 -.889 .3742 COST -.34996669 .25983471 -1.347 .1780 PARK -.23888875 .31748547 -.752 .4518 A_CARP .10402306 .45207203 .230 .8180 AGE -.03017605 .00952802 -3.167 .0015 SEX -.82329925 .29572224 -2.784 .0054 A_BUS -1.35370226 .25880361 -5.231 .0000 +------------------------------------------------------+ | Cross tabulation of actual vs. predicted choices. | | Row indicator is actual, column is predicted. | | Predicted total is F(k,j,i)=Sum(i=1,...,N) P(k,j,i). | | Column totals may be subject to rounding error. | +------------------------------------------------------+ Matrix Crosstab has 5 rows and 5 columns. CARD CARP BUS TRAIN Total +------------------------------------------------------------- CARD | 114.00000 29.00000 23.00000 22.00000 188.00000 CARP | 30.00000 28.00000 12.00000 6.00000 75.00000 BUS | 24.00000 10.00000 14.00000 6.00000 54.00000 TRAIN | 19.00000 9.00000 5.00000 24.00000 58.00000 Total | 188.00000 75.00000 54.00000 58.00000 375.00000

Page 149: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

131

Appendix – 5F Nested Logit Model (NL) with socio-demographic variable --> sample;all$ --> NLOGIT; lhs=CHOICE,CSET,MODE; CHOICES=carD,carP,bus,train; Tree=mode[Private(carD,carP),Public(bus,train)]; IVSET:(PRIVATE)=[1]; MODEL: U(carD)=A_carD+time*TRAVELTI+cost*TRIPCOST+park*PARKHOUR/ U(carP)=A_carP+time*TRAVELTI+cost*TRIPCOST+age*AGE+sex*SEX/ U(bus)=A_bus +time*TRAVELTI+cost*TRIPCOST/ U(train)= time*TRAVELTI+cost*TRIPCOST; effects:tripcost(carD,carP,bus,train); crosstab; $ Normal exit from iterations. Exit status=0. +---------------------------------------------+ | FIML Nested Multinomial Logit Model | | Maximum Likelihood Estimates | | Model estimated: Dec 21, 2005 at 11:29:25AM.| | Dependent variable CHOICE | | Weighting variable None | | Number of observations 1211 | | Iterations completed 19 | | Log likelihood function -339.8810 | | Number of parameters 9 | | Info. Criterion: AIC = .57619 | | Finite Sample: AIC = .57631 | | Info. Criterion: BIC = .61408 | | Restricted log likelihood -519.8604 | | Chi squared 359.9587 | | Degrees of freedom 9 | | Prob[ChiSqd > value] = .0000000 | | R2=1-LogL/LogL* Log-L fncn R-sqrd RsqAdj | | No coefficients -519.8604 .34621 .33909 | | Constants only. Must be computed directly. | | Use NLOGIT ;...; RHS=ONE $ | | At start values -429.9295 .20945 .20085 | | Response data are given as ind. choice. | +---------------------------------------------+ +---------+--------------+----------------+--------+---------+ |Variable | Coefficient | Standard Error |b/St.Er.|P[|Z|>z] | +---------+--------------+----------------+--------+---------+ Attributes in the Utility Functions (beta) A_CARD .44488124 .51069273 .871 .3837 TIME -.01352926 .01018200 -1.329 .1839 COST -.39840195 .20936806 -1.903 .0571 PARK -.29537137 .31799156 -.929 .3530 A_CARP .35774450 .46919161 .762 .4458 AGE -.03078216 .00950570 -3.238 .0012 SEX -.80485260 .29403670 -2.737 .0062 A_BUS -1.49335336 .31308826 -4.770 .0000 IV parameters, tau(j|i,l),sigma(i|l),phi(l) PRIVATE 1.00000000 ......(Fixed Parameter)....... PUBLIC .78047141 .14405139 5.418 .0000 +------------------------------------------------------+ | Cross tabulation of actual vs. predicted choices. | | Row indicator is actual, column is predicted. | +------------------------------------------------------+ Matrix Crosstab has 5 rows and 5 columns. CARD CARP BUS TRAIN Total +------------------------------------------------------------- CARD | 114.00000 29.00000 24.00000 22.00000 188.00000 CARP | 30.00000 27.00000 12.00000 5.00000 75.00000 BUS | 25.00000 10.00000 14.00000 6.00000 54.00000 TRAIN | 20.00000 9.00000 4.00000 25.00000 58.00000 Total | 188.00000 75.00000 55.00000 57.00000 375.00000

Page 150: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 5: Factors influencing car use

132

Appendix – 5G

Nested Logit Model (NL) with socio-demographic variable and different time parameter --> sample;all$ --> NLOGIT; lhs=CHOICE,CSET,MODE; CHOICES=carD,carP,bus,train; Tree=mode[Private(carD,carP),Public(bus,train)]; IVSET:(PRIVATE)=[1]; MODEL: U(carD)=A_carD+time_C*TRAVELTI+cost*TRIPCOST+park*PARKHOUR/ U(carP)=A_carP+time_C*TRAVELTI+cost*TRIPCOST+age*AGE+sex*SEX/ U(bus)=A_bus +time_B*TRAVELTI+cost*TRIPCOST/ U(train)= time_T*TRAVELTI+cost*TRIPCOST; crosstab; $ Normal exit from iterations. Exit status=0. +---------------------------------------------+ | FIML Nested Multinomial Logit Model | | Maximum Likelihood Estimates | | Model estimated: Dec 21, 2005 at 11:22:10AM.| | Dependent variable CHOICE | | Weighting variable None | | Number of observations 1211 | | Iterations completed 22 | | Log likelihood function -329.0757 | | Number of parameters 11 | | Info. Criterion: AIC = .56164 | | Finite Sample: AIC = .56183 | | Info. Criterion: BIC = .60796 | | Info. Criterion:HQIC = .57908 | | Restricted log likelihood -519.8604 | | Chi squared 381.5694 | | Degrees of freedom 11 | | Prob[ChiSqd > value] = .0000000 | | R2=1-LogL/LogL* Log-L fncn R-sqrd RsqAdj | | No coefficients -519.8604 .36699 .35855 | | Constants only. Must be computed directly. | | Use NLOGIT ;...; RHS=ONE $ | | At start values -331.1821 .00636 -.00689 | | Response data are given as ind. choice. | +---------------------------------------------+ +---------+--------------+----------------+--------+---------+ |Variable | Coefficient | Standard Error |b/St.Er.|P[|Z|>z] | +---------+--------------+----------------+--------+---------+ Attributes in the Utility Functions (beta) A_CARD 1.59149655 .72626926 2.191 .0284 TIME_C -.01047026 .01614767 -.648 .5167 COST -.70597135 .31151130 -2.266 .0234 PARK -.28513365 .31971276 -.892 .3725 A_CARP 1.52515097 .68412985 2.229 .0258 AGE -.03062060 .00963337 -3.179 .0015 SEX -.83448256 .29649016 -2.815 .0049 A_BUS 2.38858171 1.24844344 1.913 .0557 TIME_B -.06622050 .02918450 -2.269 .0233 TIME_T .03442719 .02387648 1.442 .1493 IV parameters, tau(j|i,l),sigma(i|l),phi(l) PRIVATE 1.00000000 ......(Fixed Parameter)....... PUBLIC .62177206 .14298972 4.348 .0000 Matrix Crosstab has 5 rows and 5 columns. CARD CARP BUS TRAIN Total +------------------------------------------------------------- CARD | 116.00000 29.00000 24.00000 19.00000 188.00000 CARP | 30.00000 28.00000 13.00000 4.00000 75.00000 BUS | 25.00000 10.00000 15.00000 5.00000 54.00000 TRAIN | 18.00000 8.00000 3.00000 28.00000 58.00000 Total | 188.00000 75.00000 55.00000 57.00000 375.00000

Page 151: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

133

CHAPTER SIX Stated Preference Survey of car travel to Perth city

Although the nested logit discrete choice study, based on available survey data, of

Chapter 5 provided estimates for trips to Perth city specifically, it left serious gaps. As

indicated at the end of Chapter 5, the travel behaviour of car users can be assessed more

adequately by obtaining their stated responses to various car suppression scenarios. The

previous chapter evaluated travel behaviour in choosing the mode of transport to the

city; however what exactly car users would do if constraints were imposed on them is

the main concern of the present chapter. The focus is to report the details of the Car

Trip Response Survey 2005, which was conducted to obtain responses to the pricing and

control measures outlined in Chapter 4, using an efficient experimental design. The

chapter first discusses the appropriate models to be developed with SP data and then the

experimental design for the survey.

Section 6.2 deals with models to estimate the expected reactions using the survey

responses. Stated Preference (SP) design, the data collection instrument, sampling and

sample frame, and data collection are discussed in Section 6.3. Descriptive information

about the sample is reported in Section 6.4.

6.1 INTRODUCTION

The principal concern of this chapter is to assess travellers’ reactions to policies

designed to reduce pollution in Perth city. The aim of the analysis is to measure the

effectiveness of strategies to achieve this. Chapter 2 has identified several strategies to

reduce air pollution, which are mainly effective in the long run. The most effective

action plans in the short run involve pricing policy. Nijkamp and Shefer (1998) have

considered several measures to address urban transport externalities. They have

categorised the measures into control measures, market-based measures, and land use

& physical planning measures. Pricing policy is considered as a market-based measure.

This chapter suggests a policy combining control measures and market-based measures,

leaving the third category aside because of its long term nature. A convenient name is

air quality control policy (AQCP).

Page 152: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

134

Many studies have used stated preference methods to observe travel behaviour with

respect to the attributes of travel time and trip costs, however not many of these have

been aimed at controlling air pollution. A study by Saelensminde (1999) used the stated

choice method to value air pollution caused by urban traffic with scenarios involving

polluted environment, travel time, and trip cost attributes, but not constraints to control

car use. The present study deals with travel responses to penalties on taking a car to the

city. The fixed charge, variable charge, parking fee and lane restriction were identified

in Chapter 4 as measures which can be applied in the city centre, and expected

responses have been investigated by applying previously estimated elasticities. The

analysis of both short and long term responses projected a significant improvement of

air quality in Perth city with reductions in CO and NOx concentrations. However the

measurement of Perth travellers’ actual reactions to this policy is the main concern of

the study. As the actual reactions of travellers could not be estimated with revealed

preference information, because the policy does not exist, a stated preference (SP)

method has been used to determine the travellers’ specific responses. This part of the

study is designed to measure the effectiveness of the AQCP by assessing the travel

responses in terms of taking a car to the city.

6.2 POLICY REACTION MODEL

To develop a model that can assess the travellers’ reaction to air quality control policy

(AQCP) a binary choice specification is used, both stated preference and socio-

demographic data being used to estimate the model. The dependent variable is choice

of taking a car to the city during a specific time of day under different situations with

regard to various price and other factors. In the mathematical models, binary choice is

the dependent variable and a few dichotomous and polychotomous variables are the

independent variables. The objective is to estimate the likelihood of taking a car to the

city under several levels of predictor variables.

The inapplicability of combining SP-RP data in this study has already been considered

in Chapter 5 and is discussed further in Section 6.2.2. However the SP model is

integrated into the RP model, developed in Chapter 5, by assuming that all charges can

be expressed as equivalent costs and incorporated into the RP model which has trip cost

and parking fee attributes. Thus in Chapter 8, the SP estimates are scaled to the RP

estimates. As noted earlier, the AQCP policy does not exist at present in Perth;

Page 153: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

135

therefore the SP approach is the only source from which travellers’ reactions can be

assessed.

6.2.1 The Model

Design must satisfy the properties of the probabilistic discrete choice model which is

hypothesised to underlie the response data (Louviere et al. 2000). A universal choice

set of {Yes, No} is used for the binary logit model to estimate the reactions of travellers

with respect to the policy measures. The random utility model is the base of this choice

model, which can be expressed as iqiqiq VU ε+= (i is one alternative, either yes or no,

for individual q), where Uiq is utility, Viq is the systematic (observed) utility component

and εiq is the random (unobserved) component. The unobserved component is that

important element in the utility function which is not measurable or understandable by

the analyst. In the binary choice model the probability of an individual choosing Yes is

expressed as:

Pn(yes) = Pr(Uyes > Uno)

= Pr(Vyes + εyes > Vno + εno)

= Pr(Vyes - Vno> εno – εyes) ........................................ (6.1)

Ben-Akiva and Lerman (1985) argued that the specification of a binary choice model

considers only the difference of random components (εno – εyes) instead of each element

separately. They also argued that there is no real difference between shifting the mean

of the random component of one alternative and shifting the systematic component by

the same amount. This implies that as long as one can add a constant to the systematic

component, the means of random components can be defined as equal to any constant

without loss of generality. Therefore the most convenient assumption is that all the

random components have zero means. As a result we can re-write equation (6.1) as:

P(yes)=Pr(Vyes > Vno) ......................................................... (6.2)

The probability of choosing the Yes alternative can be estimated using the binary logit

model. The model is expressed as:

noyes

yes

VV

V

n eeenoyesyesP+

=),|( ............................(6.3)

Page 154: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

136

As we know that the systematic component of the utility function for the No alternative

can be set to zero with no loss of generality, equation (6.3) can be rewritten as:

1),|(

+=

yes

yes

V

V

n eenoyesyesP .................................. (6.4)

Moreover, if we represent the odds of responding Yes relative to No, we see that

no

yes

yes

no

yes

yes

V

V

V

V

V

V

ee

ee

ee

noyesnoPnoyesyesP

=

+

+=

1

1),|(),|(

............................. (6.5)

Again, since Vno = 0, noVe would be one, if we take natural logarithms of both sides of

(6.5), we get,

yese Vp

pLog =⎟⎟⎠

⎞⎜⎜⎝

⎛−1

............................................................ (6.6)

where p represents the probability of choosing the yes alternative.

The systematic (observed) component Vyes is assumed to be homogeneous across the

population in terms of relative importance of those attributes contained in Vyes. This

component is also assumed to be a linear and additive function of the attributes which

determine the utility of the yes alternative. Thus the expression can be:

∑∑ +=q

qqk

kkyes ZXV αβ ....................................... (6.7)

Where βk is a vector associated with k attribute vectors, Xk; and αq is a vector associated

with q individual characteristics, Zq. We have control over the Xk by designing them to

satisfy the properties of interest; but we have less (if not no) control over Zq. These

parameters are estimated for the binary logit model using either SPSS or LIMDEP

software.

Although the logit model estimates the parameters (βk) associated with the attributes

and with the individual characteristics (αq), direct interpretation of them is difficult in a

binary logit model (Hosmer and Lemeshow 2000, Louviere et al. 2000). The marginal

effect of any variable is a meaningful way of presenting the results of a binary logit

Page 155: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

137

model. In such a model marginal effects are the marginal changes in expected

probability. Mathematically, the marginal effect for a logit model can be expressed in

equation (6.8).

)1(**)|(

qqkkqk

qkq ppX

XpE−==

∂βϕ ....................................... (6.8)

From the above expression it is evident that marginal effects for different individuals

are different. There are several ways to summarise these effects. One convenient way

of summarising them is to calculate the marginal effects of all observations in the

sample and report the mean of these effects. The interpretation of the marginal effect of

attribute k is the percentage change in probability of choosing yes (ϕk) in the choice

model for a one unit change in that attribute.

Whereas a marginal effect is the most useful behavioural output in the continuous cases,

for binary independent variables an odds ratio (OR) provides a meaningful

interpretation. An odds ratio estimates how much more likely (less likely) is an

outcome (yes) among those who have a particular attribute (present) than those who do

not have that attribute (absent). The expression of odds ratio is shown in equation (6.9).

⎟⎟⎠

⎞⎜⎜⎝

⎟⎟⎠

⎞⎜⎜⎝

=

absentCabsentCpresentCpresentC

OR

no

yes

no

yes

||

||

................................................................. (6.9)

Where (Cyes|present) is the number of observations for those who choose the yes

alternative given that the attribute is present. A similar meaning is applicable for other

parameters. An odds ratio can also be expressed as:

keOR β= ............................................................................. (6.10)

To ensure efficient estimation of the model a well-designed data collection instrument is

essential.

6.2.2 Inapplicability of combined SP-RP

As discussed in Chapter 5, the SP model should not be used alone to estimate future

impacts because of its hypothetical nature which leads to lack of “validity” and

Page 156: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

138

“stability”. On the other hand, an RP model can provide unbiased estimates as it

reflects travellers’ actual behaviour. Hence, Morikawa (1994) and others (Louviere et

al. 2000, Ben-Akiva and Morikawa 1990, Cherchi and Ortuzar 2002) have

recommended combining SP and RP data to exploit their advantages and overcome their

limitations by adjusting scale factors of the two data sets.

In this study however there was no opportunity to combine RP and SP data. Choices in

the RP study in Chapter 5 and in this SP study could not be the same, so that RP and SP

data could not be combined to estimate a single model. The RP study developed a

mode choice model with the four modes covered in the PARTS survey, whereas the SP

study focused on car driver behaviour. Even in the SP study, revealed and stated

information could not be used to develop a combined model. The choice in the SP

questions was whether respondents would take a car to the city or not, and the RP

section of the same study sought information about those who took a car to the city.

Therefore, there was no mode choice except taking a car to the city.

Using only the SP model to predict travellers’ behaviour may not be efficient and, for

the purpose of prediction, the SP coefficients are used to simulate responses within the

RP model presented in Chapter 5. Details of the simulation process are given in Section

8.3 in Chapter 8.

6.3 DESIGNING THE SP MODEL

6.3.1 Experimental design

An experiment is designed by manipulating attributes and their levels to permit rigorous

testing of certain hypotheses of interest (Louviere et al. 2000). Efficient experimental

designs are widely used in many fields of study, especially in agriculture. The approach

is also common in the transportation field. A design may be classified as fully or

fractionally factorial. A full factorial design is one where each level of each attribute is

combined with each level of all other attributes. Although this design ensures that all

effects of the attributes are captured, it is only practical when a small number of

attributes and levels are of interest. A fractional factorial design can minimise the

number of combinations by selecting a particular subset or sample of complete

factorials so that particular effects of interest can be estimated as efficiently as possible:

the orthogonality of the variables inherent in the full factorial design is still preserved in

the fractional factorial.

Page 157: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

139

Seven policy measures (attributes) were identified in chapter 4 (Table 4.2). Three of

these attributes had 3 levels of value and others had 2 levels. Therefore, if we consider

a complete factorial design the total number of combinations would be 432 (=33 X 24)

profiles, which would be difficult to collect from respondents. As a result a fractional

factorial design is a realistic option in this case. The attributes are shown in Table 6.1.

Table 6.1: Attributes and their levels used for the decision process

Attributes Levels of attributes

Units

Fuel price $1, $1.5, $2 per litre of petrol

Fixed charge $0, $1, $2 per entry into city disregarding time of day

Car size charge $0, $1 for a large car per entry

Entry time charge $0, $4 per entry into the city between 7am and 10am

Parking fee $0, $2, $5 per hour

Parking space Limited, un-limited

Lane restriction for cars yes, no between 7am and 10 am

The attributes are explained as follows.

• Fuel price is unleaded petrol price per litre. It is assumed that diesel and other

fuels move similarly.

• Fixed charge is a charge to be imposed on a car each time it enters Perth city.

The charge is assumed to be collected electronically without any delay (similar

to Melbourne City Link).

• Car size charge is a charge to be imposed electronically on a relatively large car

each time it enters the city. Examples of large cars are Toyota Camry,

Mitsubishi Magna, Mazda6, Kia Optima (2.4L), Hyundai Sonata (2.7L),

Holden Commodore, Ford Falcon, and most 4WDs.

• Entry time charge is a charge to be imposed electronically on each car entering

the city between 7am and 10am.

• Parking fee is an hourly charge imposed for parking a car in the city.

• There are two values of Parking space: ‘difficult’ or ‘easy to find’.

Page 158: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

140

• Lane restriction is defined as the left lane in the main city streets (e.g. St.

Georges Tce, William St., Barrack St.) being closed to cars and used for other

purposes such as cycling or walking.

The attributes were combined to create profiles and respondents were asked “whether

they would take a car to Perth city”. The choice response was binary, yes or no. The

simplest form of design is a main effect only design and an orthogonal design ensures

the attributes are not correlated. The smallest main effects orthogonal design in this

case contains only 16 profiles. These combinations can be in different forms. One

possible combination is shown in Table 6.2; this was adopted in the study. A structured

questionnaire was developed using these 16 scenarios and the respondents were asked

whether they would take a car to the city under each of these scenarios.

Table 6.2: Orthogonal main effects design profile

Profiles Fuel price ($ per litre)

Fixed charge ($ per entry)

Car size charge ($ per

large car)

Entry time charge

($ at morning peak)

Parking fee ($ per

hr)

Parking space

Lane restriction

for cars

1 1.5 0 1 4 2 limited yes

2 1 2 1 4 0 un-limited yes

3 1 1 0 4 2 un-limited yes

4 2 0 1 4 5 un-limited no

5 1 0 0 0 0 limited yes

6 2 0 0 4 0 limited yes

7 1 2 1 4 0 limited no

8 1.5 2 0 0 5 un-limited yes

9 1 1 0 4 5 limited no

10 1 0 1 0 5 limited yes

11 1.5 1 1 0 0 limited no

12 2 1 1 0 0 un-limited yes

13 1.5 0 0 4 0 un-limited no

14 1 0 1 0 2 un-limited no

15 1 0 0 0 0 un-limited no

16 2 2 0 0 2 limited no

6.3.2 Data collection instrument

A survey, called the Car Trip Response Survey 2005 containing three sections, was

developed. The SP section of the questionnaire was designed mainly to present the 16

scenarios and to collect responses under the scenarios. There were two other sections to

Page 159: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

141

collect responses about the traveller’s last trip to Perth city and their household. The

respondent was asked to choose from a binary choice of {yes and no} for a question

“would you take your car to Perth City?” under certain conditions. However to make

the choice set exhaustive, a follow up question was also asked for {yes} answer groups

“whether they would take the car to the city between 7am and 10am”. A complete

questionnaire is provided in Appendix 6A.

While we recognise that peoples’ choice decisions are mainly based on the attributes of

the alternatives, in many cases the decision may be influenced by socio-demographic

factors. Therefore in order to improve model fit the questionnaire asks for each

respondent’s demographic profile. Thus the three sections of the questionnaire are:

Section 1 asking about the respondent’s last car trip to Perth city, Section 2 asking for

choice of alternative under each of the 16 scenarios, and Section 3 concerning

household information.

Information about the last car trip to the city is important to analyse the present status of

the respondent’s travel activity and behaviour. In this section respondents are asked

about the purpose of their last trip to the city. They are also asked about the size of the

car they have used, entry time into the city, cost of fuel and parking. One question

included in this section was about the alternatives available if taking a car to the city

was not convenient on the last occasion. This information is considered as the no

response to each of the choice questions in Section 2.

In Section 2 containing SP questions related to the stated choice of respondents, sixteen

different scenarios were presented. At the beginning of this section detailed

explanations were given for each attribute used. As the scenarios do not exist in reality,

respondents were asked to respond to them as if they were real.

Section 3 contained only three questions regarding the respondent’s household. This

section was limited so that respondents would not find it unduly intrusive. Only the

number of cars, number of driving licences, and the suburb in which the household is

located were asked in this section. At the end of the questionnaire a Perth City map was

included to show the boundary of the city mentioned in the questionnaire.

In designing the questionnaire several issues were considered to make it more

acceptable and simple for the respondents. While the sequence of questions is a very

important issue to be considered in face-to-face interviews, it is less important for mail-

Page 160: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

142

out questionnaires (Dillman 1978, Dillman 2000, Ayidiya & McClendon 1990, and

Schwarz & Hippler 1995, Alwin 1978). Principles of questionnaire design followed

included putting demographic questions at the end, and organising responses vertically

rather than horizontally. Finally, the size of the survey booklet and the font size were

considered in order to make it acceptable to the respondents.

6.3.3 Sampling frame and sample size

“Within individuals, responses to successive profiles may depend in some way on

previous responses. Between individuals, differences in preferences lead to violation of

the Independently and Identically Distributed (IID) assumption because the joint

distribution of utility parameters (β) is not the convolution of independent random

variables” (Louviere et al. 2000). Thus logistic regression or a logit model would be

appropriate if each separate profile were randomly assigned to 100 respondents (i.e.

1600 total respondents) rather than all 16 profiles being evaluated by 100 respondents.

However, in this study a sample size of 2000 was selected and all 16 profiles provided.

Keeping the IID assumption in mind, the model is analysed as if based on panel data or

as a multi-period logit model.

The sample frame defines the universe of respondents from which a finite sample is

drawn to collect data. The objective of the study often influences the sample frame. As

the objective was to assess the reactions of travellers to Perth city to environmental

control policies, the sample frame should have been all members of households who

travel to the city. In fact, for the purpose of this study, the sample frame was taken to be

households in Perth Metropolitan area. Data from the Perth and Regional Travel Survey

(PARTS), an ongoing project, found that people come to Perth city for various purposes

from about 102 suburbs. About 70% use a car (as driver or as a passenger) as their

main mode of transport.

