University of Southern California Dissertations and Theses

Smart growth and walkability affect on vehicle use and ownership

(USC Thesis Other)

Smart growth and walkability affect on vehicle use and ownership

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

SMART GROWTH AND WALKABILITY
AFFECT ON VEHICLE USE AND OWNERSHIP

By

Derek Richard Newland

A Thesis Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
MASTER OF SCIENCE
(GEOGRAPHIC INFORMATION SCIENCE AND TECHNOLOGY)

May 2015

Copyright 2014 Derek Richard Newland

ii

DEDICATION
I dedicate this document to my parents Douglas and Susan Newland, my girlfriend Anna
Simpson and my employer Jim Minnick for their constant support and understanding, and to my
committee chair Robert Vos, who has helped me through this project in understanding the
concepts and theories behind the research and for his support in completing this project.

iii

ACKNOWLEDGEMENTS
I will be forever grateful to my mentor, Professor Robert Vos. Thank you for your patience and
understanding. Tenacity and persistence are qualities I have come to appreciate through this
process.

iv

TABLE OF CONTENTS

DEDICATION .....................................................................................................................................ii
ACKNOWLEDGEMENTS ................................................................................................................... iii
LIST OF TABLES ................................................................................................................................ vi
LIST OF FIGURES ............................................................................................................................. vii
ABSTRACT ...................................................................................................................................... viii
CHAPTER 1: INTRODUCTION ........................................................................................................... 1
1.1 Motivation ............................................................................................................................ 2
1.2 Study Overview ................................................................................................................... 3
CHAPTER 2: BACKGROUND ............................................................................................................. 6
2.1 Influence of Smart Growth non VMTs and Automobile Ownership............................. 8
2.2 Walkability studies and vehicle behavior ....................................................................... 11
CHAPTER 3: METHODOLOGY ........................................................................................................ 16
3.1: Sampling Framework ...................................................................................................... 16
3.2: Hypothesis and Independent Variables ......................................................................... 18
3.3: Neighborhood Slope and Walkability ............................................................................ 21
3.4: Proximity of Bus Stops .................................................................................................... 25
3.5: Climate and Walkability ................................................................................................. 26
3.6: Land Use and Walkability .............................................................................................. 29
3.7: Independent Smart Growth Variables .......................................................................... 29

v

3.8: Independent Regional Variables .................................................................................... 31
3.9: Dependent variables ........................................................................................................ 32
3.10: Confounding Variables ................................................................................................. 33
3.11: Geoprocessing of variables ........................................................................................... 33
3.12: Statistical analysis .......................................................................................................... 39
CHAPTER 4: RESULTS..................................................................................................................... 41
4.1 Bivariate Correlations ...................................................................................................... 41
4.2 Regional Variables ............................................................................................................ 44
4.3 Interaction Testing of Regional Variables ......................................................................... 49
4.4 Linear Regression Modeling ............................................................................................ 50
4.5 Vehicle Miles Traveled Linear Regression Model ......................................................... 50
4.6 Automobile Ownership Linear Regression Model......................................................... 53
4.7 Conclusion ......................................................................................................................... 54
CHAPTER 5: DISCUSSION AND CONCLUSION ................................................................................ 55
5.1 Assumptions and Limitations .......................................................................................... 59
5.2 Future Research ................................................................................................................ 60
REFERENCES .................................................................................................................................. 62
Appendix A .................................................................................................................................... 66

vi

LIST OF TABLES
Table 1 Hypotheses of Variables 20
Table 2 Table of Significant Correlations 42
Table 3 VMT and Automobile Ownership Mean Values 45
Table 4: Vehicle Miles Traveled Best-fit Regression Model 51
Table 5 Correlation between the 3 independent variables in the best-fit regression model 52
Table 6: Automobile Ownership Best-fit Linear Regression Model 53

vii

LIST OF FIGURES
Figure 1 San Diego, CA MSA DEM Raster Data by Sample Block Groups 23
Figure 2 San Diego, CA MSA Slope Without Water by Sample Block Groups 24
Figure 3 Example of KML Acquisition Within Google Earth for Four Sample Block Groups in
the Portland, OR MSA 26
Figure 4 Mean Temperature by Sample Block Group for Washington, DC MSA 28
Figure 5 Google Earth KML Data for Bus Stops in ArcMap for Conversion to Shapefile 35
Figure 6 Bus Stop Shapefile created from Google Earth KML Files 36
Figure 7 Scatterplot of VMTs and MSA Density 47
Figure 8 Scatterplot of Automobile Ownership and Transit Expenditure 48
Figure 9 Map of MSA Locations 66
Figure 10 Chicago MSA and Samples 67
Figure 11 Miami MSA and Samples 68
Figure 12 Portland MSA and Samples 69
Figure 13 San Diego MSA and Samples 70
Figure 14 Washington D.C. MSA and Samples 71

viii

ABSTRACT
This study tests the effects of the built environment on vehicle miles traveled (VMTs) and
automobile ownership, with specific reference to aspects of neighborhood walkability studies
and research design at nested spatial scales of metropolitan regions and neighborhoods. This
adds to existing smart growth studies as they tend to focus on Census data and non-walkability
land use variables such as rail transit infrastructure. This study looks at 75 census block group
samples within 5 metropolitan statistical areas (MSAs). The variables measured for the study
include, bus stops per square mile, jobs within 45 minute transit ride, gross activity density,
temperature, distance to retail, distance to transit, and slope among others. This study also looks
at including regionally measured variables such as people per transit station, MSA density, and
transit expenditure in conjunction with neighborhood scaled variables in order to test if there are
any interactions between neighborhood and regional variables. The variables are entered into a
multivariate regression model to find the best-fit model in order to explain the relationships
between the dependent and independent variables. The study finds that the new walkability
variables added to the research add significantly to the explanatory value of regression models
beyond studies that use just smart growth land use variables. The implications for this study are
that there is ground work laid for a new type of smart growth and walkability joint study at a
multiple region level.

1

CHAPTER 1: INTRODUCTION
The purpose of this study is to examine how smart growth strategies and walkability of the built
environment impacts vehicle behavior and automobile ownership. Vehicle behavior will be
measured by vehicle miles traveled (VMTs). While this is a dual study of smart growth and
walkability the primary emphasis will be on the walkability variables. The smart growth portion
of the study, mainly the variables and a similar analysis method are built in part from the initial
work in this area conducted by Eisman (2012).
Earlier smart growth studies, like Eisman (2012), measure the built environment by
population density, job access, connections to transit networks, and street connectivity. A high
level of street connectivity, measured as intersection nodes, is thought to be a proxy for
walkability of neighborhoods. Other studies look at rail variables such as passenger miles per
capita and mass transit infrastructure density (Cervero and Murakami 2009). Also some studies
look at jobs-housing balance along with other land use variables (Cervero and Duncan 2006).
Although the underlying sampling framework and initial data sets are built from Eisman (2012),
this study will both develop some new independent variable datasets and add to the existing set
of independent variables by utilizing the basic smart growth variables found with the literature to
measure the mix of economic uses and distance to jobs.
The walkability aspect of this study will take a different approach than most walkability
studies. This study is a multiple region study utilizing random samples of census block groups
from five Metropolitan Statistical Areas (MSA). Because of the scale of the study area, a typical
walkability study is not suitable due to the type of variables used in such a study. The issue at
hand is scale. Most walkability studies utilize variables found at medium to small scales: county,
city, and most often neighborhood or street. Datasets are independently created for various

2

regions. Because of this the data involved in a walkability study can be very detailed. Variables
for a walkability study often include, tree canopy, side walk quality, American with Disability
Act codes such as curb ramps, public safety data, street illumination and street width (Spoon
2005). These variables are difficult and time consuming to collect at neighborhood scale
samples within a multiple region study. This is because a region not only contains many
neighborhoods but also many different city and county jurisdictions which may or may not
maintain such data or have GIS/mapping programs at all. However, there are walkability
variables that can be constructed from national datasets to be utilized in a multiple region
walkability study. Some of these variables are neighborhood slope (Villanueva et al, 2013),
mass transit access, climate, and mixed land use (Spoon, 2005).
This study will conduct a multiple region smart growth study with an emphasis on
regionally favorable walkability aspects. This is something rarely seen in the literature with
multiple region level study areas, especially in so far as studies are able to utilize detailed
variables at the street level (see Appendix A for MSAs, their locations and sample locations
within the MSAs).
1.1 Motivation
The method of how we travel in the future is already becoming an issue with fossil fuel usage
ever increasing and long term threats to fuel supplies. The impact of fossil fuel consumption
from automobile transport on global warming and its effect on our lives is also a growing
concern. Large cities have always been in the news for traffic jams and general everyday
congestion. Highway construction can even lead to near Hollywood style traffic jams such as the
2011 “Carmageddon” in Los Angeles, CA (Reuters, 2011) or the 9 day 60 mile traffic jam near

3

Beijing in 2010 (Reuters, 2010). These concerns and events show that there is a need for
effectively evaluating how we design urban areas and even entire regions.
As well as global warming from vehicle emissions, there is also the issue of pollution.
Health issues from smog can be severe. Smog (lower atmosphere ozone) can cause several
respiratory issues, such as aggravating asthma, reduced lung function, and irritate the respiratory
system (EPA 1999). Along with regulation on vehicle emissions, reducing people’s need to use
their vehicle by smart urban design that promotes walking or the use of public transportation can
reduce the emission of chemicals that form smog and reduce the health hazard attributed to
vehicle as well as industrial causes of smog.
While alternative fuel for transportation holds much promise from hydrogen to biofuels
(Gajendra Babu and Subramanian, 2013), there also needs to be emphasis and effort from all
levels of government in designing urban and suburban environments that contribute to better
walkable options and efficient suitable transit networks to encourage lower usage of personal
vehicles. In 2008 California enacted California Senate Bill 375, which creates a formal process
for local planning officials to develop land use plans that reduce vehicle miles travelled (VMTs
and thus reduce greenhouse gas emissions from automobile travel (CEPA 2014). Unfortunately
there do not appear to be any penalties for not adopting CA SB 375 and it is merely an incentive
plan to encourage local governments and developers to adopt its policies to meet California’s
greenhouse gas emission goals (CEPA 2014).
1.2 Study Overview
This study seeks to aid in the evaluation of land use policies by identifying how the built
environment effects automobile transit and apply that to new development or re-development of
city centers through statistical analysis of human and geographic variables. The study employs

