MEMORANDUM
SUBJECT: 	Demographic and Social Patterns in Housing Units Near Large Highways and other Transportation Sources
FROM:	Chad Bailey, Office of Transportation and Air Quality/Assessment and Standards Division
TO:	Docket # EPA-HQ-OAR-2011-0135
   I.    Executive Summary
         1. A.       Background
The purpose of this memo is to examine the issue of environmental justice as it relates to housing near roads. EPA defines environmental justice as the fair treatment and meaningful involvement of all people regardless of race, color, national origin, or income with respect to the development, implementation, and enforcement of environmental laws, regulations, and policies.  EPA seeks to provide the same degree of protection from environmental health hazards for all people. 
Concentrations of many air pollutants are elevated near high-traffic roadways.  If minority populations and low-income populations disproportionately live near such roads, then an issue of environmental justice may be present. 
This memo explores demographic and socioeconomic differences that exist on a national level between populations near roads and populations that are not near roads. Characteristics of housing units near roads are compared to those not near roads in order to determine if there is a statistically significant difference in demographic or socioeconomic traits between the two groups. 
         1. B.       Method
Data employed in this analysis are from the American Housing Survey (AHS). Every two years, the U.S. Census Bureau conducts the AHS on a national level. The AHS is sponsored by the United States Department of Housing and Urban Development and gathers data about housing units in the United States. In this repeated survey, the same set of housing units is surveyed every odd-numbered year in order to track the changes in households and their housing. Basic information about the unit, resident characteristics, housing and neighborhood quality, size of living space, and financial information are among categories of data gathered in the AHS. The 2009 AHS used in this analysis contained responses from 73,222 households. The Census Bureau makes the data available in several different groups called modules. These modules group the data by type, with similar variables put together into one module. Data from two modules are used in this analysis: "newhouse", which contains general information about the households and housing units surveyed; and "mortg", which contains mortgage information for different households. 
Most statistical analyses were performed using SPSS 10.1.
The variable that will be focused on in the analysis is coded as etrans and contains each household's response to the question "How about any railroads, airports or highways with at least 4 lanes  -  any of these within a half block of this building [your housing unit]?" or 300 feet The Census Bureau reports the presence of these features within 300 feet (about 91 meters) of each housing unit. Concentrations of many pollutants are elevated with reduced distance to roads, with the steepest declines often occurring within about 100 meters. Select pollutants, including NOx and CO, have been found to drop to below 50% of at-road concentration outside 100 meters (m). As seen in Figure 1, for numerous pollutants, the highest concentrations and largest gradients occur within 100 m of roads. As a result, the half block (300 feet) specified in the variable etrans will represent a difference in pollutants from on-road emissions. All the statistical analysis detailed in this memo is done with respect to the binary variable etrans. The other variables used in the analyses are detailed at the outset of each section.
Figure 1  -  Regression of Normalized Pollutant Concentration against Distance from Source

Source: Karner, A. A., Eisinger, D. S., & Niemeier, D. A. (2010). Near-roadway air quality: Synthesizing the findings from real-world data. Environmental Science & Technology, 44(14), 5334-5344.
Within SPSS, we weighted the sample data from the American Housing Survey according to the specifications of the Census Bureau. As a result of the weighting, the data shows a representation of the 130,111,607 households in the United States in 2009. The weighting accounts for survey design and non-responses, ensuring all housing units nationally are properly represented in summary statistics. The statistical tests used are two-sample t-tests and logistic regression. 
Of the 130,111,607 households represented in the 2009 AHS, the sampling data characterizes households near four-lane highways, railroads, and airports in the following way:
   oo 22,075,098 households were within a half-block (17.0%)
   oo 104,936,084 were not (80.7%)
   oo 284,953 were represented by samples that didn't know (0.2%)
   oo 391,903 were represented by samples that refused to answer (0.3%)
   oo 2,423,569 were represented by samples that were not reported (1.9%)
The sum of the weight for each category was used to determine the population for each value of the variable.
         1. C.       Key Findings
The statistical analyses described below indicate significant differences between average characteristics of households residing within a half block of railroads, airports, or four-lane highways and characteristics of those residing outside that range. Specifically, minority householders, householders with lower educational attainment, and lower-income households are all more likely to be within a half block of railroads, airports, or four-or-more-lane (4+ lane) highways than households not fitting those characteristics. 
   II.    T-tests For Significance
         2. A.       Introduction
We used two-sided Student's t-tests assuming independent samples to determine the significance of pairwise contrasts in various demographic and economic characteristics between households within or outside 300 feet from railroads, airports, or 4+lane highways (the etrans response).  
The raw results of the t-tests performed in this section are in Appendix A.
         2. B.       Variables
The variables which will be examined using a t-test are ammort, ammrt2, lot, unitsf, cars, value, lmed, zinc2, and zinc. These coded names represent the following variables:
   * ammort  -  amount of first mortgage when acquired
   * ammrt2  -  amount of second mortgage when acquired
   * lot  -  area of lot in square feet
   * unitsf  -  area of the unit in square feet
   * cars  -  number of cars kept for use of members of the household
   * value  -  current market value of the unit
   * lmed  -  average area median family income
   * zinc2  -  household income
   * zinc  -  family income
We examined these variables to separately control for differences in housing unit characteristics (unisf, value), site characteristics (lot), metropolitan statistical area economic patterns (lmed), family and household income characteristics (zinc2, zinc), and household financing (ammort, ammrt2).
         2. C.       Results
Detailed statistics and t-test results for each variable are presented in Appendix A.
Overall, households with a positive etrans response, that is, those located within a half block of a railway, airport, or four-lane highway, had worse economic situations.  Households with positive etrans responses, in univariate comparisons, have lower average household income, family income, smaller homes and lots.  If they are own their homes, they have smaller first and second mortgages.  Lastly, housing units with a positive etrans response live in metropolitan areas (MSAs) with slightly higher average median incomes.
The results suggest that personal economic situations and regional economic trends affect the odds that a housing unit is located within a half block of a railway, airport, or four-lane highway (etrans response).  
Reflecting personal economic situations, households that reside away from such sources have, on average, higher incomes, more durable assets (e.g., cars), larger homes, larger lots, and, if they own their homes, have greater mortgage debt.  This suggests that people living near railways, airports, and 4+ lane highways have lower socioeconomic status (SES).
Regional economic trends also appear to influence the likelihood of living near a transportation sources, with housing units within half a block having greater average median income at the MSA level.  These differences will be examined further in the following sections.