An eventual sample size of 300 to 400 is convenient for the purpose of this study. It

was expected that the response rate would be about 20% in the case of a mail survey.

Keeping this rate in mind, a sample of 2000 households was selected from the 464,901

households listed in the White Page Telephone Book for the Perth Metropolitan area.

According to Louviere et al. (2000) this sample size is within the size range of a sample

for simple random sampling.

Page 161: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

143

6.3.4 Data collection

A complete survey set containing a survey booklet (12 pages), a covering letter (1

page), and a return envelope, was posted to each sample addressee on 20 May 2005.

There was a total of 401 responses, and the usable number was 369 for stated preference

analysis (effective response rate18%). The time pattern of responses is shown in Figure

6.1. The first group of responses was received on the third working day. The number

of responses increased on day 4 and then gradually declined over the next few days.

Almost all responses were received within 4 weeks of mail out. About 11% of the

survey was undelivered due to insufficient or incorrect addresses.

0

10

20

30

40

50

60

70

80

3 4 5 6 7 8 9 10 11 12 13 14 15 16

Late

r

Days after mail-out

# of

resp

onse

s

6.4 SURVEY OUTCOMES

The responses were entered into the computer for analyses and the data used to develop

a binary logit model on the travellers’ reactions to the policy measures. The factual

responses to sections 1 and 3 indicated that people come to the city from at least 169

suburbs. Descriptive statistics are presented in Table 6.3. The total number of

observation for descriptive statistics varies between 369 and 376 depending on non-

responses to some questions. The sample proportions correspond fairly closely to the

Census 2001 data for the Perth workforce (Table 6.3).

Figure 6.1: Response pattern of the survey

Page 162: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

144

Table 6.3: Respondents profile from their last car trip to Perth city Sample

Frequency Sample

Percentage Perth Census

2001 (percentage)

Frequency of trip to Perth city Once a week or more 112 29.8 About once a month 72 19.1 Less frequently 192 51.1 Purpose of last trip to the city Work 106 28.4 Education 6 1.6 Shopping 94 25.2 Personal business or recreation 122 32.7 Others 45 12.1 Work status in Perth City Work full time 58 15.5 17.3 Not work full time 317 84.5 82.7 Size of car used Small car 169 45.9 Large car 199 54.1 Alternative option in case of inconvenience of taking car to the city

Change mode 264 71.4 Change time of day 17 4.6 Cancel activity 40 10.8 Perform activity in another location 49 13.2 Modal split for change mode alternative Public transport (Bus/Train) [Average estimated time of

travel = 36 min] 240 90.9

Walk 3 1.1 Cycle 6 2.3 Taxi 15 5.7 Entry time into the city Before 7 am 16 4.3 Between 7 am and 10 am 151 40.9 Between 10 am and 5 pm 167 45.3 After 5 pm 35 9.5

Table 6.3 shows that the majority (58%) of people driving into Perth city did so for the

purposes of shopping, personal business and recreation, though about 28% go for work.

Half of those who go to the city to work (work purpose group) work full time. Around

half of the respondents used large cars to drive into the city.

About 71% of the respondents are willing to change mode of transport if taking a car to

the city is not convenient, yet 13% say they would change the location of the activity

Page 163: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

145

and 10% say they would cancel the activity. Almost all of those who would change

mode would choose public transport to get to the city. It is also shown in Table 6.5 that

the alternative choices for both work and non-work groups are similar. This

information suggested that people are willing to use an alternative to their car if it is

required.

Table 6.4: Summary of metric variables

Mean St. Dev. Time spent on last trip to Perth city (min) 175.98 147.83

Price of petrol per litre (cent) 101.73 5.32

Parking fee per hour ($) 1.36 1.49

Travel time on public transport (min) 36.07 20.30

Number of cars in a household 2.01 1.00

Number of driver’s licences in a household 2.04 0.81

The average time spent in the city was 6 hours for work purpose and 3 hours for non-

work. The mean metric variables (Table 6.4) were average petrol price per litre of

101.73¢, mean parking fee per hour of $1.36, average travel time on public transport of

36.07 min, average number of cars in a household 2.01, and average number of driver’s

licences 2.04.

Cross-classifications are also of interest. One of the useful groupings is into the work

purpose and non-work purpose (those who go to the city for other than work) groups.

Some of the informative comparisons are presented in Table 6.5.

Table 6.5: Classification by purpose (% of cases)

Alternative choices Work Non-work

Change mode 73 (69.5) 190 (72.2)

Change time of day 7 (6.7) 10 (3.8)

Cancel activity 13 (12.4) 27 (10.3)

Perform activity to another location 12 (11.4) 36 (13.7)

Entry time

Before 7am 13 (12.4) 3 (1.1)

Between 7am and 10 am 62 (59.0) 88 (33.5)

Between 10 am and 5 pm 23 (21.9) 144 (54.8)

After 5 pm 7 (6.7) 28 (10.6)

Page 164: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

146

In Table 6.3 we see a large group of people (45%) enter the city before 10 am, while the

remainder enter after 10 am. These two groups roughly approximate to the work and

non-work groups (see Table 6.5). About 71% (=12.4% + 59.0%) of the work purpose

group enters the city at or before the morning peak whereas about 66% (= 54.8% +

10.6%) of the non-work purpose group go after the morning peak.

The relationship between the purpose of travel to Perth city and selected alternatives if it

were not convenient to take a car to the city is shown in Table 6.6. It is observed that a

good proportion (=35.4% + 31.3%) of people whose purpose was either shopping or

personal business or recreation are willing to change the location of their activities.

Table 6.6: Relation between trip purpose and selected alternative if not taking a car (% of column)

Purpose/Alternatives Change mode

Change time of day

Cancel activity

Perform activity at another location

Work 73 (27.8) 7 (41.2) 13 (32.5) 12 (25.0)

Education 3 (1.1) 0 (0.0) 1 (2.5) 1 (2.1)

Shopping 70 (26.6) 4 (23.5) 3 (7.5) 17 (35.4)

Personal business or recreation 84 (31.9) 2 (11.8) 19 (47.5) 15 (31.3)

Others 33 (12.5) 4 (23.5) 4 (10.0) 3 (6.3)

Table 6.7 shows the relation between size of the car used and purpose of the trip.

People who travel to the city for work purposes tend to prefer (62.3%) a large car,

whereas those who go for a non-work purpose are as likely to use a small as a large car.

Table 6.7: Car size and trip purpose (% of column)

Car size/Purpose Non-work Work

Small car 128 (49.0) 40 (37.7)

Large car 133 (51.0) 66 (62.3)

A summary of SP responses is presented in Table 6.8. Percentages of yes and no

responses corresponding to the 16 scenarios are shown in the table.

Page 165: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

147

Table 6.8: Responses to SP profiles: Proportion of respondents who would take their car to the city

Attributes Response (%)

Scenario

Fuel price ($ per litre)

Fixed charge ($ per entry)

Car size charge ($ per

large car)

Entry time charge ($ at

morning peak)

Parking fee ($ per

hr)

Parking space

Lane restriction

for cars Yes No

1 1.5 0 1 4 2 Limited yes 36.8 63.2

2 1 2 1 4 0 un-limited yes 50.7 49.3

3 1 1 0 4 2 un-limited yes 48.1 51.9

4 2 0 1 4 5 un-limited no 17.3 82.75 1 0 0 0 0 Limited yes 86.2 13.8

6 2 0 0 4 0 Limited yes 51.6 48.4

7 1 2 1 4 0 Limited no 45.7 54.3

8 1.5 2 0 0 5 un-limited yes 21.6 78.4

9 1 1 0 4 5 Limited no 19.7 80.3

10 1 0 1 0 5 Limited yes 23.5 76.5

11 1.5 1 1 0 0 Limited no 67.7 32.3

12 2 1 1 0 0 un-limited yes 60.4 39.6

13 1.5 0 0 4 0 un-limited no 65.8 34.2

14 1 0 1 0 2 un-limited no 68.7 31.3

15 1 0 0 0 0 un-limited no 91.9 8.1

16 2 2 0 0 2 limited no 39.9 60.1

We can observe that almost all (91.9%) of the respondents would take a car to the city

under scenario 15, in which no charges and restrictions are imposed. On the other hand,

most of the respondents (82.7%) would not take a car to the city in scenario 4. In this

profile fuel is expensive and extreme charges are applied although parking space is

unlimited and lane capacity is not restricted. It is evident that people are less willing to

take a car to the city when parking fees are very high (scenarios 8, 9, 10), which

indicates high sensitivity to parking fees.

A similar summary is presented for the work and non-work groups in Table 6.9.

Percentages of responses to the 16 scenarios show some clear distinctions between the

two different groups but in most of the scenarios both groups have similar responses.

Scenarios 4 and 15 have extreme responses for both groups which is also evident in

Table 6.8 for aggregated responses. However, in scenarios 2, 3, 6, and 7, the two

Page 166: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

148

groups have contrasting views. The non-work group is more sensitive to charges and

restrictions.

Table 6.9: Responses to SP profiles for work and non-work groups

Attributes Work purpose

group (%) Non-work purpose

group (%)

Scenario

Fuel price

($ per

litre)

Fixed charge ($ per entry)

Car size

charge ($ per large car)

Entry time

charge ($ at

morning peak)

Parking fee ($ per hr)

Parking space

Lane restriction

for cars

Yes No Yes No 1 1.5 0 1 4 2 Limited yes 48.6 51.4 33.0 67.0

2 1 2 1 4 0 un-limited yes 67.9 32.1 44.4 55.6

3 1 1 0 4 2 un-limited yes 59.0 41.0 44.1 55.9

4 2 0 1 4 5 un-limited no 29.5 70.5 12.6 87.4

5 1 0 0 0 0 Limited yes 92.3 7.7 84.2 15.8

6 2 0 0 4 0 Limited yes 67.3 32.7 46.2 53.8

7 1 2 1 4 0 limited no 59.6 40.4 40.8 59.2

8 1.5 2 0 0 5 un-limited yes 36.5 63.5 16.2 83.8

9 1 1 0 4 5 limited no 34.6 65.4 14.2 85.8

10 1 0 1 0 5 limited yes 39.4 60.6 17.3 82.7

11 1.5 1 1 0 0 limited no 76.0 24.0 65.1 34.9

12 2 1 1 0 0 un-limited yes 72.1 27.9 56.3 43.7

13 1.5 0 0 4 0 un-limited no 76.0 24.0 62.8 37.2

14 1 0 1 0 2 un-limited no 71.2 28.8 68.6 31.4

15 1 0 0 0 0 un-limited no 92.3 7.7 92.0 8.0

16 2 2 0 0 2 limited no 57.7 42.3 33.3 66.7

Details of the model building process using this sample dataset are discussed in the next

chapter.

6.5 SUMMARY

This chapter focused on the design of the stated preference survey and summarising the

responses. Previous chapters identified the policy instruments and estimated the

impacts on air quality control based on an elasticity approach, using reported results in

one case and new econometric model for Perth in the other. However, actual responses

of travellers have been clarified by the stated choice study.

Page 167: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

149

The summaries of the socio-demographic characteristics of the respondents revealed

that most people travel to Perth for the purpose of shopping or personal business or

recreation. Therefore, a useful grouping for further analysis is work purpose and non-

work purpose. In addition to their socio-demographic profiles, these two groups have

contrasting responses to some of the stated preference scenarios. The next chapter

reports on the development of a binary logit model to assess car travel behaviour, using

the survey data.

Page 168: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

150

Appendix 6A

Car Trip Response Survey 2005 Survey Booklet

This is the Car Trip Response Survey 2005 booklet. It is asking you to provide information about your car trips to Perth City and your reactions to various car trip situations. The booklet contains 3 sections and the Perth City map.

• Section 1 is asking about your last car trip to Perth City, • Section 2 is about choosing alternatives under 16 situations, and • Section 3 contains three questions about your household.

Planning and Transport Research Centre (PATREC) Business School

University of Western Australia WA 6009

Page 169: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

151

Completion of the questionnaire is considered evidence of consent to participate in the study.

Section 1

Please tick (√) appropriate box, and write if necessary. 1. How often do you drive to Perth City?

Once a week or more How many times per week? ………… About once a month Less frequently

2. Do you work full-time in Perth City?

Yes No

3. What was the purpose of your last trip to Perth City?

Work Education Shopping Personal Business or recreation Other

4. What type of car did you use on your last trip to Perth City?

Small car Large car (or 4WD) [see box]

Some of the large cars are:

Toyota Camry, Mitsubishi Magna, Mazda6, Kia Optima (2.4L), Hyundai Sonata (2.7L), Holden Commodore, Ford Falcon, most 4WDs and similar

5. How much time did you spend on your last trip to Perth City? ………….. 6. At what time did you enter the City? …………AM …………..PM 7. What was the price per litre when you last bought fuel? …………………. 8. What was the parking fee per hour when you last parked your car in the City?

…………………… (write 0 for free parking)

9. Suppose it was not convenient to take your car on your last trip to Perth City, please suggest an alternative (tick the most likely one).

Use a different transport mode Change time of day Cancel activity Perform activity in another location

Next section is about the choice situation

Which one would you choose (at least one)? Bus or Train estimated trip time…..… Walk estimated trip time…..… Cycle estimated trip time…..… Taxi estimated trip time…..…

Page 170: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

152

Section 2 The air quality in Perth City is deteriorating due to exhaust emissions from cars. Therefore, the following questions are seeking your reactions to various scenarios. You are asked to make only 16 choices. Suppose you have a reason to take a car (private or business) to Perth City. Would you take the car under the following scenarios (even though some may not necessarily be realistic)? Please do your best to complete these questions. They are very important for the study. At the bottom of each scenario you are asked to respond to questions about taking your car to Perth City (see map at page 9). For each decision you are asked to select either YES or NO (NO means an alternative suggested in question 9 of section 1). Features used in these questions are:

• Fuel price: unleaded petrol price per litre [assume diesel and other fuels move similarly].

• Fixed charge: a charge imposed on a car each time it enters Perth City (see map). The charge would be collected electronically without any delay (similar to Melbourne City Link and other cities in the world).

• Car size charge: a charge imposed electronically on a relatively large car (see box) each time it enters the city.

Some of the large cars are:

Toyota Camry, Mitsubishi Magna, Mazda6, Kia Optima (2.4L), Hyundai Sonata (2.7L), Holden Commodore, Ford Falcon, most 4WDs and similar

• Entry time charge: a charge imposed electronically on each car entering the city between 7am and 10am.

• Parking fee: an hourly charge imposed for parking a car in the city. • Parking space: whether finding parking space is difficult or easy in the city. • Lane restriction: the left lane in the main city streets (e.g. St. Georges Tec,

William St., Barrack St.) closed to cars and used for other purposes such as cycling or walking.

Decision Scenario 1 Features values Units

Fuel price $1.50 per litre Fixed charge $0 per entry into Perth City disregarding time Car size charge $1.00 for a large car per entry Entry time charge $4.00 per entry into the city between 7am & 10 am Parking fee $2.00 per hour Parking space limited Lane restriction for cars yes between 7am and 5pm

Under these conditions would you take the car to Perth City? (tick √ ) If YES would you enter Perth City (tick √):

No Yes

between 7am and 10am at another time

Page 171: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

153

Decision Scenario 2 Features values units

Fuel price $1.00 per litre Fixed charge $2.00 per entry into Perth City disregarding time Car size charge $1.00 for a large car per entry Entry time charge $4.00 per entry into the city between 7am & 10 am Parking fee $0 per hour Parking space un-limited Lane restriction for cars yes between 7am and 5pm

Under these conditions would you take the car to Perth City? (tick √ ) If YES would you enter Perth City (tick √):

Decision Scenario 3 Features values units

Fuel price $1.00 per litre Fixed charge $1.00 per entry into Perth City disregarding time Car size charge $0 for a large car per entry Entry time charge $4.00 per entry into the city between 7am & 10 am Parking fee $2.00 per hour Parking space un-limited Lane restriction for cars yes between 7am and 5pm

Under these conditions would you take the car to Perth City? (tick √ ) If YES would you enter Perth City (tick √):

Decision Scenario 4 Features values units

Fuel price $2.00 per litre Fixed charge $0 per entry into Perth City disregarding time Car size charge $1.00 for a large car per entry Entry time charge $4.00 per entry into the city between 7am & 10 am Parking fee $5.00 per hour Parking space un-limited Lane restriction for cars no between 7am and 5pm

Under these conditions would you take the car to Perth City? (tick √ ) If YES would you enter Perth City (tick √):

No Yes

between 7am and 10am at another time

No Yes

between 7am and 10am at another time

No Yes

between 7am and 10am at another time

Page 172: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

154

Decision Scenario 5 Features values units

Fuel price $1.00 per litre Fixed charge $0 per entry into Perth City disregarding time Car size charge $0 for a large car per entry Entry time charge $0 per entry into the city between 7am & 10 am Parking fee $0 per hour Parking space limited Lane restriction for cars yes between 7am and 5pm

Under these conditions would you take the car to Perth City? (tick √ ) If YES would you enter Perth City (tick √):

Decision Scenario 6 Features values units

Fuel price $2.00 per litre Fixed charge $0 per entry into Perth City disregarding time Car size charge $0 for a large car per entry Entry time charge $4.00 per entry into the city between 7am & 10 am Parking fee $0 per hour Parking space limited Lane restriction for cars yes between 7am and 5pm

Under these conditions would you take the car to Perth City? (tick √ ) If YES would you enter Perth City (tick √):

Decision Scenario 7 Features values units

Fuel price $1.00 per litre Fixed charge $2.00 per entry into Perth City disregarding time Car size charge $1.00 for a large car per entry Entry time charge $4.00 per entry into the city between 7am & 10 am Parking fee $0 per hour Parking space limited Lane restriction for cars no between 7am and 5pm

Under these conditions would you take the car to Perth City? (tick √ ) If YES would you enter Perth City (tick √):

No Yes

between 7am and 10am at another time

No Yes

between 7am and 10am at another time

No Yes

between 7am and 10am at another time

Page 173: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

155

Decision Scenario 8 Features values units

Fuel price $1.50 per litre Fixed charge $2.00 per entry into Perth City disregarding time Car size charge $0 for a large car per entry Entry time charge $0 per entry into the city between 7am & 10 am Parking fee $5.00 per hour Parking space un-limited Lane restriction for cars yes between 7am and 5pm

Under these conditions would you take the car to Perth City? (tick √ ) If YES would you enter Perth City (tick √):

Decision Scenario 9 Features values units

Fuel price $1.00 per litre Fixed charge $1.00 per entry into Perth City disregarding time Car size charge $0 for a large car per entry Entry time charge $4.00 per entry into the city between 7am & 10 am Parking fee $5.00 per hour Parking space limited Lane restriction for cars no between 7am and 5pm

Under these conditions would you take the car to Perth City? (tick √ ) If YES would you enter Perth City (tick √):

Decision Scenario 10 Features values units

Fuel price $1.00 per litre Fixed charge $0 per entry into Perth City disregarding time Car size charge $1.00 for a large car per entry Entry time charge $0 per entry into the city between 7am & 10 am Parking fee $5.00 per hour Parking space limited Lane restriction for cars yes between 7am and 5pm

Under these conditions would you take the car to Perth City? (tick √ ) If YES would you enter Perth City (tick √):

No Yes

between 7am and 10am at another time

No Yes

between 7am and 10am at another time

No Yes

between 7am and 10am at another time

Page 174: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

156

Decision Scenario 11 Features values units

Fuel price $1.50 per litre Fixed charge $1.00 per entry into Perth City disregarding time Car size charge $1.00 for a large car per entry Entry time charge $0 per entry into the city between 7am & 10 am Parking fee $0 per hour Parking space limited Lane restriction for cars no between 7am and 5pm

Under these conditions would you take the car to Perth City? (tick √ ) If YES would you enter Perth City (tick √):

Decision Scenario 12 Features values units

Fuel price $2.00 per litre Fixed charge $1.00 per entry into Perth City disregarding time Car size charge $1.00 for a large car per entry Entry time charge $0 per entry into the city between 7am & 10 am Parking fee $0 per hour Parking space un-limited Lane restriction for cars yes between 7am and 5pm

Under these conditions would you take the car to Perth City? (tick √ ) If YES would you enter Perth City (tick √):

Decision Scenario 13 Features values units

Fuel price $1.50 per litre Fixed charge $0 per entry into Perth City disregarding time Car size charge $0 for a large car per entry Entry time charge $4.00 per entry into the city between 7am & 10 am Parking fee $0 per hour Parking space un-limited Lane restriction for cars no between 7am and 5pm

Under these conditions would you take the car to Perth City? (tick √ ) If YES would you enter Perth City (tick √):

No Yes

between 7am and 10am at another time

No Yes

between 7am and 10am at another time

No Yes

between 7am and 10am at another time

Page 175: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

157

Decision Scenario 14 Features values units

Fuel price $1.00 per litre Fixed charge $0 per entry into Perth City disregarding time Car size charge $1.00 for a large car per entry Entry time charge $0 per entry into the city between 7am & 10 am Parking fee $2.00 per hour Parking space un-limited Lane restriction for cars no between 7am and 5pm

Under these conditions would you take the car to Perth City? (tick √ ) If YES would you enter Perth City (tick √):

Decision Scenario 15 Features values units

Fuel price $1.00 per litre Fixed charge $0 per entry into Perth City disregarding time Car size charge $0 for a large car per entry Entry time charge $0 per entry into the city between 7am & 10 am Parking fee $0 per hour Parking space un-limited Lane restriction for cars no between 7am and 5pm

Under these conditions would you take the car to Perth City? (tick √ ) If YES would you enter Perth City (tick √):

Decision Scenario 16 Features values units

Fuel price $2.00 per litre Fixed charge $2.00 per entry into Perth City disregarding time Car size charge $0 for a large car per entry Entry time charge $0 per entry into the city between 7am & 10 am Parking fee $2.00 per hour Parking space limited Lane restriction for cars no between 7am and 5pm

Under these conditions would you take the car to Perth City? (tick √ ) If YES would you enter Perth City (tick √):

Next section is about some household information

No Yes

between 7am and 10am at another time

No Yes

between 7am and 10am at another time

No Yes

between 7am and 10am at another time

Page 176: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

158

Section 3

Please write for the following questions. Your household information This section is the minimum information about your household required to interpret the study results. This information will be strictly confidential. 1: The suburb you live in is ……………………………………………………….

2: The number of cars and other motor vehicles in your household is ……………

3: The number of people in your household who have a driver’s licence is ………..

End of Survey

Page 177: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 6: Stated Preference Survey

159

Perth City [final page of the questionnaire]

Page 178: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

160

CHAPTER SEVEN

Modelling the reactions of car travellers to Perth city

The task of this chapter is to develop models which best exploit the data collected in the

Car Trip Response Survey 2005. As discussed in Chapter 6, the aim is to develop a

model to assess reactions to potential pollution alleviation measures. Chapter 6 also

presented the survey and sample characteristics, which are used for model estimation in

this chapter.

Section 7.2 discusses the odds ratios for policy measure variables calculated from the

raw data. Section 7.3 starts with model estimation using the logit model. Sub-sections

of 7.3 deal with the selection of the appropriate model for this dataset. After the

development of a panel data model and a latent class model, binary logit models were

estimated for work and non-work groups. Section 7.4 reports the marginal effects of

different variables for both work and non-work groups. A separate binary logit model is

developed in Section 7.6 for the question of taking the car to the city between 7am and

10am if the respondent chooses to take a car.

7.1 INTRODUCTION

The strength of the travel modelling reported in Chapter 5 was based on actual travel

data (RP) but its weakness was in the unsatisfactory estimates of coefficients required to

assess the impact of measures to limit car use in Perth city. This chapter reports on the

results of the stated choice (SP) study which indirectly rectifies the weakness of the RP

modelling. However many analysts would regard it as unsound to apply purely SP

coefficients as estimated. In Chapter 8 they are combined with RP coefficients of

Chapter 5 using a procedure which, in effect, scales the magnitude of the SP estimates

back to those of the RP coefficients.

In this chapter, model estimation is based on three alternative structures, each giving

insights into particular aspects of traveller response to restrictive measures. Stated

choice responses to alternative scenarios were collected in the Car Trip Response

Survey 2005 (Chapter 6) which was designed to assess car traveller reaction to the

hypothesised Air Quality Control Policy (AQCP) initially identified in Chapter 4.

Page 179: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

161

The SP choice was binary (yes, no) in response to the question whether the respondent

would take a car to the city under each set of pricing and control measures. Whereas

Multinomial Logit (MNL) or Nested Logit (NL) or any of the family of multinomial

logit models are used for multiple discrete choice analysis, in the binary case a logistic

regression or simply a logit model is appropriate. Before constructing any model, odds

ratios are calculated with the raw data to provide information about the likelihood of

taking a car to the city if the values of attributes are changed.

7.2 REACTIONS TO ATTRIBUTE LEVELS

Five of seven attributes in the SP questions contain metric values and the other two

contain categorical values. All of the charges are in monetary terms; however they were

presented in polychotomous form. Consequently it would be of interest in this analysis

to know the respondents’ reaction to a change in the value from one level to the next.