4

multivariate linear regression to accomplish this. The intent is to use a multiple region study to
identify how the different smart growth and walkability variables affect VMT and automobile
ownership so that comparisons can be made for different regions and land use policies.
One of the study regions for this project is the Portland, Oregon metropolitan statistical
area (MSA). This region is particularly known for its smart growth policies. Jun (2008) studied
whether their policies were related to reduced automobile dependence and found mixed results.
However, areas with higher accessibility to light rail and bus services and more mixed land used
were associated with a higher probability of alternative modes of transportation to personal
vehicles or driving alone. However, areas with higher employment and residential densities
alone did not see a reduction in driving. The effect of accessibility of light rail and bus routes
with high mixed land use appears to show the positive effects of smart growth and walkability
polices. The inclusion of the Portland MSA then, should help within the study when utilizing
other MSAs, several with greater populations, from different regions of the country.
The assumption that various changes in land use policies and land use itself can raise or
lower VMTs deserves robust study. The analysis and method used within this study adds to the
existing research by implementing a multiple region smart growth and walkability evaluation. It
is important to consider variables both at the neighborhood and regional scales, as well as
potential interaction between variables at these two scales. For example, is a densely built
neighborhood in a region with high mass transit network density more likely to have lower
VMT’s or automobile ownership than a similar neighborhood situated in a region with low mass
transit network density? This is a detailed version of the overall research question that motivates
this study: What elements of urban built environment at the neighborhood and regional scales are
associated with reductions in VMTs or automobile ownership?

5

Following from this research question, this study proposes and tests a set of 14
hypotheses. Of these, 6 of the hypotheses rely on newly created independent variables. The
study makes extensive use of geospatial datasets and ArcGIS to develop these variables. The
hypotheses, variable definitions and construction are carefully described in the Methods Chapter.
In the analysis, many of these hypotheses are initially rejected as bivariate correlations
and some are further rejected in considering the total regression model. In its Results Chapter,
the study identifies key differences between the regions and tries to explain these differences by
testing for interactions of regional scale variables with neighborhood scale variables. Two
complete regression models are presented and need for future work is identified.

6

CHAPTER 2: BACKGROUND
Interest in research on smart growth approaches to urban development has increased along with
concerns over rising fuel cost, decreasing oil reserves, and climate change. Smart growth aims in
large part to locate and design neighborhoods in ways that decrease vehicle miles traveled and
potentially also decrease automobile ownership (Cervero and Murakami 2009). Interest in
walkability of neighborhoods has also grown in parallel to health concerns of obesity, diabetes,
and other conditions thought to be due in part to a lack of exercise (Miranda et al. 2012).
Walkability research aims to understand the built environment or neighborhood factors that may
encourage people to exercise.
The research done in this area is generally separated between studies focused on smart
growth versus walkability. The two types of studies are rarely integrated in a comprehensive
way, especially in terms of explanatory variables. Smart growth studies tend to look at transit,
population density, employment, road networks and housing density (Eisman, 2012, Cervero and
Murakami, 2009). Alternatively, walkability studies use a wide range of variables, some similar
or the same as smart growth but many are very different such as, tree cover, sidewalk conditions,
climate, terrain (slope), and street width to name a few. Walkability studies may also include
variables related to individual human behavior such as habits or family circumstances (Spoon
2005).
Aside from contrasting research variables, the theme of scale is also an important point of
difference between the smart growth and walkability literatures. Smart growth studies tend to
have small to medium sized study areas consisting of cities or urban areas, with neighborhoods
or census block groups as the focus point for data. Few studies address sample or study areas at
the regional level. Walkability studies typically use even more focused study areas than smart

7

growth research. Most if not all walkability studies deal with study areas at very local scales
such as streets, neighborhoods and cities, there was no literature found during research for this
study to indicate studies at larger scales.
Within this study, smart growth and walkability variables effect on VMT and automobile
ownership are analyzed together. Walkability variables will be analyzed along with smart
growth variables at a large regional scale utilizing Metropolitan Statistical Areas which in some
cases may consist of over 20 counties. Five different MSAs across the country will be included
with samples consisting of neighborhood scale samples measured as census block groups. The
variables chosen for measurements of smart growth consist of nearly all variables typically used
within the literature while measurements of walkability will be chosen based on aspects that are
conducive to the size of the sample (n=75) along with scale of the study areas. Variables that
require extensive fieldwork to gather data or hand digitization of image files, such as side walk
conditions and tree canopy cover are not conducive to the scale of this study.
This chapter reviews the literature on how smart growth affects VMTs and automobile
ownership, placing an emphasis on the scales used for each. Next, the chapter reviews
walkability studies focusing on walkability measures of communities, cities, and counties. In
general walkability studies do not appear to deal with walkability’s effect on VMTs or
automobile ownership. One key idea tested in this study is that neighborhoods that are more
walkable might also result in fewer VMTs and/or reduced ownership of automobiles.

8

2.1 Influence of Smart Growth non VMTs and Automobile Ownership
The majority of VMT and automobile ownership studies tend to focus on small scale study areas
such as neighborhoods, cities, counties, block groups or Census tracts. Few if any of these
studies look at very large study areas such as an MSA (Metropolitan Statistical Area), include
multiple metropolitan regions, or utilize walkability variables along with smart growth variables
to determine the effects that the combination of these variables may have on VMTs and
automobile ownership. That being said, there are two studies that address multiple regions
consisting of block groups within an MSA. However, neither includes a combined look at the
influence of smart growth and walkability effect on VMTs or automobile ownership
In his research on smart growth, Daniel Eisman (2012) looked at how smart growth
within the built urban environment affected automobile ownership and vehicle miles traveled
(VMTs). His method of analysis involved examining several land use and statistical variables at
the neighborhood level. The study involved 75 samples that consisted of census block groups as
neighborhoods and were located within 5 Metropolitan Statistical Areas (MSAs) covering 5
separate regions of the United States. The study evaluated distances to jobs, retail and transit as
well as population density, transit networks income and several other population statistics.
Eisman found based on his series of best fit regression models that there was a statistically
significant association of VMTs and automobile ownership with the built urban environment.
Eisman’s study was the starting point for the research performed in this study (Eisman 2012).
Robert Cervero and Jin Murakami (2009) have looked at effects on VMTs from the built
environment by evaluating 370 urbanized areas around the US. Urbanized areas within the study
consisted of individual and connected cities (such as the Los Angeles, California area). These
urbanized areas vary in size and the variables for the study aggregate metrics for the entire study

9

area (i.e., region). The variables used within the study were VMTs, rail variables such as
passenger miles per capita and infrastructure density, population and employment densities,
income, and areal size variables. The variables were put into a path dependent model (structural
equation model (SEM)) to test their effects on VMTs and it was found that population density
has the strongest association with VMTs at a direct coefficient of -0.604, indirect coefficient of
0.233 and a total coefficient of -0.381. The total coefficient results from the overall model,
allowing for other variables in the path. This is the result that the authors report as their major
finding. Some of their other results included automobile commute shares at a total coefficient of
.602, and road density at 0.415. Employment, size of urbanized area, and rail-transit were found
to have less effect on VMTs (Cervero and Murakami 2009). This study was one of the few
articles in the literature that looks across multiple regions. However, unlike this study, the
authors use only regional data or regionally defined variables, computed as averages of
neighborhood scale analysis. Additionally, unlike this study they do not test for the interaction
between the neighborhood and regional scales.
In a separate study Cervero and Duncan (2006) explore whether retail-housing mixing
(mixed land use) or jobs-housing balance have the greatest effect on VMTs and VHTs (vehicle
hours traveled). Jobs-housing balance is a form of land use planning that attempts to bring jobs
and residents into closer proximity in any given community. For example, one policy sometimes
used is to offer grants to workers who purchase residences close to their job. Utilizing regression
models for their variables they found that jobs-housing balance had a greater effect on VMTs
than mixed land use (Cervero and Duncan 2006). However for jobs-housing balance to be
effective, it requires incentives to employees and meaningful efforts by local jurisdictions to
support it by rezoning and finding funding for these types incentives.

10

While most studies look at land use effect on VMTs and ownership, not all studies use
land use as the variable of interest. Kim and Brownstone (2010) studied the impact of residential
density on VMTs as well as fuel consumption. The scale of the study is at a neighborhood and
midsized city or urban scale in that the data was sampled and measured utilizing census tracts
and block groups but the regions to study were chosen at the regional scale utilizing MSAs. The
housing density variables were taken from census data and covered most available population
and household variables available. These variables were population per square mile by block
group and tract level and housing units per square mile. It is not stated where the tract level
variables are considered regional variables but typically variables at the tract level are not
regional. The variables were imputed into a simultaneous equation model in order to calculate
the effects on VMTs and fuel usage.
The results of the research were that there was a statistically significant effect of land use
and population density on VMTs and fuel usage. The main result was that more densely
populated areas had few VMTs and less fuel usage. Results for the study were broken up by
variable as well based on the types of variables. Socio-demographic variables results were
broken down into individual results. A number of drivers had a strong influence on household
vehicle behavior. The number or workers had a nonlinear effect on annual mileage and fuel
usage. Income, which is a variable within this study as well, was found to be statistically
significant and fuel usage increased linearly with income. Number of children was overall
statistically insignificant. “Life cycle effects” which consisted of three variables, retired
households, single person households, and non-single person households, had different statistical
significance for each variable. Retired households had a negative direct effect on mileage

11

traveled, while single person households were statistically significant for household density, fuel
consumption and annual mileage (Kim and Brownstone 2010).
2.2 Walkability studies and vehicle behavior
This study’s main focus is aspects of walkability effect on VMTs and automobile ownership in
combination with smart growth and transit variables. There is a large literature on measurement
and indexing of walkability. However, most walkability studies focus on county, city,
neighborhood, or street study areas within a specific region. Also, the literature does not
combine walkability metrics with the more commonly investigated transit and smart growth
variables. The goal here, as identified above, is to use nested neighborhood and regional scales
of multiple regional (MSA) study areas to better understand the role of walkability in influencing
vehicle behavior.
Stephanie Chen’s (2012) study on bus route walkability along two Orange County bus
routes is scaled at the level of a single county, focusing on the neighborhoods along the lengths
of existing bus routes. The variables Chen chose for her study were population density, street
connectivity, steepness, and tree canopy, along with 3 different buffer types: half-mile radii,
route-adjacent, and stop-and-route-adjacent. The buffers were used to calculate a walkability
score and then compared to the scores for stops along each of two bus routes. She found that in
general, walking paths to bus stops routed along grid street neighborhoods were more walkable
than cul-de-sac neighborhoods (Chen 2012).
Another type of multiple location walkability assessment, was performed by Robert
Stevens who looked at determining the effectiveness of a walkability assessment type to assess
the walkability of four local area parks in Springfield, Oregon (Stevens 2005). The study area
scale for the study is the neighborhood scale; however the scale of measurement was at the street

