D3 – Marginal cost case studies for road and rail transport

GRACE D3 – Marginal cost case studies for road and rail transport

1

SIXTH FRAMEWORK PROGRAMME PRIORITY [Sustainable surface transport]

Call identified: FP6-2003-TREN-2

GRACE Generalisation of Research on Accounts and Cost Estimation

D3 – Marginal cost case studies for road and rail transport

Version 1

November 2006

Authors: Gunnar Lindberg (VTI) with contribution from partners

Contract: FP6-006222 Project Co-ordinator: ITS, University of Leeds

Funded by the European Commission

Sixth Framework Programme

GRACE Partner Organisations University of Leeds; VTI; University of Antwerp; DIW; ISIS; Katholieke University of Leuven; adpC; Aristotle University of Thessalonika; BUTE; Christian-Albrechts University; Ecoplan; IER University of Stuttgart; TNO

Inro, EIT University of Las Palmas; University of Gdansk


2

GRACE FP6-006222 Generalisation of Research on Accounts and Cost Estimation Marginal cost case studies for road and rail transport This document should be referenced as:

Lindberg, G (2006), Marginal cost case studies for road and rail transport Deliverable D 3, GRACE. Funded by Sixth Framework Programme. ITS, University of Leeds, Leeds, November 2006 24 November 2006 Version No: 1 Authors: as above. PROJECT INFORMATION Contract no: FP6-006222: Generalisation of Research on Accounts and Cost Estimation Website: www.grace-eu.org Commissioned by: Sixth Framework Programme Priority [Sustainable surface transport] Call identifier: FP6-2003-TREN-2 Lead Partner: Institute for Transport Studies, University of Leeds (UK) Partners: UNIVLEEDS, VTI, University of Antwerp (UA), DIW, ISIS, KUL, adpC, AUTH, BUTE, CAU, Ecoplan, IER, TNO Inro, EIT, Gdansk BEI (Büro für Evaluation + Innovation) is a subcontractor to DIW.

DOCUMENT CONTROL INFORMATION

Status: Final submitted Distribution: European Commission and Consortium Partners Availability: Public (on acceptance by European Commission) Filename: D3 version 1 Quality assurance: Co-ordinator’s review: Chris Nash

Signed: Date:


3

Table of contents 0 Summary ............................................................................................................................ 5

0.1 Infrastructure cost ....................................................................................................... 5 0.2 Road congestion and rail scarcity ............................................................................... 6 0.3 Accidents .................................................................................................................... 7 0.4 Air pollution and Greenhouse gases ........................................................................... 7 0.5 Noise ........................................................................................................................... 8 0.6 Sensitive areas ............................................................................................................ 8

1 Introduction ...................................................................................................................... 10 2 Infrastructure cost ............................................................................................................. 12

2.1 Methodology and definition ..................................................................................... 12 2.1.1 Econometric approach ...................................................................................... 13 2.1.2 Engineering, lifetime or duration approach ...................................................... 14

2.2 Overview of Case studies on infrastructure cost ...................................................... 15 2.2.1 Road ................................................................................................................. 15 2.2.2 Rail ................................................................................................................... 17

2.3 Results ...................................................................................................................... 20 2.3.1 Elasticity ........................................................................................................... 20 2.3.2 Average and marginal cost. .............................................................................. 27 2.3.3 Discussion ........................................................................................................ 30

2.4 Conclusions .............................................................................................................. 36 3 Congestion and scarcity ................................................................................................... 38

3.1 Road ......................................................................................................................... 38 3.1.1 Overview .......................................................................................................... 42 3.1.2 Results .............................................................................................................. 43

3.2 Rail ........................................................................................................................... 45 3.2.1 Introduction ...................................................................................................... 45 3.2.2 The case study .................................................................................................. 45 3.2.3 Results .............................................................................................................. 46

3.3 Conclusions .............................................................................................................. 48 4 Accidents .......................................................................................................................... 50

4.1 Methodology and definitions ................................................................................... 50 4.1.1 Definition of external cost of accidents ............................................................ 50 4.1.2 Methodology .................................................................................................... 52

4.2 Valuation of accidents .............................................................................................. 53 4.3 Risk perception ......................................................................................................... 55 4.4 The risk elasticity ..................................................................................................... 56 4.5 Insurance cost ........................................................................................................... 57 4.6 Conclusions .............................................................................................................. 58

5 Air pollution and Greenhouse gases ................................................................................. 60 5.1 Road Transport and Air pollution ............................................................................ 60

5.1.1 Description of Case Studies ............................................................................. 60 5.1.2 Emissions from road vehicles .......................................................................... 62

5.2 Greenhouse gases ..................................................................................................... 63 5.3 Rail Transport ........................................................................................................... 63 5.4 Results ...................................................................................................................... 64

6 Noise ................................................................................................................................. 66 6.1 Noise impacts ........................................................................................................... 66 6.2 Valuation of Annoyance ........................................................................................... 68

7 Sensitive areas .................................................................................................................. 70


4

7.1 Definition and indicators .......................................................................................... 70 7.2 Cost categories ......................................................................................................... 70

7.2.1 Air pollution ..................................................................................................... 71 7.2.2 Noise ................................................................................................................. 71 7.2.3 Visual intrusion ................................................................................................ 72 7.2.4 Accidents .......................................................................................................... 72 7.2.5 Infrastructure costs ........................................................................................... 72

7.3 Conclusion ................................................................................................................ 73 8 References ........................................................................................................................ 74


5

0 Summary This report presents the Case studies made on road and rail in the GRACE project. Each Case study contains a huge amount of specific information. This report tries to summarise the main results from the Case studies and presents an overview of each study. The original studies are presented in Annex. Below we summarise the main conclusions for each cost category considered.

0.1 Infrastructure cost • The cost-elasticity with respect to the traffic-output describes the relationship between

average cost and marginal cost such that Marginal Cost = Elasticity *Average Cost. o The elasticity for road infrastructure cost decreases as the measure changes

from renewal to maintenance and to operation. The average elasticity for renewal cost is between 0.58 and 0.87, for an aggregate of renewal and maintenance cost the elasticity is between 0.48 and 0.58 while the elasticity for only maintenance and operation are from 0.12 to zero.

o The elasticity for rail infrastructure cost is lower than the elasticity for road and doesn’t show the same difference between different measures. The average elasticity is between 0.26 and 0.30 for an aggregate of renewal and maintenance, for maintenance it is between 0.20 and 0.24 and for operation or short term maintenance it is 029 to 0.32.

o The majority of the studies suggest that the elasticity decreases with increased traffic. Thus highly used infrastructure has a lower elasticity than low volume infrastructure. All elasticities reported above are from the average traffic in the studies.

• The operation or short term maintenance is related to total trainkm or total vehiclekm while the renewal and maintenance usually are related to gross tonnekm or HGVkm.

o Few of the studies have been able to test which type of traffic influences the infrastructure cost. In general, this has been decided a priori based on other information.

• Most of the studies use an econometric approach with paneldata. However, a minority of the studies did use paneldata models but use pooled ordinary least square to estimate the cost function.

o In two studies a duration model is used where a function of the lifetime of a road pavement or railtrack is estimated. The result can be used to derive a marginal renewal cost. The rail study gave results in line with the econometric study and supported the conclusion drawn from the econometric studies that there indeed exists a marginal cost related to renewal on railways. The result was similar between the two approaches. However, the road study suggested a very low effect of traffic on the observed lifetime of a pavement. A possible explanation with some support is that the authority predicts the higher traffic volume when deciding on the pavement thickness. The marginal cost is thus not found in observed lifetime but in increased cost of the measures taken.

• The average cost is less homogenous than would be expected. o For road studies the average renewal cost is 0.036 €/HGVkm in the Swedish

all roads study and 1.59 €/HGVkm for the German motorway study. The Polish study allocated the renewal cost to all vehicles and suggests a cost of


6

0.21 €/vehkm. For the aggregate of renewal and maintenance measures the average cost is 0.059 €/HGVkm in the Swedish study. Operation is in the Swedish study allocated to all vehicles and has an average cost of 0.024 €/vkm.

o In the rail infrastructure cost study maintenace and renewal has an average cost of 0.0028 €/Gtkm in Sweden and 0.0036 €/Gtkm in Switzerland. The maintenance only average cost is 0.0021€/Gtkm in Sweden and 0.0022 €/Gtkm in Switzerland. The UK study shows an average maintenance cost of 0.0052 €/Gtkm. Operation has an average cost of 0.153 €/trainkm in Sweden. The Hungarian study suggests a cost for ‘train movement’ 3.5 €/trainkm and 0.0041 €/Gtkm.

• The marginal cost follows from the elasticities and the average costs. o The marginal cost on roads has a huge variability depending on the huge

variability in average cost. The cost on German motorways is 1.39 €/HGVkm for renewal only. Corresponding cost for all Swedish paved roads are 0.032 and 0.12 in Poland. The Swedish results for gravel roads is 0.236 €/HGVkm. Aggregating renewal and maintenance generates a marginal cost of 0.040 €/HGVkm in Sweden and 0.13 €/HGVkm in Poland. Operation is not associated with traffic volume according to the Swedish case study.

o The marginal cost in the rail sector is 0.00070 €/Gtkm inSweden for renewal and maintenance and 0.00097 €/Gtkm in Switzerland. Maintenace only has a cost of 0.00031 €/Gtkm in Sweden and 0.00045 €/Gtkm in Switzerland. The marginal cost in UK is estimated to 0.002 €/Gtkm. Operation has a marginal cost of 0.054 €/trainkm in Sweden. The Hungerian study concludes on a marginal cost of 0.22 €/trainkm for train movements.

0.2 Road congestion and rail scarcity • The main focus of the road congestion case study was to identify reasons why

previous case studies show such a huge variability in road congestion costs. It was found that these differences can be variously attributed to:

o differences in the definition of “optimal” tolls – the term is often quite loosely applied. For example; the term sometimes relates only to congestion tolls (rather than covering other externalities), sometimes allows for the cost of implementation of the tolls (and sometimes not), and sometimes relates only to simple tolls - such as cordons (rather than tolls which vary in space and time).

o differences in the way that optimal tolls (however defined) are calculated. For example, do they fully reflect the behaviour of travellers at the margin or are they derived from a theoretical representation of the marginal impacts?

o differences in the nature of the cities being studied. Factors which are particularly likely to influences the result include the degree of congestion, the availability and attractiveness of alternative modes, the drivers’ tolerance of congestion, and the capacity of the network to absorb additional demand.

o differences in the valuation of different externalities – perhaps reflecting different values of time and resource costs.

o differences in the models used to estimate system performance. • For rail transport we find that a substantial peak scarcity charge per slot is justified;

o the off-peak charge would only be 10% of this level. The results seem to confirm the view that existing variable charges for the use of infrastructure on


7

key main lines where capacity is scarce are too low as a result of the neglect of scarcity in the charges set.

o The private slot value is different from the social slot value which indicates problems with a simple market based solution. This result is an effect of high congestion cost on the road network in the CS and that this is not internalized in a road pricing regime.

o The institutional arrangements behind franchising suggest that where the dominant operator is a franchisee, data may be available for the rail regulator to perform a scarcity estimate along the lines of the GRACE case study. This is certainly true in Britain.

0.3 Accidents • This Case study only consists of an overview and state-of-the-art survey. The result is

thus not based on any new research made within the GRACE project. The following conclusions can be drawn;

o A growing consensus on the method to estimate the value of statistical life (VSL) seems to emerge. The HEATCO project suggests specific values for each Member State.

o Nevertheless, the research on VSL continues with the aim to explore the numerous biases that have been found to potentially affect the estimates.

o On the question of the proportion of internal and external cost and especially the perception of road users risk no new conclusions can be drawn. This is still an area of large uncertainty.

o However, assuming something on the perceived cost, actual databases can be used to estimate the proportion of internal cost.

o There is still no consensus on the risk elasticity. Surprisingly, many studies find decreasing risk with increasing traffic volume. This could be a problem of the studies or behaviour effects. If we do not control for infrastructure quality, we may find that roads with higher expected traffic volume are designed with a higher traffic safety standard. In addition, road users may react to a perceived increased risk by driving more carefully and slower. This is an unobserved cost component that would increase the cost.

0.4 Air pollution and Greenhouse gases • Four case studies for road transport within densely built areas have been conducted.

They are expected to complete the picture on air pollution from existing studies and to analyse the variations of environmental costs and the driving parameters. Assessing data availability and due to the fact that a broad range of European countries and local meteorological conditions should be considered, the cities selected for this purpose were Berlin, Prague, Copenhagen and Athens.

o The results show that for all vehicle types the higher marginal costs due to airborne emissions correspond to the city of Athens, followed by Berlin, Copenhagen and Prague in that order.

o The factors that seem to be more relevant for these results are the wind speed and the population density. The high share of low wind speeds for the Athenian area together with a population density close to 20 000 hab/km2 in some zones, leads to a pollutant exposure of the population which is about a factor of two higher compared to the other cities.


8

o Petrol cars cause lower cost per vehicle kilometre compared to diesel cars as they emit much less fine particles, leading to lower health impacts.

• A European abatement cost of €20 per tonne of CO2 represents a central estimate of

the range of values for meeting the Kyoto targets in 2010 in the EU based on estimates by Capros and Mantzos (2000).

o They report a value of €5 per tonne of CO2 avoided for reaching the Kyoto targets for the EU, assuming a full trade flexibility scheme involving all regions of the world. For the case that no trading of CO2 emissions with countries outside the EU is permitted, they calculate a value of €38 per tonne of CO2 avoided. It is assumed that measures for a reduction in CO2 emissions are taken in a cost effective way. This implies that reduction targets are not set per sector, but that the cheapest measures are implemented, no matter in which sector.

o Recent work has confirmed the assumption that emissions in future years will have greater total impacts than emissions today.

o For application in GRACE we recommend using a range of €14 to €51 (with a central value of €22 per tonne of CO2- equivalent emission in the period 2000 to 2009). These shadow prices were derived from Watkiss et al. (2005b), converting from ₤2000/t C to €2002 (factor prices).

0.5 Noise • This case study is in this report only based on a state-of-the-art review. Subsequent

workpackages will present new estimates. o Existing estimates show considerable non-linearities of marginal noise cost

with background noise levels. Based on case studies in Berlin, Stuttgart and Helsinki the following conclusions can be drawn.

In Berlin the average number of persons per road kilometre affected by noise is slightly higher than in Stuttgart. However, the costs are more than a factor of three lower due to the much higher number of vehicles and higher speeds on Frankfurter Allee leading to a higher background noise level. In Helsinki the population density along the route considered is lower than in Berlin and Stuttgart, furthermore the average distance from buildings is higher – leading to lower noise costs.

0.6 Sensitive areas • The impact pathway has been used to estimate a factor that relates the cost in Alpine

regions to the cost in ‘flat’ regions. o For air pollution only local effects are relevant as regional effects would be the

same in both regions. The biggest effect is found from the topographical and meteorological conditions.

o An alpine region would have a cost 5 times higher than a flat area for local air pollution for road transport with a slightly higher factor for cars than for HGV. If this factor is to be applied on all pollutions (local and regional) the factor would be around 2.5.

o The corresponding factor for rail is around 3.5.


9

o The noise cost is also estimated to be about 5 times higher in road transport and 4 times for rail transport.

o The number of accidents is higher per kilometre in Alpine regions suggesting a factor of 1.2 for road transport.

o The infrastructure maintenance cost is for the road sector about 4.5 times higher and for rail transport 1.4 times.

o In addition, a factor for visual intrusion is suggested to be around 10 due to the specific alpine conditions. This has however, no corresponding marginal cost.


10

1 Introduction The GRACE project aims to support policy makers in developing sustainable transport systems by facilitating the implementation of such pricing and taxation schemes that reflect the costs of infrastructure use. In order to carry out this mandate, five major areas of research are covered:

• New case study research to address gaps in the existing level of knowledge of marginal social costs for road, rail, air and waterborne transport.

• Development and refinement of methods to enable the use of transport accounts as monitoring instrument for the implementation of transport pricing reform in an enlarged Europe.

• Innovative research on the appropriate degree of complexity in transport charges. • Guidance on the marginal social cost of the different modes of transport in specific

circumstances and on simple and transparent methods for determining charges. • Modelling the broad socio-economic impacts of pricing reform.

This deliverable is dedicated to the first area of GRACE. It addresses the question of marginal costs of infrastructure use in the road and rail sector. The deliverable is the outcome of a number of individual case studies conducted within a general framework. The individual Case studies are presented in an Annex. The Case studies covers following cost elements (Table 1). Table: 1 Cost elements covered in this Deliverable

Chapter Cost element 2 Infrastructure cost 3 Congestion and scarcity cost 4 Accidents 5 Air pollution and greenhouse gases 6 Noise 7 Costs in sensitive areas.

The infrastructure cost case studies (Chapter 2) explore the marginal cost of motorways in Germany and a broader set of roads in Poland and Sweden. Lifetime model is developed to analyse the renewal costs on Swedish roads. The same approach is used on Swedish railways together with an econometric approach. Pioneering work in the same area is made in Switzerland, UK and Hungary. In the area of congestion and scarcity cost (Chapter 3) the aim is to clarify the variability in current estimates on congestion cost and suggest a more unified approach. Novel research is conducted in the area of rail scarcity where a modelling approach is used to derive estimates. The area of Accident cost (Chapter 4) focus on a state-of-the-art survey and the insurance market is included in the approach. Numerous studies on air pollution and green house gases exist which this deliverable expands upon, with the addition of new case studies (Chapter 5), while attempting to create a clearer picture of transferable results. The marginal cost of noise is analysed with a review (Chapter 6). Finally, the environmental cost of transport in sensitive areas has been discussed in policy documents and the concept is here further developed (Chapter 7).


11

Table: 2 Case studies 1.2A Marginal motorway infrastructure costs for Germany DIW 1.2B I Road infrastructure cost in Sweden - renewal VTI 1.2B II Road infrastructure cost in Sweden - econometric VTI 1.2C Road infrastructure cost in Poland UG and DIW 1.2D I Rail infrastructure cost in Sweden - econometric VTI 1.2D II Rail infrastructure cost in Sweden – renewal VTI 1.2E Rail infrastructure cost in Switzerland ECOPLAN 1.2F Rail infrastructure cost in Hungary BUTE 1.2G Rail infrastructure cost in UK ITS 1.3A Estimating Rail Scarcity Costs – modelling ITS/VTI 1.3B Road congestion ITS/AUTH 1.4 Accidents – State-of-the-art survey VTI 1.5A Urban case studies for road and rail IER 1.6A Noise - Urban case studies for road and rail IER 1.6B Noise - Extra-urban case studies for road and rail IER 1.7 Environmental costs in sensitive areas ECOPLAN/IER The single Case studies are presented as appendices to this main report.


12

2 Infrastructure cost The short-run marginal infrastructure cost related to an additional vehicle or train consists of four components; first, the increased wear of the infrastructure leading to additional routine maintenance, secondly, the damage to the infrastructure leading to earlier future periodic maintenance. A third component is the increased cost inflicted on other vehicles/trains. Fourthly, congestion or scarcity cost, and corresponding peak load pricing, is in many sectors a necessary part of understanding and developing cost allocation or pricing principles in the transport sector. This chapter focuses on the two first categories; routine maintenance (including operation) and renewal while congestion and scarcity is covered in Chapter 3. The remaining part of this section is organised as follows; section 2.1 discusses the methodology and definitions used, section 2.2 describes shortly each Case Study while the results are compiled in the following section 2.3. Next section, 2.4, summarise the existing literature and benchmark the GRACE results against other studies. In the last section 2.5 the conclusions are presented.