Odds ratios can explain the responses to different levels of the attributes. Some of the

attributes have 3 levels and some have 2. Table 7.1 summarises the odds ratios for

changing from the 1st level to the 2nd and in a few cases to the 3rd.

Table 7.1: Odds ratios for the attributes

Attributes Levels 2nd level 3rd level

Fuel $1, $1.5, $2 0.78 0.62

Fixed charge $0, $1, $2 0.78 0.52

Car size charge $0, $1 0.76

Entry time charge $0, $4 0.54

Parking fee $0, $2, $5 0.50 0.14

Parking space Un-limited, Limited 0.77

Lane restriction No, Yes 0.82

The interpretation of 0.78 for fuel’s 2nd level is that people are 22% (=1-0.78) less likely

to take a car to the city if fuel price increases from $1.00 to $1.50. Similarly people are

38% less likely to take a car if fuel price increases from $1.00 to $2.00. People are

again 22% less likely to take a car if a fixed charge of $1.00 is imposed on taking the

car into the city, and 48% less likely to take a car if the fixed charge is $2.00.

Sensitivity was greatest with respect to the parking fee. People are 50% less likely to

take a car to the city if the parking fee is changed from free parking to $2.00 per hour,

Page 180: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

162

and 86% less likely if it is $5.00. This set of information provides insight into the

behaviour regarding taking a car to the city under various levels of policy measure.

The study further investigated the differences in choice preferences between the work

and non-work groups. Table 7.2 shows comparative odds ratios for these two groups.

Table 7.2: Odds ratios of attributes for two groups

Work group Non-work group Attributes Levels

2nd level 3rd level 2nd level 3rd level

Fuel $1, $1.5, $2 0.80 0.72 0.78 0.57

Fixed charge $0, $1, $2 0.83 0.68 0.75 0.46

Car size charge $0, $1 0.76 0.76

Entry time charge $0, $4 0.60 0.50

Parking fee $0, $2, $5 0.46 0.17 0.51 0.11

Parking space Un-limited, Limited 0.86 0.73

Lane restriction No, Yes 0.92 0.78

As observed in the marginal effects analysis, on most of the attributes the non-work

group is more sensitive to the change in levels than the work group, particularly the

severe charges. For example, the work group is 28% less likely to take a car to the city

if fuel price increases from $1.00 to $2.00, whereas the non-work group is 43% less

likely to take a car for the same change. The lesser responsiveness of the work group is

expected as they may have less choice to avoid the situation unless they change to

public transport. The non-work group has much more flexibility.

Only in the case of the attribute parking fee is the work group more sensitive than the

non-work group. A person in the work group is 54% less likely to take a car to the city

if parking fee per hour increases from zero (free parking) to $2.00, but a person in the

non-work group is 49% less likely. One could argue that many of the work group may

not pay for parking due to employer provision of parking space so that they would be

more sensitive to paying $2.00.

7.3 MODEL ESTIMATION

A number of models are estimated with this dataset. The SP data alone are used to

estimate a model of choosing to take a car to the city. The socio-demographic data are

also used to improve model fit. The dependent variable is binary in nature (0,1). In the

Page 181: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

163

data set two dependent variables are used to estimate the models; these are from two SP

questions. The first one is whether respondents would take a car to the city under

certain conditions, and the second one is if they take the car to the city whether they will

take it at a specific time of day. These two variables are identified as Q1 and Q2 in the

database.

This study developed the models using a random utility approach, keeping in mind that

other approaches such as latent regression and conditional mean function are available.

Under the random utility approach the respondent derives the utility from the

alternatives. In estimating a logit model the dependent variable may be in grouped or

individual form. In this analysis both Q1 and Q2 are in individual form. A logit model

is developed using LIMDEP software and SPSS software is used to verify the results.

The following discussions consider the development of various logit models with Q1

and Q2 as the dependent variables.

7.3.1 Binary logit model for Q1: Whether to take the car

The SP data alone were used to estimate the model initially. In the SP data the

independent variables are the seven policy measures mentioned before. Five of these

contain metric values and the other two contain categorical values. The list of variables

is given below:

Dependent variable ⇒ Q1 binary response [1=yes, 0=no] for a question “would you take a car to the city?”

Independent variables ⇒ fuelprice (X1) metric value [fuel price per litre in $] ⇒ fixedcharge (X2) metric value [fixed charge in $] ⇒ sizecharge (X3) metric value [car size charge in $] ⇒ entrytime (X4) metric value [charge in $ to enter the

city between 7am and 10am] ⇒ parkfee (X5) metric value [hourly parking fee in $] ⇒ parkspace (X6) categorical value [parking space

limited=1 and unlimited=0] ⇒ lane (X7) categorical value [the left lane in the

main city streets is closed to cars=1 and not closed to cars=0]

The model specification can be expressed as in equation (7.1), which is essentially the

expanded form of equation (6.7). The estimated parameters for this specification using

Page 182: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

164

the logit model (Q1SP) are provided in Table 7.3 (LIMDEP output is shown in

Appendix-7A).

76543arg2arg1 XXXXXXXV laneparkspaceparkfeeentrytimeesizechefixedchfuelpriceyes βββββββα +++++++= ............................................ (7.1)

Table 7.3: Binary logit model for Q1 with only SP data (Q1SP)

Variables Coefficient Std. Err. t-ratio

Constant 2.768 0.140 19.68

βfuelprice -0.651 0.072 -8.93

βfixedcharge -0.369 0.035 -10.43

βsizecharge -0.381 0.060 -6.33

βentrytime -0.193 0.015 -12.79

βparkfee -0.419 0.015 -26.86

βparkspace -0.283 0.059 -4.75

βlane -0.204 0.059 -3.41

Sample size 5728 (358 individuals in 16 scenarios)

Log likelihood function -3357.81

Pseudo R2 0.15

χ2 1224.16

The signs of the coefficients are negative as expected, implying that an increase in the

attribute values (mostly increased charges) would reduce the likelihood of taking a car

to the city. The pseudo R2 value is not high, but the t-ratios show high reliability for the

coefficients of the independent variables. The coefficients of this model are not easy to

interpret directly as the dependent variable is in log form. However, odds ratios and

marginal effects have meaningful interpretations, which will be discussed later. This

model is used to estimate predicted probability of the odds of taking a car to the city. It

correctly predicts 68% of taking or not taking a car to the city (Appendix-7A, last table).

The SP question asks the car traveller’s choice of whether to take a car to the city under

various policies and the choice model is improved by adding socio-demographic

variables. After several investigations, the model was finalised by adding three

variables. These are – i) number of cars per licence, ii) dummy variable for work and

Page 183: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

165

non-work purpose, and iii) dummy variable for a large car. The results are shown in

Table 7.4 (LIMDEP output is shown in Appendix-7B).

Table 7.4: Binary logit model for Q1 with SP and socio-demographic data (Q1SPSD)

Variables Coefficient Std. Err. t-ratio

Constant 1.757 0.173 10.14

βfuelprice -0.675 0.074 -9.08

βfixedcharge -0.384 0.036 -10.62

βsizecharge -0.395 0.061 -6.42

βentrytime -0.202 0.015 -13.02

βparkfee -0.439 0.016 -27.21

βparkspaec -0.297 0.061 -4.87

βlane -0.214 0.061 -3.49

βcar/licence 0.726 0.103 7.03

βworkpurpose 0.711 0.068 10.40

βcarsize 0.377 0.061 6.19

Sample size 5728

Log likelihood function -3241.48

Pseudo R2 0.18

χ2 1456.81

This model (Q1SPSD) shows a somewhat better fit. The signs are as expected with

socio-demographic variables. The number of cars per licence has a positive influence

which is expected. The model also shows that the work purpose group is more likely to

take a car to the city and that people with large cars are more likely to take a car to the

city; this could be due to many reasons. Furthermore, t-ratios and other model fit

statistics of this model are better than the Q1SP model and this model predicts 71% of

respondents’ choice decisions correctly (Appendix-7B, last table).

Despite the satisfactory estimates, it is recognised that this model treats 16 responses by

one individual as if they were made by 16 different individuals. The unique

characteristics of longitudinal data (successive profile responses) with repeatable events

allow one to control for unobserved individual characteristics (Powers & Xie 2000). In

a binary logit model we assume the observations are independent across individuals, but

Page 184: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

166

not necessarily across successive responses of one individual. A previous event induces

a change in individual behaviour. Therefore a panel data model was worth investigating

in this case.

7.3.2 Panel Data Model

Louviere et al. (2000) stated that in the case of successive profiles one response may

depend in some way on previous responses. The basic formulation of the panel data

mode is:

)(

)(

1 qqt

qqt

x

x

iqt eeP αβ

αβ

+

+

+= ........................................................ (7.2)

where alternative i is chosen by individual q at the tth profile. The αq is an individual

heterogeneity term, which can be expressed as:

∑=

′−−−=iT

tqqtqqtq bxxyy

1)()(α ...................................(7.3)

Here ∑ ==

T

t qtq yy1

and ∑ ==

T

t qtq xx1

, where yqt is choice of individual q at time period t

and xqt is the vector for individual q at time period t.

A model is developed by considering panel observations. That means 16 choice

responses per respondent are considered as data for 16 steps or ‘periods’. The results of

a panel data model for binary choice are shown in Table 7.5 (LIMDEP output is shown

in Appendix-7C).

This model (Q1SPP) seems very good as compared with the base model (Q1SP), though

data for 78 individuals (out of 358) were excluded because αq is not estimable. This

group contains all responses that were the same (either all 1s or all 0s). Powers & Xie

(2000) stated that this model is affected by an incidental parameter problem because

there is a unique αq term for each individual. This unique term can be factored out

considering only those binary sequences where a change in responses in all profiles

occurs. The model is developed using eventually 280 (=358-78) individual series of

responses which are different in different ‘periods’. The model provides outstanding

fitness results because of excluding those who gave the same response to all questions.

However, the exclusions mean that the responsiveness represented in the estimated

coefficients (Table 7.5) is exaggerated.

Page 185: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

167

Table 7.5: Panel data model for Q1 with only SP data (Q1SPP)

Variables Coefficient Std. Err. t-ratio

βfuelprice -1.910 0.14 -13.36

βfixedcharge -0.866 0.06 -13.66

βsizecharge -1.074 0.11 -9.64

βentrytime -0.538 0.03 -17.68

βparkfee -1.173 0.04 -28.27

βparkspace -0.546 0.10 -5.23

βlane -0.284 0.10 -2.68

Sample size 4480

Log likelihood function -1332.00

Pseudo R2 0.60

χ2 4051.61

The coefficients of the panel data model (Q1SPP) show higher values for all of the

attributes than the Q1SP model. Each group has its own parameter vector, qi δββ +=′ ,

where δq is a scalar for the individual. This circumstance suggests that an individual

specific model, which captures individual heterogeneity, needs to be developed. The

latent class model (LCM) is such a model that can analyse heterogeneity between

individuals.

7.3.3 Latent Class Model

The underlying theory of the latent class model assumes that individual behaviour

depends on observable attributes and on latent heterogeneity that varies with factors

which are unobservable to the analyst (Greene and Hensher 2002). Segmentation or

classification is established with respect to the intrinsic preferences of the choice.

Respondents may vary their responses to the choice scenario alternatives with respect

to, say, imposed charges. This heterogeneity in charge responsiveness may be related to

socio-demographic characteristics. Even after including basic demographic

characteristics in the model, there are several sources of heterogeneity which often are

unobserved but have an important influence on the choice behaviour (Abramson et al.

1998). Swait (1994) argued that classes or segments can be defined not only on the

Page 186: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

168

basis of attitudinal and socio-demographic data, but also on observed choice behaviour

and attributes of the alternatives.

Swait (1994) suggested that classes can be characterised by variance differences, so that

members of class c have taste parameter, βc and scale λc. If the utility function is

ciqiqcccicciq XU ||| εβλαλ ++= and ciq|ε is conditionally IID extreme value type I within

class, then the choice probability for members of class c is expressed as

∑= )(

)(

| jqcc

iqcc

X

X

ciq eeP βλ

βλ

..................................................... (7.4)

If the probability of being in class c is given by Wqc, then the unconditional probability

of choosing alternative i is

qcC

c ciqiq WPP ∑ ==

1 | ..................................................... (7.5)

The probability, Wqc, is the prior probability attached (by the analyst) to membership of

being in class c. This probability is individual specific if individual characteristics

sharpen the prior probability, but in many applications Wqc is simply a constant (Wc).

There are many ways to parameterise Wqc, a convenient one being the multinomial logit

form (Greene 2005, Kamakura et al. 1994).

∑=

= C

k

Z

Zq

qcqk

c

e

eW

1

αγ

αγ

................................................................ (7.6)

Here α is the scale factor of the membership function, γc is the class parameter, and

vector Zq is an unknown segment specific parameter containing both a psychographic

construct (based on choice behaviour) and the socio-demographic information for

individual q. People with invariant characteristics will be in one class. The class

specific probabilities may be a set of fixed constants if no such characteristics are

observed. Bartholomew (1987) reported that the conditional choice probabilities can be

decomposed into weighted averages of latent choice probabilities. This model does not

impose the Independence-from-Irrelevant Alternatives (IIA) property on the observed

probabilities (Greene 2002). In latent class modelling a response to one profile is

assumed to be independent of a response to other profiles. In other words, a response to

one item tells us nothing about a response to another item once we take class

membership into account (Flaherty 2002). Another assumption of the latent class model

is that the classes are mutually exclusive and exhaustive. In principle the graphical

Page 187: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

169

structure of the latent class model is expressed in Figure 7.1, where a different segment

or class is C, and yes and no responses are i and j respectively.

When the number of classes is unknown, choosing the number of classes becomes a

model selection issue. That means one should select the best model from a set of

models with different numbers of classes. However, we cannot compare the models

using a log-likelihood ratio test because those conditional tests do not follow the χ2

distribution. Therefore the AIC (Akaike Information Criteria) and BIC (Bayes

Information Criteria) are the alternatives to select the model (Moustaki & Papageorgiou

2005, Flaherty 2002, Swait 1994).

The LIMDEP’s default number of latent classes is five, but this is fairly high (Greene

2002). This study investigated the model fit for class numbers 2, 3, and 4 using AIC and

BIC. A series of latent class models were also developed by identifying variables for

the segmentations. None of the variables produced a significantly better model.

Although a model with 4 classes provides a better model in terms of AIC and BIC,

further observations in terms of magnitudes and signs of coefficients, standard errors,

marginal effects, and t-ratios show that models with 3 classes and 4 classes are not

practically useful. Hence the model with 2 classes was selected for this dataset as the

signs of the parameters are correct and their significance levels are fairly high.

The results of the latent class model (Q1LC) developed in this study are shown in Table

7.6 (LIMDEP output is shown in Appendix-7D).

C1 C2 Cn

i j i j i j

Figure 7.1: Latent class model structure

Page 188: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

170

Table 7.6: Latent class model with SP and socio-demographic variables (Q1LC)

Latent class 1 (202 members) Latent class 2 (156 members)

Coefficient Std. err. t-ratio Coefficient Std. err. t-ratio Constant 1.910 0.21 9.10 5.259 1.41 3.73

βfuelprice -1.085 0.12 -8.95 -1.209 0.37 -3.25

βfixedcharge -0.513 0.08 -6.10 -0.725 0.26 -2.78

βsizecharge -0.683 0.13 -5.28 -0.578 0.24 -2.37

βentrytime -0.383 0.03 -14.26 -0.283 0.08 -3.53

βparkfee -0.703 0.03 -22.76 -0.858 0.07 -11.87

βparkspace -0.436 0.12 -3.77 -0.862 0.28 -3.06

βlane -0.095 0.15 -0.62 -0.799 0.36 -2.20

βcar/licence 0.723 0.06 12.27 1.360 0.10 13.35

βworkpurpose 0.738 0.07 10.45 1.627 0.15 10.63

βcarsize 0.279 0.06 4.57 1.240 0.08 15.87

Prior probabilities for class membership

0.56 0.03 20.86 0.44 0.03 16.20

Log likelihood function -2303.40

Pseudo R2 0.29

χ2 1876.18

This model produces a better fit than the basic logistic model (Q1SPSD) in terms of log-

likelihood and pseudo R2. The two classes show their uniqueness through the estimated

parameters. We can see that members in class 2 appear to be more sensitive to the fixed

charge and parking fee. Although the levels of confidence in the class 2 estimates are

not as high, some of them are significantly different from the class 1 estimates. A

significant difference test is done with the following asymptotically normal test statistic:

)()( 21

21

kk

kk

VarVarz

ββ

ββ

+

−= …………………………………… (7.7)

The effects of socio-demographic factors on class 2 are higher than on class 1.

Although these two classes are fairly similar in terms of choice attributes, except

parking fee, they are significantly different (at 95% confidence level) with respect to the

effect of socio-demographic characteristics.

Page 189: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

171

The model can certainly reveal the heterogeneity of the individuals in the dataset. Table

7.7 shows the clear distinction in terms of choice behaviour. The members in class 1

predominantly (72%) choose the no alternative, whereas members in class 2 prefer

(80%) choosing the yes alternative.

Table 7.7: Cross tabulation of class members and choice alternatives (percentage)

Class 1 Class 2

No 72 20

Yes 28 80

The main purpose of the latent class model is to segment the sample on the basis of

homogeneity. Members within a class should have one or more similar characteristics,

which may not be known to the analyst. The study further investigated the socio-

demographic profile (available with this data set) of the class members. Noteworthy

factors in the data set are i) purpose of the trip to the city, ii) car size, iii) whether full-

time worker in the city or not, iv) number of cars in household, v) entry time to the city,

which are used to identify the relationship with the class membership.

The mean values of different variables for class 1 and class 2 are presented in Table 7.8.

Only metric variables are shown in the table; other non-metric (categorical) variables

are presented in cross-tabulation form in a later section. Table 7.6 shows that class 1

and class 2 are only slightly different in terms of the variables presented. Members of

both classes drove to the city on average about 3.5 times a week, spent almost 3 hours,

entered the city around 11 am, paid about $1.00 a litre for petrol and took 36 minutes

travel time on public transport. Only in cars and licences per household did they differ

a little.

Page 190: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

172

Table 7.8: Mean values of selected variables for class 1 and class 2

Variables

Class 1

(202 members)

Class 2

(156 members)

Weekly car trips to Perth city 3.6 3.5

Time spent on last trip to Perth city (min) 174.9 177.4

Time of entry on last trip 11am 11am

Fuel price per litre (¢) 100.3 101.5

Parking fee per hour ($) 1.2 1.5

Travel time on public transport (min) 36.0 36.1

Number of cars in household 1.9 2.2

Number of licences in household 2.0 2.2

Furthermore, cross tabulations between class members and categorical variables also

provide inconclusive evidence of differentiation. To investigate the relationship

between class membership and trip purpose, a cross tabulation is presented in Table 7.9.

In both classes the majority of members are non-work travellers, and also the ratio

between work and non-work are very similar for both classes.

Table 7.9: Latent class and purpose group membership (% of class in brackets)

Non-work purpose Work purpose

Class 1 148 (73.2) 54 (26.7)

Class 2 107 (68.6) 49 (31.4)

Table 7.10 shows a cross-tabulation between class membership and selected variables.

The only variable in which there is an appreciable difference is number of cars. Class 2

has a higher proportion of 2 or more car households but the difference is not great

enough for distinctive segmentation. Frequency of trips to Perth is not a useful variable

to classify the segments. In both classes a little more or less than 30% go to the city

once a week, about 20% go once a month and about 50% go less frequently.

A relationship between class membership and alternative options available in case it is

inconvenient to take a car to the city is also inconclusive. Finally, the tabulation of class

and parking fee per hour shows that a smaller proportion of class 1 member pay more

Page 191: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

173

than $3 per hour than class 2 members, but it is not conclusive in segmenting these two

groups.

Table 7.10: Latent class and selected variables (in brackets, % of class in each case)

Class 1

(202 members)

Class 2

(156 members)

Cars in household

1car in household 77 (38.1) 31 (19.9)

2 or more car in household 125 (61.9) 125 (80.1)

Frequency of trip to Perth

Once a week 54 (26.7) 53 (34.0)

About once a month 39 (19.3) 31 (19.9)

Less frequently 109 (54.0) 72 (46.2)

Alternatives of not taking a car to Perth

Change mode 156 (77.0) 99 (63.2)

Change time of day 4 (2.0) 13 (8.4)

Cancel activity 20 (10.0) 19 (12.3)

Perform activity to another location 22 (11.0) 25 (16.1)

Parking fee per hour

Free parking 73 (36.1) 51 (32.4)

up to $0.75 7 (3.7) 3 (2.1)

$0.75 to $1.5 50 (24.6) 32 (20.4)

$1.5 to $3.0 63 (31.4) 55 (35.2)

More than $3.0 8 (4.2) 15 (9.9)

Abramson et al. (1998) stated that latent class models may indicate the variability of

consumers’ intrinsic preferences but they cannot be used to identify individuals with a

strong positive or negative response. An important finding obtained in this latent class

analysis is that the segmentation is unable to capture a behavioural difference between

members of the two classes. Intrinsic preference is not observed; however there could

be a strong relationship between class membership and other socio-demographic factors

such as income, age, gender, etc. which were not collected in this survey. Although

Page 192: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

174

Table 7.6 shows greater sensitivity to some charges by class 2 members, the differences

were not as great as the differences between the work and non-work groups. The next

section deals with models for these two groups of respondents – work purpose and non-

work purpose.

7.3.4 Binary logit models for work and non-work groups

Table 6.3 in Chapter 6 shows the differences in purposes of trips to the city. Table 7.11

(reproduced from Table 6.9) shows that substantial differences in travel behaviour are

associated with the purpose of trips. Percentages of Yes responses for the two groups

show a clear dissimilarity in the 16 scenarios.

Table 7.11: Responses to SP profiles for work and non-work groups

Attributes

% giving a Yes response

(whether to go to the city by car)

Scenario

Fuel price

($/ litre)

Fixed charge

($/ entry)

Car size charge

($/ large car)

Entry time charge ($ at

morning peak)

Parking fee ($/

hr)

Parking space

Lane restriction

for cars Work

purpose group

Non-work

purpose group

1 1.5 0 1 4 2 Limited yes 48.6 33.0

2 1 2 1 4 0 un-limited yes 67.9 44.4

3 1 1 0 4 2 un-limited yes 59.0 44.1

4 2 0 1 4 5 un-limited no 29.5 12.6

5 1 0 0 0 0 Limited yes 92.3 84.2

6 2 0 0 4 0 Limited yes 67.3 46.2

7 1 2 1 4 0 limited no 59.6 40.8

8 1.5 2 0 0 5 un-limited yes 36.5 16.2

9 1 1 0 4 5 limited no 34.6 14.2

10 1 0 1 0 5 limited yes 39.4 17.3

11 1.5 1 1 0 0 limited no 76.0 65.1

12 2 1 1 0 0 un-limited yes 72.1 56.3

13 1.5 0 0 4 0 un-limited no 76.0 62.8

14 1 0 1 0 2 un-limited no 71.2 68.6

15 1 0 0 0 0 un-limited no 92.3 92.0

16 2 2 0 0 2 limited no 57.7 33.3

Page 193: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

175

It is clear that the two groups differ in their responses. Therefore, separate binary logit

models for the work purpose and non-work purpose groups have been developed. Table

7.12 shows the results of these two models (LIMDEP output is shown in Appendix-7E).

The coefficients for car size charge, entry time charge, parking space, lane restriction,

and car per licence are not significantly different. However the important coefficients

for parking fee are significantly different at the 95% level and also for car size (99%).

The fuel price and fixed charge coefficients differ significantly at the 90% level. To

most of the policy measures the non-work group is more responsive than the work

group, except for car size charge which is also reflected in the actual car size variable.

Travel by the non-work group is generally discretionary so that these people have more

freedom of choice and are therefore more responsive. The coefficients for cars per

licence indicate that people in the work group are more inclined to take a car to the city

than those in the non-work group when this ratio is higher. More detailed discussion of

the group behaviour in terms of policy measures is in a later section.

Table 7.12: Binary logit model for the work (Q1SPSDW) and the non-work (Q1SPSDNW) purpose group with SP and socio-demographic data

Work purpose Non-work purpose

Coefficient Std. Err. t-ratio Coefficient Std. Err. t-ratio Constant 1.344 0.34 3.97 2.146 0.20 10.49

βfuelpricl -0.486 0.14 -3.53 -0.747 0.09 -8.37

βfixedcharge -0.283 0.07 -4.17 -0.420 0.04 -9.69

βsizecharge -0.415 0.12 -3.60 -0.379 0.07 -5.14

βenrtytime -0.168 0.03 -5.83 -0.214 0.02 -11.50

βparkfee -0.392 0.03 -13.88 -0.464 0.02 -23.24

βparkspace -0.196 0.11 -1.72 -0.325 0.07 -4.47

βlane -0.128 0.11 -1.12 -0.231 0.07 -3.16

βcar/licence 0.970 0.20 4.94 0.625 0.12 5.14

βcarsize 0.816 0.12 7.03 0.205 0.07 2.85

Sample size 1648 4080

Log likelihood function -927.15 -2296.25

Pseudo R2 0.15 0.18

χ2 338.08 1039.06

Page 194: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

176

7.4 MARGINAL EFFECTS ANALYSIS

The literature indicates that marginal effects (Equation 6.8) are more meaningful

measures than coefficients from a logit model (Hosmer & Lemeshow 2000). The utility

parameters are not the marginal effects (Louviere et al. 2000). The marginal effects of

Q1SPSDW and Q1SPSDNW models are provided in Table 7.13.