12

level specifically street segment. The data collected for use in the study was TIGER file street
data from the US Census and then field data collected by Stevens himself.
The appeal of this study, in part due to its study scale, is that the data can be collected
locally and by the researcher themselves. The data was collected using ArcPad software on a
PDA and utilizing a data collection method call the Pedestrian Environment Data Scan (PEDS)
(Stevens 2005). The PEDS assessment tool contained 77 walkability indicators but Stevens
chose 20 of those indicators that were found to be the most important in assessing walkability.
Being a street level study, these indicators/identifiers can be very specific such as: attractiveness
for walking, safety, traffic volume, sidewalk condition, land uses, number of traffic lanes,
presence of building setbacks, crossing aids, tree count along streets, and if there are parking lots
that can be walked through. Each street segment being assessed was given a score based on
presence of these indicators. The analysis of the data was done using ArcGIS with the sampled
street segments being aggregated to census blocks to create easy to read polygons. Next,
catchments, defined as the ½ mile area where streets are available for pedestrians to use were
created utilizing multiple centroids in the polygons to indicate park entrances (Stevens, 2005).
The hostile streets were then removed as not being suitable for walking. The results of this data
acquisition and creation were then compared against TIGER and LCOG street classifications
(Stevens, 2005).
The findings from Stevens (2005) determined which parks had better walkability
amongst the four tested. The purpose of the study was to test the method of walkability
assessment to aid in future development of Springfield, Oregon as well as determine the parks
with the best walkability. It is different than this study which seeks assess aspects of walkability
of a region. Still, Stevens’ research gives good insight into the types of variables used in

13

determining walkability and the study scales required to utilize each type of variable. There is a
tension between the scale of the study area and level of detail that can be gathered in terms of
walkability variables.
Anupama Mantri (2008) studied a GIS approach to measuring walkability within a
neighborhood and applied the variables within his GIS model to parcel level data. The measures
of walkability used by Mantri were connectivity (road network), proximity (access/proximity to
activities), density (residential density), land use mix, and safety measures (Mantri 2008). His
approach and scale of his study area allowed for in depth analysis of the neighborhood including
location information for many different activities and destinations specific to the neighborhood
such as theatres, restaurants, pharmacies, churches etc. This could be considered an ideal
method of studying walkability due to the amount of data available to a small study area such as
a neighborhood rather than analyzing larger study areas such counties or regions.
While researching walkability studies, it became apparent that most walkability studies
deal with the cities, neighborhoods, or smaller landscapes. However there are studies that deal
with these small study areas but also link several of them in a region or nationally. Horacek et
al. (2012) studied the walkability and bikeability of US postsecondary educational institutions,
which could be considered neighborhood study areas depending on the size of the campus. The
variables used for studying walkability and bikeability and which were scored on a set of
standards, were sets of criteria for “safety, path quality and path temperature comfort” (Horacek
et al. 2012, 10). Safety criterion were variables such as crosswalk quality, night time safety such
as lighting and side walk existence and quality. Path quality was made up of variables such as
path size, buffer zone from road ways, and terrain (elevation change, slope etc.) among others.
Path temperature constituted whether there was shade or not (Horacek et al. 2012). These types

14

of variables are difficult to measure and obtain for larger study areas making it difficult to
determine the walkability of multiple neighborhoods in multiple counties and in multiple
regions.
There does not seem to be a single group of walkability variables though many seem
constant in several studies. In order to determine what makes up walkability measurements and
what variables may be used in measuring walkability, Steven Spoon (2005) researched the
literature to create a relatively comprehensive list of what is considered important in defining and
measuring walkability as well as the variables that are most prevalent amongst the different
studies. The variables that are found in several different literature reviews and are to be looked
at in this paper are density, access to transit, and mixed land use. These are the variables that
make the most sense and are most accessible for the scale and diversity of the study areas of this
study.
The variables that were chosen for this study were taken from the list created from
Spoon’s research of walkability studies. These variables were chosen based predominately on
the study regions and scale of the samples along with the availability of data due to the scales.
The first variable was elevation on the prediction that flatter land is easier to walk on.
Temperature and precipitation was chosen next on the prediction that milder climates would
encourage more walking. The next variable, bus stops per square mile, was chosen as a
prediction that many bus stops make it easier to walk to access public transit networks. The next
chapter will discuss the reasons and methods of acquisition of the data for these variables.
While there are many studies and assessments of walkability within the literature, there
are also websites that allow the general public to find the walking score for their streets or
neighborhoods by simply entering an address. Walkscore.com is a website that will calculate

15

walk score utilizing a web application designed to calculate walk score based on measurements
of distance from a chosen location to surrounding amenities such as super markets, restaurants,
entertainment venues, and parks as well distance to transit stops. There are other data types used
in the calculations such as road network and route directness. Since the data is calculated on the
fly through a web app the data used in the application would only be applicable as individual
data types and thus the walk score would not be incorporable into the data sheet for the
regression models used in this study (Walk Score 2014).
After reviewing the literature for smart growth and walkability there are several
consistencies within each topic. Smart growth research uses study areas at many different scales
from regional to neighborhood and has a large number of census based variables such as
population density, income, employment, etc. Geographic variables within the studies are all
similar as well, utilizing land use and transit variables in most cases. Walkability studies use
smaller scales with the largest being at the city or county scale. The variables within these
studies cover a very wide range detail and scales. The most detailed study areas appear to be
small scaled studies at the street level that in some cases can only be collected in the field by the
researcher themselves. This study looks to complete a multiple region smart growth study with
regional and neighborhood scaled variables at neighborhood scaled samples. This study will add
to the literature by adding walkability variables that can be measured at the neighborhood scale
but can be acquired from multiple regions with some ease as variables such as sidewalk
condition are neither feasible nor cost effective for a study with region scaled study areas.

16

CHAPTER 3: METHODOLOGY
The purpose of this study is to look at how smart growth, along with aspects of walkability, may
impact vehicle miles traveled (VMTs) and automobile ownership at the neighborhood scale. The
locations for this study were five (5) metropolitan statistical areas chosen to represent regions of
the United States with different development histories and resulting urban forms. The samples
are neighborhood level and represented as census block groups as the method for applying
census data and acquiring and processing non census data.
The sampling framework and initial model of Eisman (2012) is a foundation for this
study. But this study develops new variables related to walkability, applies variables from
EPA’s Smart Location Database (EPA, 2013), and deepens analysis of regional effects. As
stated previously, the majority of walkability studies tend to only consider variables measurable
at county, city, neighborhood, or street scales and rarely compare or analyze multiple regions or
locations. Utilizing variables used within the literature and obtainable by geoprocessing for
multiple regions and jurisdictions, it is hypothesized that the predictive power of existing
regression models for vehicle miles traveled (VMTs) and automobile ownership at the
neighborhood scale may be improved.
3.1: Sampling Framework
This study adopts the sample of 75 Census block groups developed for Eisman’s (2012) study.
Based on Eisman’s logic, block groups make sense because they are essentially predetermined
neighborhoods consisting of clusters of several city blocks with the exception of more rural areas
where a block group can be many square miles. It is important to note that a few of these larger
block groups are included in the random sampling in some of the five Metropolitan Statistical

17

Areas (MSAs) that make up the study regions when the samples include block groups in the
outskirts of the urban area.
Census block groups were selected by Eisman because they are the best scale at which to
get detailed information for key variables due to their small size (Eisman 2012) and they are the
closest in scale to what could be considered a neighborhood. The new independent variables
developed for this study, however, do not use census data, and while spatially matched to the
block groups, they are not directly reliant on the block group as the spatial unit of measurement.
Still, the unit of analysis remains the census block group because several variables gathered by
Eisman (2012) and tracked in EPA’s Smart Location Database are only available at the block
group level. Using a different scale for this study could negate the attempt to apply both smart
growth and walkability into a single examination of transportation behavior by possible
eliminating the smart growth variables.
The MSAs chosen for the study were Chicago, IL; Miami, FL; Portland, OR; San Diego,
CA; and Washington DC. These areas were chosen because they represent different regions
around the United States: the Midwest, South East, North West Coast, South West Coast and the
East Coast. MSAs were chosen as the regional unit of measure because they are defined in part
by transportation needs including commuting across the area. MSAs consist of 50,000 people or
more and contain the counties that consist of the core urban area and adjacent counties that have
a high degree of social and economic integration with the urban core. This integration is
measured by the commute to work between adjacent areas and the urban core (US Census, 2014)
Eisman (2012) selected 15 sample block groups from each of the five MSA’s. In general,
the samples were selected using a process known as stratified random sampling (Eisman, 2012).
Within the selected MSAs, a “random” field within an Excel spreadsheet was assigned to all the

18

block groups. Within the added “random” field, a random number generator was used to place
random values between 0 and 1 in the column. The spreadsheet was then ordered based on the
“random” field and the lowest thirteen random numbers were chosen. The fourteenth and
fifteenth sample block groups of each sample region was determined by taking the most and least
dense block groups within each of a given MSA. Thus, it is important to note that the sample is
not a pure stratified random sample, but includes two outliers for each region with regards to
neighborhood population density.
3.2: Hypothesis and Independent Variables
Eisman hypothesized that the built environment would influence the number of automobiles
owned per household as well as the vehicle miles traveled (VMTs) and that automobile
ownership will be greater in sprawl neighborhoods (Eisman 2012). The same holds true with
this study with the added hypothesis that a built environment conducive to walkability will affect
the number of automobiles owned as well as VMTs. It is predicted that areas with greater
walkability will have reduced automobile ownership and vehicle miles traveled. See Table 1 for
research variables and their hypotheses.
The variables used in this study to determine walkability are; neighborhood and street
slope, number of accessible bus stops per square mile, climate, and land use. There have been
various studies on what defines walkability and the variables used to measure it. The variables
for this study were decided upon based in part on Spoon (2005) in which several pieces of
literature were analyzed and their variables assembled. Many of the variables used to measure
walkability are available at only small analytical scales. The included variables such as sidewalk
quality, tree cover, or walkway availability. These types of variables are not conducive to a
regional study such as this which compares several study areas across several different MSAs.