2.1 Methodology and definition The focus is exclusively on infrastructure costs and no costs for running the traffic are included in the analysis. It is thus oriented towards a transport policy which separates infrastructure management from the traffic decisions. We use here three different categories of infrastructure costs. Infrastructure operation, for example snow removal, is defined to have a very short time horizon and is undertaken to keep the infrastructure open and functioning for traffic. Maintenance activities have a longer time horizon and are preventive measures to avoid degradation. Finally, renewal activities have a longer time horizon and are undertaken to bring the infrastructure back to its original condition1. New construction and improvements are not included in the marginal cost analyses but have of course a link to the congestion and scarcity cost discussed in chapter 3. Table: 3 Definitions Measure Purpose Other name Example Operation To keep the infrastructure

open for traffic Snow removal

signals Maintenance Preventive measures

against deterioration of the infrastructure or corrective measures to repair minor damages

Routine maintenance, Preventive maintenance, annual maintenance

Crack sealing, patching, shoulder maintenance etc Tamping, ballast cleaning etc

Renewal Bringing the infrastructure back to its original condition

Periodic maintenance, Structural repair, structural maintenance

Repair, reinforcement and resurfacing Track renewal

1 In European standards (EN 13306:2001) ‘maintenance’ is defined as ‘combination of all technical, administrative and managerial actins during the life cycle of an item intended to retain it in, or restore it to, a state in which it can perform the required function’. Both the measure maintenance and renewal in the table is thus a part of the general maintenance term.


13

Two distinct approaches have been used; an econometric approach and a lifetime approach.

• In the econometric approach a cost function is estimated to describe the variability of costs as a function of infrastructure characteristics, geographical, climate information and finally traffic volumes. The observed correlation between traffic and cost is then the base to estimate the marginal cost. In this approach information on expenditure is collected over a number of years. The observation unit is in some cases a single road or rail segment while other studies use information over a larger network, usually a Maintenance Delivery Unit (MDU) where the maintenance work has been contracted out. The expenditure is expressed separately for the different measures related to operation, maintenance or renewal/repair. In some cases the expenditure information is constructed from physical information on measures taken.

• The duration model, or engineering approach, is based on the lifetime of a piece of infrastructure and is used to calculate renewal cost. This approach does not require expenditure information but lifetime information. A lifetime or duration function is estimated as a function of infrastructure characteristics, geographical and climate information and traffic volumes as in the econometric approach. The change in the lifetime as the traffic changes will affect the present value of the future renewal costs and is thus the basis for the marginal cost calculation.

2.1.1 Econometric approach The dominant approach has been the econometric approach. To fix the idea we use a double log functional form:

)ln()ln()ln()ln(...)ln()ln()ln( 2212

2111 iiBiBiAiAii PIQQQQC δγββββα ++++++= (1)

Where • iC is the cost per annum for section or zone i; • iQ is outputs for section or zone i ; here in terms of traffic with vehicles of different

types (A and B). Above is also a squared term included; • iI is a vector of fixed input levels for section or zone i – these include the

infrastructure variables i.e. track length, track quality or pavement type etc; • iP is a vector of input prices.

Given that we succeed in the estimation of the function in (1) the marginal cost can be derived as the product of the average cost (AC) and the cost elasticity ε. In the example above we included the square of the traffic variable QA which means that the elasticity with respect to vehicle type A is non-constant if β11 is non-zero.

)ln(2lnln

111 AAA

AA Q

QdCd

dQQ

CdC ββε +=== (2)

The average cost is simply the cost C divided by the relevant output variable Q. However, the average cost will depend on the traffic volume Q. Usually this is expressed as the mean in the sample. But it should be clear that the marginal cost will usually depend on the traffic volume.

[ ]A

AA Q

CQQCACMC )ln(2 111 ββεε +=== (3)


14

Two additional observations should be highlighted.

• First, while the theoretical specification above includes different outputs in terms of different vehicles the reality is more problematic. In general, the correlation between different outputs is so strong that the econometric model can not distinguish between the effect from, for example, different vehicle types. This means that we a priori need to decide on only one output variable to use in a study.

• Secondly, input prices are often assumed to be constant between sections or areas and thus not included in the studies.

2.1.2 Engineering, lifetime or duration approach This approach starts with the observation that long time series on expenditure information is difficult to find. The basic assumption is that the length of an interval between two renewal measures depends on the aggregate of traffic that has used a certain section. Existing literature (Newbery 1988b, Small et al. 1989) focuses on road and assumes that the number of standard axles that can use a road before the pavement has to be renewed is a design parameter of road construction. Lindberg (2002) however makes use of the fact that the number of standard axles which the road can accommodate after all is a function of the actual, not the predicted traffic volume. Adding or subtracting vehicles to the original prediction will therefore affect the timing of a reinvestment and there is, consequently, a marginal cost associated with variations in traffic volume. The lifetime of a pavement – the number of years between resealing – (T) is a function of the constant annual number of vehicles that pass the infrastructire (call it QA) and the strength of the infrastructure where Θ denotes the number of vehicles the infrastructure can accommodate and m indicates the climate dependent deterioration:

mTeQ − ⎥⎦

⎤⎢⎣

⎡Θ=

AQ)( T (4)

Each renewal of the infrastructure has a cost of C. The first renewal takes place at year 0. We can then calculate the present value of an infinite number of renewals as (5) if considering the cost from the perspective of the initial overlay (PVC0); r is the relevant discount rate. To study the effect of annual traffic on the cost the annualised present value of an average piece of infrastructure (ANCave) can be expressed as (6).

)e-(1

C PVC lim )eee(1 C PVC rT-on

Tn r -r2T-rT-o =+…++=

→∞ (5)

)e-(1

C e PVC rT-t)-r(T-

t = (6)

The marginal cost caused by shortening the renewal intervals due to higher traffic loads can be obtained by differentiating the annualised present value of the infrastucture with the annual traffic volume. By using the deterioration elasticity ε – the change of lifetime due to higher traffic loads (equation (7)) – and the definition of average costs AC=C/T=C/QT the marginal costs for an average road MCAverage can be expressed as (8).


15

TQ

dQdT

=ε . (7)

AC - MCAverage ε= (8)

2.2 Overview of Case studies on infrastructure cost In GRACE nine different case studies have been carried out. The table below summarise the main characteristics of the studies. The majority of the studies use an econometric approach but two studies try with an engineering or duration approach. The focus is on maintenance and renewal cost even if operation is included in some studies. The majority of studies use data on road or rail section while two studies have data on a more aggregate level (Maintenance Delivery Units - MDU). The time span is between 55 years and 1 year. Although most studies collect data as a panel database only one study succeeds in panel modelling (Case study 1.2D). The definition of the cost function is for most studies such that it allows for variable elasticities. Table: 4 Infrastructure cost Case Studies Case study Mode Method Measure Data Model Elasticity

Eco

nom

etri

c

Eng

inee

ring

Ope

ratio

n

Mai

nten

ace

Ren

ewal

Sect

ion

MD

U

Tim

e

Obs

erva

tions

pe

r ye

ar

Infr

astr

uctu

re

type

POL

S

Pane

ldat

a

Dur

atio

n m

odel

Con

stan

t E

last

icity

Var

iabl

e el

astic

ity

1.2A German

Road 1980-1999 221 Motorway

1.2B Sweden I

Road 1998-2002 145 All

1.2B Sweden II

Road 1950-2005 (142331) All

1.2C Poland

Road 2002-2004 264 National

1.2 D Sweden I

Rail 1999-2002 185 All

1.2D Sweden II

Rail 1999;2006 1400 All

1.2E Switzerland

Rail 2003-2005 371 All

1.2F Hungary

Rail 2001-2005 1 723

All

1.2G UK

Rail 2005-2006 53 All

Note: The Hungarian CS uses a different approach based on the national network (1 observation) and 723 cost categories (activities).

2.2.1 Road For road infrastructure cost four case studies (CS) have been carried out. The CS in Germany and Poland have a similar approach and are based on paneldata over expenditure by road section. In Sweden two CS have been carried out with two different approaches; one based on paneldata on expenditure in maintenance delivery units (MDU), i.e. organizational units which take care of the road maintenance in a limited geographical area, and the other on observed lifetime of pavements which results in deterioration elasticities.


16

1.2A - Germany The German CS includes one production cost oriented study (Model I) and one study focused on the influence of traffic (Model II). The studies are based on detailed physical information of renewal measures on (West) German motorway sections during 20 years. The database consisted originally of 1830 sections but only sections where renewal has taken place during the period were included in the database (221 sections)2. Based on unit costs for each type of construction3 a database on renewal expenditure was constructed. The annual data was summed up over the period which resulted in a cross-sectional database of 221 observations of total renewal expenditure over the period 1980 to 1999. The explanatory variables include factor input prices (regional), type of material used, regional dummies (i.e. information on in which region the road is located), annual daily traffic volume (AADT) of passenger cars and goods vehicles and climate data. Additional information on the age of the motorway and previous renewal expenditures was collected but excluded from the final model due to partly wrong sign and lack of significance in the statistical analysis.

1.2B - Sweden – econometric models The cost data is based on the Swedish National Road Administrations accounting system (VERA). The observation unit are 145 small areas, so called Maintenance delivery units (MDU), which were established by the Road Administration when maintenance contract were procured on the market. This means that we can find information on actual maintenance and operation expenditure by MDU directly in the accounting system. For some cost categories detailed analyses have been necessary to identify the right MDU for the expenditure. The GRACE study is focused on maintenance for paved and gravel roads, winter operation as well as ordinary operation of paved and gravel roads. The database covers the period 1998 to 2002. The traffic information for passenger cars and heavy goods vehicles for paved and gravel roads are also collected from the Road Administration and aggregated over the MDU’s. In addition data on road length of paved and gravel roads as well as roads of different categories for each MDU are collected.

1.2B – Sweden – duration analysis This CS takes an alternative approach to estimate the renewal costs. The basic idea is to analyse the interval between pavements renewal measures for road sections and find the influence of traffic on these intervals. This influence is expressed as deterioration elasticity, which accounts for the percentage change in lifetime years due to a 1% change in traffic volume. Together with information on pavement cost the deterioration elasticity is used to derive the marginal renewal cost. The basic database of this CS is extensive: it includes observations on every completed renewal interval in the Swedish national road network from 1928 to 2005. The dataset also contains information on the traffic of passenger cars and HGVs. In addition, information on speed limits, road width, road type and in which region the road is located are available. The CS excludes gravel roads and limits the period to roads which have been repaved after 1951. The subset consists of 142331 complete observations4. The CS develops the theory to take into account random elements in the lifetime function5. A

2 The omission of non-renewed sections could bias the results? 3 Bituminous concrete, bituminous mastic asphalt, bitumen binder, mastic asphalt with crushed materials, cement concrete, thin layer, others. 4 of which 46464 are censored in the sense that the last interval has not yet ended. 5 The lifetime is here assumed to follow a Weibull distribution.


17

priori we expect the lifetime to decrease with more traffic, higher speed, narrow lanes, rougher climate, higher road category which has a lower threshold value and design quality. For the latter a proxy based on intervals of traffic volume (traffic class) has been used.

1.2C - Poland The Polish CS is based on a database with all sections of national roads where renewal works have been conducted between 2002 and 2004. Renewals include rebuilding, strengthening, refurbishing or modernization. The final database consists of 264 sections with an average length of 6 km. For these sections also maintenance cost has been added6. The explanatory variables are AADT for motorbikes, passenger cars, LGVs, HGVs with and without trailer and buses. The study also includes regional dummies, information on location of the road (urban/non-urban) and whether the road has one or two lanes. The Polish database covers a short time period (3 years). As a first approach the data were treated as in the German CS, with annual expenditures summarised over the three years and the database then collapsed to a cross-section database with 264 observations. The analyses based on this dataset were not successful due to interdependencies and low significance of the estimates. Instead a dataset based on the assumption that the sections were independent between years was constructed. This dataset has 264*3=792 observations. To solve the problem that renewals of a section occurs about every ten year and the database only covers three years the observed information on renewal expenditure was divided by the average road lifetime in the region generating some kind of average renewal cost per year. Finally, the cost is expressed per kilometre. The result of this second approach was judged as reasonable although the first approach would have been favoured had it been possible.

2.2.2 Rail For rail infrastructure costs five case studies were carried out. The studies in UK, Switzerland and one of the Swedish studies have similar approaches based on econometrics. The second Swedish study is an attempt to use the duration approach in railways while the Hungarian study has a much different statistical approach.

1.2D - Sweden The collected information includes operation, maintenance and renewal costs on 185 track sections over the period 1999 to 2002 and originates from the Rail Administration accounting system. Infrastructure operation is dominated by snow removal (80 %). The traffic variables have been the most difficult information to collect and include information on gross tonnes and number of trains per section by passenger and freight trains over the four years. Combined with information on track length, this leads to traffic volumes expressed as gross ton kilometres and train kilometres. Dividing tonnage by the number of trains yields an average train weight per track section, for total traffic, and per freight and passenger train. Some track sections have no passenger services so average weight is not computable for these observations. We have also computed two freight traffic ratios, total freight train/gross tonne kilometres divided by total train/gross tonne kilometres. A wide range of technical features of the track are collected including for example rail weight, curvature, joint, switches etc. In addition regional dummies (i.e. where in which region the track is located) are available. It should however be noted that the majority of the variables 6 The omission of sections where renewal has not taken place may bias the result.


18

show very little variation over time at a track section level and are highly correlated with each other. This will most likely give rise to multicollinearity problems in our model estimations. The dataset consists thus of a panel database. In contrast to the other CS and to previous Swedish CS on the same dataset (Andersson 2006) this CS uses and succeeds with a panel data model. The panel data allows variables to vary in both the i (individual) and t (time) dimensions. A standard pooled ordinary least square regression (POLS) assumes only a common constant term. The most basic forms of panel data models can be grouped into fixed effects (FE) and random effects (RE) models. The FE model uses an individual specific constant and possibly also a time specific constant. A feature of the FE estimator is that it requires a large number of parameters to be estimated, which consumes degrees of freedom when there are a large number of individual effects to account for. The RE model does not come with this feature as it (like POLS) only estimates a common constant but assumes individual- and time-specific random elements.

1.2D - Sweden – duration model Following the idea developed for the road sector and presented in the Swedish road duration model (1.2B) this case study endeavours to apply this same approach to the railway sector. Two data samples from the track information system have been matched, the first from 1999 and the second from the end of 2005. Changes between these years can be identified through changes in the infrastructure information. From the information of the year when the track is laid, we can derive an age variable for each observation. A change during the study period results in two observations, one for the initial track that is replaced (this observation is uncensored, i.e. it contains a full lifecycle) and one for the new track that, if no change is observed, is registered as a censored observation (i.e. not a full lifecycle) at the end of 2005. Since no comprehensive traffic database exists, we need to create a time series of data based on known information. Andersson (2006) has created a database for the period 1999 – 2002. This database is extended back to 1993 based on track segment information from the main freight and passenger operating companies during this period. From 1993 and back, we extrapolate the most recent existing information back to the year the track is laid, adjusted for annual traffic growth from Swedish official statistics (SIKA Officiell Statistik). From 2002 and forward we extrapolate using traffic growth coefficients from Banverket. This method gives an estimate of annual track segment traffic for the time window of our observations. The data sample consists of 1,631 observations but missing age and traffic data reduces the number of observations to 1,493 out of which 1,333 observations are censored.

1.2E - Switzerland The data used is based on the whole railway network of Switzerland including all main lines divided into almost 500 sections. Some defined track sections are maintained by other countries, other railway companies (not SBB), some are marshalling yards or have been redefined. This results in 371 observations (track sections) per year that can be analysed with complete information for the years 2003, 2004 and 2005. A section is not strictly homogeneous, that is, between its endpoints it can vary in terms of rail and sleeper types, ballast, curvature, slope etc. The cost data contains information on: operation maintenance (e.g. cleaning, snow and ice removal), track maintenance, forestry, engineering, signal tower maintenance, wire


19

maintenance and electronic installation. Moreover, within these different cost categories SBB separates between short-run maintenance costs (“Contracting A”) that arise yearly and long-run costs which arise periodically and have the characteristics of renewal costs (“Contracting B”). Due to the fact that the data base is only available since 2003, the estimation of renewal costs is based on a relatively short time period of three years. Therefore, we do not estimate renewal costs by themselves but in combination with maintenance costs. Traffic data includes average daily data on number of trains, axle load and gross-tons per track, as well as yearly data on train kilometres, axle load kilometres and gross-ton-kilometres per track for the main lines. Infrastructure variables includes track length, switches, bridges and tunnels, level crossings, radius and slope, noise and fire protection, rail age and sleepers age as well as maximum speed.

1.2F - Hungary The cost data in the Hungarian study is based on detailed cost accounting information. The information is recorded on 723 cost items relevant for infrastructure management. The cost information is thus very detailed and includes the measures operation, maintenance and renewal. Some of the items are recorded on the level of section. The account was collected for five years, 2001 – 2005. The output measure contains data for train km, station usage and traffic performance divided into information for passenger and freight trains. Information on weights is also included. Unfortunately, the traffic information is collected on an aggregate level (nation level) and not on sections. For each output measure only five observations are thus available. The approach taken in this CS is different from the other econometric CS. For each of the cost items a model is estimated which tries to explain the cost with the different performance indicators as explanatory variables. Based on the result from the regression of the model a performance indicator is dedicated to each specific cost item. For the final analyses the cost items are aggregated into six activity groups based on the type of activity and the performance indicator (cost drivers) established in the detailed analyses.The activity groups are: train movement, path allocation, interim passenger train services, beginning/end of line passenger services, marshalling/shunting for freight wagons and consignment of freight wagons. These aggregated cost items/activities are modelled with information on the performance indicator. The marginal cost is then the derivation of the estimated function w.r.t each included output measure.

1.2G - UK In the UK study cross section data from Network Rail for 53 Maintenance Delivery Units (MDUs) for 2005/06 is used. 67% of total maintenance expenditure is available at the MDU level. The remaining expenditure (33% of the total maintenance budget) includes maintenance of electrification and plant equipment and other expenditure and can not be allocated to individual MDUs. Instead it is allocated to one of 18 Maintenance Areas or more aggregate levels. The cost categories allocated to MDU consist of signalling and telecoms (15%), Permanent way (34%) and General MDU expenditures. Traffic data is available at three levels of disaggregation; from total traffic at the highest level to intercity passenger traffic, other passenger and freight traffic at the most granular level. Efforts have been made to investigate whether, after accounting for the average weight of trains, there exist detectable differences in the wear and tear impacts of different types of


20

trains. Information on the infrastructure includes data on length by track type, maximum speed and load, signalling equipment, rail age and length of electrification. Of these variables length of track, proportion of track length with maximum axle load greater than 25 tonnes, with maximum line speed greater than 100 mph, with continuous welded rail (CWR) or proportion with rail age above 30 years and a labour price index, were included in the final model.

2.3 Results In GRACE we have tried to look at infrastructure costs in a way that ensures consistency across modes (road and rail). A general conclusion is that the approach is successful and fruitful comparisons between the results can be made. In the following, the results are presented first for the cost elasticity (2.3.1), followed by a discussion on the available average and marginal cost estimates (2.3.2).

2.3.1 Elasticity The figure below summarises the estimated (average) elasticities in our econometric case studies for road and rail infrastructure. The elasticities are divided into renewal (R), maintenance (M) and operation cost (O). The following general results should be mentioned;

• The (average) elasticity with respect to traffic is below one in all our studies • The elasticity in the rail sector is smaller than in the road sector • The elasticity decreases as we move from renewal measures to maintenance and

operation in the road sector. That pattern is not clear in the rail sector. Figure: 1 Cost elasticity with respect to traffic

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

R-Germ

any

R-Polan

d

R-Swed

en pav

ed

R-Swed

en grav

el

RM-Swed

en

RM-Switz

erlan

d

RM-Swed

en

RM-Pola

nd

M-Swed

en

M-Switz

erlan

d

M-UK (m

odel V

)

M-Switz

erlan

d

M-Pola

nd

O-Swed

en

O-Swed

en

O-Swed

en w

inter

O-Swed

en pa

ved

Elast RailElast Road


21

Additionally we observe; • In the majority of the studies the elasticity is constant or decreases with increased

traffic (β11<0). Only one study, on German Motorways, suggests an increasing elasticity.

• The Hungarian rail study uses a different classification but points at elasticities between 0.07 and 0.16.

• Renewal production shows considerable economies of scale.