Table 7.13: Marginal effects on choice of car for the work and non-work groups

Attribute Work group Non-work group

Fuel -0.11 -0.19

Fixed charge -0.06 -0.10

Car size charge -0.09 -0.09

Entry time charge -0.04 -0.05

Parking fee -0.09 -0.11

Parking space -0.05 -0.08

Lane restriction -0.03 -0.06

Car per licence 0.22 0.15

Car size 0.19 0.05

As an example (Table 7.13), a $1 increase in fuel price reduces the probability of taking

a car to the city by 11% and 19% for the work and non-work groups respectively. On

the other hand, the probability of taking a car is increased by 22% and 15% for the work

and non-work groups with one unit increase in cars per licence. In almost all of the

policy measures the non-work group is more sensitive than the work purpose group. In

contrast, the work group is more affected than the non-work group by changes in socio-

demographic variables. These results imply that the non-work group has alternatives to

taking a car to the city. They can perform their activities in other locations, change

entry time of day, use public transport or even cancel the activity in the city, whereas

the work group may not have that flexibility. A relationship between marginal effect

and odds ratio can be expressed in equation (7.8).

)()1( 2 ORLn

ORORME+

= ....................................... (7.8)

Where, ME is marginal effect and OR is odds ratio.

Page 195: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

177

Previously calculated odds ratios (shown in Table 7.2) can be converted to marginal

effects using equation (7.8). Odds ratios of the fuel attribute for both the work and non-

work groups are transformed into marginal effects for demonstration. The odds ratios

of fuel price increases from $1 to $2 were 0.72 and 0.57 for the work and non-work

groups. Marginal effects derived from these two figures (by equation 7.8) are -0.08 and

-0.13 compared to -0.11 and -0.19 in Table 7.13. The marginal effects estimated for the

same attribute from the binary logit model are not expected to be the same but close

enough to indicate consistency.

Elasticity estimation is not appropriate in those cases where the attribute’s base value is

zero (non-existence in reality). The marginal effect takes the place of elasticity. In all

cases, marginal effects play a significant role in assessing the reaction of travellers to

the air quality control policy. However, fuel price elasticity can be calculated as actual

fuel prices paid by the respondents were collected in the survey.

7.5 FUEL PRICE ELASTICITY

The models developed in this chapter produce different coefficients for the same

attribute. They are not directly comparable because of the different model

specifications. These produce different scale factors in the model so that comparison of

coefficients is inappropriate. However comparison can be made after computing

elasticities for the same attribute. Choice elasticities are probably the most useful

output from discrete choice analysis for policy purpose (Taplin et al. 1999). As

discussed in Chapter 5 (5.2.2), elasticity should be calculated using the probability

weighted mean of the sample. The direct elasticity of individual q choosing alternative i

is expressed in equation (7.9).

)1( iqfqffq PXE −= β .......................................... (7.9)

Where, βf is fuel price coefficient

Xfq is fuel price paid by q individual

Piq is probability of choosing i alternative by q individual

A weighted mean is calculated from the individual elasticities:

Page 196: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

178

∑∑ −

=

qiq

qiqiqfqf

f P

PPXE

)1(β ................................................. (7.10)

Various models developed in this chapter provide different fuel price coefficients.

Table 7.14 shows fuel price elasticities from various models with their fuel coefficients.

The estimated elasticities from various models are within a range between -0.095 and

-0.192. These figures are well within previously estimated fuel price elasticities in

various places including the capital cities in Australia and can be compared with the

short term variable elasticities presented in Table 4.9 in Chapter 4. Such results suggest

that the estimates made by SP alone may not be far out of line with actually observed

behaviour in this case.

Table 7.14: Fuel price elasticities for various models

Model

Fuel

coefficient

Fuel price

elasticity

Model with only SP data -0.650 -0.133

Model with SP and socio-demographic data -0.675 -0.132

Panel Data model -1.910 -0.192

Latent Class Model

Class 1 -1.084 -0.145

Class 2 -1.208 -0.126

Work group model with SP and socio-demographic data -0.486 -0.095

Non-work group model with SP and socio-demographic data -0.747 -0.145

Range of estimates from previous studies (excluding extremes) -0.09 to -0.3

Fuel elasticities for the work and non-work groups show that working people are less

responsive in fuel price change than non-work group. For 1% increase in fuel price,

work group is 0.095% less likely to take a car to the city, whereas non-work group is

0.145% less likely to take a car to the city.

7.6 BINARY LOGIT MODEL FOR Q2

Conditional question (Q2) followed from the answer to the first question. The question

was whether respondents would take a car to the city between 7am and 10am if they

intended to take a car. The response to this question is also in binary form. People are

expected to be influenced by two SP attributes; these are entry time charge and lane

Page 197: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

179

restriction. Binary logit model (Q2SP) gave the results in Table 7.15 (LIMDEP output

is shown in Appendix-7F).

Table 7.15: Binary logit model for Q2 with only SP data (Q2SP)

Variables Coefficient Std. Err. t-ratio

Constant 0.519 0.0603 8.6

βentrytime_charge -0.360 0.0205 -17.6

βlane_restriction -0.125 0.0795 -1.6

The Q2SP model (Table 7.15) shows the expected inverse relationships. A slightly

improved model is obtained by adding a dummy variable to represent the work purpose

group. The results are shown in Table 7.16 (LIMDEP output is shown in Appendix-

7G).

Table 7.16: Binary logit model for Q2 with SP and work purpose (Q2SPSD)

Variables Coefficient Std. Err. t-ratio

Constant 0.525 0.0604 8.69

βentrytime_charge -0.361 0.0205 -17.66

βlane_restriction -0.123 0.0796 -1.55

βworkpurp 0.0012 0.0006 1.99

Sample size 2925

Log likelihood function -1844.53

Pseudo R2 0.09

χ2 354.88

This model (Q2SPSD) shows that the dummy variable has a very small influence, but it

is logical to include it. All these models are good estimators for assessing people’s

reaction to policy measures.

7.7 SUMMARY AND CONCLUSION

The focus of this chapter is on assessing the reactions of travellers to potential policies

to control air pollution in Perth city. The collected data (Chapter 6) were used to

develop a policy reaction model. A range of models are developed which enable one to

gain the insights into travel behaviour with respect to charging policies.

Page 198: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

180

A latent class model was developed to segment groups with respect to their response

behaviour and unknown influences. It gave fairly good model fit results but failed to

identify the class members’ characteristics in terms of socio-demographic attributes.

Instead, grouping on the basis of work and non-work gave models which are most

appropriate for further analysis. Marginal effects analysis shows that the non-work

group is more sensitive to charges than the work group. The same conclusion is

obtained from fuel price elasticities. Fuel price elasticities from the various models

indicated a general consistency and reliability as they are well within the range of

previously estimated elasticities in Australia.

The results reported in this chapter indicate that the behaviour of travellers to Perth city

in response to the hypothesised Air Quality Control Policy (AQCP) would make that

policy effective in improving air quality in Perth. The majority of travellers are in the

non-work purpose group and they are more sensitive to the proposed policy. Chapter 8

combines the results of this chapter and the econometric model developed in Chapter 5.

Then the combined model’s simulation results are used to determine the impact on air

quality in Perth through the air pollution model developed in Chapter 3.

Page 199: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

181

Appendix – 7A Binary Logit Model (Q1SP) for taking a car choice with only stated preference data --> LOGIT ;lhs=q1 ;rhs=one,fuel,fixed,size,entytime,parkfee,parkspac,lane $ Normal exit from iterations. Exit status=0. +---------------------------------------------+ | Multinomial Logit Model | | Maximum Likelihood Estimates | | Model estimated: Jun 23, 2006 at 07:01:01PM.| | Dependent variable Q1 | | Weighting variable None | | Number of observations 5728 | | Iterations completed 5 | | Log likelihood function -3357.816 | | Number of parameters 8 | | Info. Criterion: AIC = 1.17522 | | Finite Sample: AIC = 1.17522 | | Info. Criterion: BIC = 1.18451 | | Info. Criterion:HQIC = 1.17845 | | Restricted log likelihood -3969.895 | | Chi squared 1224.157 | | Degrees of freedom 7 | | Prob[ChiSqd > value] = .0000000 | | Hosmer-Lemeshow chi-squared = 13.22925 | | P-value= .10420 with deg.fr. = 8 | +---------------------------------------------+ +---------+--------------+----------------+--------+---------+-------+ |Variable | Coefficient | Standard Error |b/St.Er.|P[|Z|>z] | Mean of X| +---------+--------------+----------------+--------+---------+-------+ Characteristics in numerator of Prob[Y = 1] Constant 2.76857682 .14064857 19.684 .0000 FUEL -.65056735 .07282527 -8.933 .0000 1.37500000 FIXED -.36911304 .03539368 -10.429 .0000 .75000000 SIZE -.38160019 .06026813 -6.332 .0000 .50000000 ENTYTIME -.19385051 .01516029 -12.787 .0000 2.00000000 PARKFEE -.41974356 .01562544 -26.863 .0000 1.75000000 PARKSPAC -.28348123 .05972747 -4.746 .0000 .50000000 LANE -.20478177 .05999277 -3.413 .0006 .50000000 +--------------------------------------------------------------------+ | Information Statistics for Discrete Choice Model. | | M=Model MC=Constants Only M0=No Model | | Criterion F (log L) -3357.81617 -3969.89452 -3970.34705 | | LR Statistic vs. MC 1224.15672 .00000 .00000 | | Degrees of Freedom 7.00000 .00000 .00000 | | Prob. Value for LR .00000 .00000 .00000 | | Entropy for probs. 3357.81618 3969.89452 3970.34705 | | Normalized Entropy .84572 .99989 1.00000 | | Entropy Ratio Stat. 1225.06174 .90505 .00000 | | Bayes Info Criterion 6776.20418 8000.36090 8001.26595 | | BIC - BIC(no model) 1225.06177 .90505 .00000 | | Pseudo R-squared .15418 .00000 .00000 | | Pct. Correct Prec. 68.29609 .00000 50.00000 | | Means: y=0 y=1 y=2 y=3 y=4 y=5 y=6 y>=7 | | Outcome .4937 .5063 .0000 .0000 .0000 .0000 .0000 .0000 | | Pred.Pr .4937 .5063 .0000 .0000 .0000 .0000 .0000 .0000 | | Notes: Entropy computed as Sum(i)Sum(j)Pfit(i,j)*logPfit(i,j). | | Normalized entropy is computed against M0. | | Entropy ratio statistic is computed against M0. | | BIC = 2*criterion - log(N)*degrees of freedom. | | If the model has only constants or if it has no constants, | | the statistics reported here are not useable. | +--------------------------------------------------------------------+

Page 200: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

182

+----------------------------------------+ | Fit Measures for Binomial Choice Model | | Logit model for variable Q1 | +----------------------------------------+ | Proportions P0= .493715 P1= .506285 | | N = 5728 N0= 2828 N1= 2900 | | LogL = -3357.81617 LogL0 = -3969.8945 | | Estrella = 1-(L/L0)^(-2L0/n) = .20714 | +----------------------------------------+ | Efron | McFadden | Ben./Lerman | | .19635 | .15418 | .59859 | | Cramer | Veall/Zim. | Rsqrd_ML | | .19706 | .30311 | .19242 | +----------------------------------------+ | Information Akaike I.C. Schwarz I.C. | | Criteria 6715.63512 6715.64442 | +----------------------------------------+ +---------------------------------------------------------+ |Predictions for Binary Choice Model. Predicted value is | |1 when probability is greater than .500000, 0 otherwise.| |Note, column or row total percentages may not sum to | |100% because of rounding. Percentages are of full sample.| +------+---------------------------------+----------------+ |Actual| Predicted Value | | |Value | 0 1 | Total Actual | +------+----------------+----------------+----------------+ | 0 | 1938 ( 33.8%)| 890 ( 15.5%)| 2828 ( 49.4%)| | 1 | 926 ( 16.2%)| 1974 ( 34.5%)| 2900 ( 50.6%)| +------+----------------+----------------+----------------+ |Total | 2864 ( 50.0%)| 2864 ( 50.0%)| 5728 (100.0%)| +------+----------------+----------------+----------------+

Page 201: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

183

Appendix – 7B Binary Logit Model (Q1SPSD) for taking a car choice with stated preference and socio-demographic data --> LOGIT ;lhs=q1 ;rhs=one,fuel,fixed,size,entytime,parkfee,parkspac,lane,car_lic,workpurp,... $ Normal exit from iterations. Exit status=0. +---------------------------------------------+ | Multinomial Logit Model | | Maximum Likelihood Estimates | | Model estimated: Mar 08, 2006 at 02:57:09PM.| | Dependent variable Q1 | | Weighting variable None | | Number of observations 5728 | | Iterations completed 6 | | Log likelihood function -3241.487 | | Number of parameters 11 | | Info. Criterion: AIC = 1.13564 | | Finite Sample: AIC = 1.13565 | | Info. Criterion: BIC = 1.14842 | | Info. Criterion:HQIC = 1.14009 | | Restricted log likelihood -3969.895 | | Chi squared 1456.815 | | Degrees of freedom 10 | | Prob[ChiSqd > value] = .0000000 | | Hosmer-Lemeshow chi-squared = 12.42675 | | P-value= .13315 with deg.fr. = 8 | +---------------------------------------------+ +---------+-------------+---------------+--------+---------+---------+ |Variable |Coefficient |Standard Error |b/St.Er.|P[|Z|>z] |Mean of X| +---------+-------------+---------------+--------+---------+---------+ Characteristics in numerator of Prob[Y = 1] Constant 1.75784227 .17336152 10.140 .0000 FUEL -.67569311 .07438906 -9.083 .0000 1.37500000 FIXED -.38475502 .03622523 -10.621 .0000 .75000000 SIZE -.39532308 .06154631 -6.423 .0000 .50000000 ENTYTIME -.20190196 .01550156 -13.025 .0000 2.00000000 PARKFEE -.43906150 .01613191 -27.217 .0000 1.75000000 PARKSPAC -.29737900 .06100459 -4.875 .0000 .50000000 LANE -.21422146 .06124263 -3.498 .0005 .50000000 CAR_LIC .72596102 .10319751 7.035 .0000 .99394786 WORKPURP .71067523 .06828687 10.407 .0000 .28770950 CARSIZE .37702134 .06090448 6.190 .0000 .53910615 +--------------------------------------------------------------------+ | Information Statistics for Discrete Choice Model. | | M=Model MC=Constants Only M0=No Model | | Criterion F (log L) -3241.48697 -3969.89452 -3970.34705 | | LR Statistic vs. MC 1456.81510 .00000 .00000 | | Degrees of Freedom 10.00000 .00000 .00000 | | Prob. Value for LR .00000 .00000 .00000 | | Entropy for probs. 3241.48697 3969.89452 3970.34705 | | Normalized Entropy .81642 .99989 1.00000 | | Entropy Ratio Stat. 1457.72016 .90505 .00000 | | Bayes Info Criterion 6569.50516 8026.32027 8027.22532 | | BIC - BIC(no model) 1457.72016 .90505 .00000 | | Pseudo R-squared .18348 .00000 .00000 | | Pct. Correct Prec. 70.51327 .00000 50.00000 | +--------------------------------------------------------------------+

Page 202: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

184

+----------------------------------------+ | Fit Measures for Binomial Choice Model | | Logit model for variable Q1 | +----------------------------------------+ | Proportions P0= .493715 P1= .506285 | | N = 5728 N0= 2828 N1= 2900 | | LogL = -3241.48697 LogL0 = -3969.8945 | | Estrella = 1-(L/L0)^(-2L0/n) = .24496 | +----------------------------------------+ | Efron | McFadden | Ben./Lerman | | .23007 | .18348 | .61552 | | Cramer | Veall/Zim. | Rsqrd_ML | | .23091 | .34904 | .22457 | +----------------------------------------+ | Information Akaike I.C. Schwarz I.C. | | Criteria 6482.97779 6482.99056 | +----------------------------------------+ +---------------------------------------------------------+ |Predictions for Binary Choice Model. Predicted value is | |1 when probability is greater than .500000, 0 otherwise.| |Note, column or row total percentages may not sum to | |100% because of rounding. Percentages are of full sample.| +------+---------------------------------+----------------+ |Actual| Predicted Value | | |Value | 0 1 | Total Actual | +------+----------------+----------------+----------------+ | 0 | 2038 ( 35.6%)| 790 ( 13.8%)| 2828 ( 49.4%)| | 1 | 899 ( 15.7%)| 2001 ( 34.9%)| 2900 ( 50.6%)| +------+----------------+----------------+----------------+ |Total | 2937 ( 51.3%)| 2791 ( 48.7%)| 5728 (100.0%)| +------+----------------+----------------+----------------+

Page 203: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

185

Appendix – 7C Panel Data Model (Q1SPP) for taking a car choice with stated preference data --> LOGIT ;lhs=q1 ;rhs=fuel,fixed,size,entytime,parkfee,parkspac,lane ;fixed effects ;pds=16 $ +---------------------------------------------+ | Logit Regression Start Values for Q1 | | Maximum Likelihood Estimates | | Model estimated: Mar 08, 2006 at 03:01:12PM.| | Dependent variable Q1 | | Weighting variable None | | Number of observations 5728 | | Iterations completed 10 | | Log likelihood function -3357.816 | | Number of parameters 8 | | Info. Criterion: AIC = 1.17522 | | Finite Sample: AIC = 1.17522 | | Info. Criterion: BIC = 1.18451 | | Info. Criterion:HQIC = 1.17845 | +---------------------------------------------+ +---------+-------------+---------------+--------+---------+---------+ |Variable |Coefficient |Standard Error |b/St.Er.|P[|Z|>z] |Mean of X| +---------+-------------+---------------+--------+---------+---------+ FUEL -.65056735 .07282527 -8.933 .0000 1.37500000 FIXED -.36911304 .03539368 -10.429 .0000 .75000000 SIZE -.38160019 .06026813 -6.332 .0000 .50000000 ENTYTIME -.19385051 .01516029 -12.787 .0000 2.00000000 PARKFEE -.41974356 .01562544 -26.863 .0000 1.75000000 PARKSPAC -.28348123 .05972747 -4.746 .0000 .50000000 LANE -.20478177 .05999277 -3.413 .0006 .50000000 Constant 2.76857682 .14064857 19.684 .0000 Nonlinear Estimation of Model Parameters Method=Newton; Maximum iterations=100 Convergence criteria: max|dB| .1000D-08, dF/F= .1000D-08, g<H>g= .1000D-08 Normal exit from iterations. Exit status=0. +---------------------------------------------+ | FIXED EFFECTS Logit Model | | Maximum Likelihood Estimates | | Model estimated: Mar 08, 2006 at 03:01:12PM.| | Dependent variable Q1 | | Weighting variable None | | Number of observations 5728 | | Iterations completed 7 | | Log likelihood function -1332.007 | | Number of parameters 287 | | Info. Criterion: AIC = .56530 | | Finite Sample: AIC = .57060 | | Info. Criterion: BIC = .89865 | | Info. Criterion:HQIC = .68133 | | Sample is 16 pds and 358 individuals. | | Bypassed 78 groups with inestimable a(i). | | LOGIT (Logistic) probability model | +---------------------------------------------+

Page 204: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

186

+---------+-------------+---------------+--------+---------+---------+ |Variable |Coefficient |Standard Error |b/St.Er.|P[|Z|>z] |Mean of X| +---------+-------------+---------------+--------+---------+---------+ Index function for probability FUEL -1.91066564 .14299214 -13.362 .0000 1.37500000 FIXED -.86617453 .06337699 -13.667 .0000 .75000000 SIZE -1.07445475 .11141946 -9.643 .0000 .50000000 ENTYTIME -.53848426 .03045430 -17.682 .0000 2.00000000 PARKFEE -1.17337155 .04150714 -28.269 .0000 1.75000000 PARKSPAC -.54630255 .10437333 -5.234 .0000 .50000000 LANE -.28409980 .10586189 -2.684 .0073 .50000000

Page 205: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

187

Appendix – 7D Latent Class Model (Q1LC) for taking a car choice with stated preference and trip purpose data --> LOGIT ;lhs=q1 ;rhs=one,fuel,fixed,size,entytime,parkfee,parkspac,lane,car_lic,carsize,w ;lcm ;pts=2 ;pds=16 ;fixed effects ;maxit=100 $ Normal exit from iterations. Exit status=0. +---------------------------------------------+ | Latent Class / Panel Logit Model | | Maximum Likelihood Estimates | | Model estimated: Mar 08, 2006 at 03:05:35PM.| | Dependent variable Q1 | | Weighting variable None | | Number of observations 5728 | | Iterations completed 32 | | Log likelihood function -2303.398 | | Number of parameters 23 | | Info. Criterion: AIC = .81229 | | Finite Sample: AIC = .81232 | | Info. Criterion: BIC = .83900 | | Info. Criterion:HQIC = .82159 | | Restricted log likelihood -3241.487 | | Chi squared 1876.178 | | Degrees of freedom 13 | | Prob[ChiSqd > value] = .0000000 | | Sample is 16 pds and 358 individuals. | | LOGIT (Logistic) probability model | | Model fit with 2 latent classes. | +---------------------------------------------+ +---------+-------------+---------------+--------+---------+---------+ |Variable |Coefficient |Standard Error |b/St.Er.|P[|Z|>z] |Mean of X| +---------+-------------+---------------+--------+---------+---------+ Model parameters for latent class 1 Constant 1.91004029 .27563026 6.930 .0000 FUEL -1.08477898 .13007597 -8.340 .0000 1.37500000 FIXED -.51273046 .06884833 -7.447 .0000 .75000000 SIZE -.68254282 .11081446 -6.159 .0000 .50000000 ENTYTIME -.38262654 .02761827 -13.854 .0000 2.00000000 PARKFEE -.70333564 .03991163 -17.622 .0000 1.75000000 PARKSPAC -.43555722 .10081468 -4.320 .0000 .50000000 LANE -.09454354 .10245665 -.923 .3561 .50000000 CAR_LIC .73837922 .16879266 4.374 .0000 .99394786 CARSIZE .27920491 .11171373 2.499 .0124 .53910615 WORKPURP .72255552 .11984544 6.029 .0000 .28770950 Model parameters for latent class 2 Constant 5.25887999 .72557824 7.248 .0000 FUEL -1.20853678 .18818633 -6.422 .0000 1.37500000 FIXED -.72486256 .12170503 -5.956 .0000 .75000000 SIZE -.57752262 .16948748 -3.407 .0007 .50000000 ENTYTIME -.28310161 .04422862 -6.401 .0000 2.00000000 PARKFEE -.85821722 .04637056 -18.508 .0000 1.75000000 PARKSPAC -.86171831 .16151156 -5.335 .0000 .50000000

Page 206: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

188

LANE -.79901715 .16778956 -4.762 .0000 .50000000 CAR_LIC 1.62696536 .36217601 4.492 .0000 .99394786 CARSIZE 1.23971301 .17293758 7.169 .0000 .53910615 WORKPURP 1.35995199 .17810130 7.636 .0000 .28770950 Estimated prior probabilities for class membership Class1Pr .56293364 .02797334 20.124 .0000 Class2Pr .43706636 .02797334 15.624 .0000