         2. A.       Summary
   I.    Univariate Logistic Regression
         1. A. Introduction
In order to investigate the significance of binary variables in relation to the variable etrans, the focus of this analysis, we have applied logistic regression to examine whether various qualitative factors change the odds of living within a block of a railway, airport, or 4+ lane highway.  These results are expressed in terms of how each variable affects the "odds ratio" (OR) of an affirmative etrans response.  An odds ratio reflects how the chance that a given outcome happens depends an independent variable (e.g., race, income). 
         1. B. Variables
The binary variables examined using this method are described below:
   * Householder race and ethnicity
         o hhspan  -  whether or not the householder is Hispanic
         o white  -  whether or not the householder is white
   * Household socioeconomic status
         o hhhsgrad  -  whether or not the householder graduated from high school
         o unigrad  -  whether or not the householder graduated from a four-year university
         o hhwlineq  -  whether or not the householder worked in the last week
         o qfs1  -  whether or not the household received foods stamps in the past year
         o qwelf  -  whether or not the household received welfare in the past year
         o qdiv  -  whether or not the household received stock dividends in the past year
   * Housing unit characteristics
         o proj  -  whether or not the unit is owned by a public housing authority
         o garage  -  whether or not a garage or carport is included with the unit
For the most part, these variables represent demographic and economic characteristics of the householder, or the house itself. 
         1. C. Results
The detailed tables for each variable examined are presented in Appendix B of this memorandum, along with SPSS output tables with logistic regression results.
Overall, every variable described in Section II.B had a statistically-significant OR (OR), suggesting that the demographic and economic characteristics analyzed affect whether a household lives near a major transportation source.  
Figure II-1 summarizes the ORs for univariate regressions, ranking them by magnitude.  As shown, nearly every EJ-related variable results in a substantial increase in the likelihood that a housing unit will be located in a location where traffic-related air quality is a concern.  The only SES-related variable that results in little change in the odds of a positive etrans response is hhwlineq, which indicates whether the householder worked in the past week.  This pattern suggests that living in a location with greater potential traffic-related air pollution is associated more with individual factors (e.g., race and ethnicity) and economic factors (e.g., educational attainment, housing unit features) that reflect longer-term issues.
Figure II-1


