Spatial Heterogeneity and the Geographic

Distribution of Airport Noise

The authors thank Lesli Ott for excellent research assistance and the Atlanta Department of Aviation for noise contour data. We thank meeting participants at the Federal Reserve System Committee on Regional Economic Analysis, especially Anil Kumar, for their comments. Nancy Lozano-Gracia also provided helpful comments on an earlier version of the paper. The

views expressed are those of the authors and do not necessarily reflect official positions of the Federal Reserve Bank of St. Louis, the Federal Reserve System, or the Board of Governors.

Jeffrey P. CohenAssociate Professor of Economics

University of Hartford

Cletus C. CoughlinVice President and Deputy Director of

ResearchFederal Reserve Bank of St. Louis

ASSA Meeting – January 2010Atlanta, GA

Motivation & Major QuestionsMotivation

Possibility that airport noise depends on

• neighborhood demographic characteristics

• proximity to airport (noise and distance are not necessarily

correlated)

• prices of houses

Major Questions

• Which of the preceding variables are significant

determinants of noise? Does estimation incorporating

spatial heterogeneity generate results that differ from

estimation ignoring spatial heterogeneity?

• How do the results differ across locations? Why?

ApproachOur approach

• Locally Weighted Regressions (LWR) (non-parametric

spatial specification)

• This is opposed to spatial autocorrelation and/or

spatially lagged variables (parametric approach)

Advantage of LRW

• Allows for non-uniformity in relationships between

variables

Background: Airport NoiseFederal Register (2000)

Annoyance: the adverse psychological response to noise

• 12 percent of people subjected to a DNL of 65 decibels

report that they are “highly annoyed”

• 3 percent are highly annoyed when subjected to a DNL of

55 decibels

• 40 percent are highly annoyed at a DNL of 75 decibels

Nelson (2004)

Since 1979 federal agencies have regarded land subject to 65

to 74 decibels as “normally” incompatible with residential use

and land subject to less than 65 decibels as “normally”

compatible with residential land use

Airport Noise, Proximity & Housing Prices• Standard finding is that airport noise reduces housing prices:

McMillen (2004): 9 percent reduction for houses near Chicago O’Hare for 65 db or more

Over time, noise levels around O’Hare have decreased Espey and Lopez (2000): 2 percent reduction for houses near Reno

Cannon for 65 db or more Lipscomb (2003): no effect for houses in College Park, GA, which

is near Atlanta Hartsfield-Jackson

• Proximity: must control for proximity to accurately measure effect of noise

• Proximity tends to have a positive effect: Tomkins, Topham, Twomey, and Ward (1998): Manchester McMillen (2004): O’Hare Lipscomb (2003): Hartsfield-Jackson

Other Studies of Airport NoiseCohen and Coughlin (2008): Atlanta, GA airport

• noise assumed to be exogenous – common

assumption

• houses exposed to 70 db noise faced sale prices that

were 20 percent lower than houses in the buffer

zone

Sobotta, et. al. (2007): Phoenix, AZ airport

• modeled noise as dependent variable

• found higher Hispanic neighborhoods were exposed

to significantly more noise

• Estimation by Ordered Probit

ModelY = f (X, Z, u)

Where

Y =

noise (ordered categorical variable ; can be <65db, 65db, 70db)

X =

{ log(distance), log(age), % Hispanic, % black, median household income }

Z =

{ log(fitted sale price) }

u =

normally distributed error term, zero mean and constant variance

Ordered ProbitFor ordered probit we re-define noise:

Y Noise

0 < 65 db

1 65 db but < 70 db

2 70 db

Alternative Estimation ApproachLocally Weighted Regressions (LWR)

• addresses spatial nature of the data

Background

• Spatial Econometrics (parametric approach)

• LWR (non-parametric approach)

Locally Weighted Regressions (LWR)More tractable for ordered probit

Advantages

• non-parametric approach to address spatial variation

• allows for non-linearity in relationships between

independent variables and dependent variables

• for ordered probit, can be estimated by a “pseudo

maximum likelihood” approach

Locally Weighted Regressions (LWR)With 3 regimes in ordered probit, pseudo log-likelihood

function is:

where is standard normal c.d.f.