Page 207: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

189

Appendix – 7E Binary logit model for work group (Q1SPSDW) for taking a car choice with stated preference and socio-demographic data --> Sample; all$ --> reject; workpurp=0$ --> LOGIT ;lhs=q1 ;rhs=one,fuel,fixed,size,entytime,parkfee,parkspac,lane,car_lic,carsize $ Normal exit from iterations. Exit status=0. +---------------------------------------------+ | Multinomial Logit Model | | Maximum Likelihood Estimates | | Model estimated: Mar 08, 2006 at 03:17:35PM.| | Dependent variable Q1 | | Weighting variable None | | Number of observations 1648 | | Iterations completed 5 | | Log likelihood function -927.1581 | | Number of parameters 10 | | Info. Criterion: AIC = 1.13733 | | Finite Sample: AIC = 1.13741 | | Info. Criterion: BIC = 1.17014 | | Info. Criterion:HQIC = 1.14949 | | Restricted log likelihood -1096.200 | | Chi squared 338.0841 | | Degrees of freedom 9 | | Prob[ChiSqd > value] = .0000000 | | Hosmer-Lemeshow chi-squared = 2.82893 | | P-value= .94463 with deg.fr. = 8 | +---------------------------------------------+ +---------+------------+----------------+--------+---------+---------+ |Variable | Coefficient| Standard Error |b/St.Er.|P[|Z|>z] |Mean of X| +---------+------------+----------------+--------+---------+---------+ Characteristics in numerator of Prob[Y = 1] Constant 1.34424397 .33861238 3.970 .0001 FUEL -.48623486 .13788252 -3.526 .0004 1.37500000 FIXED -.28306081 .06791429 -4.168 .0000 .75000000 SIZE -.41525462 .11520832 -3.604 .0003 .50000000 ENTYTIME -.16805016 .02881126 -5.833 .0000 2.00000000 PARKFEE -.39217934 .02825296 -13.881 .0000 1.75000000 PARKSPAC -.19641088 .11392486 -1.724 .0847 .50000000 LANE -.12806099 .11406099 -1.123 .2615 .50000000 CAR_LIC .97044876 .19654957 4.937 .0000 1.04822006 CARSIZE .81566892 .11608697 7.026 .0000 .62135922 +----------------------------------------+ | Fit Measures for Binomial Choice Model | | Logit model for variable Q1 | +----------------------------------------+ | Proportions P0= .382282 P1= .617718 | | N = 1648 N0= 630 N1= 1018 | | LogL = -927.15813 LogL0 = -1096.2002 | | Estrella = 1-(L/L0)^(-2L0/n) = .19973 | +----------------------------------------+ | Efron | McFadden | Ben./Lerman | | .19022 | .15421 | .61810 | | Cramer | Veall/Zim. | Rsqrd_ML | | .19138 | .29818 | .18547 |

Page 208: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

190

+----------------------------------------+ | Information Akaike I.C. Schwarz I.C. | | Criteria 1854.32841 1854.36122 | +----------------------------------------+ +---------------------------------------------------------+ |Predictions for Binary Choice Model. Predicted value is | |1 when probability is greater than .500000, 0 otherwise.| |Note, column or row total percentages may not sum to | |100% because of rounding. Percentages are of full sample.| +------+---------------------------------+----------------+ |Actual| Predicted Value | | |Value | 0 1 | Total Actual | +------+----------------+----------------+----------------+ | 0 | 336 ( 20.4%)| 294 ( 17.8%)| 630 ( 38.2%)| | 1 | 190 ( 11.5%)| 828 ( 50.2%)| 1018 ( 61.8%)| +------+----------------+----------------+----------------+ |Total | 526 ( 31.9%)| 1122 ( 68.1%)| 1648 (100.0%)| +------+----------------+----------------+----------------+ Binary logit model for non-work group (Q1SPSDNW) for taking a car choice with stated preference and socio-demographic data --> Sample; all$ --> reject; workpurp=1$ --> LOGIT ;lhs=q1 ;rhs=one,fuel,fixed,size,entytime,parkfee,parkspac,lane,car_lic,carsize $ Normal exit from iterations. Exit status=0. +---------------------------------------------+ | Multinomial Logit Model | | Maximum Likelihood Estimates | | Model estimated: Mar 08, 2006 at 03:18:10PM.| | Dependent variable Q1 | | Weighting variable None | | Number of observations 4080 | | Iterations completed 6 | | Log likelihood function -2296.257 | | Number of parameters 10 | | Info. Criterion: AIC = 1.13052 | | Finite Sample: AIC = 1.13053 | | Info. Criterion: BIC = 1.14599 | | Info. Criterion:HQIC = 1.13600 | | Restricted log likelihood -2815.791 | | Chi squared 1039.068 | | Degrees of freedom 9 | | Prob[ChiSqd > value] = .0000000 | | Hosmer-Lemeshow chi-squared = 10.08246 | | P-value= .25929 with deg.fr. = 8 | +---------------------------------------------+

Page 209: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

191

+---------+------------+----------------+--------+---------+---------+ |Variable | Coefficient| Standard Error |b/St.Er.|P[|Z|>z] |Mean of X| +---------+------------+----------------+--------+---------+---------+ Characteristics in numerator of Prob[Y = 1] Constant 2.14639555 .20460977 10.490 .0000 FUEL -.74703835 .08924403 -8.371 .0000 1.37500000 FIXED -.42019967 .04337119 -9.688 .0000 .75000000 SIZE -.37892330 .07373364 -5.139 .0000 .50000000 ENTYTIME -.21352361 .01856628 -11.501 .0000 2.00000000 PARKFEE -.46370507 .01995643 -23.236 .0000 1.75000000 PARKSPAC -.32462060 .07265634 -4.468 .0000 .50000000 LANE -.23134087 .07314671 -3.163 .0016 .50000000 CAR_LIC .62502113 .12169996 5.136 .0000 .97202614 CARSIZE .20460623 .07182124 2.849 .0044 .50588235 +----------------------------------------+ | Fit Measures for Binomial Choice Model | | Logit model for variable Q1 | +----------------------------------------+ | Proportions P0= .538725 P1= .461275 | | N = 4080 N0= 2198 N1= 1882 | | LogL = -2296.25679 LogL0 = -2815.7910 | | Estrella = 1-(L/L0)^(-2L0/n) = .24537 | +----------------------------------------+ | Efron | McFadden | Ben./Lerman | | .23152 | .18451 | .61843 | | Cramer | Veall/Zim. | Rsqrd_ML | | .23226 | .35004 | .22483 | +----------------------------------------+ | Information Akaike I.C. Schwarz I.C. | | Criteria 4592.51848 4592.53396 | +----------------------------------------+ +---------------------------------------------------------+ |Predictions for Binary Choice Model. Predicted value is | |1 when probability is greater than .500000, 0 otherwise.| |Note, column or row total percentages may not sum to | |100% because of rounding. Percentages are of full sample.| +------+---------------------------------+----------------+ |Actual| Predicted Value | | |Value | 0 1 | Total Actual | +------+----------------+----------------+----------------+ | 0 | 1697 ( 41.6%)| 501 ( 12.3%)| 2198 ( 53.9%)| | 1 | 699 ( 17.1%)| 1183 ( 29.0%)| 1882 ( 46.1%)| +------+----------------+----------------+----------------+ |Total | 2396 ( 58.7%)| 1684 ( 41.3%)| 4080 (100.0%)| +------+----------------+----------------+----------------+

Page 210: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

192

Appendix – 7F Binary Logit Model (Q2SP) for time of taking a car choice with only stated preference data --> LOGIT ;lhs=q2 ;rhs=one,entytime,lane $ Normal exit from iterations. Exit status=0. +---------------------------------------------+ | Multinomial Logit Model | | Maximum Likelihood Estimates | | Model estimated: Mar 08, 2006 at 03:30:58PM.| | Dependent variable Q2 | | Weighting variable None | | Number of observations 2925 | | Iterations completed 5 | | Log likelihood function -1846.748 | | Number of parameters 3 | | Info. Criterion: AIC = 1.26479 | | Finite Sample: AIC = 1.26479 | | Info. Criterion: BIC = 1.27092 | | Info. Criterion:HQIC = 1.26699 | | Restricted log likelihood -2021.975 | | Chi squared 350.4530 | | Degrees of freedom 2 | | Prob[ChiSqd > value] = .0000000 | | Hosmer-Lemeshow chi-squared = 3.80715 | | P-value= .87409 with deg.fr. = 8 | +---------------------------------------------+ +---------+-------------+----------------+--------+--------+--------+ |Variable | Coefficient Standard Error |b/St.Er.|P[|Z|>z]|Mean of X| +---------+-------------+----------------+--------+--------+--------+ Characteristics in numerator of Prob[Y = 1] Constant .51857335 .06025884 8.606 .0000 ENTYTIME -.36021110 .02046063 -17.605 .0000 1.68752137 LANE -.12531839 .07953254 -1.576 .1151 .47589744 +--------------------------------------------------------------------+ | Information Statistics for Discrete Choice Model. | | M=Model MC=Constants Only M0=No Model | | Criterion F (log L) -1846.74848 -2021.97499 -2027.45550 | | LR Statistic vs. MC 350.45302 .00000 .00000 | | Degrees of Freedom 2.00000 .00000 .00000 | | Prob. Value for LR .00000 .00000 .00000 | | Entropy for probs. 1846.74848 2021.97499 2027.45550 | | Normalized Entropy .91087 .99730 1.00000 | | Entropy Ratio Stat. 361.41406 10.96104 .00000 | | Bayes Info Criterion 3709.45905 4059.91207 4070.87311 | | BIC - BIC(no model) 361.41406 10.96104 .00000 | | Pseudo R-squared .08666 .00000 .00000 | | Pct. Correct Prec. 66.29060 .00000 50.00000 | | Means: y=0 y=1 y=2 y=3 y=4 y=5 y=6 y>=7 | | Outcome .5306 .4694 .0000 .0000 .0000 .0000 .0000 .0000 | | Pred.Pr .5306 .4694 .0000 .0000 .0000 .0000 .0000 .0000 | | Notes: Entropy computed as Sum(i)Sum(j)Pfit(i,j)*logPfit(i,j). | | Normalized entropy is computed against M0. | | Entropy ratio statistic is computed against M0. | | BIC = 2*criterion - log(N)*degrees of freedom. | | If the model has only constants or if it has no constants, | | the statistics reported here are not useable. | +--------------------------------------------------------------------+

Page 211: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

193

+----------------------------------------+ | Fit Measures for Binomial Choice Model | | Logit model for variable Q2 | +----------------------------------------+ | Proportions P0= .530598 P1= .469402 | | N = 2925 N0= 1552 N1= 1373 | | LogL = -1846.74848 LogL0 = -2021.9750 | | Estrella = 1-(L/L0)^(-2L0/n) = .11779 | +----------------------------------------+ | Efron | McFadden | Ben./Lerman | | .11647 | .08666 | .55989 | | Cramer | Veall/Zim. | Rsqrd_ML | | .11647 | .18438 | .11291 | +----------------------------------------+ | Information Akaike I.C. Schwarz I.C. | | Criteria 3693.49900 3693.50514 | +----------------------------------------+ +---------------------------------------------------------+ |Predictions for Binary Choice Model. Predicted value is | |1 when probability is greater than .500000, 0 otherwise.| |Note, column or row total percentages may not sum to | |100% because of rounding. Percentages are of full sample.| +------+---------------------------------+----------------+ |Actual| Predicted Value | | |Value | 0 1 | Total Actual | +------+----------------+----------------+----------------+ | 0 | 900 ( 30.8%)| 652 ( 22.3%)| 1552 ( 53.1%)| | 1 | 334 ( 11.4%)| 1039 ( 35.5%)| 1373 ( 46.9%)| +------+----------------+----------------+----------------+ |Total | 1234 ( 42.2%)| 1691 ( 57.8%)| 2925 (100.0%)| +------+----------------+----------------+----------------+

Page 212: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

194

Appendix – 7G Binary Logit Model (Q2SPSD) for time of taking a car choice with stated preference and socio-demographic data --> LOGIT ;lhs=q2 ;rhs=one,entytime,lane,workpurp$ Normal exit from iterations. Exit status=0. +---------------------------------------------+ | Multinomial Logit Model | | Maximum Likelihood Estimates | | Model estimated: Mar 08, 2006 at 03:32:29PM.| | Dependent variable Q2 | | Weighting variable None | | Number of observations 2925 | | Iterations completed 5 | | Log likelihood function -1844.531 | | Number of parameters 4 | | Info. Criterion: AIC = 1.26395 | | Finite Sample: AIC = 1.26396 | | Info. Criterion: BIC = 1.27213 | | Info. Criterion:HQIC = 1.26690 | | Restricted log likelihood -2021.975 | | Chi squared 354.8885 | | Degrees of freedom 3 | | Prob[ChiSqd > value] = .0000000 | | Hosmer-Lemeshow chi-squared = 102.76792 | | P-value= .00000 with deg.fr. = 8 | +---------------------------------------------+ +--------+-------------+----------------+--------+--------+----------+ |Variable| Coefficient | Standard Error |b/St.Er.|P[|Z|>z]| Mean of X| +--------+-------------+----------------+--------+--------+----------+ Characteristics in numerator of Prob[Y = 1] Constant .52544251 .06040349 8.699 .0000 ENTYTIME -.36180443 .02048678 -17.660 .0000 1.68752137 LANE -.12329480 .07959895 -1.549 .1214 .47589744 WORKPURP .00122706 .00061706 1.989 .0468 -4.43384615 +--------------------------------------------------------------------+ | Information Statistics for Discrete Choice Model. | | M=Model MC=Constants Only M0=No Model | | Criterion F (log L) -1844.53074 -2021.97499 -2027.45550 | | LR Statistic vs. MC 354.88849 .00000 .00000 | | Degrees of Freedom 3.00000 .00000 .00000 | | Prob. Value for LR .00000 .00000 .00000 | | Entropy for probs. 1844.53074 2021.97499 2027.45550 | | Normalized Entropy .90978 .99730 1.00000 | | Entropy Ratio Stat. 365.84952 10.96104 .00000 | | Bayes Info Criterion 3713.00463 4067.89312 4078.85416 | | BIC - BIC(no model) 365.84952 10.96104 .00000 | | Pseudo R-squared .08776 .00000 .00000 | | Pct. Correct Prec. 66.39316 .00000 50.00000 | | Means: y=0 y=1 y=2 y=3 y=4 y=5 y=6 y>=7 | | Outcome .5306 .4694 .0000 .0000 .0000 .0000 .0000 .0000 | | Pred.Pr .5306 .4694 .0000 .0000 .0000 .0000 .0000 .0000 | | Notes: Entropy computed as Sum(i)Sum(j)Pfit(i,j)*logPfit(i,j). | | Normalized entropy is computed against M0. | | Entropy ratio statistic is computed against M0. | | BIC = 2*criterion - log(N)*degrees of freedom. | | If the model has only constants or if it has no constants, | | the statistics reported here are not useable. | +--------------------------------------------------------------------+

Page 213: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 7: Modelling Policy Reaction

195

+----------------------------------------+ | Fit Measures for Binomial Choice Model | | Logit model for variable Q2 | +----------------------------------------+ | Proportions P0= .530598 P1= .469402 | | N = 2925 N0= 1552 N1= 1373 | | LogL = -1844.53074 LogL0 = -2021.9750 | | Estrella = 1-(L/L0)^(-2L0/n) = .11925 | +----------------------------------------+ | Efron | McFadden | Ben./Lerman | | .11778 | .08776 | .56057 | | Cramer | Veall/Zim. | Rsqrd_ML | | .11784 | .18646 | .11426 | +----------------------------------------+ | Information Akaike I.C. Schwarz I.C. | | Criteria 3689.06422 3689.07240 | +----------------------------------------+ +---------------------------------------------------------+ |Predictions for Binary Choice Model. Predicted value is | |1 when probability is greater than .500000, 0 otherwise.| |Note, column or row total percentages may not sum to | |100% because of rounding. Percentages are of full sample.| +------+---------------------------------+----------------+ |Actual| Predicted Value | | |Value | 0 1 | Total Actual | +------+----------------+----------------+----------------+ | 0 | 907 ( 31.0%)| 645 ( 22.1%)| 1552 ( 53.1%)| | 1 | 338 ( 11.6%)| 1035 ( 35.4%)| 1373 ( 46.9%)| +------+----------------+----------------+----------------+ |Total | 1245 ( 42.6%)| 1680 ( 57.4%)| 2925 (100.0%)| +------+----------------+----------------+----------------+

Page 214: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 8: Modelling the impact of policies on air quality

196

CHAPTER EIGHT Modelling the impact of policy measures on air quality in Perth

In order to calculate probable responses to policy measures, this chapter first takes the

stated preference (SP) results of Chapter 7 and integrates them into the revealed

preference (RP) results reported in Chapter 5. The objective is to assess the future air

pollution situation in Perth city if the measures are implemented and then to determine

which are the most effective relative to the costs imposed on motorists. To predict

travel behaviour, RP and SP information are combined in an efficient measure; in this

case the SP model parameters are used in the RP model to simulate results for the

purpose of prediction. Because the reliability of the required RP coefficient is low

however, the final analysis is done with SP coefficients alone.

Section 8.2 presents a set of potential policies and changes for Perth city and Section 8.3

reports their calculated impacts on car travel behaviour using the SP-RP approach and

then the estimated impact of potential policies using only SP models. In Section 8.4 the

alternative policies are assessed for impact on air quality and in Section 8.5 the effects

of the improved air quality on health outcomes. A ratio between health benefit and

financial sacrifice by motorists, a benefit-sacrifice ratio, is calculated in Section 8.6 to

compare the results of the various measures.

8.1 INTRODUCTION

The objective of testing car driver reactions to potential measures, as reported in

Chapters 6 and 7, is to assess the extent to which it is possible to influence travellers’

behaviour so that air quality can be improved. Richardson and Bae (1998) identified the

following options faced by travellers under a pricing policy:

i) Reduce trip making (for example, trip chaining, telecommuting, or simply

travelling less)

ii) No change in travel behaviour (after paying increased charges);

iii) Increasing travel (because trip times are reduced);

Page 215: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 8: Modelling the impact of policies on air quality

197

iv) Unchanged travel behaviour combined with attempts to reduce total

automobile costs (for example, keeping vehicle longer, or replacing it with a

cheaper or fuel-efficient vehicle);

v) Changing travel behaviour with the same level of trip making (for example,

changing trip time, route, or mode, such as carpools, trains); and

vi) Changes in location (for example, residence, workplace, shopping

destination).

These options will vary with geographical and political situations and the economic

condition of the travellers. The present study assumes that the first and fifth options of

Richardson and Bae (1998), reduced trip making or changed mode or time, will be the

primary response by travellers to Perth city. A number of other issues, especially

equity, are related to pricing policy; the cost of the policy would not be viewed equally

by the economically fortunate and unfortunate members of the community. This study

is not designed to investigate equity impacts but the topic is considered briefly in

Chapter 9.

The Air Quality Control Policy (AQCP) is a combination of hypothetical pricing and

control policy measures that could be applied to car travellers to Perth city. Chapters 6

and 7 dealt with expected car traveller reactions to the policy. It was found that car trips

to the city would be reduced under various policy scenarios with the rate of reduction

varying among the scenarios. This chapter discusses the implications of the policy in

terms of air quality improvement and the consequent improvement in the health of the

population of Perth.

8.2 MODEL IMPLEMENTATION

8.2.1 Estimation of policy impacts

The sixteen policy scenarios developed for the stated preference survey in Chapter 6,

were based on seven policy attributes. An orthogonal design ensured non-correlation

among the attributes. This part of the study assesses the impact of each policy in

isolation from the others. Allowing for potential fuel price increases, which are not

under local policy influence, four cases are investigated with $1 being the charge

imposed in each case. The charges and a potential fuel price increase are summarised in

Table 8.1.

Page 216: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 8: Modelling the impact of policies on air quality

198

Table 8.1: Potential policies and changes assessed

Fuel price

($/litre)

(1)

Fixed charge

($/entry)

(2)

Car size charge ($/large

car) (3)

Entry time charge ($ at

morning peak)

(4)

Parking fee

($/hr)

(5)

Policy or Change

(6) 1.02 1.0 0 0 1.3 Fixed charge policy

1.02 0 1.0 0 1.3 Car size charge policy

1.02 0 0 1.0 1.3 Entry time restriction policy

1.02 0 0 0 2.3 Parking charge policy

2.00 0 0 0 1.3 Fuel price increase

Other than the fuel price increase, which is tested at a potential $2.00 per litre, the four

policy measures are based on average fuel price ($1.02) paid at the time of the SP

survey. In order to estimate the separate effects of fixed charge, car size charge, and

entry time charge a $1 charge is set in each case; and parking charge is increased by $1

per hour from the average fee.

Two control measures, parking space and lane restriction, are not included here in the

policy formulation because of their non-monetary nature even though they were

included in the SP scenarios. They are considered in Chapter 9.

The average parking fee considered by the respondents in the survey for the paid-

parking space in the city was $1.30 per hour. Changes in fuel price and parking fee

would be changes to existing attributes. The expected impact of the various policy

measures on air quality is assessed in Sections 8.3, 8.4 and 8.5.

8.3 MODELLED IMPACT ON TRAFFIC

The plan to use a survey of actual behaviour and conduct a separate stated choice

survey, in which there could be no RP component, has been fully implemented and

reported in Chapter 5 (RP) and Chapters 6 and 7 (SP). Thus the two results can be

integrated but the parking coefficients of the RP model (-0.295) and the two SP models

(-0.392 for work and -0.464 for non-work) do not differ significantly. A simple z-test

found that they are not significantly different at the 95% confidence level. Furthermore

the estimated SP model coefficients are very reliable, especially the parking coefficients

which have t-values of 13.88 and 23.24, whereas the reliability of the parking

Page 217: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 8: Modelling the impact of policies on air quality

199

coefficient in the RP model is very low (t = 0.93), meaning that there are about 18

chances in 100 that the parking fee coefficient in the RP model is zero.

Although researchers have shown limited confidence in the direct application and

external validity of SP models, a number of studies (Kocur and Louviere 1983,

Horowitz and Louviere 1993, Swait et al. 1994, Carson et al. 1994) have reported SP

model estimates which alone have been consistent with reality. Those results do not

have any direct implications for this study but there is elasticity evidence that it may be

reasonable to draw inferences from the SP results of the Car Response Survey 2005 to

actual outcomes. As noted in Chapter 7, the elasticities of demand for fuel by people

driving to the city, derived from the SP models, of -0.09 to -0.19 are within the range of

previously estimated urban short term fuel price elasticities. Also the parking charge

elasticities of -0.24 and -0.44 for the work and non-work groups from the SP models are

within the range of previous estimates. Parking elasticities are considered further in

Section 8.4.

For these reasons, the SP coefficients alone are used in a second assessment of

responses (Section 8.3.2). However the combined procedure is applied first.

8.3.1 Combined SP and RP

The SP models in Chapter 7 are used to estimate the proportion of travellers intending

to take a car to the city under the four policy measures and the hypothetical fuel price

change (Table 8.1). Prediction is done by simulating the RP model developed in

Chapter 5 using coefficients from both the RP and SP models. The mechanics of doing

this is to apply all charge related (fixed charge, size charge, and entry time charge)

coefficients to form equivalent parking fee components and then add them to form one

equivalent fee, which is used in simulating the RP model (NL model).

In the RP models the hypothetical charges were not included as choice attributes

because there were no such charges in Perth city. Only travel time, trip cost and parking

fee attributes were used in the utility function for the car driver mode. In reality car

users consider fuel price as the cost of running a car and parking fee and other

suggested charges as the cost of taking a car to the city. Travellers can reasonably be

assumed to treat all suggested charges as being equivalent to the cost of taking a car to

the city, which is equivalent to a parking fee in the present situation. The effect of

increasing fuel price and parking fee on choosing the car mode can be estimated by

Page 218: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 8: Modelling the impact of policies on air quality

200

running a simulation of the RP model. This means using the original model’s

parameters to predict the number of travellers after a change in PARKHOUR (hourly

parking fee) attributes on only the car driver (Car A) mode. Detailed specifications of

the models are provided in Appendices 8A, 8B and 8C. The NL model developed in

Chapter 5 is shown in appendix 8A, repeated from Appendix 5F, and is used as the base

model for simulation. Appendix 8B presents the simulated model with fuel price

change and Appendix 8C with a parking charge change.

Average fuel price paid by the traveller during the survey was $1.02 per litre and

average hourly parking fee was $1.30 per hour in the city. These two values are

considered as the base level of the analysis. The study estimates the proportion of car

users under the increased fuel price and parking fee from the base case. The fixed

charge, car size charge, entry time charge, and parking fee are converted to equivalent

parking fee by equation (8.1). The coefficients of various charges shown in Table 7.12

of Chapter 7 are shown again in Table 8.2 for work and non-work group models.

Table 8.2: Model coefficients of various charges (extracted from Table 7.12)

Work group Non-work group

Charges Q1SPSDW model

coefficient

Coefficients proportional

to parking fee

Q1SPSDNW model

coefficient

Coefficients proportional

to parking fee Fixed charge -0.283 0.72 -0.420 0.91

Car size charge -0.415 1.06 -0.379 0.82

Entry time charge -0.168 0.43 -0.214 0.46

Parking fee -0.392 1.00 -0.464 1.00

( ) ( )parkingentrytimecarsizefixedparkingentrytimecarsizefixede CCCCCCCCP +++++++= 46.082.091.072.043.006.172.028.0

........................................ (8.1)

Where, Pe = Equivalent hourly parking fee ($ per hour)

Cfixed = Fixed charge per entry into Perth city ($ per entry)

Ccarsize = Car size charge for a large car per entry ($ per large car per entry)

Centrytime = Entry time charge per entry into the city between 7am and 10am ($ per entry at morning peak)

Cparking = Hourly parking fee ($ per hour)

Page 219: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 8: Modelling the impact of policies on air quality

201

The proportional coefficients are used to formulate equation (8.1) which converts all

charges to parking fee equivalent. The results of this calculation are shown in Table

8.3.