   II.    Multivariate Logistic Regression Model
         2. A. Introduction
Multivariate logistic regression is a method of finding a relationship between an OR (OR) of a binary variable resulting from a number of explanatory variables. To examine the effect of any particular variable, logistic regression allows for the control of other variables, aiding in interpretation.  
This section presents the results of four logistic regression models: one model employing a small number of EJ-related variables, one including a moderate number of variables, one including many variables that relate to both demographic and socioeconomic characteristics, and the last one readjusted to remove cases of multicollinearilty from the large model. The raw SPSS output for this section can be located in Appendix C.
         2. B. Minimal EJ Logistic Regression Model
Out of the 73,222 households surveyed, 52,217 indicated an answer to etrans. Executive Order 12898 stipulates that all federal agencies consider the impacts of their actions affecting "minority populations and low-income populations."  Thus, adhering strictly to the characteristics specifically enumerated in the E.O., this model only includes the variables hhspan (householder Hispanic status); white (householder white/nonwhite); and zinc2 (household income). 
In performing the regression, SPSS identifies which cases are valid to use. Due to the use of survey weights to make the data representative of the general population, cases with a missing or zero weight value were removed. Of the 73,222 households surveyed, 53,354 have a useable weight value. 9,385 of those cases were removed for having missing values in at least one of the explanatory variables or the response variable, leaving 43,969 cases to be weighted and analyzed as part of the regression. 
According to the model, the three explanatory variables included all had significant effect on whether or not a living unit is within a half block of a four-lane highway, a railway, or an airport.  
We assessed fit using the Hosmer and Lemeshow test.  However, a known limitation of this test is that for large sample sizes, most of these tests result in a significant "bad fit."  As the analysis contain in this section has a total weighted population of 130,111,607 households, this large sample size may contribute to the significance found in the Hosmer-Lemeshow tests. 
Table III-1 shows correlation coefficients between the variables. For correlation between two binary variables, hhspan and white, a phi coefficient was used. For correlation between zinc2 and each of the two binary variables, a point-biserial correlation coefficient was used.
	Table III-1
                                  Correlation
                               Household Income
                               White Householder
                             Hispanic Householder
                                    -0.074
                                     0.103
                               White Householder
                                     0.092
                                       

The largest correlation was between the variables white householder and Hispanic householder, but all correlations were small and unlikely to affect the model's outputs due to their low multicollinearity.  
The results of the logistic regression, that is, the variables and their corresponding ORs, are shown in Table III-2. 
	Table III-2
                                   Variable
                                      OR
                            95% Confidence Interval
                                   Constant
                                     0.283
                                       -
                             Hispanic Householder
                                     1.332
                                  1.331-1.334
                               White Householder
                                     0.678
                                  0.677-0.679
                          Household Income (10,000s)
                                     0.962
                                  0.961-0.962

As described above, the income-based OR is based on how each household's income differs from $66,200, the average level reported in the AHS.  The "constant" represents odds of a household with a non-Hispanic, nonwhite householder with a average household income living within a half block of a four-lane highway, a railway, or an airport (i.e., an affirmative etrans response). These odds are 0.283, which equates to roughly a 22% chance of an affirmative etrans response. If the householder is white, then the odds decrease to 0.192, with probability of an affirmative etrans response of about 16%. If the householder is Hispanic and nonwhite, the odds increase to 0.377, with a probability of etrans of about 27%. For housing units with Hispanic householders, the odds of living within a half block of an affirmative etrans response are about twice that of a non-Hispanic white householder.  Also, a change of $10,000 in household income results in a proportional change in odds of 4%.
Compared with the univariate analysis above, the OR for housing units with Hispanic and nonwhite householders are nearly identical, suggesting that they are confounded by neither one another nor by income.
The results of this model suggest that race/ethnicity and income status both have a significant effect on the odds of a household living within a half block of a four-lane highway, a railway, or an airport.
         2. C. Medium-level Logistic Regression Model
To address the potential for confounding by other factors and to determine which factors contribute to the probability of living near  railway, airport, or 4+lane highway, we developed a regression model including more variables.  In addition to the variables included in the above model, the following variables were used: whether the household had received welfare in the past year, whether the household had received stock dividends in the past year, number of bedrooms in the housing unit (difference from 3), and indicator variables representing the type of unit that the household lives in (nunit2). The Hosmer and Lemeshow test produced a chi-square statistic of 1598.2 for this model, indicating significance in the test. 
Table III-3 shows correlation coefficients between the variables. For correlation between two binary variables, a phi coefficient was used. For correlation between quantitative variables and binary variables, a point-biserial correlation coefficient was used. The correlation between the binary variables and the nominal variable nunit2, reports Cramer's V as the measure. The significance of contrasts in household income (zinc2) among unit types in the nominal housing unit type variable nunit2 is computed using the Scheffe method as a part of post-hoc analysis, and is reported below the table. 
Table III-3
                                  Correlation
                                   Unit Type
                     Received Stock Dividends in Past Year
                         Received Welfare in Past Year
                               Household Income
                               White Householder
                             Hispanic Householder
                                     0.111
                                     0.086
                                     0.038
                                    -0.074
                                     0.103
                               White Householder
                                     0.175
                                     0.085
                                     0.072
                                     0.092
                                       
                               Household Income
                                       *
                                    -0.080
                                     0.190
                                       
                                       
                         Received Welfare in Past Year
                                     0.085
                                     0.038
                                       
                                       
                                       
                     Received Stock Dividends in Past Year
                                     0.105
                                       
                                       
                                       