19

While MSAs such as San Diego CA, consist of only one county, the others can consist of nearly
twenty counties which make local data acquisition difficult, if not impossible.
For this reason, variables following from Spoon (2005) that can be determined from
national datasets or a single provider such as Google or the USGS are ideal as they provide
consistent or as close to consistent data across the United States. Such variables include bus
stops per square mile, temperature and precipitation raster datasets and Data Elevation Model
(DEM) raster datasets. These variables can be acquired for the entire nation and processed down
to the neighborhood scale for analysis.

20

Table 1 Hypotheses of Variables
Dependent Variables Hypothesis
Automobile Ownership
Land use and walkability aspects will influence
automobile ownership.
Vehicle Miles Traveled
VMTs will be reduced by land use and walkability
aspects.
Independent Variables Hypothesis
Neighborhood Slope
Will have a direct relationship with VMTs and
Automobile ownership.
Bus Stops per sq/mile
Will have an inverse relationship with VMTs and
Automobile ownership.
Minimum Temperature
Will have an inverse relationship with VMTs and
Automobile ownership.
Maximum Temperature
Will have an inverse relationship with VMTs and
Automobile ownership.
Minimum Temperature
Will have an inverse relationship with VMTs and
Automobile ownership.
Precipitation
Will have an inverse relationship with VMTs and
Automobile ownership.
Jobs Within 45 min Transit Ride
Will have an inverse relationship with VMTs and
Automobile ownership.
Jobs Per Household
Will have an inverse relationship with VMTs and
Automobile ownership.
Land Use Diversity (Mixed land Use)
Will have an inverse relationship with VMTs and
Automobile ownership.
Density
Will have an inverse relationship with VMTs and
Automobile ownership.
Distance to Job Centers
Will have a direct relationship with VMTs and
Automobile ownership.
Distance to Retail Centers
Will have a direct relationship with VMTs and
Automobile ownership.
Distance to Transit
Will have a direct relationship with VMTs and
Automobile ownership.
Transit Expenditure
Will have an inverse relationship with VMTs and
Automobile ownership.
People Per Transit Station
Will have an inverse relationship with VMTs and
Automobile ownership.
Regional Density (MSA Density)
Will have a direct relationship with VMTs and
Automobile ownership.
Confounding Variables Hypothesis
Income
Will have a direct relationship with VMTs and
Automobile ownership.

21

3.3: Neighborhood Slope and Walkability
Slope was chosen and developed as a variable in this study because it is hypothesized that areas
that are flatter will be more appealing and conducive to walking as a mode of transportation and
cost saving method. If it takes less physical effort for someone to walk back and forth from
work to home or to the shops it may be more likely for people to do so and perhaps to drive less
as a result. While flatter land maybe best for general walking commutes it could be suggested
that for recreation and health, steeper slopes may be appealing as well (Villanueva et al. 2013).
However, this study is focusing on walkability as a means of everyday transportation and not
primarily as a method of exercise.
This study measures slope both for the overall neighborhood and for the street network.
Both slopes were measured because: a) sidewalks are built along streets and thus should have the
same slope and streets may be graded for more gradual inclines than the surrounding areas, and
b) walkable areas do not occur only along sidewalks. There may be parks and walking paths
between buildings that can be used as shortcuts through city blocks that allow for more efficient
travel by walking. By measuring both slopes, differences in slope of possible walking space can
be measured.
Neighborhood and street slope were derived from USGS Digital Elevation Model (DEM)
10 meter raster files and were chosen as variables for walkability based on the prediction that
neighborhoods with a greater slope percentage would not be as walkable as a flatter area.
Neighborhood slope (slope without water) is the percent slope of the entire block group with a
quarter mile buffer in order to ensure measurement of rasters immediately adjacent to the
boundaries of the neighborhood. This buffer also helps to measure the percent slope of the roads
which in many cases do not fall inside the block group for the reasons stated above. Street slope

22

was calculated by extracting raster values from the neighborhood slope raster using a street
network data. Street slope was chosen along with neighborhood slope in order to cover areas
that maybe more level due to the leveling and or raising of road networks during construction.

23

Figure 1 San Diego, CA MSA DEM Raster Data by Sample Block Groups

24

Figure 2 San Diego, CA MSA Slope Without Water by Sample Block Groups

25

3.4: Proximity of Bus Stops
Bus stops per square mile was the next variable looked at to define walkability. The
hypothesis is that if there are a large number of available bus stops in a neighborhood, VMTs and
automobile ownership will be low because people can easily walk to a bus stop. Previous studies
have focused on rail access because bus stop data have been hard to develop across multiple
regions. However, this study hypothesizes that bus stops may be particularly important because
in many U.S. cities bus transportations is the only mass transit network covering most of the
region. Also, even in cities with highly developed rail networks, bus networks are critical for
access on the first and last mile of trips using the rail network. For this study, bus stop data was
acquired through Google Earth data as it was the best source of bus stop data available
nationwide and was verifiable through the use of their “street view” function to see if stops are
actually present (Google Earth, 2013). The number of stops was counted and normalized by the
areal extent of the block group.

26

Figure 3 Example of KML Acquisition Within Google Earth for Four Sample Block
Groups in the Portland, OR MSA

3.5: Climate and Walkability
The third variable to determine walkability at a regional scale is climate. The prediction is that
moderate climates encourage walking. Areas with a cooler summers as well as warmer winters
are presumed to be nicer to walk in, thus encouraging walking and reducing use of automobiles.
The data used to determine temperature were thirty year normal maximum temperature, thirty
year normal minimum temperature and thirty year mean temperature. The thirty year mean
temperature was chosen as a variable along with minimum and maximum temperature to cover
for the potential of abnormal minimum or maximum temperatures that could affect the analysis,
such as 1 year out of 30 where an area may experience an abnormally cold winter. The variables

27

were based on the averages for their respective measurements over 30 years. The reason for the
different variables for climate is to simply cover the range of possible temperature variables.
Areas may have extreme cold or heat for only a month or two, but the mean temperature may be
mild. Thus by modeling all three temperature variables there is a better chance of determining if
climate has any factor. The data came in the form of raster data from the Prism Climate Group
and the cells were extracted based on the sample block group shape file (Prism, 2014). Aside
from temperature, precipitation was also looked into for determining climate. Areas with
substantial yearly rainfall may not be as walkable, or as nice to walk around, if it rains most of
the year. The data from precipitation came from a thirty year normal precipitation raster file.
The precipitation data also came from the Prism Climate Group.

28

Figure 4 Mean Temperature by Sample Block Group for Washington, DC MSA

29

3.6: Land Use and Walkability
The final variable category used to define walkability was based on mix land use. This variable
was made up of three variables obtained through the Smart Location Database from the EPA
(EPA, 2013). These variables consisted of jobs within a forty-five minute transit ride, jobs per
household and land use diversity. Land use was chosen as a walkability variable because it is
based on the idea that areas with a diverse mixture of land use will encourage walking because
people may more easily be able to walk from home to work, school, or shopping if these
locations are within their census block group. Thus, land use diversity could lead to lower
vehicle ownership as families might forgo a car with less need to drive, or it might lead to lower
VMTs as people find work and shopping closer to home.
The variable for jobs within a forty-five minute transit ride reflects accessibility to jobs
from homes and therefore may show no need for automobile ownership or at least fewer VMTs.
This variable was chosen to reflect land use and walkability on the hypothesis that if people have
jobs close to their home they may be more likely to take transit to their job instead of using a
personal vehicle.
Jobs per household is used as an indicator of walkability on the hypothesis that areas with
a better balance of jobs to residences may promote walking to work or shorter drives to work.
Thus neighborhoods with a higher ratio of jobs per household would be predicted to have fewer
VMTs and lower rates of automobile ownership.
3.7: Independent Smart Growth Variables
As mentioned at the outset, this study seeks to add variables on smart growth and walkability to
Eisman’s (2012) initial study. The objective is to understand some of his original variables and
models in the context of new and additional independent variables. Thus, this study also

30

incorporates the majority of smart growth variables used within his study. The variables used to
determine and measure smart growth effect on VMTs and automobile ownership were density,
distance to jobs, distance to retail centers, and transit access.
Densification of neighborhood regional urban development is a leading smart growth
strategy. Dense neighborhoods often, if not always, mean multi-family residences such as multi-
story apartment complexes. These types of environments often make parking difficult and
expensive for both residents and people visiting nearby business or the families in those
apartments. Because of this walking or public transit such as light rail or bus lines, make more
sense for some people in how they travel in their daily lives. The hypothesis for density is that
the higher the density there will be less vehicles owned and less VMTs. Density was measured
as the number of households in a block group divided the acreage of the block group.
Distance to jobs is the measure of the distance from the block group to an “employment
center.” Eisman (2012) defined an employment center as Census tract in the top ten percent of
an MSA with total number of jobs. He obtained the job data from the 2000 Census
Transportation Planning Package (CTPP). The distance between sample block groups and
Census tracts was from the centroids of each polygon. The hypothesis for the variable is that the
closer somebody lives to their job, the more likely they are to take public transit and potentially
walk if they are close enough. In turn, the farther a person lives from their place of employment
the more likely they are to use their own vehicle.
On top of going to work, people need to shop for food, clothing, amenities, etc. Distance
to retail is the measure of the sample block groups to the Census tracts in the top ten percent of
the MSA in terms of aggregate retail jobs. Like distance to jobs this data was also obtained from
the CTPP. The hypothesis is that the closer the people live to retail centers the less they will

31

need to drive and may choose other modes of transportation. However, it should be noted that
some forms of shopping or long shopping trips may require the need for the use of a personal
vehicle to transport purchases to a person’s residence.
The last smart growth variable utilized from Eisman’s study is distance to transit stations.
This variable refers to rail transit only making the walkability variable of bus stops per square
mile viable and a nice addition to the walkability variables. Rail transit can be a great mode of
transportation for some people. Light rail services like trams or trolleys can be a comfortable
and cost effective way for people to get to and from work but only if the stations are accessible.
Distance to transit is the measure of the sample block groups to a transit station. The hypothesis
is that better access to transit stations will encourage lower VMTs and automobile ownership
3.8: Independent Regional Variables
Originally these independent regional variables were utilized in Eisman’s study as confounding
variables. However, this study spends more time and delves deeper into their possible effects on
VMTs and automobile ownership through scatterplot analysis and analysis of potential
interactions with block group scale variables. The independent regional variables were
individual transit expenditure, regional (MSA) density, and people per transit station (rail).
Individual transit expenditure is a variable that measures the annual consumer
expenditure on public transportation per person for a given region. This variable is measured at
the MSA level and the data came from the Bureau of Labor Statistics’ 2000-2001 Consumer
Expenditure Survey. It is expected that higher expenditures will have a negative effect on VMTs
and automobile ownership and should have a correlation with the land use and number of bus
stop variables.