Road - elasticity The average elasticity plotted against the average traffic volume (expressed in HGVs per day) used in the studies depicts the main conclusions from these case studies. The average elasticity is always below 1. The pattern whereby the elasticity decreases as we move towards short-term measures is here clear, i.e. the elasticity related to renewal (R elast) is highest followed by the elasticity related to an aggregate measure of renewal and maintenance (R+M elast). The elasticity related to operation is almost zero. Figure: 2 Average elasticity from GRACE road case studies

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 1000 2000 3000 4000 5000 6000

HGV per day

Elas

ticity R elast

R+M elastM/O elast

Note: The HGV measure is used to compare the different studies. As can be seen from the table below each study has found a specific form of traffic measure that best fit their model. The table below summarises the estimated elasticities as well as the estimated parameters in the studies. The German study suggests that the elasticity increases with more traffic (β11>0) while the Swedish studies suggests the opposite (β11<0). However, as can be seen from the figure above the Swedish and the German studies have data from road sections with very different traffic levels. The Polish studies are based on fixed elasticity models (β11=0). The German study includes an interaction term between HGV and passenger cars. To simplify the interpretation of the results in the table below we once again present the general expression for the elasticity (equation 2) derived previously.

)ln(2lnln

111 AAA

AA Q

QdCd

dQQ

CdC ββε +=== (2)


22

None of the studies was able to clearly verify which output variable predominantly drives the cost. Contrary to the rail sector these road studies need to rely on rather rough measures of traffic. The output variable distinguishes only between vehicle classes, for example passenger cars and heavy goods vehicle (HGV), and does not include any more sophisticated weight information. The correlations between these aggregate output variables are strong and usually an a priori decision has to be taken on which of them to include. However, thanks to the correlation the elasticity (but not the average cost) may be similar between different output variables. Table: 5 Road elasticities β1 β11*

lnQ β2* lnX

Mean Q

Elasticity

Output (Q)

Interaction term X

Renewal Germany R 0.15 0.38 -0.26 5002 0.87 HGV Passenger

carsC) Poland R 0.57 8592 [1403]A) 0.57 AADT No

Sweden R paved

4.95 -0.38 87594 [158]B) 0.72 HGVkm in region No

Sweden R gravel

0.68 718 [5]B) 0.68 HGVkm in region No

Sweden duration model

0.039DE HGV No

Renewal and Maintenace

Sweden R+M 3.3 -0.24 88313 [125] B) 0.58 HGVkm in region No

Poland R+M 0.48 8592 [1403]A) 0.48 AADT No

Maintenace/Operation Poland M 0.12 8592 [1403]A) 0.12 AADT No

Sweden O 0.147 -0.007 869962 [1232]B) (0.05) vkm in region No

Sweden O winter

0.21 -0.0152 869962 [1232]B) (0.007) vkm in region No

Sweden O paved

0.495 -0.034 859463 [1554]B) (0.03) vkm in region No

Sweden O gravel

1.11 -0.136 10498 [69]B) (-0.09) vkm in region No

Note: DE=Deterioration elasticity, A) Average HGV traffic B) Output measure expressed per km road. C) Mean volume 26632 (In parenthesis)= non significant estimates The first part of the German study (Model I) suggests that renewal production shows considerable economies of scale where a 1% increase in the tendered lot only increases the cost by 0.66%. The second part of the study, where renewal costs can be related to traffic volume, meets with problems related with the significance of material input and the sign of the effect of passenger cars. As a squared term of goods vehicle as well as an interaction term between goods vehicles and passenger cars is included the cost elasticity will depend on the goods traffic volume as well as the passenger car traffic volume. The cost elasticity for goods vehicles (with an average number of standard axles) was estimated between 0.005 and 1.17 as the AADT of goods vehicles increases evaluated at the mean of the passenger car volume (see figure below). The mean elasticity is 0.87.


23

Figure: 3 Cost elasticity (ratio between marginal and average costs) of trucks at German motorways

Source: Case study 1.2A. As in many other studies the Polish study shows strong correlation between different measures of traffic and the final model considers total traffic (i.e. all vehicles). The model does not include squared terms or interaction terms. This means that the estimated elasticity is constant over traffic volume. The elasticity of renewal and maintenance cost with respect to total traffic volume is 0.48. The model with only renewal cost shows the elasticity 0.57 and with only maintenance cost the elasticity is 0.12. In the Swedish CS it has a priori been assumed that maintenance cost is caused by heavy goods vehicles and operation by all vehicles. The model includes a squared term on traffic volume and both fixed and random effects models are tested but the fixed effects model finally chosen. The elasticity at mean traffic volume for operation is 0.05 but this is not statistically significant. That is true also for the disaggregating into winter operation (0.0073) and operation on paved (0.028) and gravel (-0.0955) roads. The main finding from these estimates is that we cannot conclude that operation expenditure is related to traffic volume. The marginal cost is thus zero. For maintenance cost a dynamic approach could have been appropriate to take interdependencies over time into account. However, these models where not successful and instead a simpler approach similar to the one utilised in the German and Polish CS has been used where the expenditures are expressed as an average over the time period considered. The cost elasticity for all maintenance expenditures with respect to heavy goods vehicles is 0.58 at the mean traffic volume and the elasticity is decreasing with traffic volume. For paved roads only the elasticity is 0.72 and for gravel roads 0.57.

0

0.2

0.4

0.6

0.8

1

1.2

5000 25000 45000 65000 85000

AADT trucks

MC

/ A

C


24

Figure: 4 Elasticity for maintenance cost, Sweden (within 95% confidence interval)

Source: Case study 1.2.Bi

Duration modelling The estimated models have a good overall fit. In the complete model with all variables included almost all variables are significant and their sign is consistent with the a priori assumptions. However, the flow of passenger cars has no significant influence on pavement lifetime in the complete model. This is probably due to its close correlation to traffic class dummies. As a matter of fact, the passenger car coefficient becomes significant if the traffic class dummies are dropped as is done in one of the models. The flow of HGVs on the other hand has an impact that is clearly significant, in the complete model as well as in the reduced models. In the complete model the deterioration elasticity (HGV) is -0.039. Thus an additional percent of HGVs means that the pavement lifetime decreases by 0.039 percent, quite a small number. The results also tell us that the pavement on wider roads lasts longer and that a higher speed limit causes the pavement lifetime to decrease. A comparison between


25

the complete model and the reduced model shows estimates to be stable for different model specifications.

Rail - elasticity All rail infrastructure CS show a constant or decreasing (β11<0) elasticity with traffic volume. The elasticities are in the same range for all measures and in the range of 0.2 and 0.3 for the econometric models. In addition, we find significant deterioration elasticity in the duration model approach which suggests that renewals are affected by traffic volume and thus connected with a marginal cost. This is consistent with the econometric studies which suggest a higher elasticity when renewals are added to the maintenance measure. Table: 6 Rail elasticity β1 β11*ln

Q β2*lnX Mean Q Elasticity lnQ Inter-

action term (X)

Renewal Sweden (duration) 0.109DE GT Freight 0.146DE GT Passenger

Maintenance and Renewal Sweden 1.567 -0.0844 7445989 0.302 Grosse Tonnes Switzerland (A+B) 0.265 0.265 Grosse Tonnes

Maintenace Sweden 1.47 -0.0844 7445989 0.204 Gross Tonne Switzerland (A) 0.200 0.200 Gross Tonne UK (model V) 5.834 -0.1818 4809570 0.239 Gross Tonne Switzerland (part of A) 0.285 0.285 Gross Tonne

Operation Sweden 3.314 -0.79 0.0495 15499 0.324 Trains lnQ*lnQ

Hungarytrain movement 0.063 Train km path allocation 0.085 interim passenger train service

0.081 No pass.train stops

beg/end of line pass train servic

0.108 No of pass train

marshall/shunt for freight wagons

0.161 No of wagons

consigment of freight wagons

0.090 No of consigned wagons

DE=Deterioration elasticity; GT=Grosse Tonne The Swedish econometric study continues the work in Andersson (2005) and succeeds in estimating a paneldata model. The results are very similar to previous analyses with POLS and can thus be seen as stable. The new analysis reinforces the conclusion that renewal has a significant marginal cost associated to it as well as operation and maintenance. All of the estimated elasticities decrease with the output volume. The elasticity is 0.302 for maintenance and renewal, 0.204 for only maintenance and 0.324 for operation. The figure below depicts the function for the elasticity of maintenance and renewal cost.


26

Figure: 5 Maintenance and renewal elasticity, Sweden

Source: Case study 1.2DI Following the findings of Gaudry and Quinet (2003), who found a difference between the impact of additional gross tonnes resulting from heavier trains, as opposed to that resulting from more trains of the same weight, the UK Case study adopts both a measure of the density of train miles per track-km and the average weight of those trains. While the CS finds significant elasticities of both density and average weight it cannot be concluded that they differ. Consequently, the CS cannot support the idea that there is a difference between more trains of the same weight and heavier trains in the same number. The models are therefore estimated with Gross tonnekm as output variable. The CS also tries to distinguish between freight and passenger trains with some interesting but not totally robust results. The final model includes a square term on traffic and the result is a mean elasticity of 0.239 with a negative b11 coefficient. Similarly to the experience of the CS for Sweden the Swiss CS does not face any serious problems adding renewal costs to maintenance costs. The inclusion of regional dummies (districts) has a relatively small effect on the explanatory power of the estimated models. Most coefficients (not the regional dummies) are significant at the 1 percent level and mostly have expected signs. Unexpected signs were noted for tunnel distance and speed which both reduced the cost. The elasticities are constant within a rather small range from 0.2 for maintenance, 0.265 for maintenance and renewal and 0.285 for a limited part of the maintenance cost. The Hungarian CS has a totally different approach but can nevertheless derive elasticities for the six activity groups. The result of this model is very low elasticities compared to the other econometric studies7, ranging from 0.063 for the cost item train movement related to train kilometre up to 0.16 for the cost of marshalling related to wagon. 7 The CS does not report t-value for the estimated parameters.


27

Duration modelling The Swedish case study on lifetime of railways generated results that are consistent with the econometric studies. There is a significant and price relevant cost related to rail renewal in line with what has been found in econometric studies of renewal cost data. The CS estimates marginal costs for freight and passenger trains separately in the range of €0.00012 – 0.00028 per gross tonne kilometre. A weighted marginal cost based on gross tonnes per observation gives an estimate for freight traffic at € 0.00012 per gross tonne kilometre. The equivalent for passenger traffic is € 0.00028. Somewhat surprisingly, the marginal cost for passenger trains is higher than for freight trains, but a possible explanation to this is higher quality demands for passenger trains. There is also a significant ageing effect after controlling for traffic loads. This indicates that there are climate and weather effects that reduce the life of Swedish railway tracks. It was suggested above that one possible explanation for the low deterioration elasticity in the road sector was the possibility to adapt the measure taken, for example the thickness of the pavement, as the expected traffic increases. It could be the case that the technology of rail infrastructure is less flexible and similar measures, for example type of track, are taken irrespectively of the expected traffic volume.

2.3.2 Average and marginal cost. The estimated elasticities resulted in a reasonably narrow range of estimates. These elasticities can then be multiplied by the average cost to give the marginal cost. Our estimates of the average cost vary substantially between the case studies.

Road The marginal cost can be calculated as the product of the average cost and the elasticity. Although the elasticity is constant the average cost is not and it falls with increasing traffic volume. Consequently, the marginal cost decreases with increasing traffic. Applying the elasticity on the average cost per goods vehicle kilometre in the German study (1.59 €/vkm) results in a marginal cost of 0.08 €/HGV-km at very low traffic volume increasing to 1.87 €/HGV-km on roads with the highest traffic volume. Evaluated at the average traffic volume the cost is 1.39 €/HGV-km. In Poland, evaluated at the average traffic volume, the renewal and maintenance average cost is 0.27 €/vkm (for all vehicles) and the marginal cost 0.13 €/vkm. The corresponding number for renewal only is 0.21 €/vkm and the MC is 0.12 €/vkm. The Swedish CS reports a significant difference between the marginal cost of paved and gravel roads. The former have a marginal cost of 0.032 €/HGV-km and the latter almost ten times higher marginal cost, 0.24 €/HGV-km. An aggregate of renewal and maintenance suggests an average cost of 0.040 €/HGV-km. The model for operation cost does not show a significant marginal cost and it can then be assumed to be zero. The resulting average cost per vkm differs largely between the regions in Sweden with the highest cost in north of Sweden for both maintenance (0.182 €/HGV vkm) and operation (0.011 €/vkm) with the lowest cost in the south (0.013 and 0.003).


28

The table below summarise the average and marginal cost for the road case studies. Observe that the mean traffic volume is very different between the studies with the highest traffic volume in the German study and the lowest in the Swedish study (see table 5). Table: 7 Average and marginal cost in the road sector AC MC Outputvariable €/Xkm €/Xkm Q

Renewal Germany R 1.590 1.390 HGV Poland R 0.210 0.120 All veh Sweden R paved 0.036 0.032 HGV Sweden R gravel 0.415 0.236 HGV Sweden duration model - 0.0013 HGV

Reneval and MaintenaceSweden R+M 0.059 0.040 HGV Poland R+M 0.270 0.130 All veh

Maintenace/OperationPoland M Na na All veh Sweden O 0.024 (0.002) All veh Sweden O winter 0.015 (0.001) All veh Sweden O paved 0.003 (0.001) All veh Sweden O gravel 0.066 (0.010) All veh

Duration model Based on a unit cost of 7.05 €/m2 pavement an average cost over a pavement interval can be estimated to 0.028 €/vkm. Applying the elasticities and a correction factor following the choice of a Weibull distribution on these average costs suggests a marginal cost of 0.0013 €/vkm with an interval from almost zero to 0.004. The reason for this low marginal cost in this approach is of course the low elasticity. Previous research has assumed that this is due to a weather/climate effect but this CS can reject any such influences. One possible explanation not further analysed in the CS is that the responses from the Road authority are such that the unit cost per sqm differs depending on traffic volume. If a higher volume is expected a more expensive pavement measure is taken. This is supported by the Road Administrations own data (Zarghamp 2002) from which the following simple function on maintenance cost per square meter and measure (SEK/ m2)8 has been estimated9. Ln(SEK/m2) = 2.466 (0.271) + 0.212 (0.0345) * ln(Q) The traffic volume Q increases the cost per square meter with the elasticity 0.2. If we believe this information two things happen as the traffic volume increase; first the lifetime of the pavement is reduced and secondly, the road authority responds to this expected reduction in lifetime with a more expensive maintenance measure.

8 1 SEK = 0.1085 € 9 Data on SEK/sqm over 40 years for ’built roads’ divided by number of pavement measures in different traffic classes in region Mitt, Mälardalen and counties F and M. R2 = 0.67.


29

Rail The average and marginal cost in the Swedish and Swiss CS are rather similar, regarding both average cost and marginal cost. The marginal cost related to maintenance is in Sweden 0.31 €/1000GTkm and in Switzerland 0.45 €/1000GTkm. Adding renewals to the maintenance increases the marginal cost to 0.70 €/1000GTkm in Sweden and 0.97 €/1000GTkm in Switzerland. In addition, the Swedish study finds a marginal cost for operation which is 0.054 per trainkm. The CS from UK reports both higher average costs and marginal costs compared to the other studies. The maintenance marginal cost is estimated to 2.0 €/1000GTkm. The Hungarian CS results in much higher average and marginal costs as presented in the table below. However, that study is based on a different approach. Table: 8 Average and marginal cost in the rail sector AC MC Outputvariable €/Xkm €/Xkm X Renewal Sweden – duration model 0.00028 Gross Tonne (Passenger) 0.00012 Grosse Tonne (Freight) Maintenance and Renewal Sweden 0.00285 0.00070 Grosse Tonnes Switzerland (A+B) 0.00364 0.00097 Grosse Tonnes Maintenace Sweden 0.00209 0.00031 Gross Tonne Switzerland (A) 0.0022 0.00045 Gross Tonne UK (model V) 0.00517 0.001978 Gross Tonne Switzerland (part of A) 0.00133 0.00038 Gross Tonne Operation Sweden 0.153 0.054 Trains Hungary train movement 3.5 0.22 Train path allocation 29.8 2.52 No of train interim passenger train service 13.4 1.09 No pass.train stops beg/end of line pass train servic 17.1 1.85 No of pass train marshall/shunt for freight wagons 5.03 0.81 No of wagons consigment of freight wagons 8.22 0.74 No of consigned wagons

Duration models The Swedish case study on lifetime of railways generated results that are consistent with the econometric studies. There is a significant and price relevant cost related to rail renewal in line with what has been found in econometric studies of renewal cost data. We estimate marginal costs for freight and passenger trains separately in the range of € 0.00012 – 0.00028 per gross tonne kilometre. Somewhat surprisingly, the marginal cost for passenger trains is higher than freight trains, but a possible explanation to this is higher quality demands for passenger trains.


30

2.3.3 Discussion The aims of this section is to presents a survey of the literature and try to understand in a wider context the result of the GRACE case studies.

Road The table below summarise studies from US, Canada and Australia on the share of infrastructure cost that can be attributed to traffic load (standard axles). It can be compared to the elasticities reported from the GRACE case studies. The table includes both rehabilitation (renewal) and routine maintenance. We can conclude that the load shares come in a number of different forms but that they always are below 1 (or equal to). The load share seems to be higher for rehabilitation than for routine maintenance which reinforces the conclusion from the GRACE case studies. In addition, we note that flexible pavement has a lower impact of load related factors than rigid pavements in these studies from US. Table: 9 Load shares in US-studies Study Year Flexible JCP CRC Composite Rehabilitation 10 Li et.al. (2001) 1995-1997 0.28 0.78 0.38 Indiana HCAS 1984 0.42 0.78 0.38 ARRB Study (Australia) 0.88 0.88 0.88 Federal HCAS 1997 0.84-0.89 0.78-0.86 0.84-0.89 Routine maintenance11 Li et.al. (2002) 1995-1997 0.257 0.357 0.632 0.28 Indiana HSC Approach 1984 0.21 0.54 1.00 0.29 Ontario study 1990 0.25-0.33 Note: HCSA = Highway Cost Allocation Study; JCP=joint concrete pavement, CRC=continuously reinforced concrete and Composite Hajek et.al. (1993) estimates the marginal pavement cost of truck damage in Ontario defined as the cost of providing pavement structure for one additional standard axle. In general the approach is to define a minimum thickness of the layer and then allocate the extra cost for additional thickness to the extra load. The method is similar to the federal method used in US but this study focus on the marginal effect while the general approach is to analyse the average effect. The study states that ‘a small increment in thickness permits a significant increase in traffic loads’ (p 52). The economies of scale in accommodating additional load results in a decreasing average cost function for additional standard axles. The marginal cost derived with this method is depicted in the figure below.

10 Source: Li et.al (xxxx) 11 Source: Li et.al. (2002)


31

Figure: 6 Marginal cost for rehabilitation in Ontario on new and in-service asphaltic concrete pavements

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

0.80

0.90

1.00

0 10 000 20 000 30 000 40 000 50 000 60 000 70 000

ESAL per year per lane

€/km

MC New pavementMC In-service pavement

Source: own figure based on Hajek et.al. (Exchange rate 0.7013 €/Canadian$) The marginal cost is decreasing as in numerous of the GRACE studies. However, this study has a totally different approach. The driving force behind this decreasing cost function is the economies of scale in layer thickness. If the authority predicts a higher traffic load in deciding on the measure we will, due to an adaptation of technology, observe a decreasing average and marginal cost ex post if the prediction is fulfilled. In a situation where the technology is fixed we may expect a totally different form of cost function. Such a fixed technology can be detected with more detailed information of the infrastructure characteristic. Lindberg (2002) analysed the marginal cost on Swedish roads with an engineering approach. The data originated from the Swedish Long Term Pavement Performance program and contained detailed data on infrastructure characteristics. The resulting marginal cost is shown in the table below. For a given road strength (SCI) the marginal cost is increasing with increasing traffic load. However, if the probability of observing a road with higher road strength is higher when the traffic volume increases we may observe a function with a shape more as the line.


32

Figure: 7 Marginal cost per standard axle on Swedish roads depending on road strength (surface curvature index =SCI).