Table 8.3: Conversion of policy measures to parking fee equivalents

Fixed charge

($/entry)

Car size charge ($/large

car)

Entry time charge ($

at morning peak)

Parking fee

($/hr)

Converted to parking fee

equivalent $ by equation (8.1)

Policy Measure

1.0 0 0 1.3 2.16 Fixed charge

0 1.0 0 1.3 2.19 Car size charge

0 0 1.0 1.3 1.75 Entry time charge

0 0 0 2.3 2.30 Parking charge

Equation (8.1) says that 28% of respondents were from the work group and 72% from

the non-work group. The coefficients of the charges are the model (Q1SPSDW and

Q1SPSDNW) parameter equivalents of a parking fee. The study would have estimated

the policy impact for the work and non-work groups which were clearly identified in

Chapter 7. However, as mentioned in Chapter 5, separate NL models for the two groups

could not be developed from the RP data because of small sample size. Parking

equivalent costs are estimated for four policy measures using equation (8.1). These

costs and suggested increase in fuel prices for four measures are used to simulate the RP

model and estimate the number of respondents choosing the car mode to go to Perth

city. These increases in trip cost and parking fee would reduce the utility of car use.

The utility function for car driver is expressed in equation (8.2) (similar to the function

shown in Section 5.5.1).

UcarD = αcarD + αtt +βtc * (Ctripcost+ Cfuel_price_increase) + βp* Pe ................ (8.2)

Where, αcarD is alternative specific constant for car driver mode. αtt is travel time coefficient considered as a constant because in the simulation

this value does not change. βtc is the coefficient of trip cost (-0.398). βp is the coefficient of parking fee (-0.295).

Ctripcost is trip cost ($ per trip). Cfuel_price_increase is increase in fuel price ($ per litre). Pe is equivalent hourly parking fee ($ per hour).

Page 220: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 8: Modelling the impact of policies on air quality

202

Equation (8.2) is used in RP model simulations where positive values of Cfuel_price_increase

and Pe make the utility with car mode less attractive; as a result a portion of car trips

would switch to other modes (shown in Appendices 8B and 8C). As discussed in

Chapter 5, the estimated trip cost coefficient (βtc) is fairly reliable with t-value 1.9,

whereas the parking fee coefficient (βp) is relatively poor with t-value 0.93. The effect

of applying Equation (8.2) is to scale the responses to charges, thus making a moderate

reduction in the projected impacts of the hypothetical policy measures.

Table 8.4 shows the impact of suggested policies on car use for the entire sample

(LIMDEP output for fuel price and parking fee increase are provided in Appendices 8B

and 8C as examples). Although LIMDEP output (Appendix-8B) shows that the

percentage change in car driving would be -7.5%, the percentage taking a car to the city

shown in Table 8.4 is calculated as the percentage of car drivers relative to the base

case, e.g. 176 as a percentage of 204 (column 10, last row).

Table 8.4: Estimated impacts on car use to the city Policy or Change

(1)

Fuel price ($)

(2)

Fixed charge

($)

(3)

Car size charge

($)

(4)

Entry time

charge ($) (5)

Parking fee ($)

(6)

Parking fee

equivalent ($/hr)

(7)

Equivalent increase in

parking fee ($/hr)

(8)

Increase in fuel price ($) (9)

Taking a car (% of

base case) (10)

Base case 1.02 0 0 0 1.3 1.30 0 0 100

Fixed charge 1.02 1.0 0 0 1.3 2.16 0.86 0.00 91

Car size charge 1.02 0 1.0 0 1.3 2.19 0.89 0.00 91

Entry time charge 1.02 0 0 1.0 1.3 1.75 0.45 0.00 96

Parking charge 1.02 0 0 0 2.3 2.30 1.00 0.00 90

Fuel price 2.00 0 0 0 1.3 1.30 0 0.98 86

Table 8.4 indicates that one dollar imposed as a fixed charge would have the same

impact on motorists as a car size charge, with a 9% reduction (10th column) in car use

from the base case. A $1 per hour increase in parking fee to $2.30 per hour would

reduce car use by 10% and a $2.00 per litre fuel price would reduce car use to the city

by 14%. A later section discusses the sensitivity of car use to the parking fee. Fuel

price changes are omitted from subsequent discussion because these are not controlled

by any State agency.

Page 221: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 8: Modelling the impact of policies on air quality

203

8.3.2 Application of the SP model alone

As discussed, the estimated parking coefficient in the RP model (-0.295) and the two SP

models (-0.392 for work and -0.464 for non-work) are not significantly different and the

confidence levels of the SP model parking coefficients are high. Therefore this section

uses the unmodified SP results as an alternative way to assess the impact of policy

measures on air pollution.

An advantage of this procedure is that both the work and non-work SP models can be

applied to assess the separate responses. An equivalent parking fee is not required to

apply the SP models on their own; individual charges of $1 each for the fixed charge,

car size charge and entry time charge, and $1 increase in parking fee can be directly

used to estimate the impacts. Table 8.5 shows the impact of car use under four

alternative policies using the SP models. The results are estimated by using the work

and non-work SP model coefficients with $1 increase in each variable at a time. The

impact for the entire sample is estimated by adding the work and non-work groups

proportionately.

Table 8.5: Policy responses on car use using SP models

Taking a car (% of base)

Policy Measures work non-work Combined

(28% work and 72% non-work)

Fixed charge ($1 per entry) 91 83 85

Car size charge ($1 per large car per entry) 87 85 85

Entry time charge ($1 per entry at morning peak) 95 91 92

Parking charge ($1 increase per hour) 88 81 83

The applied SP models show more response in car use than the combined SP-RP

approach. It is also evident in Table 8.5 that the non-work group is more responsive to

the charges than the work group. Both groups would react strongly to the $1 per hour

increase in parking charge but that takes account of average parking times. It was found

in the SP survey that average hours of parking by the work and non-work groups are 6

and 3 hours respectively. As the proportions of the two groups are 28% and 72%, the

calculated average parking time is 3.84 hours.

Page 222: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 8: Modelling the impact of policies on air quality

204

The combined responses are applied in the further analysis of air pollution. The

proportions of travellers taking a car to the city under the four policy measures are

similar to the proportions estimated using the SP-RP combined approach in Section

8.3.1. However, the proportions estimated with only the SP model will be used for the

further analysis in view of this model’s greater reliability.

8.4 ESTIMATED TRAFFIC IMPACT ON AIR QUALITY

A causal relationship has been established in Chapter 3 between air pollution and traffic

level in Perth city. Hourly levels of carbon monoxide (CO) and nitrogen oxides (NOx)

can be accurately estimated by the models (CM2CO and CM2NOx) developed. Hourly

levels of CO and NOx fluctuate with the variations of wind speed, wind direction,

pollution levels in the previous period, and traffic volume. The pollution level can be

reduced by reducing the traffic volume. The suggested policies, discussed above, would

reduce the traffic volume in Perth city and lead to a reduction in the levels of CO and

NOx. The CM2CO model is used to estimate hourly CO level in the city on an average

weekday. CO levels for different measures are represented graphically in Figure 8.1. It

appears that the parking charge policy would be the most effective in reducing CO level

but that is due to the average parking time in Perth city being 3.84 hours.

CO level

0.00

0.20

0.40

0.60

0.80

1.00

1.20

1.40

100

200

300

400

500

600

700

800

900

1000

1100

1200

1300

1400

1500

1600

1700

1800

1900

2000

2100

2200

2300

2400

hour

ppm

Fixed chargeCar size chargeEntry time chargeParking chargeBase Case

Figure 8.1: Hourly CO level under alternative policy measures: a new or added

$1.00 charges in each case

Page 223: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 8: Modelling the impact of policies on air quality

205

The CM2NOx model (see Chapter 3) is also used to estimate the impact of the suggested

policies on hourly NOx level in Perth city. Just as for CO reduction, all policy scenarios

would reduce the number of cars and therefore lower the average hourly NOx level

throughout a day, shown in Figure 8.2.

NOx level

0.00

2.00

4.00

6.00

8.00

10.00

12.00

14.00

16.00

100

200

300

400

500

600

700

800

900

1000

1100

1200

1300

1400

1500

1600

1700

1800

1900

2000

2100

2200

2300

2400

hour

pphm

Fixed chargeCar size chargeEntry time chargeParking chargeBase Case

Another way of presenting the impact on air quality of different policies is the

percentage reduction of daily average pollution level in Perth city, which is shown in

Figure 8.3. Because the $1 per hour increase is applied for 3.84 hours, the parking

charge policy reduces CO more than 14% and NOx 12% from daily average level. The

effects of the $1 fixed charge and car size charge correspond to the proportions of

respondents prepared to take their car to the city shown in Table 8.5.

Figure 8.2: Hourly NOx level under alternative policy measures: a new or added $1.00 charges in each case

Page 224: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 8: Modelling the impact of policies on air quality

206

Daily Average Pollution Reduction

0.0%

2.0%

4.0%

6.0%

8.0%

10.0%

12.0%

14.0%

16.0%

Fixed charge(($1/entry)

Car size charge($1/large car/entry)

Entry time charge($1/entry at morning

peak)

Parking charge ($1increase per hour)

% o

f red

uctio

nCONox

Sensitivity to the parking charge policy is presented in Figure 8.4 which shows the

calculated impact on car use to the city of varying the parking fee from $1.00 to $10.00.

An hourly parking fee beyond $10.00 is presumed to be unrealistic in Perth at present.

The reduction in the percentage of people taking the car to the city from the base case

follows a slightly curvilinear form. Demand responses are not expected to be linear and

the shape of the curve in Figure 8.4 provides a realistic estimate of travel behaviour.

The base case parking fee is $1.30 per hour.

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

110%

1.00

2.00

3.00

4.00

5.00

6.00

7.00

8.00

9.00

10.00

Parking fee ($/hour)

% o

f res

pond

ents

taki

ng c

ar

Figure 8.4: Impact of parking fee on car use in the city

Figure 8.3: Average daily CO and NOx reduction under alternative policy measures

Page 225: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 8: Modelling the impact of policies on air quality

207

The parking charge elasticity provides a test of the model estimates by comparing

previous elasticity estimates. The SP-RP combined approach produces a parking charge

elasticity of -0.15 using equation (8.3) (previously presented in Chapter 4), whereas the

SP models for the work and non-work groups produce -0.24 and -0.44 respectively.

( )( )PP

VVLnLnLnLn

e21

21

−= …………………………………….. (8.3)

Where, e is own price elasticity

V1 is initial traffic

V2 is reduced traffic

P1 is initial price/cost

P2 is final price/cost

The parking charge elasticities of -0.24 and -0.44 estimated using the SP models can be

compared with the average of -0.30 from previous studies presented in Table 4.9

(Chapter 4) and is well within the range (-0.07 at Portland CBD to -0.68 at Los Angeles

CBD) of previous estimates. Less confidence can be placed in the elasticity of -0.15

based on the combined SP-RP estimation because the RP model coefficients are less

reliable.

After the estimation of the degree to which various policies would reduce pollution

levels in Perth city, the last step is to transform the policy impacts into actual benefits.

The major beneficial impacts are on health and these are the basis for the final policy

assessment.

8.5 HEALTH IMPACT OF AIR QUALITY IMPROVEMENT

8.5.1 Mortality and morbidity from air pollution

The effect of air pollution on health has been of concern for hundreds of years,

especially in countries where coal was widely used. For example it was reported that in

1257 Queen Eleanor was unable to remain in Nottingham because of the smoke

(Holland 1972). A number of studies have been conducted in recent years in Australia

and overseas with the objective of costing the health and environment impacts of air

pollution (Amoako et al. 2003). Recently epidemiological studies have tried to

establish the relationships between air pollution and the mortality and/or morbidity rate.

The contribution of air pollution to deaths caused by respiratory and cardiovascular

Page 226: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 8: Modelling the impact of policies on air quality

208

diseases is particularly high. Epidemiological evidence suggests that different

pollutants (such as CO, NOx, PM10, SO2, Lead) have different influences on mortality as

well as morbidity. Many recent studies have tried to estimate the impact on mortality

due to particulates (Amoako et al. 2003, Kunzli et al. 2000, Gouveia and Fletcher 2000,

Finkelstein et al. 2003); at the same time other studies have tried to find a link between

mortality rate and NOx or CO (Chen et al. 2004, Scoggins et al. 2004, Ballester et al.

2001).

Among these studies, Amoako et al. (2003) and the Melbourne Mortality Study (2000)

have considered the Australian situation. The finding of these two studies is that air

pollution, essentially from motor vehicles, is associated with increases in mortality in all

capital cities in Australia. Amoako et al. (2003) quantified the health and economic

impacts of air pollution levels in capital cities in Australia, and the Melbourne Mortality

Study (2000) tried to estimate the relative risks (RR) of deaths due to increases in

various pollutants in Melbourne. As was mentioned in Chapter 2 the relative risk of

death is defined as the probability of death due to an event relative to the absence of that

event. The Melbourne study found a strong association of mortality with ozone and

nitrogen dioxide. Two other studies (Finkelstein et al. 2003, Gouveia and Fletcher

2000) tried to find an association between mortality, air pollution, and income or socio-

economic status. Both studies reported a positive link between lower socio-economic

status and increased relative risk of mortality. One study found an association between

residential proximity to traffic and mortality (Finkelstein et al. 2004). The study argued

that people living closer to a highway or major road have a higher probability of non-

accidental deaths. A study in China (Chen et al. 2004) found increased risk of mortality

and hospital admission for COPD (chronic obstructive pulmonary disease) with

increased SO2 and TSP (total suspended particles). A few other studies in Australia

have estimated relative risks of mortality and hospital admissions due to air pollution in

Sydney, Brisbane, and Melbourne (Morgan et al. 1998a, Simpson et al. 1997,

Petroeschevsky et al. 2001, Morgan et al. 1998b).

Different epidemiological studies report the effect of air pollution in different forms.

Some report RR (relative risk) of mortality due to an increase in 10 μg/m3 of any

pollutant; others report percentage increase in mortality due to an increase in hourly (or

daily) concentration of any pollutant from the10th to 90th centile. Another variation in

these studies is in the method used to estimate the relationship between mortality and air

Page 227: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 8: Modelling the impact of policies on air quality

209

pollution. Some used time-series analysis to build the relationship, others used cohort

analysis to develop either a pollutant model or combined pollutants effect model.

Whatever format is used in the studies the bottom-line is that the increase in pollution

will increase the mortality rate in any urban area.

The present study has adopted the approach of estimating relative risk (RR) of mortality

due to 1 ppm increase in CO and 1 ppb increase in NOx. The study averages the relative

risks of mortality from eight estimates for 1 ppb increase in NOx and two estimates for

1ppm increase in CO (Table 8.6)

Table 8.6: Relative Risks of death from individual studies

Study Relative Risk (RR) due to 1ppb increase

in NOx

Relative Risk (RR) due to 1ppm

increase in CO

Melbourne Mortality Study 2000 1.00100 1.03790

Scoggins et al. 2004 (Auckland) 1.00290

Simpson et al. 2005 (Australia) 1.00120

Hoek et al. 2002 (The Netherlands) 1.01700

Bremner et al. 1999 (London) 1.00300 1.01125

Gouveia and Fletcher 2000 (Sao Paulo) 1.00013

Chen et al. 2004 (China) 1.00290

Ballester et al. 2001 (Valencia, Spain) 1.00125

Average Relative Risk 1.00367 1.02460

Table 8.6 shows the relative risks from different studies and the averages of them. The

average RR of 1.00367 for NOx means that for 1 part per billion (ppb) increase in NOx

the probability of death is 0.37% higher and the RR of 1.0246 for CO means that for 1

part per million (ppm) increase in CO the probability of death is 2.46% higher. These

are global mortality averages. In addition, separate rates for respiratory and

cardiovascular diseases have been applied, using average RRs of death due to 1 ppb

increase in NOx and 1 ppm increase in CO. The average RRs for respiratory death are

1.004 and 1.0566 for NOx and CO respectively. The RRs for cardiovascular death are

1.00311 and 1.02568 for NOx and CO respectively. Details are shown in Table 8.7.

Page 228: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 8: Modelling the impact of policies on air quality

210

Table 8.7: Relative Risks for respiratory and cardiovascular deaths from individual studies

Respiratory disease Cardiovascular disease Death from Due to 1 ppb

increase in NOx Due to 1 ppm increase in CO

Due to 1 ppb increase in NOx

Due to 1 ppm increase in CO

Relative Risk (RR) from different studies

1.0031a, 1.0025b, 1.0064c

1.0882a, 1.025c 1.00138a, 1.00154b, 1.0064c

1.03385a, 1.0175c

Average RR 1.004 1.0566 1.00311 10.2568 a Melbourne Mortality Study 2000 b Wong et al. 2002 (Hong Kong) c Bremner et al. 1999 (London) These figures are used to estimate the health impact of the suggested policy measures

for Perth city, as discussed at the beginning of this chapter.

8.5.2 Health impact of policy implementation

This study is limited to life saved by reducing pollution in the limited area of central

Perth. Only about 10,469 (ABS June 2004) residents live in Perth city whereas the

daytime population is about 98,353 (ABS 2001). These people come to the city from

the suburbs and are usually exposed to the polluted air when they arrive in the city or

walk around for shopping or other purposes. The study estimates the number of lives

that could have been saved annually under the eight hypothetical policies. The specific

cases of respiratory and cardiovascular diseases are also considered.

The total number of deaths in Western Australia in 2004 was 10,402 excluding

accidents, poisonings, and violence; among them 893 were due to respiratory disease

and 3,682 were due to cardiovascular disease. Other causes of deaths include malignant

neoplasms, diabetes mellitus, mental and behavioural disorder, diseases of the digestive

system, congenital malformations, etc. It is known that 72% of the total Western

Australian population live in the metropolitan area and 5.3% form the daytime

population in Perth city. If the number of deaths in Perth city is assumed to be

proportional to the population then number of deaths occurring in Perth city is scaled

down to 562 for all deaths, 48 for respiratory deaths, and 199 for cardiovascular deaths.

There would be a small reduction in these deaths if the suggested policies were

implemented. Figure 8.5 shows a comparative picture of lives that could have been

Page 229: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 8: Modelling the impact of policies on air quality

211

saved under alternative policies in the year 2004, considering only non-accidental

deaths.

0.00

0.40

0.80

1.20

1.60

Fixed charge(($1/entry)

Car size charge($1/large car/entry)

Entry time charge($1/entry at morning

peak)

Parking charge ($1increase per hour)

num

ber

all death savedrespiratory death savedcardiovascular death saved

The parking charge policy could have saved more than 1.2 lives in 2004. The

suggested policies could not only save lives but also save people from pollution induced

disability. Amoako et al. (2003) report that another study (Mathers et al. 1999)

estimated approximately 9% of total life expectancy at birth for both males and females

is lost due to disability in Australia. The study also indicated that 1.2% of life

expectancy is lost due to respiratory illness and 8.9% is lost due to cardiovascular

illness.

According to the BTRE (2003b) the value of a statistical life (VOSL) in Australia is

A$1.9 million in year 2000 values. Therefore the annual benefit from lives saved can

be expressed in monetary terms as in Table 8.8.

Table 8.8: Annual value of statistical life saved from different policies or charges (in ‘000 A$)

Fixed charge

($/entry)

Car size charge ($/large car/entry)

Entry time charge ($ at

morning peak)

Parking fee

($/hr)

All deaths 2,177 2,178 1,558 2,373

Respiratory deaths 405 405 290 441

Cardiovascular deaths 788 789 564 859

Figure 8.5: Lives saved under different policy measures in 2004

Page 230: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 8: Modelling the impact of policies on air quality

212

The fixed charge and car size charge produced the same benefit as expected. The

values of statistical life saved from various policies provide the monetary value of each

policy measure; however benefits should also be assessed in relation to the costs

incurred.

8.6 BENEFIT-SACRIFICE RATIO

The study has estimated the health benefit for various potential policies in terms of

value of life, which is a part of total benefit that could be achieved by implementing the

policies. The suggested policies have the potential to reduce congestion in the city,

improve the traffic flow, and increase average speed along with reduced air pollution.

Although the congestion benefits would be substantial, this study has concentrated only

on the health benefits.

The annual health benefits from various policies have been presented in monetary terms

but to make them strictly comparable the charges (i.e. financial sacrifice) must be taken

into account. A benefit-sacrifice ratio is calculated on the basis of the financial sacrifice

that would be imposed on motorists by each policy. The sacrifice has been taken to

include the direct cost to the motorist who continues to take a car to the city and one

half of the direct cost for any who are priced out of their car (applying the ‘rule-of-half’

method). If a conventional benefit-cost ratio were applied, it would be extremely large

because the true social costs would be small. The charges are merely a transfer and not

a social cost.

The ‘rule-of-half’ takes account of the fact that some motorists are already on the verge

of giving up using a car to go to the city while, at the other extreme, there are those who

would only be induced to give up using the car by the full amount of the charge. It is a

reasonable approximation to assume that the intermediate cases can be interpolated

linearly so that the average cost perceived by those who give up using their cars is the

average of the two extremes or half of the full charge on that number of motorists. This

method was formalised by Neuburger (1971) and Blackshaw (1975) and is applied

widely in benefit-cost evaluation (e.g. Sugden and Williams 1978). The ‘rule’ applies in

this case to those who are priced out of their car.

The direct cost incurred by the motorists, in this case incremental cost (IC), would be

$1 per trip for fixed charge, car size charge or entry time charge and $1 per hour

increase in parking fee (from the average). Implementation and other associated

Page 231: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 8: Modelling the impact of policies on air quality

213

resource costs would only be a small part of the amount paid and the rest would be a

transfer payment or a redistribution of funds from a socio-economic viewpoint.

It was reported in Chapter 6 that 54.1% of the respondents used a large car and 40.9%

entered the city during the morning peak. As the health benefits are expressed annually,

the costs are also calculated annually. The benefit-sacrifice ratio is calculated in the

following steps:

Step 1: Incremental cost per trip per day (IC), taking account of the differential

costs for large cars and morning peak arrivals

Step 2: Total incremental costs per day (TIC) = IC * number of continuing trips.

Step 3: Total annual costs (TAC) = TIC * 260 (assuming 260 weekdays in a

year).

Step 4: Total financial sacrifices (TFS) = TAC for continuing car trips + ½ of

TAC for discontinued car trips (applying the rule-of-half calculation of

perceived cost to those who give up taking a car to the city).

Step 5: Benefit-Sacrifice ratio = Value of lives saved (VLS) factored down to

260 from 365 days / Total financial sacrifice (TFS).

Table 8.9 shows the incremental costs and health benefits for four policy measures. The

first row shows the proportions of respondents who would continue to take their car to

the city and the second row shows the proportions who would give up (extracted from

the 4th column of Table 8.5). The third row shows the value of life saved in million

dollars (extracted from Table 8.8) and this is factored down to 260 weekdays in the

fourth row. The uniform $1 incremental charge in the fifth row is converted to cost per

trip in the sixth to take account of the 3.84 hours of average parking time. Then in the

seventh row the total incremental cost per day is calculated for those who are taking a

car to the city, using the number with large cars or arriving in the morning peak for the

related charges. The ‘rule-of-half’ calculation of perceived cost for those priced out of

their cars is presented in row eight. The sum of the motorists’ sacrifices is shown in

row nine and the benefit-sacrifice ratio in row ten.

Page 232: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 8: Modelling the impact of policies on air quality

214

Table 8.9: Benefit-Sacrifice Ratio calculation for various policies

Fixed charge

($1/entry)

Car size charge ($1/large car/entry)

Entry time charge

($1/entry at morning peak)

Parking charge

($1 increase per hour)

(1) % of respondents continuing to take a car to the city (from Table 8.4) 85 85 92 83

(2) % of respondents priced out of their cars 15 15 08 17

BENEFITS

(3) Value of lives saved (VLS) (M$) (from Table 8.7) 2.177 2.178 1.558 2.373

(4) Value of lives saved (VLS) factored to 260 days (M$) 1.551 1.551 1.110 1.690

SACRIFICES

(5) Charge imposed ($) 1.00 1.00 1.00 1.00

(6) Incremental cost per trip per day ($) (from Table 8.4) 1.00 1.00 1.00 3.84a

(7) Total cost per day for those who continue taking a car to the city ($) 314 170b 139c 1,176

(8) Perceived cost per day for those priced out of their cars (rule-of-half) ($) 55 55 30 241

(9) Total sacrifice (M$) 0.089 0.051 0.040 0.337

(10) Benefit-Sacrifice Ratio 17.5 30.2 27.8 5.0a average hours of parking was 3.84 b proportion of large car use was 54.1% c proportion of travellers enter the city in the morning peak was 40.9%

The ratio for the parking charge policy is low compared to the other three policies

because of the high total incremental cost per day. The parking charge policy measure

of $1 increase per hour becomes $3.84 per day when applied to average parking time.

Page 233: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 8: Modelling the impact of policies on air quality

215

0.0

5.0

10.0

15.0

20.0

25.0

30.0

35.0

Fixed charge(($1/entry)

Car size charge($1/large car/entry)

Entry time charge($1/entry at morning

peak)

Parking charge ($1increase per hour)

Ben

efit

Sac

rific

e R

atio

0.000

0.050

0.100

0.150

0.200

0.250

0.300

0.350

0.400

cost

(M$)

Benefit Sacrifice Ratio Financial sacrifice

A graphical comparison of the cost effectiveness of the suggested policies is shown in

Figure 8.6. Although it was found that the parking charge policy is the most effective

in reducing air pollution and saving lives, it would impose high costs on motorists so

that its cost effectiveness is low. It is shown in Figure 8.6 that the car size charge

would be the most cost effective of the suggested policies. Although it may not save

many lives, the cost associated with this policy is low compared with other policies.