32

People per transit station is the measure of the total MSA population divided by the total
number of transit stations in the MSA. It is expected that the fewer people there are per transit
station the lower VMTs and automobile ownership will be, because fewer people per transit
station means that there more of an abundance of transit stations for people to have access to.
The last variable, regional density is a measure of the MSA population per square mile of
area. It is expected that more dense regions will have lower VMTs and automobile ownership as
a more dense region is likely to have larger more densely populated urban areas (Cervero and
Murakami 2009). People living in a highly populated area may be more likely to use bus transit,
rail transit, or live in areas where walking is a quicker and more economical mode of
transportation. By looking at the regional density as well as neighborhood density of the samples
a better picture of the population within the entirety of the study area.

3.9: Dependent variables
This study explores whether neighborhood walkability affects vehicle miles traveled and
automobile ownership. Automobile ownership was calculated by taking 2000 census data and
using a variable for total number of vehicles available in a block group. This number is then
divided by the total number of households in the corresponding block group. This calculation
gives a ratio measure of the average number of automobiles owned within a block group by
household.
Vehicle miles traveled is a measurement used to determine the number of miles traveled
by people going to and from work, shopping, school, etc. It was derived from data taken from
the 2001 National Household Travel Survey (NHTS) conducted by the Federal Highway
Administration. Utilizing a model by Hu et al. (2007), VMTs at the Census tract level were

33

measured using household size, household income, and employment rate. In order to get the
VMTs for Census block groups, they were assigned the VMTs from their corresponding Census
Tract. This method was used by Eisman to calculate the VMTs as NHTS data is not available at
the block group level. The estimates for the VMTs are per household on an average weekday,
and the tract level estimates were given based on vehicles available to a household and the size
of the household (Eisman 2012).
3.10: Confounding Variables
Other variables in this study used from Eisman’s work are the confounding variables of
household income and age of neighborhood population. Confounding variables are variables
that may affect the outcome of the study and are not based on the primary tested variables.
However, there may be relationships within the built environment that may exist between these
the variables and the rest and there needs to be control for these unrelated variables in order to
see these relationships properly.
The income variable was measured as the average income of a block group. Differences
in income may affect vehicle ownership based on whether people can afford a single vehicle or
multiple vehicles. For example, lower income households may have no vehicle or a single
parent household might have lower income and also use only a single vehicle. In contrast, higher
income households could have multiple vehicles, for example, in cases where two parents are
working or there are households with multiple drivers.
3.11: Geoprocessing of variables
The data for this study was processed using Esri ArcGIS 10.1 desktop software. Where data
from Eisman’s work was used in this study, no additional processing was done with his data. On
the new independent variables developed for this study, the analysis was performed on data

34

acquired from several different sources: Google Earth, the USGS, the PRISM Climate Group and
the EPA. The data were geoprocessed to develop the measures described above before the
values were taken in and added to a spreadsheet for analysis.
The first step in the process was acquiring the bus stop data. To acquire bus stop data,
the original method searched for data from each county in which the sample block groups were
contained. This proved to be time consuming as well as fruitless, as many counties that may
have bus routes did not have the GIS data for them. To solve the problem, the Google Earth
software and its extensive spatial database was utilized. As Google has already aggregated
nationwide bus data displaying bus stops and routes as well as schedules, it was a great source
for this hard to acquire data. Additionally due to the street view feature it was possible to check
through samples to see if bus stops did in fact exist at the specified locations. A ¼ mile buffer
file was created around each sampled block group file and imported as a KML file into Google
Earth. Each sample was then checked to see if any bus stations exist within each buffer and
those that were, were selected and individually exported as KML files. These files were then
exported into shape files and merged with each other in their corresponding regions. Bus stops
were calculated by measuring total bus stops divided by the area of the sample and its ¼ buffer
and recorded.

35

Figure 5 Google Earth KML Data for Bus Stops in ArcMap for Conversion to Shapefile

36

Figure 6 Bus Stop Shapefile created from Google Earth KML Files

37

The elevation variable was acquired utilizing United States Geological Survey, Digital
Elevation Model (DEM) raster data at 1/3 arc (10 meter) resolution (USGS, 2014). To extract
the elevation values for each sample, a modified ¼ mile buffer was created. Since the data is a
walkability variable, water was deemed “not walkable” and needed to be removed from the
elevation raster. This was done by the use of a modified ¼ mile buffer. The buffer which was
used to locate bus stations was clipped with a North American Water Polygons layer found on
ArcGIS online and produced by Esri. The water layer was clipped with the ¼ mile buffer to
create a ¼ mile buffer with water removed. After the new ¼ mile buffer was created, the
“extract by mask tool” from the “Spatial Analyst Toolbox” within ArcMap was used to create
new a DEM raster for each sample, the result was 75 DEMs (i.e., one for each study area) with
water removed.
The digital elevation models are used in this study to measure slope. As mentioned
above, two different slope measurements were created for each census block group. One
measurement describes how hilly the entire block group land is (omitting only water) and the
other measure focuses just on the hilliness of the road network. In order to calculate slope for
the entire block group, the “slope” tool also within in the “Spatial Analyst Toolbox” was used
with the “percent rise” measurement. This calculation was performed on each raster cell within
each sample DEM for each sample area. Next, to calculate slope for just the road network, the
road network was described using shapefile data from US Census Tiger files. The road network
polylines were used with the “extract by mask” tool to create slope data for only roads. In both
cases the mean slope of the raster data was recorded for analysis.
The climate variables included 30 year normals for maximum temperature, minimum
temperature, mean temperature and precipitation and was acquired from the Prims Climate

38

Group (NACSE 2014). To process these 4 raster datasets, the normal ¼ mile buffer was used to
extract the raster data for each sample area from each dataset. Next the mean temperature or
precipitation for all the raster grids within each study area was recorded in the data table.
The last variable analyzed was for mixed land use. The data to use for these variables
was difficult to determine as to what would work best for the large scale of the study. Acquiring
individual land use data for counties or regions proved difficult and time consuming as many of
the study areas do not have their own GIS data available for this type of data.
Fortunately, there is a pre-existing data set that estimates the mix of land use at the level
of the census block group for the entire United States: the Smart Location Database (SLD) from
the United States Environmental Protection Agency (EPA 2013). This database was created
using data from the 2010 Census, American Community Survey, Longitudinal Employer-
Household Dynamics, InfoUSA, NAVTEQ, PAD-US, TOD Database, and GTFS (General
Transit Feed Specification). The specific variables within the Smart Location Database that
were chosen to best determine mixed land use were activity density, jobs per household, and jobs
within a 45 minute transit ride.
Activity density is measured by employment plus housing units and jobs per household is
measured by total employment divided by households (TotEmp/HH). Employment is measured
in the SLD by using the Longitudinal Employer-Household Dynamics (LEHD) which consists of
US Census LEHD Origin-Destination Employment Statistics (LODES) tables that summarizes
employment at the census block level for all 50 states and territories. Additionally within the
LODES there are Work Area Characteristics (WAC) tables which are used for employment
tabulations. Household units are calculated by block group utilizing population, housing or
employment within a block group and fall under a density category within the SLD. This data is

39

calculated using the US Census data. Jobs within a 45 minute transit ride was calculated based
on walk network travel time and GTFS schedules (Ramsey & Bell 2014).
One issue with using the Smart Location Database in this study is that all of the data are
based on the 2010 census and 2010 census tiger files while this study is based on 2000 census
data. This creates an important limitation in that the measured state of the study areas does not
match precisely in time. The assumption is that the land use mix did not change drastically in a
ten-year period in any of the study areas.
Another aspect of the mismatch of the land use data in time is that the block group data
does not overlap to the precise spatial boundaries of the polygons found in the 2000 census data
used in this study. When the SLD data was clipped with the sample block groups, it was found
that block groups from the samples overlaid across multiple block groups from the SLD. In
order to deal with this when determining values, the 2010 polygon sharing the largest area with
the sample polygon was selected. This is an acknowledged limitation of the data but in most
sample areas there are not be significant change in structures and there is fairly close overlap
between block group polygons as defined for the 2000 and for 2010 Census.
3.12: Statistical analysis
Statistical analysis for this study was performed utilizing SPSS for linear regression modeling.
In order to create linear regression models the data needed to be assembled into a table. This
was done by taking the values from the data in ArcMap and importing them into the Excel sheet
with one row for each sample. Once the data was in the Excel sheet, it was loaded into SPSS v
21 where analysis of the variables was done. Histograms and descriptive statistics were drawn
for each variable. The variables, excluding regional variables, were all non-normally distributed.
To make the variables fit better into the regression models based on “normal” distributions, the

40

natural log was taken for each of these variables. Next in order to determine if the variables had
a correlation between each other and especially the dependent variables, a bivariate correlation
table was created in SPSS.
Once the data was set up and analyzed, the multivariate linear regression models were
developed within SPSS. Variables were added and removed within the models to achieve the
best fit based on the significance of the relationship within the regression model. To determine
the strength of the relationships between the independent variables and the dependent variable in
the model, the adjusted R squared was used.