0

0.005

0.01

0.015

0.02

0.025

0.03

0.035

0 100 200 300 400 500 600 700 800 900 1000

Standrad axles per day

€/km

SCI 50SCI 75SCI 100SCI 125SCI 150SCI 175SCI 200SCI 250

1 € = 8.92 SEK Source: Lindberg (2002) An increasing cost function is, as has been seen, also the result of the German GRACE study. This study is a two-model study where the technology is modelled separately (with economies of scale as a result). The remaining marginal cost takes a form more similar to the result in the table above. We may suspect that the studies on renewal cost doesn't control for the technology in a similar way. This may create an uncontrolled variation between different studies. The US Federal Highway Cost Allocation Study (1997 and addendum 2000) allocates the cost for the federal Highway program to different user categories. For each category an average cost is the calculated. The table below summarise the result also in € per km. Table: 10 US Federal Highway Cost Allocation Study 2000 cent/mile €/km Car 0.8 0.0039 Pickup/Van 0.76 0.0037 Buses 3.2 0.0157 Single Unit Truck (weight in pounds) - <25000 2.2 0.0108 25001-50000 5.46 0.0268 >50000 18.12 0.0888 All Single Units 4.38 0.0215 Combination Trucks (weight in pounds) - <50000 3.43 0.0168 50001-70000 5.21 0.0255 70001-75000 7.62 0.0374 75001-80000 8.65 0.0424 80001-100000 15.32 0.0751 >100001 20.28 0.0994 All Combinations 8.43 0.0413 All Trucks 6.74 0.0330


33

The ‘All trucks’ result from the US cost allocation study is similar to the result from the Swedish econometric study on renewal and maintenance (0.033 compared to 0.040). This is only a coincidence but suggests that the US figures above are on the low end of the result from the GRACE studies as the Swedish values are low compared to the other GRACE studies. The table below presents the result from a series of research studies and is an extension of the table in Case study 1.2A. Table: 11 Marginal infrastructure cost around the world

Source Scope of the study Type of cost considered Marginal cost (MC)

estimate Cost elasticity MC/AC

Hajek et.al. (1993)

Ontario Urban Freeway Major Arterial Minor Arterial Collector Local

Rehabilitation 0.002 $/HGVkm 0.007 0.012 0.031 0.461

Herry and Sedlacek 2002

Austria Maintenance and renewals

2.17 €Cents/vkm (per HGV)

Lindberg (2002) Total road network Sweden

Cost of pavement resurfacing

0.77 ... 1.86 € Cents/vkm

0.1…0.8

Link (2002) Germany Renewal 0.05 – 2.70 €Cents/vkm (per HGV)

Newbery (1988a)

Tunisian roads Cost of pavement resurfacing

0.13 … 2.58 US$/ESAL km

0.19…1.07

Newbery (1990) UK road network Cost of pavement resurfacing

0.035 pence/ ESAL km n.a.

Ozbay et al. (2000)

Highways Northern New Yersey


0.062 US$/ vehicle mile n.a.

Schreyer et al 2002

Sweden Maintenance, renewals and upgrades

3.62 – 5.17 €Cents/vkm (per HGV)

Small and Winston (1988)

US highways Cost of pavement resurfacing

0.022 … 0.023 US$/ESAL mile

n.a.

Small et al. (1989)

US rural and urban freeways


0.0148 … 0.0432 US$/ ESAL mile

n.a.

Note: € per $; 0.7888; km per mile; 1.609. In the figure below we have converted all the estimates to a common unit, €/HGVkm or €/vkm. We have assumed 1.5 ESAL per HGV and current exchange rates12. The comparison is rough. In the figure we have not presented the full value of the GRACE German Case study (should be 1.39 €/HGVkm). For some studies we present a lower and a higher value.

12 € per $ = 0.7888; € per £= 1.48; km per mile=0.62, ESAL per HGV = 1.5.


34

Figure: 8 Rough comparison between the GRACE case studies and other estimates

0.000 0.025 0.050 0.075 0.100 0.125 0.150 0.175 0.200

Ontario Urban freeway

Ontario Major Arterial

Ontario Minor Arterial

Ontario Collector

Ontario Local

Herry and Sedlacek 2002

Lindberg (2002)

Lindberg (2002)

Link (2002)

Link (2002)

Newbery (1990)

Ozbay et al. (2000)

Schreyer et al 2002

Schreyer et al 2002

Small and Winston (1988)

Small et al. (1989)

Small et al. (1989)

GRACE 1.2A (Germany Renewal)

GRACE 1.2C (Poland Renewal)

GRACE 1.2Bi (Sweden Renewal paved)

GRACE 1.2Bi (Sweden Renewal and Maintenace)

GRACE 1.2C (Poland Renewal and Maintenace)

US HWCA all truck

US HWCA all combinations

€/Xkm

All vehHGV

Note: 1.5 ESAL per HGV has been used Grace 1.2A ; 1.39 €/HGVkm; Ontario Local; 0.323 €/HGVkm.

Railway In the railway sector the difference between different studies are not as big as for the road sector. One explanation could be that the technology is more homogenous and easier to control which makes studies less vulnerable to the problem with unobserved variables. Another, less positive, explanation could be that studies are less common in the railway sector and still starts from a similar approach. The table below summarise a number of current studies.


35

Table: 12 Results from other studies compared against the estimated models Study (maintenance costs only) / Model estimated

Country Marginal Cost Estimates (Average) Euro per Thousand

Gross Tonne-km

Elasticity of cost with respect to tonne-km

Johansson and Nilsson (2004)

Sweden 0.127 0.169 (average)


Finland 0.239 0.167 (average)

Tervonen and Idstrom (2004)

Finland 0.18 0.133-0.175

Munduch et al (2002) Austria 0.55 0.27 Gaudry and Quinet (2003)

France Not reported 0.37 (average)

Andersson (2005) Sweden 0.293 (pooled OLS model) 0.272 (random

effects model)

0.1944 (average pooled OLS model) 0.1837 (average Random

effects model) Booz Allen & Hamilton (2005)

UK Approx 1.5 Proportion of maintenance cost variable with traffic: 0.18; 0.24

for track maintenance Source: UK CS The figure below summarise the GRACE case studies where results are presented per Gtkm and the results from the survey presented in the table above. Figure: 9 Rough comparisons between the GRACE case studies and other estimates

€/Gtkm

0 0.5 1 1.5 2 2.5



Tervonen and Idstrom (2004)

Munduch et al (2002)

Andersson (2005)

Booz Allen & Hamilton (2005)

GRACE 1.2Dii (Sweden, duration model passenger)

GRACE 1.2Dii (Sweden, duration model freight)

GRACE 1.2Di (Sweden Maintenace and Renwal)

GRACE 1.2E (Switzerland Maintenace and Renwal (A+B))

GRACE 1.2Di (Sweden Maintenace)

GRACE 1.2E (Switzerland, Maintenace)

GRACE 1.2G (UK, Maintenace)

GRACE 1.2E (Switzerland, Operation/Maintenace)

€/Xkm

€/Gtkm


36

It can be observed that the highest cost can be found for two studies in UK, the GRACE study 1.2G and Booz Allen & Hamilton (2005). Renewal has a significant impact on the cost as can be seen when comparing the Grace studies in Sweden and Switzerland.

2.4 Conclusions This GRACE report summarises 9 Case studies on the marginal infrastructure cost. The individual CS are presented in appendices. Each CS contains interesting information and the summary in this report cannot cover all topics presented in these studies. However, we have tried hereafter to present the general picture and the common results. These can be discussed in five different areas, i) methodology, ii) elasticity, iii) differentiation, iv) average cost and v) marginal cost.

i. Most of the studies use an econometric approach and collect paneldata. However, a minority of the studies use paneldata models. In two studies a duration model is used where a function of the lifetime of a road pavement or railtrack is estimated. The result can be used to derive a marginal renewal cost. The rail study gave results in line with the econometric study and supported the conclusion drawn from the econometric studies that there indeed exists a marginal cost related to renewal on railways. The result was similar between the two approaches. However, the road study suggested a very low effect of traffic on the observed lifetime of a pavement. A possible explanation with some support is that the authority predicts the higher traffic volume when deciding on the pavement thickness. The marginal cost is thus not found in observed lifetime but in the increased cost of the measures taken.

ii. The cost-elasticity with respect to the traffic-output describes the relationship between average cost and marginal cost such that Marginal Cost = Elasticity *Average Cost. The elasticity for road infrastructure cost decreases as the measure changes from renewal to maintenance and to operation. The elasticity for rail infrastructure cost is lower than the elasticity for road and doesn’t show the same difference between different measures. In addition, the majority of the studies suggest that the elasticity decreases with increased traffic. Thus highly used infrastructure has a lower elasticity than low volume infrastructure. All elasticities reported above are for the average traffic in the studies.

iii. The operation or short term maintenance is related to total trainkm or total vehiclekm while the renewal and maintenance are usually related to gross tonnekm or HGVkm. Few of the studies have been able to test which type of traffic predominantly influences the infrastructure cost. In general, this has been decided a priori based on other information. However, one study conducted a test on the difference between additional trains of the same weight or additional weight of the same number of trains but could not find any significant difference.

iv. The average cost is less homogenous than what could be expected. For road studies the average renewal cost is 0.036 €/HGVkm in the Swedish all roads study and 1.59 €/HGVkm for the German motorway study. The Polish study allocates the renewal cost to all vehicles and suggests a cost of 0.21 €/vehkm. For the aggregate of renewal and maintenance measures the average cost is 0.059 €/HGVkm in the Swedish study. Operation is in the Swedish study allocated to all vehicles and has an average cost of 0.024 €/vkm. In the rail infrastructure cost study maintenance and renewal has an average cost of 0.00285 €/Gtkm in Sweden and 0.00364 €/Gtkm in Switzerland. The maintenance only average cost is 0.00209€/Gtkm in Sweden and 0.0022 €/Gtkm in


37

Switzerland. However, the UK study shows an average maintenance cost of 0.00828 €/Gtkm. Operation has an average cost of 0.153 €/trainkm in Sweden. The Hungarian study suggests a cost for ‘train movement’ 3.5 €/trainkm and 0.0041 €/Gtkm.

v. The marginal cost follows from the elasticities and the average costs. The marginal cost on roads has a huge variability depending on the huge variability in average cost. The cost on German motorways is 1.39 €/HGVkm for renewal only. The corresponding cost for all Swedish paved roads is 0.032 and 0.12 in Poland. The Swedish result for gravel roads is 0.236 €/HGVkm. Aggregating renewal and maintenance generates a marginal cost of 0.040 €/HGVkm in Sweden and 0.13 €/HGVkm in Poland. Infrastructure operation costs do not vary with traffic volume according to the Swedish case study. The marginal cost in the rail sector is 0.00070 €/Gtkm inSweden for renewal and maintenance and 0.00097 €/Gtkm in Switzerland. Maintenance only has a cost of 0.00031 €/Gtkm in Sweden and 0.00045 €/Gtkm in Switzerland. The marginal cost in UK is estimated in 0.002 €/Gtkm. Operation has a marginal cost of 0.054 €/trainkm in Sweden. The Hungarian study concludes on a marginal cost of 0.22 €/trainkm for train movements.


38

3 Congestion and scarcity Infrastructure has a long lifetime and is difficult to adjust for fluctuations in demand. The capacity will therefore be limited and a problem of allocation will occur when traffic increases. On the road the result is congestion – increased traveltime for all users – and in the rail sector the main result is that other operators will not be able to get the slot they want – scarcity. This section summarises the CS on road congestion (3.1) and rail congestion (3.2). The former focuses on the question – can we explain the difference in estimates that has been reported in the past? The latter develops an approach to evaluate the cost of scarcity.

3.1 Road In contrast to other cost categories and modes, there have been a host of studies involving the estimation of marginal road congestion cost. The difficulty is that the available estimates vary considerably between different studies, and not always in the ways one might anticipate. The table below provides a range of results from reviewed studies and illustrates this variation. The original values estimated by the studies reviewed are presented, along with the same values updated to 2003 prices. It can be difficult to compare values between cordon schemes and distance related schemes. In order to facilitate a comparison of the values it is assumed that the average car trip length is around 10.5 kms, as suggested by the UK DfT (2003) for trips in medium urban areas. It can be seen that there are wide ranging differences between the values presented by the six studies for what are essentially medium sized cities. The highest values are put forward by Newberry and Santos (2003) whose present two sets of values, one calculated from an area wide speed-flow curve and the other using SATURN. The area wide speed flow values are consistently higher than the SATURN values (ranging from 10% to nearly 60% higher) with a highest value of close to 526 pence for Northampton and a lowest value of 12.74 pence for Bedford. At the other end of the scale are the values calculated by Milne (2002) which for similar size cities produce values of less than 1 pence per car unit km but these appear to be averages over very wide areas, whereas the cordon charge studies will only charge radial trips to the centre.


39

Table: 13 Comparison of Values from previous studies Study Values Measured Values in 2003 Prices Pence Sansom et al. (2001) Value Measured – Short run MEC: values without brackets are the low

estimates & figures in brackets are high estimates. Value Measured – Short run MEC: values without brackets are the low estimates & figures in brackets are high estimates.

Central London: Motorways: 53.75 Trunk & Principal: 71.09 Other: 187.79 Outer London: Motorways: 31.09 Trunk & Principal: 28.03 Other: 39.66 Outer Conurbation: Motorways: 35.23 Trunk & Principal: 12.28 Other: 0.00 Urban 15-25 km2 Trunk & Principal: 7.01 Other: 0.00 Urban 5-10 km2 Trunk & Principal:2.94 Other: 0.00 Rural: Motorway: 4.01 Trunk & Principal: 8.84 Other: 1.28

Inner London: Motorways: 20.10 Trunk & Principal: 54.13 Other: 94.48 Inner Conurbation:: Motorways: 53.90 Trunk & Principal: 33.97 Other: 60.25 Urban>25 km2 Trunk & Principal:10.13 Other: 0.72 Urban 10-15 km2 Trunk & Principal: 0.00 Other: 0.00 Urban 0.01-5 km2 Trunk & Principal: 1.37 Other: 0.00

Central London: Motorways: 57.08 Trunk & Principal: 75.49 Other: 199.41 Outer London: Motorways: 33.01 Trunk & Principal: 29.77 Other: 42.11 Outer Conurbation: Motorways: 37.41 Trunk & Principal: 13.04 Other: 0.00 Urban 15-25 km2 Trunk & Principal: 7.44 Other: 0.00 Urban 5-10 km2 Trunk & Principal:3.12 Other: 0.00 Rural: Motorway: 4.26 Trunk & Principal: 9.00 Other: 1.36

Inner London: Motorways: 21.34 Trunk & Principal: 57.48 Other: 100.33 Inner Conurbation:: Motorways: 57.24 Trunk & Principal: 36.07 Other: 63.98 Urban>25 km2 Trunk & Principal: 10.76 Other: 0.76 Urban 10-15 km2 Trunk & Principal: 0.00 Other: 0.00 Urban 0.01-5 km2 Trunk & Principal: 1.45 Other: 0.00

Unit – Per Car Unit Km (1998 prices & values – pence) Unit – Per Car Unit Km


40

Comparison of Studies – Values – Continued………

Study Values Measured Values in 2003 Prices and Values – Pence Newberry & Santos (2003)

Values Measured – MEC. 1st figures calculated from Area Wide Speed-flow Curves; Figures in brackets calculated using Saturn. Northampton: 495 (315) Kingston Upon Hull: 209 (166) Cambridge: 80 (71) Norwich: 16 (14) Lincoln: 78 (67) York: 60 (44) Bedford: 12 (11) Hereford: 72 (57) Unit-Per Car Unit Km (1998 prices & values-pence)

Values Measured – MEC. 1st figures calculated from Area Wide Speed-flow Curves; Figures in brackets calculated using Saturn. Northampton: 525.64 (334.50) Kingston Upon Hull: 221.94 (176.28) Cambridge: 84.95 (75.39) Norwich: 16.99 (14.87) Lincoln: 82.83 (71.15) York: 63.71 (46.72) Bedford: 12.74 (11.68) Hereford: 76.46 (60.53) Unit-Per Car Unit Km

Milne (2002) Values Measured – MEC Helsinki: 0.26 Edinburgh: 0.65 Salzburg: 0.92 Unit-Per Car Unit Km (1998 Prices & values-pence)

Values Measured – MEC Helsinki: 0.28 Edinburgh: 0.69 Salzburg: 0.98 Unit-Per Car Unit Km

May et al. (2002) Values Measured – MEC. 1st best pricing based on Saturn. Top 10 links with uniform charges: 0.80 Top 10 links with two levels of charges: 0.50 & 2.00 Unit-Per Car Unit/ Trip (2000 Prices- £s) Values Measured - Judgemental Cordons. Inner 1 – 0.50 Inner 2 – 0.75 Outer 1 – 2.25 Outer 2 – 0.75 Unit-Per Car Unit/ Trip (2000 prices & values- £s)

Values Measured – MEC. 1st best pricing based on Saturn. Top 10 links with uniform charges: 83 (7.9) Top 10 links with two levels of charges: 52 & 208 (5.0 &19.8) Unit-Per Car Unit/ Trip Values Measured - Judgemental Cordons. Inner 1 – 52 (5.0) Inner 2 – 78 (7.4) Outer 1 – 234 (22.3) Outer 2 – 78 (7.4) Unit-Per Car Unit/ Trip (Unit-Per Car Km)


41

Comparison of Studies – Values – Continued………

Study Values Measured Values in 2003 Prices Pence Santos (2004) Value Measured – MEC based on an optimal toll that maximises social

surplus: defined as total utilities of all trips minus sum of total costs of all trips. Northampton: 3.47 Kingston upon Hull: 3.73 Cambridge: 1.60 Lincoln: 1.07 Norwich: 0.80 York: 1.60 Bedford: 1.60 Hereford:1.60 Unit – Optimal Toll Per Car Unit/Trip (2002 prices & values- £) for a single cordon scheme. Northampton: 2.40 & 2.40 Kingston upon Hull: 3.20 & 0.53 Cambridge: 0.80 & 2.67 Lincoln: 0.80 & 1.07 Norwich: 0.80 & 0.80 York: 1.07 & 1.33 Bedford: 2.7 & 2.40 Hereford:1.07 & 1.07 Unit – Per Car Unit/Trip (2002 prices & values- £) for a double optimal toll

Value Measured – MEC based on an optimal toll that maximises social surplus: defined as total utilities of all trips minus sum of total costs of all trips. Northampton: 352 (33.5) Kingston upon Hull: 378 (36.0) Cambridge: 162 (15.4) Lincoln: 108 (10.3) Norwich:81 (7.7) York: 162 (15.4) Bedford: 162 (15.4) Hereford: 162 (15.4) Unit – Optimal Toll Per Car Unit/Trip (Per Car Unit Km) Northampton: 243 & 243 (23.1 & 23.1) Kingston upon Hull: 324 & 54 (30.9 &5.1) Cambridge: 81 & 271 (7.7 & 25.8) Lincoln: 81 & 108 (7.7 & 10.3) Norwich: 81 & 81 (7.7 & 7.7) York: 108 & 135 (10.3 & 12.9) Bedford: 274 & 243 (26 & 23.1) Hereford: 108 & 108 (10.3 & 10.3) Unit – Per Car Unit/Trip (Per Car Unit Km)

Santos (2000) Values Measured – MEC Cambridge-Morning Peak: 61.4 Cambridge-Evening Peak: 51.0 York-Morning Peak: 48.9 York-Evening Peak: 49.9 York-Off Peak: 42.7 Unit-Per Car Unit Km (1996 prices & values-pence)

Value Measured – MEC Cambridge-Morning Peak: 65.20 Cambridge-Evening Peak: 54.16 York-Morning Peak: 51.93 York-Evening Peak: 52.99 York-Off Peak: 45.34 Unit-Per Car Unit Km


42

We can find a number of reasons for the difference in results from previous studies. The figure below suggests a number of different measures on the cost that could be attributed to congestion. In the figure we have a Free flow cost that is independent of the traffic volume. As the traffic increases the private marginal cost (PMC) increases as the delay becomes more severe. The key concept in pricing is that the social marginal cost (SMC) increases faster. The external cost is the difference between these to cost curves. In addition, we have indicated the existence of other externalities. The demand (D) is decreasing as cost increases and the private optimal is in the figure around 80 ‘traffic units’ while the social optimal is approximately 70 units. Figure: 10 The concept

0

50

100

150

200

250

0 20 40 60 80 100 120

Index Traffic

Inde

x Co

st p

er k

m FreeFlowPMCSMCOtExtD

The aim of this GRACE report is to show estimates of the external marginal cost. Focusing on congestion the appropriate measure is the difference between the social marginal cost and the private marginal cost (the arrow). If the purpose is to find the optimal congestion price to be paid we need to look at the external marginal cost at the optimal traffic level, i.e. at Q*. However, in addition to the congestion externality also other externalities exists, such as accidents, air pollution, climate change and noise which should be included in the price. The optimal price to pay is thus p*+e* (e= OtExt). As simpler concept is the measure in relation to the free flow speed. This is the delay cost. However, it should be clear that delay cost also exists at the optimal level of traffic. Free flow is never, or seldom, the optimal condition for infrastructure use.