                                       
*Scheffe post-hoc test indicates significant differences for each contrast between unit types.
The largest correlation was between the variables household income and Welfare in Past Year. However, 0.190 is a very low correlation coefficient. None of the variables in this model are very correlated with the others.
The variables and their corresponding ORs are as shown in Table 36.
	Table III-4
                                   Variable
                                      OR
                            95% Confidence Interval
                                   Constant
                                     0.200
                                       -
                             Hispanic Householder
                                     1.166
                                  1.164-1.168
                               White Householder
                                     0.783
                                  0.782-0.784
                          Household Income (10,000s)
                                     0.983
                                  0.983-0.983
                         Received Welfare in Past Year
                                     1.022
                                  1.019-1.026
                     Received Stock Dividends in Past Year
                                     0.864
                                  0.862-0.865
                                   Bedrooms
                                     0.865
                                  0.865-0.866
                                   Unit Type
                                       
                                       
                          One-unit building, attached
                                     1.364
                                  1.361-1.367
                          Building with 2+ apartments
                                     2.028
                                  2.025-2.031
                             One-unit Mobile Home
                                     1.158
                                  1.156-1.161

The value of the constant represents the odds of living within a half block of a four-lane highway, a railway, or an airport for a household with a non-Hispanic, nonwhite householder with a household income of $66,200 that hasn't received welfare or stock dividends in the past year, in a one-unit, detached building with three bedrooms. These odds are 0.198, which equates to roughly a 16.5% probability of an affirmative etrans response. The odds change depending on the differing characteristics present and their corresponding ORs, above.
         2. D. Summary of Logistic Models

Comparisons between the univariate ORs, the "minimal" multivariate model, and the larger model containing more variables, there are several notable commonalities.  First, the most directly EJ-related variables, race, Hispanic status, and income, maintain statistically significant ORs.  Second, between the univariate and "minimal" multivariate models, the ORs (and associated confidence intervals) for race and ethnicity change very little in magnitude, suggesting that they are confounded neither by one another, nor by income.  Third, comparing the expanded model, including housing unit characteristics and added economic indicators, with the "minimal" and univariate models, the ORs for Hispanic status and race are attenuated to some extent, though they remain significant.  Fourth, comparing the "minimal" model with the expanded model, the OR for household income remains very similar, despite the inclusion of indicators of high-income and low-income status (e.g., receiving dividend income and welfare).

The above observations suggest that income may be a non-linear effect on the probability of residence near an airport, railway, or 4+ lane highway, with larger effects at the "tails" of the income distribution.  Though nonlinear income terms were not examined in this analysis, we plan to examine them in forthcoming analyses.

Furthermore, it appears that accounting for very high and very low income indicators, in addition to housing unit characteristics (e.g., bedroom number and unit type), account for the attenuation in race and ethnicity indicators.  This phenomenon will be examined in greater detail in subsequent analyses. 

   III. Discussion
The above analyses generally support a conclusion that households near airports, railways, and 4+ lane highways are more likely to be of a racial or ethnic minority and lower in income.  People on welfare are also more likely to live near such large transportation sources.  Housing units near facilities named in etrans are also smaller and more often attached structures (e.g., multi-unit dwellings).
In subsequent analyses, we plan to address other issues that affect how demographics and SES affect responses to etrans.

.
Appendix A: T-test Data and Results

A.1	Data Summary
Mortgage
t-tests were run to determine if there were significant differences in the values of mortgages for households located within/outside of a half block of a railway, airport, or four-lane highway.  One t-test examined the initial value of the first mortgage (variable ammort), and another t-test examined the initial value of the second mortgage (variable ammrt2).  As not every household assumes one or more mortgage, there are missing values for each mortgage measurement.  Households without mortgages may be renters or own their homes outright.  Out of a total of 130,111,607 households represented in the AHS, 47,995,888 households have values for the first mortgage with 82,115,719 missing, and 5,517,534 values for the second mortgage with 124,594,073 values missing. Additional cases are left out due to missing values in the variable etrans. The summary statistics of these groups are shown in Tables A-1 and A-2 below:
Table A-1
                                First Mortgage
                             Original Sample Size
                             Weighted Sample Size
                                 Weighted Mean
                          Weighted Standard Deviation
                  Within (1/2) block of airport/railway/ highway
                                     2488
                                   6,339,776
                                  $147,719.59
                                 $135,168.162
                 Outside (1/2) block of airport/railway/ highway
                                    15,788
                                  40,410,963
                                  $168,015.23
                                 $154,991.759
                          Difference (within-outside)
                                   (t-value)
                                       
                                       
                                  -$20,295.63
                                  (t=-344.23)
                                       