41

CHAPTER 4: RESULTS
This chapter reports the results from the data analysis, starting with bivariate correlations of each
independent variable with the two dependent variables. Next, the chapter reports the results of
exploration of the possible regional effects. Last, the chapter reports the results of linear
regression models for both vehicle miles travelled (VMTs) and average household automobile
ownership.
4.1 Bivariate Correlations
The first part of the analysis process involved creating a bivariate correlation table in order to
determine if there were any significant correlations between the dependent variables (VMTs and
automobile ownership) and the independent and confounding variables. The following table
(Table 4.1) shows the variables with significant correlations with the two dependent variables. A
full matrix of the bivariate correlation of each variable in the study was created. It was inspected
both for correlations with the dependent variable and for possible correlations among the
independent variables. High correlation among the independent variables is important to
consider in building regression models because it can make it difficult to compare and
understand the explanatory power of any individual predictor in the overall regression model,
i.e., the problem of “multicolinearity” (Allison 1999).

42

Table 2 Table of Significant Correlations
Vehicle Miles
Traveled
Automobile
Ownership
Density Pearson Correlation -.596** -.638**
Sig. (2-tailed) .000 .000
Distance to Job
Center
Pearson Correlation .501** .499**
Sig. (2-tailed) .000 .000
Distance to Retail
Center
Pearson Correlation .349** .353**
Sig. (2-tailed) .002 .002
Distance to
Transit
Pearson Correlation .489** .547**
Sig. (2-tailed) .000 .000
Bus Stops per
Square Mile
Pearson Correlation -.699** -.681**
Sig. (2-tailed) .000 .000
Mean Slope Pearson Correlation .275** .280*
Sig. (2-tailed) .017 .015
Minimum
Temperature
Pearson Correlation -.315** No Correlation
Sig. (2-tailed) .006
Mean
Temperature
Pearson Correlation -.261* No Correlation
Sig. (2-tailed) .024
Gross Activity
Density
Pearson Correlation -.702** -.739**
Sig. (2-tailed) .000 .000
Jobs Within a 45
Min Transit Ride
Pearson Correlation -.554** -.633**
Sig. (2-tailed) .000 .000
MSA Density Pearson Correlation No Correlation -.234*
Sig. (2-tailed) .043
Transit Spending Pearson Correlation .248* No Correlation
Sig. (2-tailed) .032
People Per Transit
Station
Pearson Correlation -.278* No Correlation
Sig. (2-tailed) .016
**. Correlation is significant at the 0.01 level (2-tailed).
*. Correlation is significant at the 0.05 level (2.tailed).

43

For ease of presentation, the following (Table 2) shows the variables with significant correlations
with the two dependent variables. From the correlation tables it was found that the dependent
variable for the average household VMTs in each census block group has many significant
correlations with both smart growth and walkability variables. Slightly more of the independent
variables have significant bivariate correlations with the VMTs than automobile ownership.
There are thirteen variables where VMTs show a correlation with most being significant at the
.01 level of significance. Automobile ownership had significant correlations with ten of the
testing variables. Most variables had a significant correlation with both dependent variables, and
as expected the two dependent variables are themselves correlated with a Pearson’s R of (-.875*
at the .000 level (i.e., households with more vehicles also drive more).
It is an interesting note that of the few variables with correlations with only one or more
dependent variables, most are measured at the regional level. For example, MSA density, as
measured at the regional level does not correlate with VMTs but does correlate with automobile
ownership. The four variables significant with only VMTs are transit spending, people per
transit station, minimum temperature, and mean temperature. Transit spending and people per
transit station are explicitly measured at the regional level. Also, the variations in climate among
the study regions mean the temperature variables are implicitly regional, though each census
block group has a slightly different measure of temperature due to microclimates within regions.
In looking at bivariate correlations with regional variables, an important caveat to bear in mind is
that there are really just 5 independent measures of each variable, even though the total number
of cases is 75.
VMTs had the strongest significant correlation with gross activity density (Pearson’s R=
-.702**), bus stops per square mile (Pearson’s R = -.699**), Density (Pearson’s R = -.596**)

44

and jobs within a 45 min transit ride (Pearson’s = -.554**). Gross activity density is the
measurement for mixed land use or land use diversity (shops mixed with residences etc.). The
significant correlation shows that as the land use becomes more mixed (i.e., economically
diverse) the number of VMTs goes down. This supports the original hypothesis for the variable.
The significant correlation results for the other three variables show the same inverse
relationship. As bus stops increase, VMTs go down and the same goes for jobs within a 45
minute transit ride. These results also support the hypothesis for the use of the variables within
the study and are promising for use within the linear regressing model.
Automobile Ownership had the strongest significant correlation with gross activity
density (Pearson’s R = -.739**), bus stops per square mile (Pearson’s R = -.681**), Density
(Pearson’s R = -.638**), and jobs within a 45 min transit ride (Pearson’s R = -.633**). Like
VMTs, these variables show an inverse relationship. It should be noted that the top four
strongest significant correlations for both dependent variables are all the same variables in the
same order. The reason for this is unclear but shows that these four variables are strong
indicators both for VMTs and automobile ownership, again pointing to significant overlap in
what these two variables measure.
4.2 Regional Variables
In addition to the variables at the individual census block group level, there are three regional
variables. Regional variables only consist of five values, one value per region, instead of the
usual fifteen independent values per region for each sample (i.e., one value per census block
group). Because of the nature of the regional variables there was further evaluation of the
variables to explore their significance. The mean values by region for each dependent variable
were calculated and difference of means testing was performed on them. Table 3 shows the

45

means grouped by their statistically significant differences. Scatterplots were also drawn of the
regional variables with each of the dependent variables to inspect for discernible trends that
might point to relationships.
Table 3 VMT and Automobile Ownership Mean Values
MSA Mean VMTs MSA Mean Automobile Ownership
Washington DC 59.62219998 San Diego 1.851924968
San Diego 52.3365337 Washington DC 1.801894624
Portland 50.83582958 Portland 1.744232531
Chicago 49.58853866 Chicago 1.507537922
Miami 36.88243465 Miami 1.49282606

Mean VMTs differ greatest between Washington DC and Miami with DC having the highest
average vehicle miles traveled in its neighborhoods at 59.6 VMTs and Miami having the least at
36.8 VMTs. Automobile ownership has little difference between San Diego, Washington DC
and Portland who all are at or close to 1.8 vehicles per household, while Chicago and Miami
have the least with both at or near 1.5 vehicles per household.
A scatter plot was drawn for each of the three regional variables with each dependent
variable. The regional variables were transit spending, MSA density, and people per transit
stations. The purpose of creating scatter plots for these variables is to perform visual inspection
to identify potential relationships between the regional and dependent variables. To read a
scatterplot, a line is drawn through the clusters of points on the individual columns. Depending
on the angle of the line, a relationship between the two variables can be determined. For
example in Figure 7, VMTs tend to drop as MSA density increases. This shows there is a
possible relationship between VMTs and MSA density even though San Diego is fairly spread
along its column there is still a decreasing trend. Additionally this scatterplot shows Chicago
and Portland having similar VMTs, this can be easily explained in that Portland has an active

46

smart growth development plan and promotes mixed land use and walkability (Jun, 2008).
Automobile ownership and transit expenditure shows no relationship (Figure 8), the clusters
along the columns are relatively close and have a minimal rise and drop as transit expenditure
goes up. This rise and fall shows no constant trend and thus indicates there may be no
relationship between the variables. After creating the scatterplots for the 3 regional variables, it
was concluded that VMTs had potential relationship with MSA density and people per transit
station. VMTs had no relationship with transit expenditure. Automobile ownership had a
potential relationship with MSA density and no relationship with people per transit station or
transit expenditure.

47

Figure 7 Scatterplot of VMTs and MSA Density

48

Figure 8 Scatterplot of Automobile Ownership and Transit Expenditure

49

4.3 Interaction Testing of Regional Variables
In order to test the regional model’s effect within the regression model, interactions were
performed between the regional variable MSA density and the independent variables found in
the best fit regression model. An interaction is performed by multiplying the regional variable
with the neighborhood variable, this gives a new variable to be included in the regression model
along with the variables that were used to create the interaction. By doing this, the study
attempts to find strong relationships with the dependent variable.
MSA density was chosen for the interaction due to its possible relationship with both
dependent variables. The independent variables chosen for the interaction were bus stops per
square mile and gross activity density (mixed land use) for the VMT regression model and jobs
within a 45 min transit ride and gross activity density for the automobile ownership regression
model. These variables were chosen for the interactions because they were most likely to be
affected by MSA density.
The interaction between MSA density and the independent variables showed no
relationship and did not add to the regression model. When the interactions were included within
the linear regression model (one at a time) the interaction with bus stops per square mile had a
significance of .455 which shows no significance in the model, and gross activity density had a
significance of .396, also showing no significance. The interactions within the automobile
ownership best fit regression model were more interesting, as the interaction between jobs within
45 minute transit ride was just significant at .052 but within the model made the jobs within 45
minute transit ride variable not significant by raising it to a significance of .588. This makes the
interaction void as the variable is eliminated from the model. Gross activity density was also not
significant at .215.