3.1.1 Overview The investigation is designed to throw particular light on the reasons for some unexpected differences in the results reported for different cities in previous studies. In order to retain experimental control the work is based on a carefully specified series of “city scenarios” rather than on a set of real cities which differ from each other in a myriad of ways.

Q*


43

Results are presented for four city scenarios – a reference case and three variant scenarios each of which differs from the reference case in only one respect. The reference city has a built up area of approximately 10 km2 and has a population of about 800 000 producing somewhat over 100 000 car movements in the morning peak – creating significant peak period congestion. The modelled area is about 30 km across (thus extending well beyond the main built-up area) and the modelled road network comprises three types of link “principal”, “other”, and “special” – the latter, which includes links close to schools and hospitals, was introduced as a class of links which are particularly susceptible to externalities caused by road traffic)- which add up to nearly 500 kilometres. The first variant city has 20% greater population, the second has a more restricted network (with about 25% fewer road kilometres – though leaving the principal road network largely intact) and the third has an enhanced network (with an additional orbital route, comprising links of type “other” which adds about 20% extra road kilometres to the network). Table: 14 City scenarios

• Base – city of 10 km2 with 800,000 inh • Higher – 20% greater population • Fewer – 25% fewer road kilometre • Extra – 20% more road kilometre

The results for each city scenario relate to a typical morning peak hour and have been produced using a SATURN model. Results are produced for each city for a “without tolls” run and a “with tolls” run. The “without tolls” results are based on user equilibrium assignment. The “with tolls” results show the equilibrated situation after application of optimal tolls reflecting the externalities caused by the traffic. The model allows the imposition of these tolls to affect route choice and the decision on whether to travel by car during the peak (the net effect of decisions on trip frequency, trip timing and mode).

3.1.2 Results The Case study in Annex gives detailed results from the different scenarios. We have here focused on, and calculated, cost per kilometre based on the definitions in the figure above. The non optimal case is marked as the 0-case while the optimal case is marked with an asterisk (*). The first and basic observation is related to the type of city. The table below uses the optimal base case (Base *) as reference case. We can expect a difference of up to 40% only because the type of city differs. Table: 15 Cost between different cities (relation to optimal base case)

Cost element Base 0 Base * High 0 High * Fewer 0 Fewer * Extra 0 Extra * p 1.00 1.37 0.98 1.02 p+e 1.00 1.29 0.97 1.02 e 0.89 1.00 0.90 0.91 0.90 0.91 0.94 0.99

The marginal external congestion cost is 37% higher than the base case in the high demand city. The congestion price per trip is lowest in the city with extra links. However, per


44

kilometre the costs become slightly higher. It should be noted that the distance differs between the cities. Adding other externalities to the congestion cost does not change the relation between different cities dramatically. In addition to the cost of congestion and other externalities presented above, we have calculated the ‘delay’ cost per kilometre based on the Case study. The figure below depicts the kilometre cost for the different cases. Figure: 11 Cost per kilometre (€/vkm) – the optimal congestion toll (p), the cost of other externalities (e) and the delay cost (d) at optimum (*) and non-optimum (0).

0.00

0.05

0.10

0.15

0.20

0.25

0.30

0.35

0.40

Base 0 Base * High 0 High * Fewer 0 Fewer * Extra 0 Extra *

Cost

per

km

(€/v

km)

pde

Besides the observation that city structure strongly affects the result we make three observations:

• The first concerns the importance of the definition. If the delay cost is included in a study the external marginal congestion cost is overestimated by a factor of 4. If other externalities are included the overestimation is around 3.7.

• The second observation is on the importance of other externalities. If these are not included the external marginal cost is underestimated by around 20%.

• The third observation is on the importance of measuring at the optimal situation. The table below summarises the measure of other externalities and the delay cost.

Table: 16 The importance of the definition

Base* High * Fewer * Extra * Inclusion of delay cost

(d+p)/p* 4.01 4.16 4.30 4.08 (d+p+e)/(p+e)* 3.50 3.78 3.77 3.57

Inclusion of other externalities p/(p+e)* 0.83 0.88 0.84 0.83

Measure at non-optimum e/e* 0.89 0.99 0.99 0.96 d/d* 1.27 1.26 1.22 1.26 p/p* na na na na


45

3.2 Rail Congestion is only the appropriate capacity cost where the train in question represents an additional train with respect to what would otherwise have been run; where the train in question runs instead of some other train the appropriate capacity cost is the opportunity cost of trains forced off the system by lack of capacity.

3.2.1 Introduction Charging for scarce capacity would require estimation of the opportunity cost of a slot. The most attractive solution to this problem in theory is to 'auction' scarce slots. There are many practical difficulties however, including the complicated ways in which slots can be put together to produce a variety of types of service, and the fact that the value of a particular slot for a particular use depends on how other slots are being used (in terms of the operation of complementary or competing trains). Nilsson (2002) provides a more detailed consideration of auctioning and argues that it is a feasible solution. He argues that train operating companies could be asked to bid on the basis of what they are willing to pay for their most desired slot, indicating also the discount they would require per minute earlier or later than the ideal their slot is. An optimisation algorithm would then produce the best feasible solution, and train operating companies would be given a chance to revise their bids. Nilsson accepts, however, that this might not converge where train operators are competing in the same market, as their bids will be heavily dependant on what slots other operators get. Actually charging operators on the basis of the second highest bid would both give an incentive to correct revelation of willingness to pay, and ensure that charges actually reflected opportunity cost. A different approach, recommended by NERA (1998), is to identify sections of infrastructure where capacity is constrained and to charge the long run average incremental cost of expanding capacity. However, this is a very difficult concept to measure (the cost of expanding capacity varies enormously according to the exact proposal considered, and it is not easy to relate this to the number of paths created, since they depend on the precise number and order of trains run). An alternative considered in the GRACE CS is for the track charging authority to attempt to calculate directly the costs involved in depriving another operator of the slot. For instance, if a train has to be run at a different time from that desired, it is possible to use studies of the value people place on departure time shifts to estimate the value to its customers of the cost involved. Similarly, the costs of slower speeds may be estimated from passengers' values of time.

3.2.2 The case study This case study concerns the stretch of the East Coast Main Line from London to Doncaster. It is heavily used, particularly between London and Doncaster, which is where the main lines to Leeds, Hull and an important route to Scunthorpe and Grimsby branch off. There is one main operator of long distance passenger services on this route. A few years ago a new open access operator was granted access rights to operate through trains between London and Hull. The line from London to Doncaster also carries freight traffic.


46

The basis of the approach taken here is that operators should be charged for the capacity they use in accordance with the social opportunity cost of that capacity. In order to implement this approach it is necessary first to measure the amount of capacity used by each train run, and then to estimate its opportunity cost. Thus the approach investigated in the CS is the construction of a tariff based on the opportunity cost of the slot to the existing operator. If the new operator requires capacity that would deprive the existing of more than one slot then they would be charged for the appropriate number of slots. Since the existing operator is known, and is required to make data available to the regulator, this approach to charging should be feasible. Of course, if there are several other operators competing for the slot, and they all have higher values than the franchisee, then this will understate the true opportunity cost of the slot. However, basing charges on the identity of unknown possible new entrants appears difficult, at least until they start operating and data becomes available. The opportunity cost of a slot for this type of service can be estimated as the sum of:

• the additional amount of traffic attracted to rail by the presence of this train multiplied by the price it pays

• the consumers’ surplus to rail users as a result of the additional quality and capacity provided by the train

• the savings of external costs to road users and the public at large from the train attracting passengers from road.

• Less the train operating, infrastructure and external cost savings from failing to run this train.

3.2.3 Results Unfortunately, the results can not be expressed in absolute cost terms due to the use of secret information. Instead, the results are related to the ‘Total benefit of the existing operator at peak’. The value of the slot for the ‘existing peak operator’ is thus 100%. The value of the slot for the existing operator off-peak is in total 6% of the value at peak. The main positive value is related to effects on other modes (mainly congestion) as can be seen in the figure below. The negative value for the rail modes depends on a negative profit and a negative tax revenue effect. A new operator at peak generates a net value of 10% above the loss when removing the existing operator from the slot. The main positive (net) benefit is once again the reduced congestion on other modes while lost tax revenues give a negative net effect on the rail mode. A new operator off-peak has a negative net-value of 14% depending on a substantial loss to the rail operator.


47

Table: 17 Summary of slot value (Full value for existing operator at peak=100) Existing operator

at peak Existing operator

off peak New operator

at peak New operator

off peak Rail Other Rail Other Rail Other Rail Other Env+Safety -0.9 13.4 -0.9 2.9 -0.9 4.8 -0.9 2.0 Infrastructure costs 0.0 1.0 0.0 0.2 0.0 0.4 0.0 0.1 Tax revenues -12.4 -18.2 -3.1 -3.9 -4.8 -6.5 -1.1 -2.7 Consumer surplus 18.8 0.0 2.6 0.0 0.1 0.0 0.7 0.0 Congestion 0.0 52.7 0.0 11.3 0.0 18.8 0.0 7.7 Mohring 0.0 -1.7 0.0 -0.4 0.0 -0.6 0.0 -0.2 Operators profit 51.2 -4.1 -1.9 -0.9 2.0 -1.4 -19.3 -0.6 Full value 100.0 6.0 11.9 -14.1 The results suggest a substantial scarcity charge for peak slots, the charge for off peak slots would only be some 10% of this value. In terms of net social benefits, the existing operator’s use of the peak path gives the highest values for passenger use. The off-peak slot and the use of the peak slot by the new operator have much smaller, positive social values. This is driven by the small overall changes in total passenger demand arising from these two scenarios, and the small changes in operators’ profits. In the case of the new operator, the increase in the new operator’s profits is at the expense of the existing operator, and in the case of the off-peak slot actually reduced overall.


48

Figure: 12 Private, modal and Societies benefits

-30 -20 -10 0 10 20 30 40 50 60 70 80

Existing Peak

Existing Off-Peak

New Peak

New Off-Peak

Percentage of benefit Existing operator at peak

Other modesOther effects rail modeProfit operator of slotNet profit all operators

The figure above presents the elements of the slot value and in addition the profit to the operator of the slot. The profit to the operator that got the slot is always positive and higher than the net profit of all affected operators. The highest value is to the existing operator (67% of the benefit at peak for existing operator). The new operator will have a profit of 41% at peak and 8% off-peak. The lowest profit is for the existing operator off-peak (1%). Comparing the private value for the slot operator (the profit) and the social effect (the sum of the effect on other modes, the other effects on the rail modes and the net profit for operators) suggests that very large differences exist. This in turn suggests that a market solution (for example auctioning) without taking into account the effect especially on other modes would not necessarily give the social optimum result. However, appropriate pricing of other modes would considerably change this supposition.

3.3 Conclusions Our overall conclusion regarding road congestion, based on the theoretical investigations and modelling work, is that it is not surprising that the reported performance of “optimal” road user tolls differs in different studies. We conclude that these differences can be variously attributed to:

• differences is the definition of “optimal” tolls – the term is often quite loosely applied. For example; the term sometimes relates only to congestion tolls (rather than covering other externalities), sometimes allows for the cost of implementation of the tolls (and sometimes not), and sometimes relates only to simple tolls - such as cordons (rather than tolls which vary in space and time).

• differences in the way that optimal tolls (however defined) are calculated. For example, do they fully reflect the behaviour of travellers at the margin or are they derived from a theoretical representation of the marginal impacts?

• differences in the nature of the cities being studied. Factors which are particularly likely to influences the result include the degree of congestion, the availability and attractiveness of alternative modes, the drivers’ tolerance of


49

congestion, and the capacity of the network to absorb additional demand. Even for a single city These

• differences in the valuation of different externalities – perhaps reflecting different values of time and resource costs.

• differences in the models used to estimate system performance. Key issues include:

the representation of traveller response (which responses are represented?, what degree of equilibrium is assumed?)

the representation of the time dimension (where tolls vary over time, how accurately does the model reflect behaviour at the margin of different toll levels?)

the degree of detail with which the network is represented and the number of differently behaving groups included in the model (a greater degree of disaggregation will lead to a less volatile aggregate result).

For rail transport we find that a substantial peak scarcity charge per slot is justified; the off-peak charge would only be 10% of this level. The results seem to confirm the view that existing variable charges for the use of infrastructure on key main lines where capacity is scarce are too low as a result of neglecting scarcity in the charges set. The private slot value is in the CS far away from the social slot value which indicates problems with a simple market based solution. This result is an effect of high congestion cost on the road network in the CS that is not internalized in a road pricing regime. The institutional arrangements behind franchising mean that we have data for the franchisee from which we can calculate the opportunity cost of the use of each slot and thus the scarcity value. Our approach is that if any operators wish to use capacity not required for the specified minimum level of service they should pay the opportunity cost of the use of the path by the franchisee. Clearly in trying to evaluate what would be the outcome of the imposition of a scarcity charge we have had to use information on potential entrants. The CS has used typical industry data applied with a detailed rail passenger simulation model which produces estimates of revenue, costs, consumer surplus and diversion to/from other modes. Estimates of changes in external costs are then made to derive results for the overall social benefits of alternative allocations of capacity. The CS suggests that the imposition of scarcity charges based on the value of slots to the franchisee is both feasible and likely to be socially beneficial. However more work is needed on exactly what the tariff should look like and what its overall effects would be.


50

4 Accidents The external marginal accident cost is an important component in the pricing of transport. This chapter first presents a definition and discusses different methods to estimate the external marginal accident cost (4.1). Based on this the remaining part of the chapter discusses the current knowledge around valuation of accidents (section 4.2), internal and external cost (4.3) and risk elasticity (4.4). Section 4.5 discusses the insurance externality approach and 4.6 concludes with a summary.

4.1 Methodology and definitions

4.1.1 Definition of external cost of accidents The total annual cost of accidents (TC), where vehicle type j has been involved, can be written as equation (1) where A is the number of accidents and (a+b+c) the cost components discussed below. By “involved” we mean that the vehicle has been one of the parts in the accident, irrespectively of who was hurt or who was at fault. The risk (r) of category j to be involved in an accident (2) may be affected by an increase in the volume of traffic of category j (Q). This effect is expressed as a risk-elasticity (E), equation (3). The marginal cost with respect to the traffic volume for a vehicle of category j can be written as equation (4). We derive the external marginal cost as equation (5), where PMC is the private marginal cost already internalised. If we introduce θ as the share of the accident cost per collision that falls on category j (6) the external marginal accident cost can conveniently be expressed as (7).

)()(T cbarQcbaAC j ++=++= (1)

QAr = (2)

rQ

QrE

∂∂

= (3)

))(1()(MC cbaErcbadQ

Aj +++=++

∂= (4)

jj PMCMC −=eMC (5)

( )bar

PMC j

+=θ (6)

( )[ ] ( )ErcEbar +++−+= 11MC ej θ (7)

Liability The theory presented above does not explicitly discuss liability. Assume that one group (A) is injured and the second (B) is the other party in the accident. Without any liability, the injured user (A) will bear all costs and the other part (B) will not bear any cost. The external marginal cost for each group is then:

cArAEcbaArbaArAEcbaAreAMC +++=+−+++= ])[()(]1)[( (8)

]1)[( BEcbaBreBMC +++= (9)


51

The final expression of the general theory (equation 7) is a weighted sum of these two expressions, where θ expresses the probability of being the injured user. Under a negligence rule, users in group B will not bear any cost as long as they behave legally. If they break the law, they will be responsible for some of the costs as compensation (d) to user A or as a fine (M). These costs will be included in their private marginal cost and the external cost will decrease. At the same time, the compensation (d) will reduce the expected cost of an injured user in group A; consequently the external marginal cost of group A will increase. While this conclusion at first looks disturbing, it should be noted that the criminal user B will have a higher generalised cost, than the legal user B. The criminal user has to pay fine, compensation and external marginal cost. Legal user B => ]1)[( BEcbaBr

eBMC +++= (10)

Criminal user B => )(]1)[( MdBrBEcbaBreBMC +−+++= (11)

Not compensated user A => cArAEcbaAreAMC +++= ])[( (12)

Compensated user A => )(])[( dcArAEcbaAreAMC ++++= (13)

With strict liability for user B, he or she will always pay the cost in the form of compensation (d) or a fine (M) in the case of an accident – in principle he or she is always ‘criminal’ as in equation 11 above. Assume that both A and B are car users and that we cannot ex ante identify the criminal user; we have to assume that the probability of being in either group is 50/50. Consequently, while the marginal cost of group B is reduced, it is increased for group A through the compensation (d); the effect on the joint marginal cost for all car users disappears and the external marginal cost can be written once again as in the general theory (7). However, a fine (M) will affect the result. This theory assumes that users perceive the compensation and fine as a part of their cost ex ante. An interesting case is if the victim (user A) is guilty of the accident, for example a pedestrian that crosses the street illegally and is hit by a car. Depending on the legal situation the car driver should ex ante be charged as equation (11) or (12) and the pedestrian as (13). Consequently, the innocent car driver shall pay a charge ex ante.

Risk-avoiding behaviour Most of the empirical work suggests that the risk decreases with traffic volume (E<0) – see section below. This highlights one of the problems of the presented approach – risk-avoiding behaviour. The user may react in a number of different ways when he perceives that the risk level has changed. Peltzman (1975) developed the hypothesis of risk compensation and presented evidence showing that the user when given a safer environment compensates this with a higher degree of risk taking. In the same way a more unsafe environment may be compensated by the user with reduced exposure to risk. This reaction generates a cost to the user and reduces the observed change in risk. The cost of this risk-avoiding behaviour has to be included in the external marginal cost. Peirson et al (1994) introduces a form of risk avoiding behaviour where the users, when they selfishly adapt their behaviour, reduce the risk for all other users. The risk avoiding behaviour includes an element of positive externality. Johansson (1996) shows how this can be internalised through the accident externality charge in a second-best situation where the behaviour is not subsided per se. The marginal cost above is estimated based on the change in risk. This change in risk can be influenced by traffic safety behaviour, which the increased number of vehicles has forced the user to take. The cost of this behaviour is an externality. In the following we divide the users into two groups, the first group (A) is injured and the second (B) is the other part in the accident. We assume that the level of safety (s) on a given trip is associated with a cost (g), which increases as the level of safety increases. The total annual cost (TC) for accidents and traffic safety can then be written as equation (14).


52

)()()( BsgBQAsgAQcbaATC ++++= (14)

We allow only for risk-avoiding behaviour that increases the ‘internal’ safety, i.e. the user’s own safety13. Users that expect to be victims (user A; e.g. unprotected road users) will adjust behaviour (sA) to protect themselves from an accident. To the marginal accident cost, the cost of all victims’ risk-avoiding behaviour has to be added. The total external cost for category B, the unharmed user (θ=0) will be:

[ ]BdQAds

Adsdg

AQBEcbaBreBMC ++++= 1)( (15)

The last term in equation (15) is the cost of user group A’s risk-avoiding behaviour triggered by an increased number of trips of category B. A part of the risk avoiding behaviour, lower speed, can be traced to the congestion cost and handled as such. Another part can be found in infrastructure cost, where a higher number of, for example flights, increases the necessary number of safety staff and the dimension of the rescue capacity at the airport. However, we have not identified all of these effects.