is parameter vector for observation I

Doj = 1 if obs. j = 0, = 0 otherwise

D1j = 1 if obs. j = 1, = 0 otherwise

D2j = 1 if obs. j = 2, = 0 otherwise

• impact of spatial weights enters in a non-parametric way

0 1 2log ( ' ) log ( ' ) log ( 'j ij j i j j i i j j i i jw D X D y X D y X

i

WeightsSomewhat different – we use Gaussian weight function:

where is standard normal density function

= distance between house i and house j

= standard deviation of distances between

house i and all other houses j

= bandwidth

• We use cross-validation to choose preferred b: vary b to be 0.4, 0.6, 0.8, 1.0

b = 0.4 maximizes the pseudo log-likelihood function

ijij

i

dW

s b

ijd

is

b

DataAtlanta airport

Noise Contours:Atlanta Department of Aviation

Housing Prices & Characteristics:2003

Demographics:% Hispanic, % black, medianhousehold income from U.S.Census Bureau, 2000.

Summary Statistics – 508 Observations   Count Percentage

House Sales in the buffer zone – 2003 contours 343 67.5

House Sales in 65 db zone -- 2003 contours 146 28.7

House Sales in 70 db zone -- 2003 contours 19 3.7

House Sales in Atlanta 49 9.6

House Sales in College Park 147 28.9

House Sales in Conley 60 11.8

House Sales in East Point 66 13.0

House Sales in Forest Park 136 26.8

House Sales in Hapeville 50 9.8

1 story 425 83.7

2 or more stories 83 16.3

2 or less bedrooms 138 27.2

3 bedrooms 258 50.8

4 bedrooms 99 19.5

5 or more bedrooms 13 2.6

1 bathroom 246 48.4

2 bathrooms 151 29.7

3 or more bathrooms 111 21.9

0 or 1 fireplace 494 97.2

2 or more fireplaces 14 2.8

Summary Statistics – 508 Observations   Mean Range

Price (dollars) 128,442 32,378-460,500

Distance (miles) 3.29 1.06-6.06

Acres 0.37 0.03-3.88

Age (years) 39.85 0-100

B1kHH00 (percent) 56.96 0-97.5

HispHH00 (percent) 8.64 0-30.1

MedHHInc (hundreds of dollars) 319.4 116.7-606.3

Comparison of OP and OPLWR ResultsVariable Standard Ordered

Probit1

Locally WeightedOrdered Probit2

AgeLog -0.236(-4.15)

-0.401(0.192)

[-1.187,-0.179]

DistanceLog -1.564(-2.96)

0.291(0.665)

[-0.873,2.036]

PriceLog-fitted -0.421(-1.44)

-0.415(0.008)

[-0.436,-0.386]

B1kHH00 0.030(8.14)

0.039(0.010)

[-0.094,0.058]

HispHH00 0.034(3.20)

-4.928(13.920)

[-117.043,0.112]

MedHHInc 0.003(4.15)

0.002(0.001)

[0.000,0.008]

Log likelihoodObservations

-311.40508

-1539.52508

1 Parameter estimates with t-statistics in parenthesis.

2 The average of the 508 parameter estimates for the variable is listed on the first of the three lines, the standard deviation in parenthesis is on the middle line, and the range of parameter estimates in brackets is provided on the third line. The log-likelihood value is the sum of the log likelihoods for the 508 regressions. Bandwidth = 0.4.

OPLWR• Heterogeneity in the parameters for different

houses

• Sign switches: log of distance

% Hispanic in block group

• Examine the geographic aspects of these variables

more clearly

Distance Coefficients and Location of Houses

b = .4

b = .8

Hispanic Coefficients and Location of Houses

Positive Coefficients

Negative Coefficients

Conclusions• spatial effects matter

• ignoring them can lead to biased results of

assessing determinants of noise

• find evidence of heterogeneity in effects of

Hispanic population, and distance from airport,

on probability of being in the “buffer zone”