50

In addition to performing interactions with the MSA density regional variable, a dummy
variable was created for regions based on the means tests for VMTs and automobile ownership.
The variable was created by assigning a single number to all the block groups in an MSA based
on the MSA’s ranking from highest to lowest mean for VMTs and then for automobile
ownership. The dummy variable was then entered into the best-fit models to test for regional
variation relationship with the independent or dependent variables.
Both the interactions with MSA density and the regional dummy variable showed no
impact or significant relationship. Because of this, the null hypothesis for regional variation as a
predictor within the models was accepted.
4.4 Linear Regression Modeling
Best-fit regression models were created for both VMTs and automobile ownership taking into
account the new independent variables developed for this study. The adjusted R-squared was
used to compare the predictive power of the different models as this is a common method of
determining the strength of relationships in linear regression models with relatively small
numbers of observations (Allison 1999).
4.5 Vehicle Miles Traveled Linear Regression Model
The best-fit linear regression model for VMTs utilized bus stops per square mile, distance to
retail centers, gross activity density and income (table 4.6). With the exception of the
confounding variable of income all other variables used the natural log in the model. The
adjusted R-squared for the model was .598 indicating a near 60% explanation for variation in
VMTs from the four variables. Within this model, income is an important control variable:
wealthier neighborhoods drive more. However, even when controlling for income the other

51

variables are all significant predictors for VMTs. In other words, on average households in
wealthy neighborhoods with more bus stops, longer distance to major retail centers, and mixed
land use drive less than wealthy households in neighborhoods without these features.
Table 4: Vehicle Miles Traveled Best-fit Regression Model
Vehicle Miles Traveled

Adjusted R-Squared 0.598
Model Intercept Significance
Constant 11.842 .000
Bus Stops Per Square Mile -2.684 .009
Distance to Retail Center -2.08 .041
Gross Activity Density -4.603 .000
Income 2.579 .012

Eisman’s (2012) best-fit model had an adjusted R-squared of .404 with population
density by census block group and transit network (distance to rail stops) being the predictive
variables. It was found when running models with his original best-fit as the starting point that,
when using the new variables developed for this study, the distance to rail and population density
variables dropped out and another of the variables Eisman (2012) developed (distance to retail
center) became significant in the model. Bus stops per square mile superseded distance to rail
and in fact on its own, bus stops per square mile had an adjusted R-squared of .482 making it a
very strong predictor of VMTs. Population density which is one of the top four in the bivariate
correlation is superseded within the regression model by bus stops per square mile, gross activity
density, and distance to retail. It is interesting to see that distance to retail is significant with
these new variables from this study. The intercept for distance to retail has also changed from
the bivariate correlation meaning that even as distance to retail goes down, VMTs still go up.

52

This could be from the nature of shopping trips to retail centers, people may just be buying more
or larger items than they can walk with or carry on public transit.
An important caveat to mention here is that three independent variables had a strong
correlation with each other. Table 5 shows this relationship between the three variables. These
levels of correlation between the predictor variables may indicate issues with multicolinearity,
making it difficult to specify the precise strength of explanation from each variable. Income had
no bivariate correlation with either these variables or the dependent variables.
Table 5 Correlation between the 3 independent variables in the best-fit regression model

Bus Stops per
Square Mile
Gross Activity
Density
Distance to
Retail
Bus Stops per Square
Mile
Pearson Correlation 1 .770* -.525**
Sig. (2 tailed .000 .000
Gross Activity Density
Pearson Correlation .770* 1 -.675**
Sig. (2 tailed .000 .000
Distance to Retail
Pearson Correlation -.525** -.675** 1
Sig. (2 tailed .000 .000

There were other variables which showed promise when creating the best-fit models.
The top four of the bivariate correlation table were each significant within the model but were
superseded by other variables. The variable for jobs within a 45 min transit ride was significant
within the model but was superseded by distance to retail centers. This is interesting as it plays a
role in the model for automobile ownership. Again, the caveat with regards to multicolinearity
applies because both of these variables have a strong correlation between the dependent variables
as well as with each other (Pearson’s R= -.304** at the .008 level) and with other variables
within the best-fit model.

53

4.6 Automobile Ownership Linear Regression Model
The best-fit linear regression model for automobile ownership utilized jobs within a 45 min
transit ride, gross activity density, and income. The adjusted R-squared was .641 indicating a
64% explanation of automobile ownership. Like the best-fit for VMTs, income is again a control
variable.
Table 6: Automobile Ownership Best-fit Linear Regression Model
Automobile Ownership

Adjusted R-Squared 0.641

Model Intercept Significance
Constant 23.592 .000
Jobs Within a 45 min Transit
Ride
-3.887 .000
Gross Activity Density -0.612 .000
Income 3.102 .003

Eisman’s best-fit model had an adjusted R-squared of .445 with density, distance to
transit, and income being the predictive variables. Again when running the linear regression
models his model was a starting point. Density which shows up in both of this models was
superseded by gross activity density (mixed land use) and jobs within a 45 min transit ride
superseded distance to transit. Interestingly income remained as a control variable within the
model.
Most of the variables had little significance on automobile ownership including bus stops
per square mile which was so strong for VMTs. With the exception of income which had no
significant correlation with either the dependent or independent variables, again, like the VMTs
best-fit model, the independent variables had a significant relation to each other.

54

4.7 Conclusion
This analysis has looked at both smart growth variables from a previous study and new variables
intended to measure aspects of walkability. These new variables show great promise with
determining a smart growth and walkability design plan for urban and suburban development.
While not all variables show promise there are some stand out predictor variables: bus stops per
square mile, gross activity density, and jobs within a 45 min transit ride.
Bus stops per square mile is a strong indicator of a walkable environment and the linear
regression model appears to confirm the hypothesis that a greater number of bus stops, and one
might thus hypothesize a larger bus network overall, promotes less driving and allows people to
conduct their business without driving or spending money on gas. Gross activity density is not
only a walkability indicator but also a smart growth indicator. The models show that a more
diverse land use promotes fewer VMTs as well as reduces the need to own and vehicle. Jobs
within a 45 min transit ride appears to support the hypothesis that nearby transit can reduce the
need for owning a vehicle or at least multiple vehicles, though it does not add to the explanation
in the best-fitting multivariate regression model for predicting VMTs.

55

CHAPTER 5: DISCUSSION AND CONCLUSION
The evidence from the analysis of the variables in this study supports the hypothesis that smart
growth and walkability elements within the built environment promote lower VMTs and lower
automobile ownership.
The findings of this study support the findings from the other smart growth studies,
though in some cases in different ways as some of the studies used different variables and
methods to conduct their research. Cervero and Murakami (2009) found that population density
was a factor in reducing VMTs, while other variables such as transportation did not have as
strong of an effect on VMTs as population density. This study found that the number of bus
stops and land use have a greater effect on VMTs than population density. This study also tested
interaction between neighborhood and regional scales more thoroughly than Cervero and
Murakami (2009).
Cervero and Duncan (2006) used a different method and variable types for their study.
Instead of testing individual land use and census variables, they tested two different urban
development types/concepts: jobs-housing balance and retail-housing mixing (mixed land use).
They found that jobs-housing balance had a greater effect than mixed land use. That is, living
closer to the place of employment through assistance of grant programs is more effective at
reducing VMTs than mixed land use of housing and retail. This is a different outcome from this
study which found that mixed land use (gross activity density) did have a significant effect on
VMTs. I would argue however, that the very strong significant association of bus stops per
square mile with reduced VMTs supports the findings with jobs-housing balance. If a person
lives close to their job and there are nearby bus systems, this would allow people to commute
without using a personal vehicle.

56

Kim and Brownstone found that land use and population density had statistically
significant effects on VMTs with the emphasis of their study being on population and socio-
demographic variables. This study found that population density while significant, is less
significant than other variables, bus stops and land use being the most significant.
This is a similar result found by Eisman (2012). Eisman’s study found that VMTs were
significantly associated with the neighborhood’s population density and extent of region’s transit
(rail) network. As density and transit network increased, VMTs decreased. Arguably the
association is significant and helps prove the hypothesis, however, regional transit and density do
not give a clear picture of the built environment as regional transit is a regional variable that does
not study the neighborhood level and thus does not give very thorough explanation of the built
environment, additionally rail is not the only mode of mass transit and runs to fixed locations
that are generally less flexible and available than bus networks.
This study uses Eisman’s (2012) original work and makes an effort to expand on his idea
while utilizing walkability aspects to expand the regional smart growth study. The results from
this study supports Eisman’s findings while giving new explanations for reduced VMTs and
automobile ownership. The best-fit regression model for VMTs from this study showed a
significant association between the built environment and VMTs. Specifically, bus stops per
square mile, distance to retail centers, gross activity density (mixed land use), and income.
While two of the indicators (distance to retail centers and income) are from Eisman’s original
study, the other two important variables are new and come from the determined aspects of
walkability variable group. In addition to having a greater R-squared than Eisman’s best-fit
model, the variables are more descriptive in their assessment of the built environment and its
effect on VMTs. The analysis shows that as the number of bus stops increases and land uses

57

become more mixed, VMTs decrease which supports the hypothesis. Curiously the model also
shows that as distance to retail centers increase, VMTs decrease, that is counter to the intial
hypothesis. It is assumed that the closer to a retail center a person lives, the less they would
drive. Income, while not a land use measurement it is a strong control variable that can also help
explain people’s vehicle usage. As income goes down, VMTs go down and when income goes
up VMTs go up. Essentially VMTs go up for people who have the money to buy vehicles and
afford to use them. This of course isn’t the only explanation for reduced VMTs, but helps
explain them in conjunction with the other variables.
Within the best-fit linear regression model for automobile ownership, Eisman found that
there was a significant association with neighborhood population density, distance to transit, and
income. His findings support the hypothesis that smart growth has an effect on lower automobile
ownership. The same holds true for this study as well, and it is interesting in that the variables
within Eisman’s best-fit regression model are similar to those within the best-fit regression
model for automobile ownership within this study. While density did not play a role in the
model for this study, income and jobs within a 45 minute transit ride did. While both distance to
transit (rail) and jobs within 45 min transit (all modes) ride do not have exactly the same
measure, they are both transit variables that measure similar things. The fact that both variables
show up within the two different models suggests that availability of public transit is an
important factor in determining how many cars, if any, are owned by individual households.
With many two income households, this finding makes sense because with availability of transit,
it is perhaps possible for at least one person to get to work without a car.
Income is again an important control variable in this study and in Eisman’s. Income
directly effects whether people can afford to own and operate a vehicle. Gross activity density

58

(mixed land use) is another variable that has a statistically significant association within both
best-fit regression models. This means that areas with a more diverse land use and suitable
access to transit either have less of a need to own a vehicle or more than one.
Of the new variables added to this study, a few stood out as highly significant: bus stops
per square mile, jobs within 45 minute transit ride, and gross activity density were all highly
significant during the development of the models. Bus stops per square mile was associated
significantly with VMTs and had influence on all other variables within the models. This
suggests that access to bus stops and in turn bus routes is important when people make decisions
about driving.
Jobs within 45 minute transit ride was another significant variable that in early models for
VMTs was significant. However when put into a model with bus stops it lost its significance. It
was however stronger in models regarding automobile ownership and held its significance in
most models it was incorporated into.
Lastly, gross activity density was the third variable that was very significant within the
models. The mixed land use variable had a significant association with both dependent variables
and ended up in the best-fit regression models for both. While bus stops per square mile was
extremely significant within the VMT regression model, the importance of mixed land use
cannot be ignored. The idea of mixed land use is one of the aspects of smart growth. To have it
be a significant part of both best-fit models supports the idea of smart growth as well as
supporting the hypothesis within this study.

