Airport Noise and Home Values JP Nelson 2004 MetaAnalysis of Airport Noise and Hedonic Property Values Problems and Prospects Journal of Transport Economics and Policy Vol 38 Part 1 pp 128 ID: 384701
Download Presentation The PPT/PDF document "Meta-Analysis and Meta-Regression" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Meta-Analysis and Meta-Regression
Airport Noise and Home Values
J.P. Nelson (2004). “Meta-Analysis of Airport Noise and Hedonic Property Values: Problems and Prospects,”
Journal of Transport Economics and Policy
, Vol. 38, Part 1, pp. 1-28.Slide2
Data Description
Results from 20 Studies (containing 33 separate estimates), relating home prices to airport noise. All studies in US and Canada, from 1967 to present
Regressions control for other factors including: structural variables (e.g. size), locational variables, local taxes, government services, and environmental quality.
Primary Variable: Noise Depreciation Index (NDI) and its Regression coefficient (effect of increasing airport noise by 1 decibel on house cost). Positive coefficient implies that as noise increases, home value decreases. The units are percent depreciation.Slide3
Study Specific Variables / Models
For each study (with several exceptions), there are:
Noise Depreciation Index (NDI) and its estimated standard error
Mean Real Property Value (Year 2000, US $1000s)
An indicator of whether accessibility (to airport) adjustment was made (1 if No Adjustment, 0 if Adjustment was made)
Sample Size (log scale)
I
ndicator of whether the response (price) scale was linear (1 if Linear, 0 if Log)
Indicator of whether airport was in Canada (1 if Canada, 0 if US)
Models Considered
Fixed and Random Effects Meta-Analyses with no covariates
Meta-Regressions with predictors: Ordinary Least Squares with robust standard errors and Weighted Least squaresSlide4
Data
Note: Due to missing data, analyses will be based on only 31 or 29 airports.Slide5
Meta-Analysis with No Covariates
Fixed Effects Model – Assumes that each airport has the same true NDI, and that all variation is due to sampling error
Random Effects Model – Allows true NDIs to vary among airports along some assumed Normal Distribution.
Test for Homogeneity (Fixed Effects) can be conducted after estimating the mean (Hedges and
Olkin
, 1985, pp.122-123).Slide6
Estimates and TestsSlide7
Estimates and Tests - ResultsSlide8
Meta-Regressions
Regressions to determine which (if any) factors are associated with NDI
Three Models Fit:
Ordinary Least Squares with robust standard errors (White’s
heteroscedastic
-consistent standard errors)
Weighted Least Squares with weights equal to the inverse variance of the NDI:
w
i = 1/s
2
{d
i
}Weighted Least Squares with weights equal to the inverse
standard error
of the NDI:
w
i
=
1/s{d
i
}Model 1 based on k = 31 airports (2 have no Mean property values)
Models 2 and 3 based on k = 29 airports (2 have no weights)Slide9
Specification Tests Conducted on Models - I
Ramsey’s RESET Test – Used to test whether the model is correctly specified and does not involve any nonlinearities among the
regressors
.
Step 1: Fit the Original Regression with all Predictors
Step 2: Fit Regression with same predictors and squared (and possibly higher order) fitted values from first model.
Conduct F-test or t-test on polynomial fitted value(s)Slide10
Specification Tests Conducted on Models - II
White’s Test for
Heteroscedasticity
Step 1: Fit the Original Regression with all Predictors
Step 2
: Fit Regression relating squared residuals from step 1 to the same predictors and squared values for all numeric predictors (other version includes interactions for general specification test)
Compare nR
2
with Chi-Square(df
= # Predictors in Step 2)Slide11
Specification Tests Conducted on Models - IIISlide12
Ordinary Least Squares with Robust Standard ErrorsSlide13
Model 1 – OLS with Robust Standard Errors - ISlide14
Model 1 – OLS with Robust Standard Errors - I
Jarque-Bera
Test
White’s
TestSlide15
Weighted Least Squares – Models 2 and 3
Clearly Model 1 provides a poor fit (non-significant F-Statistic (p=.0826), R
2
=.3086)
Models 2 and 3 Use Weighted Least Squares with weights equal to the Variances and the Standard Errors, respectively, of the NDI estimates from each studySlide16
Weighted Least Squares – Model 2 – w
i
= 1/s
2
{d
i
}Slide17
Weighted Least Squares – Model 2 – w
i
= 1/s
2
{d
i
}Slide18
Model 2 – Specification TestsSlide19
Model 3 – WLS –
w
i
= 1/s{
d_i
}
This is a more traditional weighting scheme than Model2
The fit however, for this analysis is not as good:R
2 = 0.4131Fobs
= 3.2380, P = .0234
While for Model 2:
R
2 = 0.5389
F
obs
=
5.3756,
P = .
0020