Table A-2
                                Second Mortgage
                             Original Sample Size
                             Weighted Sample Size
                                 Weighted Mean
                          Weighted Standard Deviation
                  Within (1/2) block of airport/railway/ highway
                                      286
                                    741,548
                                  $52,521.57
                                  $70,785.283
                 Outside (1/2) block of airport/railway/ highway
                                     1874
                                   4,733,338
                                  $57,022.88
                                  $72,479.515
                          Difference (within-outside)
                                       
                                       
                                  -$4,501.30
                                  (t=-50.75)
                                       

Households that are not within a half block of an airport, railroad, or four-lane highway assume a larger first and second mortgage, on average. Table A-1 shows the difference in first mortgages between the two groups (95% CI: -$20,180 - -$20,411). Table A-2 shows the difference in second mortgage values (95% CI: $4327-$4675).
Lot and Unit Area
The variable measuring total lot area (lot) has 95,876,348 valid households analyzed as part of the mean of either near road households or non-near road households. Values of this variable less than 200 ft[2] were listed as 200 ft[2], and values over 999,997 ft[2] were recorded as 999,997 ft[2].[,]
Unit area for each household (unitsf), which gives a measure of total interior livable area, had 119,404,448 valid responses and 10,707,159 missing. Values under 99 ft[2] were coded at 99 ft[2], and values at over 24,870 ft[2] were recorded at 24,870 ft[2].
The summary statistics of the groups of the two different types of area are shown below in Tables A-3 and A-4:
Table A-3
                                   Lot Area
                             Original Sample Size
                             Weighted Sample Size
                                 Weighted Mean
                          Weighted Standard Deviation
                  Within (1/2) block of airport/railway/ highway
                                     4983
                                  12,532,783
                                42,437.64 ft[2]
                               130,819.283 ft[2]
                 Outside (1/2) block of airport/railway/ highway
                                    31,890
                                  81,021,724
                                76,202.05 ft[2]
                               191,377.678 ft[2]
                          Difference (within-outside)
                                       
                                       
                                 -33,764 ft[2]
                                  (t=-791.98)
                                       








Table A-4
                                   Unit Area
                             Original Sample Size
                             Weighted Sample Size
                                 Weighted Mean
                          Weighted Standard Deviation
                  Within (1/2) block of airport/railway/ highway
                                     8378
                                  20,345,461
                                 1548.24 ft[2]
                                2072.338 ft[2]
                 Outside (1/2) block of airport/railway/ highway
                                    38,609
                                  96,458,011
                                 1906.16 ft[2]
                                2240.453 ft[2]
                          Difference (within-outside)
                                       
                                       
                                 -357.92 ft[2]
                                  (t=-697.77)
                                       

t-tests were run again with the same hypothesis and variables, only with the top-coded and bottom-coded values removed.  Qualitatively, results were similar and statistically significant, with mean lot size difference of -18,389 ft[2] and unit area difference of 335.13 ft[2].  
Vehicle Ownership
A t-test was performed to determine whether there was a significant difference in number of cars (cars variable) for households with positive and negative etrans responses. 
Of the 130,111,607 households in the United States in 2009, there were 18,250,711 households represented by the response "Not Applicable" and thus were left out of the analysis. The means were calculated from a split sample of a total of 111,860,896 households.  The variable cars is top-coded at 5, indicating that 5 or more cars are available for household use.
Table A-5 shows summary statistics by proximity to transportation sources.
Table A-5
                                     Cars
                             Original Sample Size
                             Weighted Sample Size
                                 Weighted Mean
                          Weighted Standard Deviation
                  Within (1/2) block of airport/railway/ highway
                                     8004
                                  19,621,627
                                     1.123
                                     0.872
                 Outside (1/2) block of airport/railway/ highway
                                    35,965
                                  89,422,881
                                     1.265
                                     0.930
                          Difference (within-outside)
                                       
                                       
                                     0.142
                                  (t=-645.31)
                                       

Housing Unit Value
A t-test was performed to determine whether there was a significant difference in value, the variable indicating the unit value for households with different etrans responses. 
The variable value contains the estimated market value of the unit at the time of the survey. This variable only includes housing units which are owner-occupied or is otherwise non-rental.  87,944,404 households are represented by the responses in the survey. Table A-6 presents the summary statistics:
Table A-6
                           Market Value of the Unit
                             Original Sample Size
                             Weighted Sample Size
                                 Weighted Mean
                          Weighted Standard Deviation
                  Within (1/2) block of airport/railway/ highway
                                     4784
                                  11,844,689
                                  $201,147.67
                                 $224,696.916
                 Outside (1/2) block of airport/railway/ highway
                                    30,082
                                  76,099,715
                                  $243,271.65
                                 $293,497.539
                          Difference (within-outside)
                                       
                                       
                                  -$42,123.98
                                  (t=573.53)
                                       