The morning peak entry time charge is almost as cost effective and the charge could be

much greater than $1. In fact the choice scenarios from which the coefficients were

estimated sought the response to a $4 entry time charge.

8.7 CONCLUSION

The results of the discrete choice modelling have been applied to estimate the impacts

of the charges on car trips to the city. The reduced proportion of car use ensures

reduced air pollution in the city. The new level of pollution is estimated by applying the

pollution model developed in Chapter 3. Then the impacts on health from reduced air

pollution were estimated. A maximum of 1.25 deaths could have been saved in Perth

city in 2004 under the fairly severe parking charge policy of increasing the fee from

$1.30 to $2.30 per hour.

If cost effectiveness, where the cost is to motorists, is the criterion then the picture

would be different. Of the four different policy measures presented in this chapter to

estimate the impact on air quality in Perth city, the study results indicate that the car

Figure 8.6: Benefit-Sacrifice Ratios and financial sacrifices for different measures

Page 234: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 8: Modelling the impact of policies on air quality

216

size charge would the most effective policy with the highest ratio of benefits to the

financial sacrifices imposed on motorists. The morning peak entry time charge would

be almost as effective.

Page 235: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 8: Modelling the impact of policies on air quality

217

Appendix – 8A

Base Case Model (repeat from Appendix 5F) This model is used to simulate with TRIPCOST and PARKHOUR change. The simulated models specifications are shown in Appendices 8B and 8C. --> NLOGIT; lhs=CHOICE,CSET,MODE; CHOICES=carA,carB,bus,train; Tree=mode[Private(carA,carB),Public(bus,train)]; IVSET:(PRIVATE)=[1]; MODEL: U(carA)=A_carA+time*TRAVELTI+cost*TRIPCOST+park*PARKHOUR/ U(carB)=A_carB+time*TRAVELTI+cost*TRIPCOST+age*AGE+sex*SEX/ U(bus)=A_bus +time*TRAVELTI+cost*TRIPCOST/ U(train)= time*TRAVELTI+cost*TRIPCOST; crosstab; $ Normal exit from iterations. Exit status=0. +---------------------------------------------+ | FIML Nested Multinomial Logit Model | | Maximum Likelihood Estimates | | Model estimated: Dec 21, 2005 at 11:43:23AM.| | Dependent variable CHOICE | | Weighting variable None | | Number of observations 1211 | | Iterations completed 19 | | Log likelihood function -339.8810 | | Number of parameters 9 | | Info. Criterion: AIC = .57619 | | Finite Sample: AIC = .57631 | | Info. Criterion: BIC = .61408 | | Info. Criterion:HQIC = .59046 | | Restricted log likelihood -519.8604 | | Chi squared 359.9587 | | Degrees of freedom 9 | | Prob[ChiSqd > value] = .0000000 | | R2=1-LogL/LogL* Log-L fncn R-sqrd RsqAdj | | No coefficients -519.8604 .34621 .33909 | | Constants only. Must be computed directly. | | Use NLOGIT ;...; RHS=ONE $ | | At start values -429.9295 .20945 .20085 | | Response data are given as ind. choice. | +---------------------------------------------+ +---------------------------------------------+ | Notes No coefficients=> P(i,j)=1/J(i). | | Constants only => P(i,j) uses ASCs | | only. N(j)/N if fixed choice set. | | N(j) = total sample frequency for j | | N = total sample frequency. | | These 2 models are simple MNL models. | | R-sqrd = 1 - LogL(model)/logL(other) | | RsqAdj=1-[nJ/(nJ-nparm)]*(1-R-sqrd) | | nJ = sum over i, choice set sizes | +---------------------------------------------+ +---------------------------------------------+ | FIML Nested Multinomial Logit Model | | The model has 2 levels. | | Nested Logit form:IV parms = tauj|i,l,si|l | | and fl. No normalizations imposed a priori. | | p(alt=k|b=j,l=i,t=l)=exp[bX_k|jil]/Sum | | p(b=j|l=i,t=l)=exp[aY_j|il+tauj|ilIVj|il)]/ | | Sum. p(l=i|t=l)=exp[cZ_i|l+si|lIVi|l)]/Sum | | p(t=l)=exp[exp[qW_l+flIVl]/Sum... | | Number of obs.= 375, skipped 0 bad obs. | +---------------------------------------------+

Page 236: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 8: Modelling the impact of policies on air quality

218

+---------+--------------+----------------+--------+---------+ |Variable | Coefficient | Standard Error |b/St.Er.|P[|Z|>z] | +---------+--------------+----------------+--------+---------+ Attributes in the Utility Functions (beta) A_CARA .44488124 .51069273 .871 .3837 TIME -.01352926 .01018200 -1.329 .1839 COST -.39840195 .20936806 -1.903 .0571 PARK -.29537137 .31799156 -.929 .3530 A_CARB .35774450 .46919161 .762 .4458 AGE -.03078216 .00950570 -3.238 .0012 SEX -.80485260 .29403670 -2.737 .0062 A_BUS -1.49335336 .31308826 -4.770 .0000 IV parameters, tau(j|i,l),sigma(i|l),phi(l) PRIVATE 1.00000000 ......(Fixed Parameter)....... PUBLIC .78047141 .14405139 5.418 .0000

Page 237: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 8: Modelling the impact of policies on air quality

219

Appendix – 8B Fuel Price Change Simulated Model --> NLOGIT; lhs=CHOICE,CSET,MODE; CHOICES=carA,carB,bus,train; Tree=mode[Private(carA,carB),Public(bus,train)]; IVSET:(PRIVATE)=[1]; MODEL: U(carA)=A_carA+time*TRAVELTI+cost*TRIPCOST+park*PARKHOUR/ U(carB)=A_carB+time*TRAVELTI+cost*TRIPCOST+age*AGE+sex*SEX/ U(bus)=A_bus +time*TRAVELTI+cost*TRIPCOST/ U(train)= time*TRAVELTI+cost*TRIPCOST; crosstab; simulation=carA,carB,bus,train; scenario: tripcost(carA)=[+]0.98/ parkhour(carA)=[+]0; $ +------------------------------------------------------+ |Simulations of Probability Model | |Model: FIML: Nested Multinomial Logit Model | |Simulated choice set may be a subset of the choices. | |Number of individuals is the probability times the | |number of observations in the simulated sample. | |Column totals may be affected by rounding error. | |The model used was simulated with 375 observations.| +------------------------------------------------------+ ---------------------------------------------------------------------- Specification of scenario 1 is: Attribute Alternatives affected Change type Value --------- --------------------------- ------------------- --------- TRIPCOST CARA Add value to base .980 PARKHOUR CARA Add value to base .000 ---------------------------------------------------------------------- The simulator located 375 observations for this scenario. Simulated Probabilities (shares) for this scenario: +----------+--------------+--------------+------------------+ |Choice | Base | Scenario | Scenario - Base | | |%Share Number |%Share Number |ChgShare ChgNumber| +----------+--------------+--------------+------------------+ |CARA | 54.472 204 | 46.970 176 | -7.502% -28 | |CARB | 22.224 83 | 26.039 98 | 3.815% 15 | |BUS | 9.827 37 | 11.507 43 | 1.680% 6 | |TRAIN | 13.476 51 | 15.483 58 | 2.007% 7 | |Total |100.000 375 |100.000 375 | .000% 0 | +----------+--------------+--------------+------------------+

Page 238: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 8: Modelling the impact of policies on air quality

220

Appendix – 8C Parking Charge Simulated Model --> NLOGIT; lhs=CHOICE,CSET,MODE; CHOICES=carA,carB,bus,train; Tree=mode[Private(carA,carB),Public(bus,train)]; IVSET:(PRIVATE)=[1]; MODEL: U(carA)=A_carA+time*TRAVELTI+cost*TRIPCOST+park*PARKHOUR/ U(carB)=A_carB+time*TRAVELTI+cost*TRIPCOST+age*AGE+sex*SEX/ U(bus)=A_bus +time*TRAVELTI+cost*TRIPCOST/ U(train)= time*TRAVELTI+cost*TRIPCOST; crosstab; simulation=carA,carB,bus,train; scenario: tripcost(carA)=[+]0.2/ parkhour(carA)=[+]1.00; $ +------------------------------------------------------+ |Simulations of Probability Model | |Model: FIML: Nested Multinomial Logit Model | |Simulated choice set may be a subset of the choices. | |Number of individuals is the probability times the | |number of observations in the simulated sample. | |Column totals may be affected by rounding error. | |The model used was simulated with 375 observations.| +------------------------------------------------------+ --------------------------------------------------------------------- Specification of scenario 1 is: Attribute Alternatives affected Change type Value --------- --------------------------- ------------------- --------- TRIPCOST CARA Add value to base .200 PARKHOUR CARA Add value to base 1.000 --------------------------------------------------------------------- The simulator located 375 observations for this scenario. Simulated Probabilities (shares) for this scenario: +----------+--------------+--------------+------------------+ |Choice | Base | Scenario | Scenario - Base | | |%Share Number |%Share Number |ChgShare ChgNumber| +----------+--------------+--------------+------------------+ |CARA | 54.472 204 | 47.272 177 | -7.200% -27 | |CARB | 22.224 83 | 25.884 97 | 3.660% 14 | |BUS | 9.827 37 | 11.437 43 | 1.610% 6 | |TRAIN | 13.476 51 | 15.407 58 | 1.930% 7 | |Total |100.000 375 |100.000 375 | .000% 0 | +----------+--------------+--------------+------------------+

Page 239: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 9: Conclusions

221

CHAPTER NINE

Conclusions

After a brief review, the study concludes by summarising the assessment and impacts of

policy measures to ameliorate environmental pollution by cars in Perth city. Section 9.1

reports key findings and 9.2 presents implications of the study.

It was noted at the beginning of this thesis that for a government to undertake

unpalatable measures in order to combat air pollution it needs to be reasonably assured

that a specific impost on motorists will produce a fairly certain result. This study set out

to establish that such certainty can be achieved by accurately measuring the CO and

NOx contribution per vehicle and then determining one or more appropriate charges

based on estimates of behavioural responses.

The four major questions that have been addressed are:

• What is the nature of the variation in air pollution in Perth and what are the

causal factors that influence this variation?

• How would urban people respond to policies specifically designed to

reduce air pollution?

• Which policies would be most effective in reducing air pollution in the

urban area?

• What would be the health benefits of the reduced pollution?

Perth might be thought the last city in Australia to be concerned about air pollution

because it has a year round sea breeze which blows away the pollutants emitted from

urban sources. However the proportion of commuters to Perth city who use cars is

higher than in any other large city in Australia and the exhaust gases tend to be trapped

by the city buildings in the downtown area. Also there are some adverse wind effects.

Wind direction in Perth has a great influence on the concentration of gases. The major

gaseous pollutant in the downtown area is CO which is emitted by motor vehicles and is

observed at relatively high concentration in Perth city. The pollutants categorised as

NOx originate from motor vehicle and industry exhausts. The geographical location of

Page 240: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 9: Conclusions

222

Perth city and the unique wind pattern cause most of the northern suburbs, starting from

Perth city, to be polluted. The afternoon sea breeze brings back to the city and northern

suburbs the polluted morning air which has been blown out to sea, as well as major

industry exhausts from the southern suburbs.

9.1 RESEARCH FINDINGS

An important general finding is that pollutants accumulate in the city throughout the day

so that maximum concentrations are in the afternoon. The sources of this accumulation

are emissions from morning peak and off-peak motor vehicles and the polluted air

blown back from the sea. Although Perth city does not have as many high-rise

buildings as other major cities in Australia, it still experiences street canyon effects. All

these characteristics together mean that Perth city is exposed to substantial air pollution

and the situation will deteriorate with increased motor vehicle use.

The first step in this study was to estimate physical models of air pollution and then to

link these to travel behaviour. The essence of the policies is to penalise the car

travellers who create most of the pollution in the city in order to improve air quality.

The initial application of the physical modelling results was to measure the impact of

suggested policies by using previously estimated elasticities with the new pollution

coefficients. Each of the four suggested policies was estimated to reduce pollution

levels in Perth city, as shown in Figure 9.1.

Page 241: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 9: Conclusions

223

Annual Reductions

012345678

Fixed Charge VariableCharges

ParkingMeasures

LaneRestriction

(tonn

es)

NOx CO

a fixed charge represents $1 per entry into the city, b from Table 4.11, variable charges represent a composite of $0.45/km in the morning

peak, $0.40/km between peaks, and $0.35/km for the rest of the day. c parking measures represent parking fee of $4.0/hour in the morning peak and

$3.0/hour during off-peak. d lane restriction represents the left lane of any 4-laned road in the city would be closed

to cars.

Among these four policies the lane restriction measure would have the greatest impact

on air quality improvement. This part of the study was based partly on secondary

information about responses, in elasticity form, which may or may not reflect the actual

behaviour of Perth travellers. Therefore discrete choice methods were used to

investigate actual transport mode selection by people who travel to Perth city. The

pollution models are summarised briefly and then the discrete choice models.

9.1.1 The pollution models

The physical relationships between climate, vehicles and pollution have been

mathematically modelled for CO and NOx levels in Perth city. Table 9.1 shows that in

both cases the wind variables together contribute more to the explanation than the

traffic, as does the cross product of previous wind speed and pollutant level. However

the traffic contribution is substantial and the estimated traffic coefficients are highly

reliable (t-ratios of 29.5 and 26.8). Thus wind dominates the outcome but traffic makes

a large contribution and the impact of each vehicle on CO and NOx has been accurately

estimated.

Figure 9.1: Initial estimates of annual reduction of air pollution in Perth city under four suggested policies (from Chapter 4)

ab c d

Page 242: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 9: Conclusions

224

Table 9.1: Relative contributions of variables explaining CO and NOx levels

Standardised Beta Coefficient

CO NOx

Wind speed -0.109 -0.034

Previous period’s wind speed 0.406 0.199

Previous period’s north-east wind speed -0.200 -0.135

Previous period’s south-east wind speed -0.317 -0.206

Previous period’s south-west wind speed 0.059 0.104

Cross product of previous period’s wind speed and pollutant level -0.979 -0.853

Traffic 0.508 0.490

As shown by the signs of the standardised coefficients in Table 9.1, the study found that

the south-west wind has a positive impact and south-east and north-east winds have

negative impacts on levels of CO and NOx in Perth city. Because of the geographical

position of the William Street canyon, where the pollution monitoring receptor is

located, the south-east wind has a greater negative impact on air pollution than the

north-east wind.

9.1.2 The behavioural models

The main results of the revealed preference (RP) and stated preference (SP) model

estimations are shown in Table 9.2. The RP and SP coefficients are not directly

comparable because the RP dealt with choice of travel mode whereas the SP was limited

to ‘would you take the car’. The orthogonal experimental design in the SP case

virtually guarantees reliable estimates – as reflected by the t-ratios in brackets –

whereas the RP estimates suffer the vagaries of actual choices. Also the RP model is

based on the PARTS (Perth and Regional Travel Survey) data and it was necessary to

impute the set of choices faced by each person.

The SP survey was specifically designed to assess whether there are different degrees of

reaction to the four alternative types of charge: per entry, on a large car, on entry in the

morning peak, or an increased parking charge. The results indicate that responsiveness

does vary by type of charge (Table 9.2).

Page 243: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 9: Conclusions

225

Table 9.2: Coefficients of the RP model for trips by all modes and the binary SP models for car only: trips to Perth city (t-ratios in brackets)

RP: Trips to Perth City by Car or Bus or Train

SP: Car to Work in Perth City

SP: Car to Perth City for Non-work

Travel time -0.0135 (1.3) Trip cost -0.398 (1.9) Fuel price -0.486 (3.5) -0.747 (8.4) Fixed entry charge -0.283 (4.2) -0.420 (9.7) Charge on large car -0.415 (3.6) -0.379 (5.1) Peak entry charge -0.168 (5.8) -0.214 (11.5) Parking fee -0.295 (0.9) -0.392 (13.9) -0.464 (23.2) Less parking space -0.196 (1.7) -0.325 (4.5) Lane restriction -0.128 (1.1) -0.231 (3.2) Cars per licence 0.970 (4.9) 0.625 (5.1) Uses large car 0.816 (7.0) 0.205 (2.8) Age -0.031 (3.2) Male person -0.804 (2.7) Car drivera 0.444 (0.9) Car passengera 0.357 (0.8) Bus passengera -1.493 (4.8) a Alternative specific constants are with respect to train, which is the reference mode

The coefficients and other findings from the RP mode choice modelling show that:

i) Private and public transport mode users behave differently.

ii) Other than trip cost and travel time attributes, hourly parking fee was one of

the most influential variables on the car as a driver mode, while age and

gender influence the behaviour of car as a passenger travellers.

iii) Mode switching was most sensitive with respect to bus fare and was least

sensitive to car trip cost, as shown by the cross-elasticities.

iv) Value of travel time savings (VTTS) was $2.04 her hour for all travellers.

The stated preference (SP) survey data showed that about 70% of the respondents went

to the city for purposes other than work and more than half went infrequently. About

45% of all respondents arrived in the city before 10am. Ninety percent were willing to

use public transport if taking a car became inconvenient.

Page 244: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 9: Conclusions

226

The estimated model coefficients, marginal effects, and odds ratios all established that

the non-work group would be more responsive to the potential charges than the work

group. The majority of the non-work group come to the city for the purpose of personal

business and recreation followed by shopping.

As discussed in Chapter 8, the study plan was to integrate the SP coefficients into the

RP model in order to exploit the strengths of both. This plan was carried through but

the very low reliability of the crucial RP parameter for parking (Table 9.2) led to the

choice of the SP coefficients alone as the basis of the calculations of benefit from the

alternative air quality improvement measures.

9.2 IMPLICATIONS AND INFERENCES

The final step in this study was to measure the impact of the suggested car restriction

policies in saving lives. Although these policies would also reduce congestion, the

benefit assessment is limited to life saved by reducing pollution in Perth city. There are

less than 11,000 residents but the daytime population of the city is about 100,000.

People coming in from the suburbs are exposed to the polluted air when they walk

around for shopping or other purposes.

9.2.1 Financial assessment

In assessing the merit of alternative measures the benefits have been related to the

charges that would be imposed on motorists who currently drive a car to the city. In

economic terms, these are almost entirely transfer payments and only a very small

proportion would be to recover the resource costs of administration and maintenance.

The charges would be a sacrifice made by motorists for the social good so that it is

appropriate to formulate a ratio of benefits to the motorists’ sacrifice. The benefits are

calculated on the basis of the value of saving a life, taken to be $1.9 million (in dollars

of year 2000). The relative risk of mortality for specific increases in CO and NOx levels

has been averaged over a number of recent studies.

The benefit-sacrifice ratios for the four policy measures are shown in Figure 9.2. A car

size charge of $1 for large car per entry would confer the greatest benefits in relation to

the magnitude of the actual and perceived costs but a $1 charge imposed on all cars

entering the city in the morning peak would be nearly as effective. Only the parking

charge increase in $1 per hour would be appreciably less cost effective.

Page 245: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 9: Conclusions

227

0.0

5.0

10.0

15.0

20.0

25.0

30.0

35.0

Fixed charge(($1/entry)

Car size charge($1/largecar/entry)

Entry timecharge ($1/entry

at morningpeak)

Parking charge($1 increase per

hour)

bene

fit-s

acrif

ice

ratio

The physical limitation measures, reduced parking space or closing the kerbside lane to

cars would also be effective but these measures have not been costed. The odds ratios

of Chapter 7 (Table 7.2) indicate that they would be somewhat less effective than a

fixed charge of $1 and the marginal effects calculations (Table 7.13) indicate

comparable responsiveness by motorists to the entry time charge. As with most of the

other charges, the non-work group would be considerably more responsive. Both the

odds ratios and the marginal effects show that the non-work group would react more

strongly than the work group to either of these physical measures.

9.2.2 Other implications and inferences

All of the results have shown that the policies would have a greater effect on people

whose trips to the city are for purposes other than work. A relatively large proportion of

them would be diverted or deterred from making a trip to the city. To a large extent this

means that shoppers and others would shift their activities elsewhere.

There is a contrast between the suggested measures and Ramsey pricing which

discriminates against people with the least elastic demand; pricing to reduce pollution

would primarily target those with more elastic demand, the non-work predominantly

off-peak motorists. Any of the suggested charges would also reduce congestion in the

peaks but the response by the work group who are the main peak travellers would be

much less than the response of the non-work off-peak travellers. However the latter

Figure 9.2: Benefit-sacrifice ratios for four potential policy measures

Page 246: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

Chapter 9: Conclusions

228

group contribute to pollution throughout the day and their response would provide much

of the justification for imposing a charge.

In this study, the value of money has been viewed from a different viewpoint than has

been taken by many transport studies of the value of life. Generally transport studies

have considered a ‘selfish’ value of life by quantifying the willingness to pay to reduce

the probability of one’s own death. However this study deals with the value of lives

trapped in the equivalent of a passive smoking situation. Many countries are trying to

eliminate passive smoking by banning smoking in public places. This study is taking

the same line by suggesting effective measures to control air pollution from transport in

order to benefit the unfortunates who breathe the air.

Page 247: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

References

229

REFERENCES

Abdel-Aty, M. A. 2001, “Using ordered probit modelling to study the effect of ATIS of transit ridership”, Transportation Research Part C, Vol. 9, pp. 265-277.

Abramson, C., Buchmueller T. and Currim I. 1998, “Models of health plan choice”, European Journal of Operational Research, Vol. 111, pp. 228-247.

ABS 2004, Motor Vehicle Census, March 2004, Australian Bureau of Statistics, 9309.3, Canberra.

Adamovicz, W., Swait J., Boxall P., Louviere J. and Williams M. 1997, “Perceptions versus objective measures of environmental quality in combined revealed and stated preference models of environmental valuation”, Journal of Environmental Economics and Management, Vol. 32, pp. 65-84.

AGO 2005, “Australian methodology for the estimation of greenhouse gas emission and sink 2003”, Australian Greenhouse Office, Canberra.

Akcelik, R. and Besley M. 2003, “Operating cost, fuel consumption, and emission models in aaSIDRA and aaMOTION”, 25th Conference of Australian Institutes of Transport Research (CAITR 2003), Adelaide.

Alwin, D. F. (ed.) 1978, Survey design and analysis: Current issues, Sage Publication, London.

Amoako, J., Ockwell A. and Lodh M. 2003, “The economic consequences of the health effects of transport emissions in Australian capital cities”, 26th Australian Transport Research Forum, New Zealand.

Arentze, T. Hofman F. and Timmermans H. 2004, “Predicting multi-faceted activity-travel adjustment strategies in response to possible congestion pricing scenarios using an Internet-based stated adaptation experiment”, Transport Policy, Vol. 11, pp. 31-41.

Australian Government 2003, Fuel Consumption Guide, ISBN 0 642549 354.

Ayidiya, S. A. and McClendon M. J. 1990, “Response effects in mail surveys”, Public Opinion Quarterly, Vol. 54, pp. 229-247.

Ballester, F., Tenias J. M. and Perez-Hoyos S. 2001, “Air pollution and emergency hospital admissions for cardiovascular diseases in Valencia, Spain”, Journal of Epidemiol Community Health, Vol. 55, pp. 57-65.

Bartholomew, D. J. 1987, Latent variable models and factor analysis, Oxford University Press, New York, USA.

Ben-Akiva, M. and Morikawa T. 1990, “Estimation of switching models from revealed preferences and stated intentions”, Transportation Research Part –A, Vol. 24A, No. 6, pp. 485-495.

Page 248: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

References

230

Ben-Akiva, M. and Morikawa T. 2002, “Comparing ridership attraction of rail and bus”, Transport Policy, Vol. 9, pp. 107-116.

Ben-Akiva, M. and Lerman S. R. 1985, Discrete Choice Analysis: theory and application to travel demand, The MIT Press, London, UK.

Ben-Akiva, M. Morikawa T., and Shiroishi F. 1991, “Analysis of the reliability of preference ranking data”, Journal of Business Research, Vol. 23, pp. 253-268.

Berkowicz, R., Hertel O., Larsen S., Sorensen N. and Nielsen M. 1997, “Modelling traffic pollution in streets”, National Environmental Research Institute, Denmark.

Berkowicz, R. 1998, “Street scale models” in Urban air pollution – European aspect, eds. Fenger J., Hertel O. and Palmgren F., Kluwer Academic Publishers, Netherlands, pp. 223-251.

Berkowicz, R. 2000, “OSPM – A parameterised street pollution model”, Environmental Monitoring and Assessment, Vol. 65, pp. 323-331.

Blackshaw, P. W. 1975, “The treatment of cross-modal effects in transport evaluations”, Transport Economics and Operational Analysis, Vol. 1, pp. 34-44.