4.1.2 Methodology Based on own research (for example in UNITE) and surveying the literature we have identified three approaches to estimate the external marginal accident cost in principle consistent with the definitions discussed above: i) UNITE, ii) Insurance externality, and iii) Computable General Equilibrium models (CGE). In the UNITE approach equation 7 is used and each element is estimated separately and added together. The critical elements are the value of statistical life (VSL), the proportion of internal costs, the risk and the risk elasticity. In studies of the ‘Insurance externality’ the relationship between the traffic flow and the insurance premium are estimated based on aggregate data. The underlying precondition is that the insurance covers all cost. The average driver then pays the average accident cost either in the form of an insurance premium or by bearing accident risk. An additional distance driven by a driver will increase the insurance premium by a small amount. However, as all users are affected the externality will be substantial. The method is most suitable for non-fatal accidents where VSL does not play such a dominant role. Finally, a more general framework such as CGE would be able to cover also the effect of risk avoiding behaviour and could include secondary income effects through the loss to the economy of accidents. However, this approach is dependent on the same detailed information on elasticities etc as the first approach. If behaviour adjustments are included also this could be covered but the underlying knowledge on behaviour adaptation has to come from other sources. The table below summarises these three methods

13 In the literature (Johansson (1996) there also exists a discussion on a possible positive externality in relation to this risk avoiding behaviour if it is not only internal safety that is affected (see PETS (1998) for a summary).


53

Table: 18 Different methods to estimate the marginal external accident cost

Name Definition Includes VSL

Includes the risk elasticity

Risk avoiding behaviour

i) UNITE Uses eq. 7 and estimates each component

Yes Yes No

ii) Insurance Estimates the effect on insurance premiums of increased traffic density

No Yes No

iii) CGE In a CGE framework feedback effects into the economy at large as well ass more behaviour adaptations can be included.

Yes Yes Possible

Note: CGE=computable general equilibrium

4.2 Valuation of accidents The valuation of an accident can be divided into direct economic costs, indirect economic costs and a value of safety per se. The direct cost is observable as expenditure today or in the future. This includes medical and rehabilitation cost, legal cost, emergency services and property damage cost. The indirect cost is the lost production capacity to the economy that results from premature death or reduced working capability due to the accident. However, these two components do not reflect the well-being of people. People are willing to pay large amounts to reduce the probability of premature death irrespectively of their production capacity. The willingness-to-pay estimates the amount of money people are willing to forgo to obtain a reduction in the risk of death. Two biases in recent CVM studies have to be highlighted, the first is the hypothetical bias and the second is what we call here the scale bias. The underlying problem in relation to risk reductions appears to be that people have not formed their preferences yet (Kahneman and Tversky (2000)). When people are confronted with a hypothetical question the answer does not always reflect their actual behaviour and the hypothetical WTP often exceeds the actual WTP. Respondents can be uncertain about their true valuation; a yes-response can be a yes-maybe and a yes-definitely. Several studies have shown that the preference uncertainty can be a key to reduce the hypothetical bias (Li, Löfgren and Hanemann (1996)). Recent studies that explore preference uncertainty suggest that the ‘certain’ WTP could be 50% to 60% of the WTP expressed by all respondents (Blumenschein et.al. (2005), Hultkrantz et.al. (2005)). The ‘certain’ WTP is in some of these studies comparable with the revealed actual WTP. The scale bias refers to the tendency of the respondents to report the same WTP irrespectively of the size of the risk reduction. These effects can be owing to what is known as the “warm glow effect”, that is, the responses “reflect the willingness to pay for a moral satisfaction of contributing to public goods, not the economic value of these goods” (p.57 Kahnman and Knetch (1992)). Another possible explanation is that it can be difficult for the respondents to


54

understand small changes in small probabilities; this can be called scale bias. The result is that respondents report the same WTP for a larger safety improvement as for a smaller improvement. According to standard economic theory the increase in the WTP should be approximately proportional to the size of the risk reduction (Hammit and Graham (1999)). If the responses are only weakly dependent of the magnitude of the risk reduction almost any VSL can be derived from the studies. Attempts have been made to overcome the scale bias through improved visual aid (Corso, Hammit and Graham (2001)) or by presenting the effect as the number of reduced victims instead of reduced risk - frequencies of occurrence rather than probabilities (Beattie (1998), Lindberg (2003)). Nevertheless, the problem seems to remain between samples. Results from Hammit and Graham (1999) indicate that preference uncertainty could explain both the hypothetical bias and the scale bias. Newer studies (Hultkrantz et.al (2005)) do not support this conclusion. Alberini et.al. (2004) added question on the respondent’s certainty in their response in addition to training the respondents to safety questions. They found indication on a WTP almost proportional to the level of the risk reduction when using only the most certain responses. However, the risk level was expressed over ten years (annually 1/10000 and 5/10000) far above what is common in studies on road accidents. Nevertheless, this recent research indicates that well executed CV studies may overcome the problems discussed above. Throughout the world empirical estimates of VOSL diametrically differ between different studies, ranging from a value of less than 200 000 to 30 million US dollars (Blaeij (2003)). Making meta-analysis of this material is difficult and it is important to focus on the reliable studies. Carthy et. al. (1999) use the contingent valuation method to estimate a WTP for less severe outcomes and a risk/risk analysis to link this WTP to fatality. This is the approach used in the UNITE project and in the recent HEATCO project. The HEATCO project has made a survey of the current European practice. The result is depicted in the figure below. The variability in accident cost is huge between different member states. Figure: 13 The current practice in use of VOSL (HEATCLO)

SE

DK

NL

CH

UKFI

FR-road

DE

IT

ESPT

CZ

HU

SK

LTLV

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

35 45 55 65 75 85 95 105 115 125 135

GDP/Capita 2002 EU-25 PPS

Mill

ion

EUR

per f

atal

ity

FR

East

South

North/West

Source: Heatco


55

Irrespectively of this, or depending on this, the HEATCO makes recommendations on methods to adopt when estimating VSL. In addition, based on the work in UNITE the project recommends default values per member states in situations where no available up to date value exists. Table: 19 Recommended values from HEATCO (€2002, factor prices) Country Fatality Severe injury Slight Injury Austria 1,683,000 231,300 18,300 Belgium 1,606,000 244,000 15,700 Cyprus 1,012,000 129,900 9,600 Czech Republic 935,000 118,100 8,800 Denmark 1,672,000 210,300 16,500 Estonia 627,000 79,500 5,900 Finland 1,551,000 208,600 15,600 France 1,551,000 217,800 16,400 Germany 1,496,000 209,400 17,100 Greece 1,067,000 136,500 10,500 Hungary 803,000 103,000 7,600 Ireland 1,837,000 235,100 18,000 Italy 1,496,000 190,700 14,700 Latvia 539,000 67,700 5,100 Lithuania 572,000 73,000 5,400 Luxembourg 2,915,000 432,700 27,200 Malta 1,133,000 142,800 10,700 Netherlands 1,672,000 223,600 18,000 Norway 2,057,000 307,000 21,500 Poland 627,000 79,500 5,900 Portugal 1,056,000 137,400 9,700 Slovakia 704,000 89,100 6,600 Slovenia 1,023,000 130,000 9,700 Spain 1,298,000 160,900 12,100 Sweden 1,573,000 239,300 17,000 Switzerland 1,804,000 262,800 20,100 United Kingdom 1,617,000 211,100 16,800 Notes: Material damages not included. Value of safety per se based on UNITE (see Nellthorp et al., 2001): fatality €1.50 million (market price 1998 – 1.25 million factor costs 2002); severe/slight injury 0.13/0.01 of fatality; Direct and indirect economic costs: fatality 0.10 of value of safety per se; severe and slight injury based on European Commission (1994). For a general approach these values can be used as a in equation 7. Direct and indirect economic costs have to be estimated separately and have to be split into internal and system external (c ). In addition, it should be noted that the values above are expressed at factor price.

4.3 Risk perception The question of internal and external accident cost can be broken into two parts; i) do users consider their own risks and ii) do they consider the risk of others? The straightforward assumption is yes to the first question and no to the second. When VSL is estimated users are asked about their trade-off between accident risk and money. The reply is used to derive VOSL. We thus believe they can value changes in hypothetical risk. If, in a real situation, they also understand and value risk changes, the VSL will be internal. If they do not understand the risk related to their decision we have an information failure. It has often been


56

shown that individuals overestimate small risk and underestimate large risks – “In general, rare causes of death [are] overestimated and common causes of death [are] underestimated”14. This is also the result of a recent study on road users’ risk where the perceived risk is compared with the objective risk of the same road user group (Andersson and Lundborg (2006)). If this result is used to discuss how one person perceives different risk levels the suggestion is that they underestimate risk changes15. Other results suggest that with a more detailed definition of the risk, the timing of the risk16

or using risk defined for relevant age groups17, the difference between actual and perceived risk diminish or disappear. But, exactly how individuals assess the marginal risk related to a change in driven distance or a new trip is unclear. Much of the analysis is based on the assumption that users understand risk changes. This is not a trivial assumption18.

4.4 The risk elasticity It is well known that when the traffic volume increases on a road the speed goes down and the average travel time increases. But what about the accident risk? As the number of vehicles increases the number of accidents will most probably increase; we have not seen any evidence on the opposite effect. However, exactly how the number of accidents increases is important; will the number of accidents increase in proportion to the increase in traffic volume, or will the increase be progressive or degressive? If the number of accidents increases in proportion to the traffic volume the risk, i.e. the number of accidents per vehicle or vehicle kilometer, will be constant; the risk elasticity (E) will take the value nil. If the increase is degressive the accident risk will decline and the elasticity will be negative. This means that an additional user reduces the risk for an accident for all other users. Finally, if the number of accidents increases progressively the risk will increase. An additional vehicle will impose an increased threat to all other vehicles and the external effect will be larger, the elasticity will be positive. As the number of vehicles increases the number of possible interactions increases with the square. This suggests that the risk should increase with traffic volume. Dickerson, Peirson and Vickerman (2000) find that the accident elasticity varies significantly with the traffic flow. They argue that the accident externality is close to zero for low to moderate traffic flows, while it increases substantially at high traffic flows. This is also found by Fridstrøm et al (1995). Winslott Hiselius (2005) concludes also from other literature that the accident risk involving only motor vehicles on urban-road links is independent of the traffic flow. At intersections the evidence is increasing accident risk. However, she also concludes that the estimates on rural roads show a great variation. Vitaliano and Held (1991) show in their estimation that the relationship between accidents and flows is nearly proportional and thus the risk elasticity is close to zero. In an overview of six international studies, Chambron (2000) finds a less than proportional increase in injury and fatal accidents. This has also been found by Hauer and Bamfo (1997) and a majority of the results review in Ardekani et al. (1997). Edlin (2003) studied the effect of traffic density on insurance premiums as well as on fatalities accident only. He found that fatalities decrease

14 Slovic, Fischhoff, and Lichtenstein 1982, p 467 15 See Viscusi (1998) for a short discussion on this topic. 16 Viscusi et.al. (1997). 17 Benjamin et al (2001) but this was not the result of Andersson and Lundborg (2006). 18 Our assumption is close to the risk homeostasis theory (Wilde 1981, 1982) where the user acts as a utility maximizer and tries to keep his risk in equilibrium with his target risk. The zero risk model (Näätänen and Summala 1974, 1976) suggests that in most circumstances the traffic risk is perceived to be equal to zero.


57

with traffic density in low density states but increases in high density state. He found the same pattern for insurance premiums. Ozbay et.al estimated the full marginal costs of highway transportation in New Jersey. From these estimates the elasticites in the figure below can be derived. Property damage and injury accidents increase with traffic volume in urban areas while fatality accidents decline. On freeway and expressway also property damage and injury accidents decline with traffic volume while they increase on interstate roads with increased traffic volume. Figure: 14 Riskelasticity New Jersey

-0.5

0

0.5

1

1.5

2

Prop.Damage Fatality Injury

Ris

kela

stic

ity

Arterial-Local-CollectorFreeway and ExpresswayInterstate

Winslott Hiselius (2005) estimates the relationship between accident and traffic flow on 83 Swedish road sections with information on hourly traffic flow. When the traffic is treated as homogenous (i.e. cars and lorries added together) the result is a decreasing accident risk, i.e. a negative elasticity. However, when car are studied separately the result suggests that the accident rate is constant or increases. However, the result with respect to lorries is reversed, indicating a decreasing number of accidents as the number of lorries increases. This is also the result from the study in UNITE (Lindberg (2003)). Unfortunately, the survey of the literature does not give one single recommendation on the magnitude and the sign on the risk elasticity. The most surprising result is that so many studies find negative elasticities. This is true also for studies that seem to be well executed and control for infrastructure quality etc.

4.5 Insurance cost Edlin (2003a and 2003b) estimates the effect on average insurance premiums of increased traffic density on state level in US. He also has information on insurance cost. This approach includes an aggregate measure of accidents, i.e. insurance cost or premiums, with covers both accident frequencies and accident severity. The underlying precondition is that the insurance system covers all cost, i.e. no underinsurance exists. However, as Edlin notes, fatalities is a specific problem where we can expect underinsurance.


58

The results in Edlin (2003b) suggest that the insurance premium increases strongly in high density states with increased traffic. In California it is estimated that the extra insurance premium could be between $1271 (+- 490) and $2432 (+-670) per year per driver depending on model specification. In low density states the effect is much smaller and sometimes negative. In South Dakota the yearly cost changes with between negative $60 (+-28) to positive $94 (+-36) per driver per year depending on specification. Figure: 15 Insurance externality, US states (lowest, middle and highest) for different model specifications.

-500

0

500

1000

1500

2000

2500

3000

3500

4000

NorthDakota

SouthDakota

Montana Kentucky SouthCarolina

California NewJersey

Hawaii

US

dolla

rs/d

rive

r

Quad.PremLin.PremQuad.CostLin.Cost

4.6 Conclusions This Case study only consists of an overview and state-of-the-art survey. The result is thus not based on any new research made within the GRACE project. We start with a review of the principle of external marginal accident cost as developed for example in UNITE. The principle lends itself easily to an approach where each element in turn is estimated and the marginal cost constructed as a product of these estimated parameters. On the other hand, the insurance externality approach tries to estimate a kind of cost function where the dependent variable is the average insurance premium or insurance cost. This approach presupposes that no underinsurance exists. The third approach mentioned in the introduction is to model the external accident cost in a CGE framework which will cover more of the behavioural and feedback effects discussed in the first approach. However, this approach depends on separate estimates on each component. On each component we may conclude; • A growing consensus on the method to estimate the value of statistical life (VSL)

seems to emerge. The HEATCO project suggests specific values for each Member State.

• Nevertheless, the research on VSL continues with the aim to explore the numerous biases that are found in the currently available estimates.

• On the question on the proportion of internal and external costs and especially the perception of road users risk no new conclusions can be drawn. This is still an area of large uncertainty.


59

• However, making assumptions on the perceived cost, the available databases can be used to estimate the proportion of internal costs.

• There is still no consensus on the risk elasticity. Surprisingly many studies find decreasing risk with increasing traffic volume. This could be a problem of the studies or due to behaviour effects. If we do not control for infrastructure quality, we may find that roads with higher expected traffic volume are designed with a higher traffic safety standard. In addition, road users may react to a perceived increased risk by driving more carefully and slower. This is an unobserved cost component that would increase the cost.


60

5 Air pollution and Greenhouse gases This chapter is divided into four sections. Section 5.1 deals with air pollution from road transport, Greenhouse gases are discussed in section 5.2 and air pollution from railways in sections 5.3. The results are presented in section 5.4.

5.1 Road Transport and Air pollution Several case studies were undertaken in order to estimate the marginal costs due to airborne pollution and greenhouse gases originated by road and rail transport. In order to improve the knowledge regarding the influence of the local conditions and site specific parameters on the calculation of external costs, this document presents the results of the cases studies conducted in four European cities, namely Berlin, Copenhagen, Prague and Athens, which cover a broad range of European countries and local meteorological conditions. The methodology used follows the Impact Pathway Approach, bottom-up methodology developed in the ExternE project series. The starting point for the bottom-up approach for quantification of marginal cost is the micro level, i.e. the traffic flow on a particular route segment. Then, the marginal external costs of one additional vehicle are calculated for a single trip on this route segment.

5.1.1 Description of Case Studies Four case studies for road transport within densely built areas have been conducted. They are expected to complete the picture on air pollution from existing studies and to analyse the variations of environmental costs and the driving parameters. Assessing data availability and due to the fact that a broad range of European countries and local meteorological conditions should be considered, the cities selected for this purpose were Berlin, Prague, Copenhagen and Athens.

Berlin The population of Germany’s capital has stabilised at 3.39 million since 2000, following a slight dip in previous years. The number of commuters between Berlin and Brandenburg has risen slowly in recent years. In comparison with other major cities in western Germany, however, total commuter numbers are low. Of the approximately 1.27 million motor vehicles in Berlin, about 81% are passenger cars and the level of motorisation - i.e. the number of motor vehicles per head of the population - has risen steadily since 1970 but at 322 cars per 1 000 inhabitants is still well below the average of 480 vehicles per 1 000 inhabitants in the old federal states (Department or Urban Development, 2004). When applied to the area of the city, this means 1 400 motor vehicles per square kilometre. For dispersion modelling on the local scale data sets based on 10 year’s averages of 3-hourly measured data by the German meteorological service were used. Detailed population data was also used to model the exposure from atmospheric dispersion of the pollutants on a local scale.


61

Copenhagen Copenhagen is the capital city of Denmark and with its population of more than a million inhabitants it is also the largest of this country. Copenhagen is located on the eastern shore of the island of Zealand (Sjælland) and partly on the island of Amager. It faces to the east the Øresund, the strait of water that separates Denmark from Sweden, and that connects the North Sea with the Baltic Sea. On the Swedish side of the Øresund, directly across from Copenhagen, lie the towns of Malmö and Landskrona. Copenhagen is also a part of the Øresund region, which consists of the eastern part of Zealand in Denmark and the western part of Skåne in Sweden.

For dispersion modelling on the local scale data sets of meteorological data from the Danish THOR system from the National Environmental Research Institute (NERI) were used. Geo-coded population data from NERI was also used for Copenhagen.

Athens Athens is Greece’s capital and largest city and its administrative, economical and cultural centre. It is located in a basin of approximately 450 km2. and is surrounded on three sides by fairly high mountains (Mt. Parnis, Mt. Pendeli, Mt. Hymettus and Mt. Aegaleon), while to the SW it is open to the sea. These mountains are physical barriers with small gaps between them, being the opening of the basin to the sea toward the Saronic Gulf. The City of Athens lies at the heart of the conurbation, with around a quarter of its population (745,514 inhabitants = 23.39%; National Statistical Service of Greece, 2001 Census). Almost the entire basin could be considered as an urban area, characterized by a high concentration of industry (about 50% of the Greek industrial activities) and high motorization (about 50% of the registered Greek cars).

For dispersion modeling on the local scale, data sets based on values of averaged 10-Minute interval measured data for the year 2000 were used. Due to the lack of information regarding not only wind speed and direction but also mixing height and stability classification, data modeled with the NCAR / Penn State Mesoscale Model (MM5) calculated by Vautard (2006) was used for this case. The information was also compared with data provided by the automatic meteorological station of the National Technical University of Athens (NTUA). Population data from the National Statistical Service of Greece (2001) and the 2001 census tables for Eurostat were used to generate the required Geo-coded data.

Prague Prague is the capital and largest city of the Czech Republic. Situated on the Vltava River in central Bohemia, it is home to approximately 1.2 million people. Reflecting the trend in the new Member States, the number of motorized vehicles has been growing over the last years with an increasing traffic density, phenomenon which health issues affects the population and the urban environment substantially. The automobile traffic in Prague is not as heavy as in some other European cities; the number of cars for 100 inhabitants is lower and the public transport network is well developed. The total number of motor vehicles registered on the Prague territory has been continuously growing, the major portion of the motor vehicle number increment goes to passenger cars. In 2005 the number of registered vehicles increased by about 14 500, yielding the total number of registered vehicles of over 750 000 at the years end so there was one car per 1.6 inhabitants in Prague (ÚDI 2006).