Income
Separate tests were performed to examine differences in responder-level income (family and household income) and metropolitan statistical area-level income.  These tests reflect economic issues at different geographic scales.  
The summary statistics and t-test results are displayed in Table A-7, Table A-8, and Table A-9 for each of the three types of income (household, family, and median MSA):
Table A-7
                               Household Income
                             Original Sample Size
                             Weighted Sample Size
                                 Weighted Mean
                          Weighted Standard Deviation
                  Within (1/2) block of airport/railway/ highway
                                     8004
                                  19,621,627
                                  $53,236.85
                                  $56,373.439
                 Outside (1/2) block of airport/railway/ highway
                                    35,965
                                  89,422,881
                                  $67,882.66
                                  $68,911.201
                          Difference (within-outside)
                                       
                                       
                                  -$14,645.81
                                  (t=-998.68)
                                       

Table A-8
                                 Family Income
                             Original Sample Size
                             Weighted Sample Size
                                 Weighted Mean
                          Weighted Standard Deviation
                  Within (1/2) block of airport/railway/ highway
                                     9305
                                  22,075,098
                                  $44,896.56
                                  $54,558.41
                 Outside (1/2) block of airport/railway/ highway
                                    42,912
                                  104,936,084
                                  $55,792.51
                                  $67,048.362
                          Difference (within-outside)
                                       
                                       
                                  -10,895.95
                                  (t=-817.42)
                                       

Table A-9
                          Average Area Median Income
                             Original Sample Size
                             Weighted Sample Size
                                 Weighted Mean
                          Weighted Standard Deviation
                  Within (1/2) block of airport/railway/ highway
                                     9305
                                  22,075,098
                                  $65,765.70
                                  $11,537.298
                 Outside (1/2) block of airport/railway/ highway
                                    42,912
                                  104,936,084
                                  $64,801.81
                                  $11,712.817
                          Difference (within-outside)
                                       
                                       
                                    $963.89
                                  (t=355.85)
                                       

The difference in the average incomes of individual households (family or household income) within a half block of a railway, airport, or 4+ lane highway from those outside half a block suggests that personal economic situations affect the likelihood of living near transportation sources.
As indicated in Table A-9, MSA-level median incomes tend to be higher for housing units within half a block of these transportation sources.  This finding suggests that regional economic patterns influence the likelihood of homes being built near these transportation sources.
A.2	Summary Statistics and T-Tests
     
     Table A-12

For Table A-2 and Table A-3, "Levene's Test for Equality of Variances" refers to a significance test performed to determine whether it can be assumed that the variances for each subgroup (the explanatory variable divided based on etrans) are equal. The null hypothesis being tested is that they are equal. As each test found a significant difference in the values of the variances, the second row ("Equal variances not assumed") in each variable should be the one referenced. 











Table A-2
Table A-3
Table B-3
Appendix B: Univariate Logistic Survey Data and Regression Results
This appendix presents the survey data from the 2009 AHS and the SPSS outputs from the logistic regression procedure.  Section B.1 presents the survey data and ORs for each included variable.  Section B.2 presents the logistic regression results.
B.1	Survey Data
B.1.1	Householder Race and Ethnicity
hhspan is a binary variable that measures whether the householder was Hispanic. Of the 73,222 households surveyed, there were 43,969 responses to both hhspan and etrans. Table B-1 describes the survey responses for these variables. 
Table B-1
                                       
                                Original Sample
                                Weighted Values
                                  Households
                             Householder Hispanic
                           Householder  Not Hispanic
                             Householder Hispanic
                           Householder Not Hispanic
                  Within (1/2) block of airport/railway/ highway
                                     1115
                                     6889
                                   2,743,936
                                  16,877,692
                 Outside (1/2) block of airport/railway/ highway
                                     4062
                                    31,903
                                   9,748,796
                                  79,674,084

22.0% of Hispanic householders and 17.5%  of non-Hispanic householders respond affirmatively to etrans. The OR for living "nearby" corresponding to a householder being Hispanic is 1.329 (95% CI 1.327-1.330), meaning that the a Hispanic householder is about 33% more likely to live near 
B.1.2	Householder Race
Of the 73,222 households surveyed, 43,969 both answered whether the household was within a half block of a four-lane highway, railway, or airport and indicated the race of the householder.  For this analysis, we coded a variable white, indicating whether the householder was white or nonwhite.  Table B-2 describes the survey responses for these variables.
Table B-2
                                       
                                Original Sample
                                Weighted Values
                                  Households
                               Householder White
                            Householder  Not White
                               Householder White
                             Householder Not White
                  Within (1/2) block of airport/railway/ highway
                                     6,137
                                     1,867
                                  14,937,801
                                   4,683,826
                 Outside (1/2) block of airport/railway/ highway
                                    29,807
                                     6,158
                                  74,064,319
                                  15,358,562