Boubel, R., Fox D., Turner D. and Stern A. 1994, Fundamentals of air pollution, 3rd edition, Academic Press, USA.

Boxall, P. C., Englin J. and Adamowicz W. L. 2003, “Valuing aboriginal artifacts: a combined revealed-stated preference approach”, Journal of Environmental Economics and Management, Vol. 45, pp. 213-230.

Bray, D. and Tisato P. 1997, “Broadening the debate on road pricing”, 21st Australian Transport Research Forum, Adelaide.

Bremner, S., Anderson H., Atkinson R., McMichael A., Strachan D., Bland J. and Bower J. 1999, “Short term associations between outdoor air pollution and mortality in London 1992-4”, Occupational and Environmental Medicine, Vol. 56, No. 4, pp. 237-244.

Brockwell, P. and Davis R. 1996, An introduction to time series and forecasting, Springer-Verlag, New York.

BTCE 1996a, Transport and Greenhouse: Costs and Options for Reducing Emissions, Bureau of Transport and Communications Economics, Report 94, Canberra.

BTCE 1996b, Traffic congestion and road user charges in Australian capital cities, Bureau of Transport and Communications Economics, Report No. 92, Canberra.

BTE 1993, Industry Commission survey of car travel elasticities, Urban Transport, Draft Report, Vol. 2, AGPS, Canberra.

BTRE 2003a, “Urban pollutant emissions from motor vehicles: Australian trends to 2020”, Report for Environment Australia, BTRE, Canberra.

BTRE 2003b, “Rail accidents costs is Australia”, Bureau of Transport and Regional Economics, Report 108, Canberra.

Page 249: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

References

231

Button, K. 1993, Transport Economics, second edition, Edward Elgar, UK.

Calthrop, E. 2002, “Evaluating on-street parking policy”, Working paper series, Center for Economic Studies, Katholieke Universiteit Leuven.

Caldas, M. A. F. and Black I. 1997, “Formulating a methodology for modelling revealed preference discrete choice data – the selectively replicated logit estimation”, Transportation Research part B, Vol. 31, No. 6, pp. 463-472.

Camagni, R., Capello R. and Nijkamp P. 1998, “Towards sustainable city policy: an economy-environment technology nexus”, Ecological Economics, Vol. 24, pp. 103-118.

Carnovale F., Tilly K., Stuart A., Carvalho C., Summers M. and Eriksen P. 1996, “Metropolitan air quality study air emissions inventory”, final report, prepared by EPA Victoria for Environment Protection Authority of New South Wales, Sydney.

Carson, R., Louvier J., Arabie P., Bunch D., Hensher D., Johnson R. Kuhfeld W., Steinberg D., Swait J., Timmermans H. and Wiley J. 1994, “Experimental analysis of choice”, Marketing Letters, Vol. 5, No. 4, pp. 351-368.

Chaaban, F. B., Nuwayhid, I. and Djoundourian, S. 2001, “A study of social and economic implications of mobile sources on air quality in Lebanon”, Transportation Research Part D, Vol. 6, pp. 347-355.

Chen, B., Hong C. and Kan H. 2004, “Exposures and health outcomes from outdoor air pollution in China”, Toxicology, Vol. 198, pp. 291-300.

Cherchi, E. and Ortuzar J. D. D. 2002, “Mixed RP/SP models incorporating interaction effects”, Transportation, Vol. 29, pp. 371-395.

Chipman, J. 1960, “The foundations of utility”, Econometrica, Vol. 28, pp.193–224.

City of Perth 2004, 2004-2005 Parking Fee Schedule, City of Perth, Western Australia.

Colls, J. 1997, Air pollution: an introduction, E & FN Spon Publisher, London.

Colvile, R. N., Hutchinson, E. J., Mindell, J. S., and Warren, R. F. 2001, “The transport sector as a source of air pollution”, Atmospheric Environment, Vol. 35, pp. 1537-1565.

Dabberdt, W., Ludwig F. and Johnson W. 1973, “Validation and applications of an urban diffusion model for vehicular pollutants”, Atmospheric Environment, Vol. 7, pp. 603-618.

De Borger, B. and Wouters S. 1998, “Transport externalities and optimal pricing and supply decisions in urban transportation: a simulation analysis for Belgium”, Regional Science and Urban Economics, Vol.28, No.2, pp.163-197.

Debreu, G. 1960, “Review of R.D. Luce individual choice behaviour”, American Economic Review, Vol. 50, pp. 186–188.

Page 250: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

References

232

Deng, X. 2006, “Economic costs of motor vehicle emissions in China: A case study”, Transportation Research Part D, (in press).

Dia, H., Panwai S., Boongrapue N., Ton T. and Smith N. 2005, “Implementation of power-based fuel consumption and environmental emission models in traffic simulation”, 27th Conference of Australian Institutes of Transport Research (CAITR 2005), Brisbane.

Dillman, D. A. 1978, Mail and telephone surveys: the total design method, A Wiley-Interscience Publication, Canada.

Dillman, D. A. 2000, Mail and Internet Surveys: The tailored design method, 2nd edn, John Wiley & Sons Publication, Canada.

Eerens, H., Sliggers C. and Hout K. 1993, “The CAR model: the Dutch method to determine city street air quality”, Atmospheric Environment, Vol. 27B, pp. 389-399.

EPA 2001, “MOBILE6 On-road motor vehicle emission model”, US Environmental Protection Agency.

EPA NSW 1995, “Metropolitan Air Quality Study – air emissions inventory”, Draft Report, Environment Protection Authority NSW.

EPA Victoria 1999, “Measurements of motor vehicle pollutants and fleet average emission factors in Melbourne”, Environment Protection Authority, Victoria.

EPA Victoria 2003, “Australian motor vehicle emission factors system – version 1.0 (AusVeh 1.0) Manual”, Environment Protection Authority, Victoria.

Finkelstein, M. M., Jerrett M., Deluca P., Finkelstein N., Verma D. K., Chapman K. and Sears M. R. 2003, “Relation between income, air pollution and mortality: a cohort study”, Journal of Canadian Medical Association, Vol. 169, pp. 397-402.

Finkelstein, M. M., Jerrett M. and Sears M. R. 2004, “Traffic air pollution and mortality rate advancement periods”, American Journal of Epidemiology, Vol. 160, pp. 173-177.

Flaherty, B. P. 2002, “Assessing reliability of categorical substance use measures with latent class analysis”, Drug and alcohol dependence, Vol. 68, pp. S7-S20.

Fulton, L. M., Noland R. B., Meszler D. J. and Thomas J. V. 2000, “A statistical analysis of induced travel effects in the U. S. mid-Atlantic region”, Journal of Transportation and Statistics, Vol. 3, No. 1, pp. 1-14.

Goodwin P.B. 1992, “A review of new demand elasticities with special reference to short and long run effects of price changes”, Journal of Transport Economics and Policy, vol.26, no.2, pp.155-186.

Gouveia, N. and Fletcher T. 2000, “Time series analysis of air pollution and mortality: effects by cause, age and socioeconomic status”, Journal of Epidemiology and Community Health, Vol. 54, pp. 750-755.

Greene, W. H. 2002, NLOGIT reference guide, Electronic Software, Inc. USA.

Page 251: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

References

233

Greene, W. H. 2005, “Reconsidering heterogeneity in panel data estimators of the stochastic frontier model”, Journal of Econometrics, Vol. 126, pp. 269-303.

Greene, W. H. and Hensher D.A. 2002, A latent class model for discrete choice analysis: contrasts with mixed logit, Working paper, Institute of Transport Studies, Sydney.

Gudmundsson, H. and Hojer M. 1996, “Sustainable development principles and their implications for transport”, Ecological Economics, Vol. 19, pp. 269-282.

Harvey, A. C. 1980, “On comparing regression models is levels and first differences”, International Economic Review, Vol. 21, No. 3, pp. 707-720.

Hensen, M. and Huang Y. 1997, “Road supply and traffic in California urban areas”, Transportation Research Part – A, Vol. 31, No. 3, pp. 205-218.

Hensher, D. and Bradley M. 1993, “Using stated response choice data to enrich revealed preference discrete choice models”, Marketing Letters, Vol. 4, No. 2, pp. 139-151.

Hensher, D., Louviere J. and Swait J. 1999, “Combining sources of preference data”, Journal of Econometrics, Vol. 89, pp. 197-221.

Hensher, D. and King J. 2001, “Parking demand and responsiveness to supply, pricing and location in the Sydney central business district”, Transportation Research Part – A, Vol. 35, pp. 177-196.

Hensher, D., Rose J. and Greene W. 2005, “Applied Choice Analysis – A primer”, Cambridge University Press, UK.

Hensher, D. and Young 1990, Fuel Demand Forecasts and Long-Term Price Elasticities of Demand, Bureau of Transport and Communications Economics, Canberra.

Hess, D. B. 2001, “The effect of free parking on commuter mode choice: evidence from travel diary data”, Working paper series, UCLA, School of Public Policy and Social Research, No. 34, April 2001.

Hess, S., Bierlaire M. and Polak J. W. 2005, “Estimation of value of travel-time savings using mixed logit models”, Transportation Research Part A, Vol. 39, pp. 221-236.

Hoek, G., Brunekreef B., Goldbohm S., Fischer P. and Brandt P. 2002, “Association between mortality and indicators of traffic-related air pollution in the Netherlands: a cohort study”, The Lancet, Vol. 360, pp. 1203-1209.

Holland, W. W. (ed.) 1972, “Air pollution and respiratory disease”, Technomic Publication, USA.

Horowitz, J. and Louviere J. 1993, “Testing predicted choice against observations in probabilistic discrete choice models”, Marketing Science, Vol. 12. No. 3, pp. 270-279.

Page 252: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

References

234

Hosmer, D. W. and Lemeshow S. 2000, Applied Logistic Regression, 2nd edn, John Wiley & Sons Inc., New York, USA.

Huang, J., Haab T. C. and Whitehead J. C. 1997, “Willingness to pay for quality improvements: should revealed and stated preference data be combined?”, Journal of Environmental Economics and Management, Vol. 34, pp. 240-255.

Hunter, L. Watson I. and Johnson G. 1991, “Modelling air flow regimes in urban canyon”, Energy and Buildings, Vol. 15-16, pp. 315-324.

Hunter, l., Johnson G. and Watson I. 1992, “An investigation of three-dimensional characteristics of flow regimes within the urban canyon”, Atmospheric Environment, Vol. 26B, No. 4, pp. 425-432.

ICF Consulting Group 1997, Opportunities to Improve Air Quality Though Transportation Pricing, Office of Mobile Sources, EPA.

Industry Commission 1993, Urban Transport, Draft Report, Vol. 2, AGPS, Canberra.

Johnson, W., Ludwig F., Dabberdt W. and Allen R. 1973, “An urban diffusion simulation model for carbon monoxide”, Journal of the Air Pollution Control Association, Vol. 23, pp. 490-498.

Johannesson, M. 1996, “The willingness to pay for health changes, the human-capital approach and the external costs”, Health Policy, Vol. 36, pp. 231-244.

Johansson, O. and Schipper L. 1997, “Measuring the long run fuel demand of cars”, Journal of Transport Economics and Policy, Vol. 31, No. 3, pp. 277-292.

Jones, P. and Lucas K. 2000, “Integrating transport into ‘joined-up’ policy appraisal”, Transport Policy, Vol. 7, pp. 185-193.

Joumard, R. and Serie E 1999, Modelling of cold start emissions for passenger cars, INRETS Report LTE 9931, France, December 1999.

Kallos, G. 1998, “Regional/Mesoscale models”, in Urban Air Pollution – European Aspects, eds. Fenger J., Hertel O. and Palmgren F., Kluwer Academic Publishers, Netherlands.

Kamakura, W., Wedel M. and Agrawal J. 1994, “Concomitant variable latent class models for conjoint analysis”, International Journal of Research in Marketing, Vol. 11, pp. 451-464.

Ketzel, M., Wahlin P., Berkowicz R. and Palmgren F. 2003, “Particle and trace gas emission factors under urban driving conditions in Copenhagen based on street and roof-level observation”, Atmospheric Environment, Vol. 37, pp. 2735-2749.

Kocur, G. and Louviere J. 1983, “The magnitude of individual level variations in demand coefficients: A Xenia, Ohio, Case Example”, Transportation Research, Vol. 17, pp. 363-374.

Koppelman, F. S. and Wen C. H. 1998, “Alternative nested logit models: structure, properties and estimation”, Transportation Research Part B, Vol. 32, No. 5, pp. 289-298.

Page 253: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

References

235

Kunzli, N., Kaiser R., Medina S. 2000, “Public-health impact of outdoor and traffic-related air pollution: a European assessment”, The Lancet, Vol. 356, pp. 795-801.

Lettice, J. 2004, “London’s charge zone: blueprint for road pricing ‘success’?”, The Register, 17th February.

Leung, D. Y. C. and Williams D. J. 2000, “Modelling of motor vehicle fuel consumption and emissions using a power-based model”, Environmental Monitoring and Assessment, Vol. 65, pp. 21-29.

Lide, D. (ed.) 2005, CRC Handbook of Chemistry and Physics, CRC Press, USA.

Louviere, J. J., Hensher D. A. and Swait J. D. 2000, Stated Choice Methods: Analysis and Application, Cambridge University Press, UK.

Luk, J. and Hepburn S. 1993, New Review of Australian Travel Demand Elasticities, ARRB Report ARR 249, Australian Road Research Board, Vermont South.

Luk, J. 1999, “Electronic road pricing in Singapore”, Road and Transport Research, Vol. 8, No. 4, pp. 28-30.

Makridakis, S., Wheelwright S. and Hyndman R. 1998, Forecasting: methods and applications, 3rd edition, John Wiley & Sons, Inc., USA.

Matas, A. and Raymond J. L. 2003, “Demand elasticity on tolled motorways”, Journal of Transportation and Statistics, Vol. 6, No. 2/3, pp. 91-108.

Mathers, C., Vos T. and Stevenson C. 1999, “The burden of disease and injury in Australia”, Australian Institute of Health and Welfare, Canberra.

Mayeres, I. 2000, “The efficiency effects of transport policies in the presence of externalities and distortionary taxes”, Journal of Transport Economics and Policy, Vol. 34, Part. 2, pp. 233-260.

McGaughey, G., Desai N., Allen D., Seila R., Lonneman W., Fraser M., Harley R., Pollack A., Ivy J. and Price J. 2004, “Analysis of motor vehicle emissions in a Houston tunnel during the Texas Air Quality Study 2000”, Atmospheric Environment, Vol. 38, pp. 3363-3372.

McNulty, S. 2000, “Singapore pays price of enjoying streets free from traffic gridlock”, Financial Times, 4 August, p. 04.

Melbourne Mortality Study 2000, “Effects of ambient air pollution on daily mortality in Melbourne 1991-1996”, Environment Protection Authority Victoria, Melbourne.

Mensink, C., Lefebre F., Janssen L. and Cornelis J. 2006, “A comparison of three street canyon models with measurement at an urban station in Antwerp, Belgium”, Environmental Modelling & Software, Vol. 21, pp. 514-519.

Monitoring Plan for Western Australia, 2001, “National Environment Protection Measure for Ambient Air Quality”, State Government of WA.

Page 254: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

References

236

Morgan, G., Corbett S., Wlodarczyk J. and Lewis P. 1998a, “Air pollution and daily mortality in Sydney, Australia, 1989 through 1993”, American Journal of Public Health, Vol. 88, No. 5, pp. 759-764.

Morgan, G., Corbett S., and Wlodarczyk J. 1998b, “Air pollution and hospital admission in Sydney, Australia, 1990 through 1994”, American Journal of Public Health, Vol. 88, No. 12, pp. 1761-1766.

Morikawa, T. 1994, “Correcting state dependence and serial correlation in the RP/SP combined estimation method”, Transportation, Vol. 21, pp. 153-165.

Moustaki, I. and Papageorgiou I. 2005, “Latent class models for mixed variables with applications in Archaeometry”, Computational Statistics and Data Analysis, Vol. 48, pp. 659-675.

Neuburger, H. 1971, “User benefit in the evaluation of transport and land use plans”, Journal of Transport Economics and Policy, Vol. 5, No. 1, pp. 52-75.

Newman, P. and Kenworthy J. 1996, “The land use-transport connection”, Land Use Policy, Vol. 13, pp. 1-22.

Newman, P. and Kenworthy J. 1999, Sustainability and Cities – Overcoming automobile dependence, Island Press, Washingto D.C., USA.

Nijkamp, P., Rodenburg C. and Wagtendonk A. 2002, “Success factors for sustainable urban brownfield development – A comparative case study approach to polluted sites”, Ecological Economics, Vol. 40, pp. 235-252.

Nijkamp, P. and Shefer D. 1998, Urban transport externalities and Pigouvian taxes: a network approach, in Button, K. J. and Verhoef, E. T. (Eds.) Road pricing, traffic congestion and the environment: issues of efficiency and social feasibility, Edward Elgar, UK, pp. 171-189.

Noland, R. B. 2001, “Relationships between highway capacity and induced vehicle travel”, Transportation Research Part – A, Vol. 35, pp. 47-72.

Noland, R. B. and Cowart W. A. 2000, “Analysis of metropolitan highway capacity and the growth in vehicle miles of travel”, Transportation, Vol. 27, pp. 363-390.

Oke, T. 1988, “Street design and urban canopy layer climate”, Energy and Building, Vol. 11, pp. 103-113.

Ortuzar, J. D. D., Cifuentes L. A. and Williams H. C. W. L. 2000, “Application of willingness-to-pay methods to value transport externalities in less developed countries”, Environment and Planning A, Vol. 32, pp. 2007-2018.

Ortuzar, J. D. D. and Iacobelli A. 1998, “Mixed modelling of interurban trips by coach and train”, Transportation Research Part A, Vol. 32, No. 5, pp. 345-357.

Oum, T. H., Waters W. G. II. and Yong J. S. 1990, “A survey of recent estimates of price elasticities of demand for transport”, Policy Planning and Research, Working paper, World Bank.

Page 255: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

References

237

Palmgren, F., Berkowicz R., Ziv A. and Hertel O. 1999, “Actual car fleet emissions estimated from urban air quality measurements and street pollution models”, The Science of the Total Environment, Vol. 235, pp. 101-109.

Papakyriazis, A. and Papakyriazis P. 1998, “Optimal environmental policy under imperfect information”, Kybernetes, Vol. 27, pp. 137-154.

Perth Air Quality Management Plan 2000, Department of Environment, Western Australia.

Perth Metropolitan Transport Strategy 1995-2029, Department of Transport, Western Australia.

Perth Photochemical Smog Study 1996, Western Power Corporation and Department of Environmental Protection, Perth.

Petroeschevsky, A., Simpson R. W., Thalib L. and Rutherford S. 2001, “Association between outdoor air pollution and hospital admission in Brisbane, Australia”, Archives of Environmental Health, Vol. 56, No. 1, pp. 37-52.

Pigou, A. 1920, The Economic Welfare, First edition Macmillan.

Pokharel, S., Bishop G. and Stedman D. 2002, “An on-road motor vehicle emissions inventory for Denver: an efficient alternative to modelling”, Atmospheric Environment, Vol. 36, pp. 5177-5184.

Powers, D. A. and Xie Y. 2000, Statistical methods for categorical data analysis, Academic Press, USA.

Richardson, H. W. and Bae C. H. C. 1998, “The equity impacts of road congestion pricing” in Road pricing, traffic congestion and the environment, eds K. J. Button and E. T. Verhoef, Edward Elagr Publishing Ltd., USA, pp. 247-262.

Robinson, N., Pierson W., Gertler A. and Sagebiel J. 1996, “Comparison of MOBILE4.1 and MOBILE5 predictions with measurements of vehicle emission factors in Fort McHenry and Tuscarora mountain tunnels”, Atmospheric Environment, Vol. 30, pp. 2257-2267.

Saelensminde, K. 1999, “Stated choice valuation of urban traffic air pollution and noise”, Transportation Research Part D, Vol. 4, pp. 13-27.

Sapkota, V. 1999, “Efficient congestion tolls in the presence of unpriced congestion: A case with non-identical road users”, 23rd Australian Transport Research Forum, Perth, Western Australia.

Schwarz, N. and Hippler H. 1995, “Subsequent questions may influence answers to preceding questions in mail surveys”, Public Opinion Quarterly, Vol. 59, pp. 93-97.

Scoggins, A., Kjellstrom T., Fisher G., Connor J. and Gimson N. 2004, “Spatial analysis of annual air pollution exposure and mortality”, Science of the Total Environment, Vol. 321, pp. 71-85.

Page 256: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

References

238

Shiftan, Y. and Suhrbier J. 2002, “The analysis of travel and emission impacts of travel demand management strategies using activity-based models”, Transportation, Vol. 29, pp. 145-168.

Siddique, S. 2004, “Estimating the relationship between traffic and air pollution”, Road and Transport Research, Vol. 13, No. 1, pp. 76-84.

Simpson, R., Williams G., Petroeschevsky A., Morgan G., Denison L., Hinwood A., Neville G. and Neller A. 2005, “The short-term effects of air pollution on daily mortality in four Australian cities”, Australian and New Zealand Journal of Public Health, Vol. 29, No. 3, pp. 205-212.

Simpson, R. W., Williams G., Petroeschevsky A., Morgan G. and Rutherford S. 1997, “Association between outdoor air pollution and daily mortality in Brisbane, Australia”, Archives of Environmental Health, Vol. 52, No. 6, pp. 442-454.

State Government of WA 1995, Perth Metropolitan Transport Strategy 1995-2029.

Stopher, P. R. 2004, “Reducing road congestion: a reality check”, Transport Policy, Vol. 11, pp. 117-131.

Sugden, R. and Williams A. 1978, The Principles of Practical Cost-Benefit Analysis, Oxford University Press, New York.

Swait, J. 1994, “A structural equation model of latent segmentation and product choice for cross-sectional revealed preference choice data”, Journal of Retailing and Consumer Services, Vol. 1, No. 2, pp. 77-89.

Swait, J., Louviere J. and Williams M. 1994, “A sequential approach to exploiting the combined strengths of SP and RP data: application to freight shipper choice”, Transportation, Vol. 21, No. 2, pp. 135-152.

Taplin, J. H. E. 1982, “Inferring Ordinary Elasticities from Choice Elasticities”, Journal of Transport Economics and Policy, Vol. 16, No. 1, pp. 55-63.

Taplin, J. H. E. 2004, “Demand for hybrid cars: elasticity approaches and a fuzzy logic contribution”, 26th Conference of Australian Institute of Transport Research, Melbourne.

Taplin, J. H. E., Hensher D. A. and Smith B. 1999, "Preserving the symmetry of estimated commuter travel elasticities", Transportation research Part B, Vol. 33, pp. 215-232.

Taylor, M. A. P. and Taplin J. H. E. 1998, “Travel demand, energy and pollution impacts of alternative congestion pricing regimes”, International Conference on Transport into the Next Millennium, Nanyang Technological University, Singapore, September, 1998.

The Committee of the Royal College 1970, Air Pollution and Health, The Royal College of Physicians of London, London.

The World Bank 2001, Environment Matters at the World Bank, The World Bank, Washington D.C., USA.

Page 257: Development of Policies to Ameliorate the …...Development of Policies to Ameliorate the Environmental Impact of Cars in Perth City, Using the Results of a Stated Preference Survey

References

239

Train, K. E. 2003, Discrete Choice Methods with Simulation, Cambridge University Press, UK.

Tretvik, T. 2003, “Norway’s toll rings: Full scale implementations of urban pricing”, IMPRINT-EUROPE seminar, Budapest University of Technology and Economics, October 2003.

US Department of Transportations 2003, Transportation Statistics: Annual Report, Bureau of Transportation Statistics, October.

Vardoulakis, S., Fisher B., Gonzalez-Flesca N. and Pericleous K. 2002, “Model sensitivity and uncertainty analysis using roadside air quality measurements”, Atmospheric Environment, Vol. 36, pp. 2121-2134.

Vardoulakis, S., Fisher B., Pericleous K. and Gonzalez-Flesca N. 2003, “Modelling air quality in street canyons: a review”, Atmospheric Environment, Vol. 37, pp. 155-182.

Weatherley, N. and Timmis R. 2001, “The atmosphere in England and Wales: an environmental management review”, Atmospheric Environment, Vol. 35, pp. 5567-5580.

Wilson, R. W. 1992, “Estimating the travel and parking demand effects of employer-paid parking”, Regional Science and Urban Economics, Vol.22, No. 1, pp. 133-145.

Willson, R.W. and Shoup D.C. 1990, “Parking subsidies and travel choices: Assessing the evidence”, Transportation, vol.17, no.2, pp.141-157.

Wong, T., Tam W., Yu T. and Wong A. 2002, “Association between daily mortalities from respiratory and cardiovascular diseases and air pollution in Hong Kong, China”, Occupational and Environmental Medicine, Vol. 59, No. 1, pp. 30-35.

Yang, H., Meng Q. and Lee D. H. 2004, “Trail-an-error implementation of marginal cost pricing on networks in the absence of demand functions”, Transportation Research Part B, Vol. 38, pp. 477-493.