62

For dispersion modeling on the local scale, data sets based on values of averaged 10-Minute interval measured data for the year 2000 were used. Due to the lack of information regarding not only wind speed and direction but also mixing height and stability classification, data modeled with the NCAR / Penn State Mesoscale Model (MM5) calculated by Vautard was used for this case. The information was also compared with data provided by the Czech Hydro meteorological Institute in Prague. Population data from the Czech Statistical Office (2001), Population and Housing Census (2001) and Večerková et. al (2006) were used to generate the required Geo-coded data.

5.1.2 Emissions from road vehicles Road vehicle types covered comprise passenger cars, light and heavy duty vehicles (LDV, and HDV respectively) with both petrol and diesel fuelled engine, except for HDV (diesel only). Vehicle emissions were modelled taking into account driving patterns and traffic situations common in city centres.

The emission factors are mainly provided by COPERT III (Computer Programme to calculate Emissions from Road Transport) and HBEFA (Handbook of emission factors for road transport).

In addition to exhaust emissions, non-exhaust emissions due to tyre and brake wear and road dust suspension should be considered for an accurate calculation of fine particulate matter emissions. However, knowledge on specific non-exhaust emissions per different road classes, vehicle categories and driving conditions is scarce and still in the process of scientific discussion. Available information on tyre and brake wear emission factors was recently published in (EEA 2003). Meanwhile first measurements and derived emission factors have been published for different European driving conditions that give a more accurate assessment of non-exhaust emissions (e.g. Düring & Lohmeyer 2004, Gehrig et al. 2003). Besides the emissions from the vehicle operation, the emissions due to fuel provision were also considered. It is assumed that the petrol and diesel are produced in refineries under representative European conditions and with average production technology. Table: 20 Emissions caused by fuel production processes in g/kg fuel

Type of fuel CO2 NOx NMVOC SO2 PM10 Petrol 560 0.105 1.80 1.90 1.10 Diesel 400 0.047 0.62 1.40 0.96

Source: IFEU (1999) Friedrich and Bickel (2001) Furthermore, emissions associated with fuel production are valued with average damage factors for emissions in the corresponding country. These damage factors were calculated based on the assumption that the emission source is not located within densely populated areas.

Table: 21 EU25 Average damage factors for emissions from refineries

€ per tonne emitted Pollutant NOx NMVOC SO2 PM10

EU25 2750 830 3100 15600 Source: Own calculations.


63

5.2 Greenhouse gases The method of calculating costs due to the emission of greenhouse gases (usually expressed as CO2 equivalents) basically consists of multiplying the amount of CO2 equivalents emitted by a cost factor. Due to the global scale of the damage caused, there is no difference how and where in Europe the emissions of greenhouse gases take place. For this reason we recommend to apply the same values in all countries.

The CO2 equivalent of a greenhouse gas is derived by multiplying the amount of the gas by the associated Global Warming Potential (GWP). The GWP for methane is 23, for nitrous oxide 296, and for CO2 it is 1.

A European abatement cost of €20 per tonne of CO2 represents a central estimate of the range of values for meeting the Kyoto targets in 2010 in the EU based on estimates by Capros and Mantzos (2000). They report a value of €5 per tonne of CO2 avoided for reaching the Kyoto targets for the EU, assuming a full trade flexibility scheme involving all regions of the world. For the case that no trading of CO2 emissions with countries outside the EU is permitted, they calculate a value of €38 per tonne of CO2 avoided. It is assumed that measures for a reduction in CO2 emissions are taken in a cost effective way. This implies that reduction targets are not set per sector, but that the cheapest measures are implemented, no matter in which sector.

However, there is a need to strive for more stringent reduction targets than Kyoto. The EU target of limiting global warming to an increase of 2°C of the earth’s average temperature above pre-industrial levels may lead to marginal abatement costs as high as about €95/t CO2. However it is an open question whether such an ambitious goal with such high costs will be accepted by the general population.

Recent work has confirmed the assumption that emissions in future years will have greater total impacts than emissions today (see e.g. Watkiss et al.; 2005a). In a recent report for the Social Cost of Carbon Review on behalf of UK’s Defra, Watkiss et al. (2005b) derive shadow price values, taking into account the expected future development of damage costs and abatement costs. This study is the most current and comprehensive exercise providing consistent values for CO2 emissions. Whereas the damage cost estimates do not rely on specific assumptions for the UK, the abatement cost estimates are based on the UK’s government long-term goal of meeting a 60% CO2 reduction in 2050 (which is broadly consistent with the EU’s 2°C target). On the one hand the costs for reaching a domestic reduction of 60% are higher than implementing a more flexible reduction scheme. On the other hand, the abatement costs only influence the cost curve for later years (starting around 2030) when uncertainties are higher. In addition, the damage cost estimates do not include some important risks.

For application in GRACE we recommend using a range of €14 to €51 (with a central value of €22 per tonne of CO2- equivalent emission in the period 2000 to 2009). These shadow prices were derived from Watkiss et al. (2005b), converting from ₤2000/t C to €2002 (factor prices).

5.3 Rail Transport The rail transport options for passengers in the four urban locations already presented were also analyzed. The relevant options considered are tram, metro (underground train) and light train, all with electrical traction. Two assumptions were analysed for the provision of electricity for an additional train:


64

a) The electricity is produced in a coal-fired power plant, which is common for supplying additional energy demand,

b) The electricity is bought on the European electricity market, assuming the UCTE power production mix.

Emission factors for the coal power plant were based on German emission factors for the electricity production in the year 2000 (UBA 2005). Emission factors for power plants in the Czech Republic, Denmark and Greece were modified based on data from the European Pollutant Emission Register (EPER) for the reporting year 2001. Emission factors for fuel production were taken from European Commission (1999b) The combination of the emission factors with country-specific damage factors per tonne of emission for the pollutants NOx, SO2, NMVOC and PM10 resulted in the cost factors presented in Table 3.7. The differences result from different emission factors (mainly for NOx and CO2) and variations in the cost per tonne of emission between countries. Global warming values were calculated using the range of values recommended per tonne of CO2-equivalent. Furthermore, emissions originated by the wear of overhead wires, rail and tyres were also considered.

5.4 Results The table below presents the results for selected vehicles for each city. The marginal costs include vehicle use, up- and downstream processes and greenhouse gases. A detailed description of the data and results can be found in the appendices.

0.00 5.00 10.00 15.00 20.00 25.00

Athens

Berlin

Copenhagen

Prague

EURO

Diesel HDV Euro IVPetrol LDV EURO VDiesel, LDV EURO IIPetrol, car EURO VDiesel car EURO IVDiesel, car EURO IIPetrol, car EURO II

Figure: 16 Marginal costs due to airborne emissions in EUR /100 vkm The results show that for all vehicle types the higher marginal costs due to airborne emissions correspond to the city of Athens, followed by Berlin, Copenhagen and Prague in that order. The factors that seem to be more relevant for these results are the wind speed and the


65

population density. The high share of low wind speeds for the Athenian area together with a population density close to 20 000 hab/km2 in some zones, leads to a pollutant exposure of the population which is about a factor of two higher compared to the other cities. Petrol cars cause lower cost per vehicle kilometre compared to diesel cars as they emit much less fine particles, leading to lower health impacts.

Table: 22 Marginal cost for urban passenger rail operation, assuming the electricity is produced in a coal-fired power plant and UCTE Mix

City Train type

Energy use per vehicle kWh/km

Cost factor Electricity production (EUR / kWh) PM10 EUR/km Total cost

EUR/ 100 vkm

With UCTE Mix

Athens

Tram 3.81 0.0404 1.05 15.4 4.5Metro 3.22 0.0404 1.05 13.0 3.8Light Train 5.42 0.0404 1.05 21.9 6.4

Berlin


Copenhagen


Prague


It can be noticed that the environmental costs associated with electric trains depends on the sources used to produce the electricity. The UCTE mix considered is shown in the figure below, being the high participation of nuclear energy in the electricity production evident. This fact leads to costs which are a factor of almost three lower than the costs if the electricity is produced in a coal-fired power plant.

Lignite

Coal

Oil

Natural GasNuclear

Hydro

Other

Industry gas

Figure: 17 UCTE electricity production mix in the year 2000


66

6 Noise The perception of sound follows a logarithmic scale, which results in considerable non-linearities of the impacts and associated costs due to a change in noise levels (in the following we refer to the equivalent noise level LAeq). The background noise level plays an important role: whereas in a quiet neighbourhood (40 dB (A)) an additional 40 dB(A), i.e. a doubling of the noise, results in a total level of 43 dB(A), the same noise increment of 40 dB(A) only leads to a total noise level of 60.04 dB(A) in a noisy environment with a background noise level of 60 dB(A). Besides this peculiarity of energetic addition of noise levels the perception, in particular the disturbance caused by changes in the noise level have to be considered. This, together with the very local character of noise makes impact assessment a challenging task; and the models used to quantify noise exposure must be able to map the environment (receptors, buildings), the vehicle technology (PC, HGV etc.) and the traffic situation (e.g. speed and traffic volume) adequately.

First approaches to quantify costs due to noise were using general values per dB, which mostly were derived from hedonic pricing studies. Such studies established a relationship between rents or house prices and properties of the flat or house, one of which was the noise exposure. The next step towards a more differentiated assessment was the inclusion of health effects caused by noise in the analysis. However, annoyance effects were usually still valued based on hedonic pricing studies. This was the case as well in UNITE (see Bickel et al., 2003). In the meantime the available knowledge has improved and allows going a step further towards following the principles of the impact pathway approach.

6.1 Noise impacts

Two major impacts are usually considered when assessing noise impacts:

- Annoyance, reflecting the disturbance which individuals experience when exposed to (traffic) noise.

- Health impacts, related to the long term exposure to noise, mainly stress related health effects like hypertension and myocardial infarction.

It can be assumed that these two effects are independent, i.e. the potential long term health risk is not taken into account in people's perceived noise annoyance.

A large amount of scientific literature on health and psychosocial effects considering a variety of potential effects of transport noise is available. For instance, De Kluizenaar et al. (2001) reviewed the state of the art, reporting risks due to noise exposure in the living environment. They identified quantitative functions for relative and absolute risks for the effect categories presented in the table below.


67

Table: 23 Categorisation of effects and related impact categories (source: De Kluizenaar et al., 2001). Category Measure given ImpactsStress related health effects RR Hypertension and ischemic heart disease Psychosocial effects AR AnnoyanceSleep disturbance AR Awakenings and subjective sleep quality RR = relative risk; AR = absolute risk A more recent study undertaken in Switzerland (Bundesamt für Raumentwicklung, 2004) reviewed additional empirical studies and concluded that for impacts from road and rail noise only few evidence has emerged in addition to De Kluizenaar et al. (2001), which was the basis for calculations in the UNITE project (see Bickel et al. 2003). Figure 6.1 shows the exposure-response functions predicting annoyance reactions on the population level as recommended by European Commission (2002).

Air

0

20

40

60

80

100

45 50 55 60 65 70 75

LDEN

Perc

enta

ge o

f adu

lts

Little annoyed

Annoyed

Highly annoyed

Road

0

20

40

60

80

100

45 50 55 60 65 70 75

LDEN

Perc

enta

ge o

f adu

lts

Little annoyed

Highly annoyed

Annoyed

Rail

0

20

40

60

80

100

45 50 55 60 65 70 75

LDEN

Perc

enta

ge o

f adu

lts

Highly annoyed

Little annoyed

Annoyed

Figure: 18 Percentage of adult population feeling little annoyed, annoyed and highly annoyed as a function of noise levels (source: European Commission 2002).

The general procedure for taking into account the site and technology specific characteristics when calculating marginal noise costs is the following: Two scenarios are calculated: a reference scenario reflecting the present situation with traffic volume, speed distribution, vehicle technologies etc., and the case scenario which is based on the reference scenario, but includes the changes due to the project alternative considered. The difference in damage costs between both scenarios represents the noise costs due to the project assessed. It is important to quantify total exposure levels and not only exposure increments, because for certain impacts


68

thresholds have to be considered. For instance, some exposure-response functions for health impacts are applicable only above a threshold of 70 dB(A) (see De Kluizenaar et al., 2001).

Depending on the exposure-response relationships available different noise indicators are required for the quantification of impacts. Examples of indicators that are commonly used are equivalent noise levels for different times of day, e.g. LAeq(7.00-19.00), LAeq(19.00-23.00), LAeq(23.00-7.00) and the compound day-evening-night noise indicator LDEN (see European Commission, 2002 for details on noise indicators). Usually noise levels are calculated as incident sound at the façade of the buildings

6.2 Valuation of Annoyance Given its high importance for the results and the challenges in its measurement, the value of annoyance caused by noise requires particular consideration. The main cost component of annoyance is disutility experienced, for which no market exists. Stated preference (SP) and revealed preference (RP) methods have been employed to estimate the economic value of changes in noise levels. The noise valuation literature is dominated by Hedonic Price (HP) studies (most of them old) on road traffic and aircraft noise of varying quality. HP studies analyse the housing market to explore the extent to which differences in property prices reflect individuals´ willingness-to-pay (WTP) for lower noise levels. Resulting values seem to be problematic to transfer, however, both theoretically and in practice (Day 2001).

The number of SP studies on road traffic noise is increasing, but only a few present WTP in terms of “euro per annoyed person per year” for different annoyance levels (little annoyed, annoyed and highly annoyed), which correspond to the endpoints of exposure-response functions. Due to the low number of studies that can be used for this approach, a “second-best” alternative was to evaluate the SP studies available with regards to quality (e.g. avoid using studies with scenarios based on changes in exposure rather than annoyance and health impacts), choose the best ones, and calculate a value in terms of “euro per dB per person per year”. This was done by Navrud (2002) to establish an EU-value.

To enable the application of the exposure-response functions predicting annoyance reactions on the population level as recommended by European Commission (2002), the project HEATCO’s carried out stated preference surveys in five European countries (see Navrud et al. 2006). Based on surveys in Germany, Hungary, Norway, Spain, Sweden and the UK, values for application in Europe were derived for the annoyance levels highly annoyed, annoyed and little annoyed.


69

Table: 24 Annual willingness-to-pay by annoyance level for reducing annoyance (€2002 factor costs per person).

€2002 factor costs per person Road, aircraft – little annoyed 30 Road, aircraft – annoyed 68 Road, aircraft – highly annoyed 68 Rail – little annoyed 30 Rail – annoyed 48 Rail – highly annoyed 48

Existing estimates show considerable non-linearities of marginal noise cost with background noise levels. The figure below presents exemplary results of the UNITE project for noise costs from passenger cars in urban areas. Costs are increasing from day to night, reflecting the higher disturbance effect of noise during night time. In Berlin the average number of persons per road kilometre affected by noise is slightly higher than in Stuttgart. However, the costs are more than a factor of three lower due to the much higher number of vehicles and higher speeds on Frankfurter Allee leading to a higher background noise level. In Helsinki the population density along the route considered is lower than in Berlin and Stuttgart, furthermore the average distance from buildings is higher – leading to lower noise costs.

0 1 2 3 4 5

day

night

day

night

day

night

Hel

sink

iB

erlin

Stu

ttgar

t

EUR / 100 vkm

Figure: 19 Marginal noise costs due to a passenger car in Helsinki, Berlin and Stuttgart (source: Bickel et al. (2003)).


70

7 Sensitive areas The main purpose of this CS is to explain and to assess the differences in the transport costs per unit of transport performance (vehicle- or train-kilometre) between a sensitive and an “insensitive” area. These cost differential factors – and not new cost rates – are the main output of the case study. The factors can be applied to existing estimates of average and marginal environmental costs to assess the differences between the cost rates in a sensitive and in an “insensitive” area in absolute terms.

7.1 Definition and indicators The notion of sensitive areas is often used in connection with environmental policy questions. The underlying idea is that there are some areas that require stronger protection than others. The Eurovignette Directive (1999/62/EC and 2006/38/EC) allows for the possibility to apply mark-ups to tolls in the case of roads in sensitive areas, in particular in mountain regions (Alps, Pyrenees, etc.) for cross-financing the investment costs of other transport infrastructures of a high EU interest in the same corridor and transport zone. However, despite its frequent use and its intuitive appeal, there is no commonly agreed definition of what constitutes a sensitive area. Although there are many attempts to define sensitive areas, there exists no clear EU-wide definition as yet. At least, most definitions give a concrete idea of what is meant by sensitivity: Thus we define sensitive areas as areas • where damages are higher;

− because of higher environmental pressures − and / or because of more damaging effects of the same pressure level

• and possibly where unique natural resources or cultural heritages are in danger.

None of the analysed definitions in the literature mentions the traffic volume. Hence, a high traffic volume alone cannot make an area sensitive, but contributes to higher environmental pressures. Most the definitions are not operationalised so as to allow differentiating between sensitive and “insensitive” areas. A common rescue is to cite examples of areas which are sensitive such as protected areas (national parks, landscape protection reserves, nature conservation areas, natural monuments, biosphere reserves, and forest reserves), UNESCO World Heritage Sites, mountain areas (area covered by the Alpine Convention), densely populated areas, wetlands or coastal zones, certain marine areas, and urban areas.19 However, this does not mean that all other areas are “insensitive”. Hence, what is meant by sensitivity seems pretty clear, but an exact definition which allows drawing a borderline between sensitive and insensitive areas is still missing.

7.2 Cost categories In this chapter we estimate the cost differentials between an Alpine area and a flat, “insensi-tive” area for road and rail transport and the reasons behind them. As the main result we derive factors between the costs in Alpine and flat areas – differentiated for passenger and goods transport.

19 See for example European Commission (2003), Sensitive areas and transport, p. 2 and T&E (2005), Sustainable Freight Transport in Sensitive Areas, p. 33-34.


71

The method is based on the impact pathway approach. For each step in the pathway a comparison is made between a Alpine area and a flat area is made. The factors for each step are added together to suggest a total cost difference between the Alpine and the flat area. The impact pathway steps considered is Emissions, Concentration and Impacts.

7.2.1 Air pollution Pollutants which are formed at considerable distances from the emission source (e.g. nitrate aerosols from NOx) or are transported over large distances are important to consider when analysing the full costs of air pollution, but not when we analyse the costs of traffic through mountain areas. The only pollutant with local effects is PM10. The effects of primary PM10 (with local effects) and secondary PM10 (with regional effects) has, however, to be disentangled. It follows that the costs per vkm or trainkm for crop losses and forest damages are equal in flat and Alpine areas, since these effects are caused by regional pollutants. The factors we derive apply to the health costs20 and damages to buildings which are caused locally. Table: 25 Results for local air pollution Impact pathway Cost driver Road factor Rail factor Emissions Gradients 1.06

(1.02 – 2.28) 1

Altitude 1.35 (1.10-1.60)

1

Concentration Topographical and meteorological conditions

4.22 (2.50 – 6.25)

4.22 (2.50 – 6.25)

Impacts Population density 0.87 0.83 Total 5.25

2.55 – 19.8) 3.5

(2.08 – 5.19) The overall factor is 5.25 for road transport. For cars the factor is slightly higher (5.35) than for HGVs (5.15). The difference between cars and HGV is explained by the large emissions of cars on steep gradients, while for the HGVs the emissions increase less when the gradient rises. However, the interval is large, reaching from 2.4 to 19.8. This shows the large uncertainties involved in the calculations. For rail transport the overall factor is smaller: 3.5. The reason is that the higher emissions due to the gradients are not emitted along the rail track, but at the location of electricity production. Thus only the factor for population density and higher concentration for the same emissions (due to abrasion and whirling up which seem to be identical in flat and Alpine areas) apply.

7.2.2 Noise For road noise we find higher motor noise emissions due to gradients. Furthermore, noise propagation conditions are better in mountain valleys than in flat areas due to temperature inversion and amphitheatre effects and reflections. Due to these effects a much larger distance from the road is necessary to reduce noise to a certain level along mountainsides than in a flat area.