16.8% of housing units with white householders and 23.4% of those with nonwhite householders responded affirmatively to etrans.  The OR corresponding to a householder being white is 0.661 (95% CI .661-.662). The odds of a household being within a half-block of a four-lane highway, railway, or airport decrease by 34% if the householder is white.
B.1.3	Householder High School Graduation
The variable hhhsgrad indicates whether the householder graduated high school. Of the 73,222 households surveyed, 43,969 responded to both hhhsgrad and etrans. Table B-3 describes the survey responses for these variables.
Table B-3
                                       
                                Original Sample
                                Weighted Values
                                  Households
                             Householder Graduated
                          Householder  Not Graduated
                             Householder Graduated
                          Householder  Not Graduated
                  Within (1/2) block of airport/railway/ highway
                                     6,746
                                     1,258
                                  16,592,399
                                   3,029,228
                 Outside (1/2) block of airport/railway/ highway
                                    31,259
                                     4706
                                  77,593,559
                                  11,829,321

17.6% of housing units with a high school graduate as householder responded affirmatively to etrans, as did 20.4% of those with householders who were not graduates.  The OR corresponding to a householder having graduated from high school is 0.835 (95% CI 0.834-0.836). The odds of a household being within a half-block of a four-lane highway, railway, or airport decrease by 16.5% if the householder graduated high school.
B.1.4 Householder University Graduation
The second variable examined in relation to education is unigrad, the binary variable of whether the householder graduated from a university. Of the 73,222 households surveyed, 43,969 answered whether the household was within a half block of a four-lane highway, railway, or airport and indicated the householder's level of education. Table B-4 describes the survey responses for these variables.
Table B-4
                                       
                                Original Sample
                                Weighted Values
                                  Households
                             Householder Graduated
                          Householder  Not Graduated
                             Householder Graduated
                          Householder  Not Graduated
                  Within (1/2) block of airport/railway/ highway
                                     2,112
                                     5,892
                                   5,004,499
                                  14,617,128
                 Outside (1/2) block of airport/railway/ highway
                                    11,340
                                    24,625
                                  27,586,625
                                  61,836,256

15.3% of units whose householders graduated from university and 19.1% of other units responded affirmatively to etrans. The OR corresponding to a householder having graduated from a university is 0.768 (95% CI 0.767-0.769), meaning that a college degree reduces one's chance of being in a "nearby" housing unit by 23.2%.
B.1.5	Current Householder Employment
hhwlineq designates whether the householder worked in the past week. Of the 73,222 households surveyed, 43,412 responded to both etrans and hhwlineq. Table B-5 describes the survey responses for these variables.
Table B-5
                                       
                                Original Sample
                                Weighted Values
                                  Households
                              Householder Worked
                           Householder  Did Not Work
                              Householder Worked
                           Householder  Did Not Work
                  Within (1/2) block of airport/railway/ highway
                                     4,664
                                     3,259
                                  11,830,420
                                   7,592,285
                 Outside (1/2) block of airport/railway/ highway
                                    21,089
                                    14,400
                                  53,717,301
                                  34,528,774

18.05% of housing units with householders working in the past week and 18.02% of those with householders that did not work in the last week responded to etrans affirmatively.  The OR corresponding to a householder having worked in the past week is 1.002 (95% CI 1.001-1.003).  Relative to the ORs of longer-term economic indicators, this short-term does not address most of the etrans responses.
                 0. 3. 1.                      Public Housing
proj is the binary variable indicating whether the household resided within a building owned by a public housing authority. Of the 73,222 households surveyed, 4721 both answered both proj and etrans. Table B-6 describes the survey responses for these variables.
Table B-6
                                       
                                Original Sample
                                Weighted Values
                                  Households
                                Public Housing
                              Not Public Housing
                                Public Housing
                              Not Public Housing
                  Within (1/2) block of airport/railway/ highway
                                      243
                                      958
                                    559,723
                                   1,693,293
                 Outside (1/2) block of airport/railway/ highway
                                      536
                                     2984
                                   1,192,188
                                   5,231,271

Households in units owned by public housing authorities had a 32.0% chance of living within a half block of an airport, railroad, or 4+ lane highway (i.e., affirmative etrans response).  Households living in units not owned by public housing authorities had a 24.5% chance of an affirmative etrans response.  The OR corresponding to a household having resided within a building owned by a public housing authority is 1.450 (95% CI 1.445-1.456). 
                 0. 3. 2.                      Food Stamps
Another socioeconomic variable examined is qfs1, the binary variable of whether the household received food stamps in the past year. Of the 73,222 households surveyed, 12,286 both answered whether the household was within a half block of a four-lane highway, railway, or airport and whether the household had received food stamps in the past year. Table B-8 describes the survey responses for these variables.
Table B-8
                                       
                                Original Sample
                                Weighted Values
                                  Households
                              Receive Food Stamps
                          Do Not Receive Food Stamps
                              Receive Food Stamps
                          Do Not Receive Food Stamps
                  Within (1/2) block of airport/railway/ highway
                                      727
                                     2102
                                   1,815,348
                                   5,115,085
                 Outside (1/2) block of airport/railway/ highway
                                     1917
                                     7540
                                   4,766,553
                                  18,844,023

27.5% of households receiving food stamps lived in areas within a half block of an airport, railroad, or 4+ lane highway (i.e., affirmative etrans response).  19.8% of households not receiving food stampes had affirmative etrans responses.  The OR corresponding to a household having received food stamps in the past year is 1.403 (95% CI 1.400-1.406).
                 0. 3. 3.                      Welfare
qwelf indicates whether the household received welfare in the past year. Of the 73,222 households surveyed, 43,969 both answered whether the household was within a half block of a four-lane highway, railway, or airport (etrans) and whether the household had received welfare in the past year (qwelf). Table B-9 describes the survey responses for these variables.
Table B-9
                                       
                                Original Sample
                                Weighted Values
                                  Households
                               Received Welfare
                             Not Received Welfare
                               Received Welfare
                             Not Received Welfare
                  Within (1/2) block of airport/railway/ highway
                                      199
                                     7805
                                    482,598
                                  19,139,029
                 Outside (1/2) block of airport/railway/ highway
                                      601
                                    35,364
                                   1,492,707
                                  87,930,173

17.9% of households not receiving welfare in the past year responded affirmatively to etrans.  24.5% of households receiving welfare in that time frame responded affirmatively.  The OR corresponding to a household having received welfare in the past year is 1.485 (95% CI 1.480-1.490), meaning that the odds of living near an airport, railroadd, or 4+lane highway were 48.5% greater for households receiving welfare in the past year.
                 0. 3. 4.                      Stock Dividends
 qdiv reports whether the household received stock dividends in the past year. Of the 73,222 households surveyed, 43,969 both answered qdiv and etrans. Table B-10 describes the survey responses for these variables.
Table B-10
                                       
                                Original Sample
                                Weighted Values
                                  Households
                                Stock Dividends
                              No Stock Dividends
                                Stock Dividends
                              No Stock Dividends
                  Within (1/2) block of airport/railway/ highway
                                      550
                                     7454
                                   1,269,987
                                  18,351,641
                 Outside (1/2) block of airport/railway/ highway
                                     3643
                                    32,322
                                   8,492,672
                                  80,930,208

Among households not receiving dividends, 18.5% responded affirmatively to etrans; 13.0% of those receiving dividends did as well.  The OR corresponding to a household having received stock dividends in the past year is 0.660 (95% CI 0.659-0.661), meaning that the odds of a household being within a half-block of a four-lane highway, railway, or airport decrease by 34% if the household received stock dividends in the past year.
                 0. 3. 5.                      Garage
Of the 73,222 households surveyed, 52,177 both answered etrans and whether the household had had a garage or carport available for their use. Table B-11 describes the survey responses for these variables.
Table B-11
                                       
                                Original Sample
                                Weighted Values
                                  Households
                                    Garage
                                   No Garage
                                    Garage
                                   No Garage
                  Within (1/2) block of airport/railway/ highway
                                     4,846
                                     4,455
                                  11,827,620
                                  10,240,417
                 Outside (1/2) block of airport/railway/ highway
                                    27,753
                                    15,123
                                  68,740,753
                                  36,115,500

22.1% of households with no garage or carport available for their use responded affirmatively to etrans.  14.7% of households with access to a garage or carport responded affirmatively.  The OR for a garage or carport is 0.607 (95% CI 0.606-0.607), meaning odds of a household with a garage or carport was 39% lower than if the household had no garage or carport available for their use.
B.2	Logistic Regression Results
                                       
Table B-12


Table B-13


Table B-14


Table B-15


Table B-16


Table B-17


Table B-18


Table B-19


Table B-20


Table B-21


            Appendix C  -  Multivariate Logistic Regression Models
                                       
For the strict model, shown first, there are three tables shown. The first two tables shown are the results of the Hosmer and Lemeshow test, and the third table shows the results of the logistic regression. In the other three models, there are four tables displayed for each. Before the table showing the results of the logistic regression, there is a table indicating the representations of the categorical variables.

                                 Strict Model

Table C-1


Table C-2


Table C-3




                              Medium-level Model

Table C-4


Table C-5


Table C-6













Table C-7


                               Wide-scope Model

Table C-8


Table C-9



Table C-10
Table C-11




                          Wide-scope Model, Adjusted

Table C-12


Table C-13











Table C-14
Table C-15