20 For residents


72

Table: 26 Results for noise Impact pathway Cost driver Road factor Rail factor Emissions Gradients 1.15

(1.06 – 1.82) 1

Concentration Topographical and meteorological conditions

5 (2.5 – 12.5)

5 (2.5 – 12.5)

Impacts Population density 0.87 0.83 Total 5.0

(2.3 – 19.8) 4.15

(2.1 – 10.4) Due to the lower population density in Alpine areas and the higher emissions, the final result for road noise is also a factor of 5 (2.3 – 19.8). For rail the results are similar; noise propagation conditions are identical, the population density along the Gotthard rail line is slightly lower than along the Gotthard motorway, but emissions seem not to be higher in Alpine areas: The main noise source of rail traffic is not the motor of the engine but the noise from moving rolling stock (wheels on the rail track). In the literature we could only find some hints on higher noise emissions on gradients, but no quantifications. Thus the factor for rail is 4.2.

7.2.3 Visual intrusion Visual intrusion is more severe in Alpine areas where the traffic routes can be seen from much farther away (from the mountain flanks) than in a flat area. However, visual intrusion is rather irrelevant for the GRACE-perspective (marginal costs), but a relevant alpine-specific cost factor (average costs). The CS makes a pioneering attempt to quantify the extra cost due to visual intrusion.

7.2.4 Accidents It is well known that accidents in tunnels and on bridges can have more serious consequences than accidents on a “normal” traffic route. Moreover, on descending slopes the braking distance is larger. However, to our knowledge no evaluations on accident rates for Alpine and flat areas exist. Therefore the CS evaluated detailed accident data from the Swiss motorways to fill this gap. In a comparison between the Gotthard motorway and the main motorway in the flat area of Switzerland the causality rate (casualties per vkm) on motorways was 1.22 times higher in the Alpine area. In contrast, we have to assume that rail accidents are identical in Alpine and flat areas, because the external accident costs of rail freight transport are almost negligible. Hence, we could not find any evidence that the costs are higher in an Alpine environment.

7.2.5 Infrastructure costs It seems clear that infrastructure costs are higher in Alpine areas than in flat areas: on the one hand more tunnels and bridges are necessary and on the other hand the road or rail track must adjust to the Alpine topography which means more curvy roads or rail tracks and thus longer traffic routes. This is especially clear when planning new infrastructures in the Alpine area since the investment costs tend to be higher. The exact amount of costs, however, is very project specific and need not be determined here as investment costs are always taken into account when planning new roads or rail tracks.


73

Instead, we concentrate on maintenance costs. Road maintenance costs are higher in Alpine areas due to bridges, tunnels, and rutting of slow HGV traffic. Therefore we evaluated data on motorway maintenance costs per canton, the most comprehensive data we could find. A rough estimation shows that, although the traffic volume in the Alpine area is about 3 times lower, the maintenance costs per kilometer motorway are about 1.5 times higher. Hence, the factor for road maintenance costs per vkm between Alpine and flat areas is about 4.5. For rail maintenance costs the Grace CS 1.2E has been used and the result is that the costs in Alpine regions are 1.4 times higher.

7.3 Conclusion The figure below summarizes all the results for the factors between Alpine and flat areas (where we use the reduced factors for total instead of local air pollution). For road transport the highest factor of more than 10 is observed for visual intrusion. For noise and infrastructure costs a factor of 5 is estimated. Effects of local air pollution are also in that magnitude. But due to the regional air pollutants the factor is about halved to 2.1. The factor for accidents of 1.2 is again about half of this. Figure: 20 Factors Alpine / flat for the different effects for road (car and HGV) and rail transport (passenger and freight transport)

0

1

2

3

4

5

6

7

8

9

10

11

air pollution noise visualintrusion

accidents infrastructure total

fact

or A

lpin

e / f

lat

car HGV passenger transport freight transport


74

8 References Alberini, Anna. (2004). ”Robustness of VSL Values from Contingent Valuation Surveys,” Fondazione Eni Enrico Mettei, Nota di lavoro 135.

Andersson, Henrik. (2005). Willingness to Pay for Reduction in Road Mortality Risk: Evidence form Sweden. Lund: Lund Economic Studies Number 126.

Andersson, M. (2005) Econometric Models of Railway Infrastructure Costs in Sweden 1999-2002, Proceedings of the third Conference on Railroad Industry Structure, Competition and Investment. Stockholm School of Economics (Sweden).

Andersson, Mats (2006), Case study 1.2D I: Marginal railway infrastructure cost estimates in the presence of unobserved effects, Annex to Deliverable D 3 Marginal cost case studies for road and rail transport, Information Requirements for Monitoring Implementation of Social Marginal Cost Pricing. Funded by Sixth Framework Programme. ITS, University of Leeds, Leeds, March 2006

Andersson, Mats (2006), Case study 1.2D II: Marginal railway renewal costs - A survival data approach, Annex to Deliverable D 3 Marginal cost case studies for road and rail transport, Information Requirements for Monitoring Implementation of Social Marginal Cost Pricing. Funded by Sixth Framework Programme. ITS, University of Leeds, Leeds, March 2006

Ardekani, S., E. Hauer and B. Jarnei (1997) in: Gartner, N et.al. (eds) ‘Traffic Impact Models in Traffic Flow Theory – A State-of-the-art Report’, Oak Ridge National Laboratory.

Bak M., Borkowski P., Musiatowicz-Podbial G., Link H. (2006), Case study 1.2C: Road infrastructure cost in Poland, Annex to Deliverable D3, Marginal cost case studies for road and rail transport. Funded by Sixth Framework Programme. ITS, University of Leeds, Leeds, October 2006

Beattie J, Covey J, Dolan P, Hopkins L, Jones-Lee M, Loomes G, Pidgeon N, Robinson A and Spencer A (1998) On the Contingent Valuation of Safety and the Safety of Contingent Valuation: Part 1-Caveat Investigator. Journal of Risk and Uncertainty, 17:5-25

Bickel, P., Schmid, S., Tervonen, J., Hämekoski, K., Otterström, T., Anton, P., Enei, R., Leone, G., van Donselaar, P., Carmigchelt H. (2003), Environmental Marginal Cost Case Studies, UNITE Deliverable 11, Stuttgart.

Blumenschein, Karen, et al. (2005). “Eliciting WTP without Bias: Evidence from a Field Experiment,” University of Kentucky Working Paper (Feb. 23, 2005).

Bonsall PA et al (2006) Understanding Variation in Estimates of Road Congestion Costs

Booz, Allen & Hamilton with TTCI UK (2005). Review of Variable Usage and Electrification Asset Usage Charges: Final Report. Report to the Office of Rail Regulation, London.

Bossche M A van den, Certan C, Goyal P, Gommers M and Sansom T (2000), Marginal Cost Methodology. UNITE (UNIfication

Bundesamt für Raumentwicklung (2004), Externe Lärmkosten des Strassen- und Schienenverkehrs in der Schweiz, Aktualisierung für das Jahr 2000, Bern, Switzerland.

Carthy,T, Chilton,S. Covey J, Hopkins L, Jones-Lee M, Loomes G, Pidgeon N, and Spencer A (1999) On the Contingent Valuation of Safety and the Safety of Contingent Valuation: Part 2-The CV/SG ‘chained’ approach. Journal of Risk and Uncertainty, 17:3 187-213


75

Chambron N (2000), Comparing Six DRAG-Type Models. In: Gaudry M and Lassarre S (eds.), Structural Road Accident Models - The International Drag Family, 205-224.

Christoph Lieb, Stefan Suter, Peter Bickel. Environmental costs in sensitive areas. GRACE (Generalisation of Research on Accounts and Cost Estimation). Funded by Sixth Framework Programme. ITS, University of Leeds, Leeds. August 2006.

Corso, Phaedra S., James K. Hammitt, and John D. Graham. (2001). “Valuing Mortality-Risk Reduction: Using Visual Aids to Improve the validity of Contingent Valuation,” The Journal of Risk and Uncertainty 23(2), 165-184.

Day, B. (2001), The theory of Hedonic Markets: Obtaining welfare measures for changes in environmental quality using hedonic market data. March 12th 2001, Economics for the Environment Consultancy (Eftec), London.

de Blaeij, Arianne, et al. (2003). “The Value of Statistical Life: a Meta Analysis,” Accident Analysis and Prevention 35(6), 973-986.

De Kluizenaar, Y., Passchier-Vermeer, W., Miedema, H.M.E (2001), Adverse effects of noise exposure on health – a state of the art summary, TNO report 2001.171, Leiden.

Department for Transport (web accessed 2006) Guidance on the conduct of transport analysis studies www.webtag.org.uk

Dickerson AP, Peirson JD and Vickerman RW (2000), Road Accidents and Traffic Flows: An Econometric Investigation. Economica 67, 101-121.

Duncan, C.S., A.J. Khattak, and F.M. Council (1998) ‘ Applying the Ordered Probit Model to Injury Severity in Truck-Passenger

Düring, I., W. Bächlin, A. Baum, A. Hausmann & A. Lohmeyer (2005), Emission factors for vehicle induced non exhaust P, Paper presented at 14th International Symposium Transport and Air Pollution, 1-3 June 2005, Graz, Austria.

Edlin, A and P. Karaca-Mandic. (2003) The accident externality fm driving. Working paper No E03-332. University of California, Berkeley.

European Commission (2002) Position paper on dose response relationships between transportation noise and annoyance. EU's Future Noise Policy, WG2 – Dose/Effect. Office for Official Publications of the European Communities, Luxembourg.

Fridstrøm L, Ifver J, Ingebrigtsen S, Kulmala R and Thomsen LK (1995), Measuring the contribution of randomness, exposure, weather, and daylight to the variation in road accident counts. Accident Analysis and Prevention 27, 1-20.

Friedrich, R., Bickel, P. (eds.), (2001), Environmental External Costs of Transport, Springer Verlag, Heidelberg.

Gaudry, M., and Quinet, E., 2003. Rail track wear-and-tear costs by traffic class in France. Universite de Montreal, Publication AJD-66.

Hajek (1993), Allocation of Pavement Damage Due to Trucks Using a Marginal Cost Method, Transportation Research record 1613.

Hammitt, J.K. and J.D. Graham (1999): Willingness to pay for health protection: Inadequate sensitivity to probability? Journal of Risk and Uncertainety, 18, 33-62.

Haraldsson, Matttias (2006) Case study 1.2BI:Marginal cost for road maintenance and operation – a cost function approachg. Annex to Deliverable D3, Marginal cost case studies


76

for road and rail transport. Funded by Sixth Framework Programme. ITS, University of Leeds, Leeds, October 2006

Haraldsson, Matttias (2006) Case study 1.2BII:The marginal cost for structural repair of roads - a duration analysis approach. Annex to Deliverable D3, Marginal cost case studies for road and rail transport. Funded by Sixth Framework Programme. ITS, University of Leeds, Leeds, October 2006

Hauer, E. and B.N. Persaud (1987): “How to estimate the safety of rail-highway grade crossings and the safety effects of warning devices”, Transportation Research Record, 1987

Haur, E. and J, Bamfo (1997) in: Proceedings of the ICTCT conference ‘Two tools for finding what Function Links the Dependent Variable to the Explanatory Variables’, Lund.

HEATCO (2006), http://heatco.ier.uni-stuttgart.de/ 2006-10-25

Heike Link (DIW), Case study 1.2A: Marginal motorway infrastructure costs for Germany, Annex to Deliverable D 3 Marginal cost case studies for road and rail transport. Funded by Sixth Framework Programme. ITS, University of Leeds, Leeds, March 2006

Highway Research Board, 1961. The AASHO-Road-Test - History and Description of Project. Special Report 61 A, Washington D.C.

Hultkrantz, L., Lindberg, G. & Andersson, C. (2006) The value of improved road safety. Journal of Risk and Uncertainty, 32, 151-170. Johansson, P. and Nilsson, J. (2002). “An economic analysis of track maintenance costs”. Deliverable D10_AnnexA3. UNITE. Reprinted in 2004. Transport Policy 11(3), pp. 277-286.

Johansson, O (1996), Welfare, externalities, and taxation; theory and some road transport applications. Gothenburg University.

Johansson, P., and Nilsson, J. E. 2002. An Economic Analysis of Track Maintenance Costs. UNITE (UNIfication of accounts and marginal costs for Transport Efficiency) Deliverable 10, Annex A3. Funded by EU 5th Framework RTD Programme. ITS, University of Leeds, Leeds. http://www.its.leeds.ac.uk/projects/unite/.

Johnsson, D. and C.Nash (2006), Case study 1.3A: Charging for Scarce Rail Capacity in Britain, Annex to Deliverable D 3 Marginal cost case studies for road and rail transport. Funded by Sixth Framework Programme. ITS, University of Leeds, Leeds, October 2006

Kahneman, Daniel, and Jack Knetch. (1992). “Valuing public goods. The purchase of moral satisfaction,” Journal of Environmental Economics and Management 22(1), 57-70.

Li, Z., K.C.Sinha, and P.S.McCarthy (2001), A determination of load and non-load shares of highway pavement routine maintenance expenditure. Road and Transport Research, June 2002.

Li, Z., K.C.Sinha, and P.S.McCarthy (2002) Methodology to determine load- and non-load-related shares of highway pavement rehabilitation expenditures. Transportation Research Record 1747.

Li, Chuanzhong, Karl-Gustaf Löfgren, and W. Michael Hanemann. (1996). ”Real Versus Hypothetical Willingness to Accept. The Bishop and Heberlein Model Revisited,” Umeå Economic Studies No. 420, Umeå University 1996.

Lindberg, G. (2006), Case study 1.4: State-of-the-art of external marginal accident costs, Annex to Deliverable D 3 Marginal cost case studies for road and rail transport. Funded by Sixth Framework Programme. ITS, University of Leeds, Leeds, March 2006


77

Lindberg, G., 2002. Marginal Costs of road maintenance for heavy goods vehicles on Swedish roads. UNITE (UNIfication of accounts and marginal costs for Transport Efficiency) Deliverable 10, Annex A2. Funded by EU 5th Framework RTD Programme. ITS, University of Leeds, Leeds. http://www.its.leeds.ac.uk/projects/unite/.

Link, H. (2002) Deliverable 10: Case Studies on Marginal Infrastructure Costs, version 1.3. Project Co-ordinator: ITS, University of Leeds (Leeds).

Link, H. and Nilsson, J. (2005). “Infrastructure”. Nash, C. and Matthews, B. Editors (2005). Measuring the Marginal Social Cost of Transport. Research in Transportation Economics Volume 14. Elsevier (Amsterdam).

Link, H., Dodgson, J., Maibach, M., Herry, M., 1999. The Costs of Road Infrastructure and Congestion in Europe. Physica/Springer, Heidelberg.

Marti, Michael, Neuenschwander René (2006), Case study 1.2E Track Maintenance Costs in Switzerland, Annex to Deliverable D 3, Marginal environmental cost case studies for road and rail transport. Funded by the Sixth Framework Program. IER, Universität Stuttgart, Stuttgart, March 2006

Munduch, G., Pfister, A., Sogner, L. and Stiassny, A. (2002) “Estimating Marginal Costs for the Austrian Railway System.” Working paper No. 78, Vienna University of Economics.

Nash, C., and Wheat, P. (2006) “Structure of Costs and Charges Review – review of work on avoidable costs and on cost variability”. ITS, University of Leeds (Leeds).

Nash, C.A. and Matthews, B. with contributions from partners (2002). British Rail Infrastructure Case Study. UNITE (UNIfication of accounts and marginal costs for Transport Efficiency) Working Funded by 5th Framework RTD Programme. ITS, University of Leeds (Leeds).

National economic research associates (NERA) (2000) Review of Overseas Railway Efficiency: A Draft Final Report for the Office of the Rail Regulator. NERA (London).

Navrud, S. (2002) The State of the Art on Economic Valuation of Noise. Report prepared for the European Commission, DG Environment. April 14th 2002. http://europa.eu.int/comm/ environment/noise/ 020414noisereport.pdf

Navrud, S., Trædal, Y., Hunt, A., Longo, A., Greßmann, A., Leon, C., Espino, R., Markovits-Somogyi, R., Meszaros, F. (2006) Economic values for key impacts valued in the Stated Preference surveys. HEATCO Deliverable 4.

Newbery, D. M., 1988a. Road Damage externalities and Road User Charges. Econometrica 56, 295-316.

Newbery, D. M., 1988b. Road Users Charges in Britain. The Economic Journal 98, 161-176.

Newbery, D. M., 1990. Pricing and Congestion: Economic principles relevant to pricing roads. Oxford Review of Economic Policy 6 (2), 22-38.

Oum, T. H. and Waters II, W. G., 1998. Recent Developments in Cost Function Research in Transportation. In: De Rus, G. and Nash, C. (Eds.). Recent Developments in Transport Economics. Ashgate Publishing, Aldershot, 33-73.

Ozbay, K., Bartin, B., Berechman, J., 2001. Estimation and evaluation of full marginal costs of highway transportation in New Yersey. Journal of Transportation Statistics, Vol 4, No. 1.

Peirson, J. Skinner, I. and Vickerman, R. (1994) The microeconomic analysis of the external costs of road accidents. Discussion Paper 94/6, CERTE, May 1994


78

Peltzman, S. (1975), The effects of autmobile safety regulations, Journal of Political Economy, vol. 83(4), page 677 - 725.

Peter Bickel, Sandra Torras Ortiz, Ulrike Kummer (IER), Case study 1.5 Air pollution and Greenhouse gases, Annex to Deliverable D 3, Marginal environmental cost case studies for road and rail transport. Funded by the Sixth Framework Program. IER, Universität Stuttgart, Stuttgart, September 2006

Sansom, T., Nash, C., Mackie, P. Shires, J. and Watkiss, P (2001) “Surface Transport Costs and Charges, Great Britain 1998”, Report By ITS and AEA technology on behalf of the Department for Transport, Environment and the Regions.

Schreyer, C., Schmidt, N., Maibach, M., 2002. Road econometrics – Case study motorways Switzerland. UNITE (UNIfication of accounts and marginal costs for Transport Efficiency) Deliverable 10, Annex A1b. Funded by EU 5th Framework RTD Programme. ITS, University of Leeds, Leeds. http://www.its.leeds.ac.uk/projects/unite/.

Shephard, R. W., 1970. The Theory of Cost and Production. Princeton: Princeton University Press.

Shepherd, S.P., Koh, A., and May, A D. (2006) “Investigating a Select Link Analysis approach to cordon design : Executive Summary”. Unpublished report to UK Department for Transport (June 2006).

Small, K. A., Winston, C., Evans, C. A., 1989. Road work: a new highway pricing and investment policy. The Bookings Institution, Washington.

Smith, A., P. Wheat (2006), Case study 1.2G : Assessing the marginal infrastructure wear and tear costs for Great Britain’s railway network, Annex to Deliverable D 3 Marginal cost case studies for road and rail transport. Funded by Sixth Framework Programme. ITS, University of Leeds, Leeds, November 2006

Talvitie A.P. and Sikow, C., 1992. Analysis of Productivity in Highway Construction using Alternative Average Cost Definitions. Transportation Research 26B, No.6, 461-478.

TÁNCZOS, K., Á. DÉNESFALVY, F. MÉSZÁROS, P. RÓNAI, and Á. TÖRÖK, (2006), Case study Case study 1.2F: Rail Infrastructure Cost in Hungary – Operational model for infrastructure cost and charge calculation, Annex to Deliverable D 3 Marginal cost case studies for road and rail transport Deliverable. Funded by Sixth Framework Programme. ITS, University of Leeds, Leeds, October 2006

Tervonen, J. and Idström, T. (2004). Marginal Rail Infrastructure Costs in Finland 1997-2002. Report from the Finnish Rail Administration. Available at www.rhk.fi [accessed 20/07/2005].

US Department of transportation (2000), Addendum to the 1997 Federal Gighway Cost allocation study. www.fhwa.dot.gov/policy/hcas/addendum.htm. (2006-08-17)

Večerková J., e. Vojtová, j. Pospíšilová. (2006), General characteristics, causes and factors of environmental changes, Yearbook Prague Environment 2005, Prague.

Vitaliano DF and Held J (1991), Road Accident External Effects: An Empirical Assessment. Applied Economics 23, 373-378.

Winslott Hiselius, L. (2005) External Cost of Transports Imposed on Neighbours and Fellow Road Users, Lund Economic Studies Number 130, University of Lund.

D3 – Marginal cost case studies for road and rail transport

Documents

Transcript of D3 – Marginal cost case studies for road and rail transport