59

5.1 Assumptions and Limitations
Within this study there were some data issues and limitations. As stated earlier, the data from
the Smart Location Database were based on the 2010 census because the SLD does not have
2000 Census data available. This is an acknowledged limitation and could skew the data.
However, major development generally has not occurred within the selected study are areas in
the intervening decade. This was checked utilizing historical aerial imagery through Google
Earth, showing that significant new structures or road network were non-existent or rare in the
study areas. Additionally, where 2010 Census block groups have changed since the 2000 Census
data, this was corrected for, by taking the values from the 2010 block group that was the majority
within the 2000 Census sample block group.
The second limitation and assumption on data was the bus stops. While it can be
confirmed that bus stops exist in 2014, there was no way of confirming their existence in 2000.
Attempts were made but the sheer number of bus lines and operators as well as jurisdictions
involved made confirming installation periods difficult if not impossible to do in any reasonable
amount of time. In future studies of smart growth, the availability of bus stop data will be
critical because this variable is very strong and an important aspect of a walkability and smart
growth study. In future studies, data on the dependent variables may also be updated,
eliminating the concern with projecting bus stops back to the year 2000.
Sorting out data between 2000 and 2010 data proved to be a difficult. This makes
conducting a longitudinal study with this array of variables difficult and extremely time
consuming. This is in part because of the number of different variables, the difference in
variables, the number of jurisdictions involved in a multiple region study and that many counties
and cities within these regions don’t maintain their own GIS data or have GIS systems. Bus

60

stops, for example, were extremely difficult to find until the Google Earth data was adopted. In
order to perform a longitudinal study at these study scales, national datasets need to be used in
order to obtain all the data. Unfortunately those types of datasets are hard to come by, especially
a dataset with both historical and current values.
The last of the limitations within this study, and I would argue Eisman’s as well, is the
nature of the regional variables, mainly the extent of region transit network and people per transit
stations. These measurements may not be as comprehensive as they should be and may need to
be looked at in the future. It may be that there are important regional effects or important
interactions between regional and neighborhood variables. Indeed, the difference of means
results for the regions suggest this. However, the regional variables explored here do not offer
good explanations, it may be that the regional variables need more robust measurement than was
developed for this study.
5.2 Future Research
This method of comparing smart growth and walkability at the neighborhood level amongst
regions is a good way to identify the importance of smart growth as well as identifying regions
that implement it. However, there is some variable identification and measurement that need to
be refined as well.
One of the variables that was originally assumed to be important in determining
walkability was climate. For this study, the 30-year normal measurements for minimum,
maximum and mean temperature were used assuming that this would be a suitable method of
measurement. This data had no real affect within the regression models. There may be however,
a better method for measuring climate within multiple regions. While the climate data raster
files were clipped down to the neighborhood level for this study, it may make more sense to turn

61

it into a regional variable. Whether or not that would have any significant impact within this
study is unknown and something to be tested in the future. There may also be more robust ways
of indexing the climate variables to describe the real effects of climate on pedestrians.
As stated within the limitations section, regional variables need a revamping in how they
are measured. There may need to be considerable thought on the matter to determine the best
ways to measure regional data and which variables make the most sense to be measured
regionally. Transit network data may need further development to test interaction between
regional and neighborhood scales. The area of network analysis may provide possible, more
complex methods at measuring the impact of transit.
Another aspect that should be looked at is the counter to the hypothesis that distance to
retail centers has with VMTs. That is, as VMTs decrease, distance to retail centers increase.
There could be some simple explanations to this such as that people generally use their vehicle
when shopping in order to haul their purchases. Large retail centers with big box stores tend to
promote large or bulk purposes which would require the use of a vehicle to haul these purchases
back to a person’s home. If this is the case, then that would show that there is no really way that
urban design around large retail centers would have any effect on reducing VMTs.
Lastly bus stops per square mile may be an even stronger determinant of walkability and
smart growth if other bus data types are acquired. Bus route length with and number of stops per
line may give a better description about how useful a bus network really is in the built
environment. There was an attempt in the study to locate this data but no reliable source could
be found to consolidate bus network data.

62

REFERENCES
Allison, Paul D. 1999. Multiple Regression: A Primer. Thousand Oaks, California: Pine Forge
Press.
Chen, Stephanie. 2012. “Investigating Bus Route Walkability: Comparative Case
Study in Orange County, California.” Master’s thesis, University of Southern California.
Accessed March 3, 2013
http://digitallibrary.usc.edu/cdm/ref/collection/p15799coll3/id/98075
CEPA (California Environmental Protection Agency). 2014. “Sustainable Communities.”
Accessed November 19, 2014. http://www.arb.ca.gov/cc/sb375/sb375.htm
Cervero, Robert and Michael Duncan. 2006. “Which reduces vehicle travel more: Jobs-housing
balance or retail-housing mixing?” Journal of the American Planning Association. 72:
475–490.
Cervero, Robert and Jin Murakami. 2010. “Effects of built environments on vehicle
miles traveled: Evidence from 370 urbanized areas.” Environment and Planning.
42: 400-418.
Eisman, Daniel Currie. August 2012. “Spatial Analysis of Urban Built Environments and
Vehicle Transit Behavior”. Master’s thesis, University of Southern California.
Accessed February 1, 2013
http://spatial.usc.edu/wp-content/uploads/2014/04/EismanDanielThesis.pdf

63

EPA (Environmental Protection Agency). 1999. “Smog-Who Does it Hurt? What You Need to
Know About Ozone and Your Health” Accessed February 17, 2014.
http://www.epa.gov/airnow/health/smog.pdf
———, 2013. “Smart Location Database”. Accessed January 15, 2014
http://www.epa.gov/smartgrowth/smartlocationdatabase.htm
Google Earth. Bus stop KML data. Google Inc. Accessed January 10, 2014.
Horeck, Tanya M. et al. 2012. “Sneakers and Spokes: An Assessment of the Walkability and
Bikeability of U.S. Postsecondary Institutions.” Journal of Environmental Health. 74: 8-
15.
Jun, Myung-Jin. 2008. “Are Portland’s Smart Growth Policies Related to Reduced Automobile
Dependence?” Journal of Planning Education and Research. 28: 100-107.
Kim, Jinwon and David Brownstone. 2010. “The impact of residential density on vehicle
usage and fuel consumption.” University of California: Department of
Economics. http://escholarship.org/uc/item/31m0w2x3.
Mantri, Anupama. 2008. “A GIS Based Approach to Measure Walkability of a Neighborhood.”
Master Thesis, University of Cincinnati. Accessed May 20, February 25, 2013
https://etd.ohiolink.edu/ap/10?0::NO:10:P10_ACCESSION_NUM:ucin1205012111
M.K. Gajendra Babu and K.A. Subramanian. 2013. Alternative Transportation Fuels:
Utilization in Combustion Engines. Boca Raton, Florida: CRC Press
Prism Climate Group. 2014. Oregon State University. Climate raster data. Accessed October 18,
2014 http://www.prism.oregonstate.edu/normals/

64

Ramsey, Kevin and Alexander Bell. 2014. Smart Location Database: Version 20. User Guide.
Accessed January 15, 2014
http://www.epa.gov/smartgrowth/pdf/sld_userguide.pdf
Reuters. 2010. “Chinese Divers sit it out in 9-day traffic jam.” Accessed October 12, 2014.
http://in.reuters.com/article/2010/08/23/idINIndia-51010420100823
Reuters. 2011. “Los Angeles braces for weekend of ‘Carmageddon.’” Accessed October 12,
2014. http://www.reuters.com/article/2011/07/14/us-carmageddon-losangeles-
idUSTRE76D2D720110714
Spoon, Steven Chadwick. 2005. “What Defines Walkability: Walking Behavior Correlates.”
Masters Project, University of North Carolina at Chapel Hill. Accessed November 15, 2013
https://cdr.lib.unc.edu/indexablecontent/uuid:384e8784-d3aa-4d8e-b15c-0f01083f5c16
Stevens, Robert David. 2005. “Walkability Around Neighborhood Parks: An Assessment of Four
Parks in Springfield, Oregon.” Master’s Thesis, Graduate School of the University of
Oregon. Accessed September 3, 2014
https://scholarsbank.uoregon.edu/xmlui/bitstream/handle/1794/1288/stevens_thes
is.pdf?sequence=1
U.S. Bureau of the Census. 2014. Metropolitan and Micropolitan Statistical Areas Main.
Accessed November 20, 2014
http://www.census.gov/population/metro/
USGS (United States Geologic Survey). 2014. National Elevation Dataset. Accessed March 8,
2014. http://ned.usgs.gov/
Villanueva, Karen et al. 2013. “People living in hilly residential areas in metropolitan Perth have
less diabetes: spurious association or important environmental determinant?”
International Journal of Health Geographics 12: 59.

65

Walk Score. 2014. Live Where You Love. Accessed November 23, 2014. www.walkscore.com

66

Appendix A
Figure 9 Map of MSA Locations

67

Figure 10 Chicago MSA and Samples

68

Figure 11 Miami MSA and Samples

69

Figure 12 Portland MSA and Samples

70

Figure 13 San Diego MSA and Samples

71

Figure 14 Washington D.C. MSA and Samples

Abstract (if available)

Abstract This study tests the effects of the built environment on vehicle miles traveled (VMTs) and automobile ownership, with specific reference to aspects of neighborhood walkability studies and research design at nested spatial scales of metropolitan regions and neighborhoods. This adds to existing smart growth studies as they tend to focus on Census data and non-walkability land use variables such as rail transit infrastructure. This study looks at 75 census block group samples within 5 metropolitan statistical areas (MSAs). The variables measured for the study include, bus stops per square mile, jobs within 45 minute transit ride, gross activity density, temperature, distance to retail, distance to transit, and slope among others. This study also looks at including regionally measured variables such as people per transit station, MSA density, and transit expenditure in conjunction with neighborhood scaled variables in order to test if there are any interactions between neighborhood and regional variables. The variables are entered into a multivariate regression model to find the best-fit model in order to explain the relationships between the dependent and independent variables. The study finds that the new walkability variables added to the research add significantly to the explanatory value of regression models beyond studies that use just smart growth land use variables. The implications for this study are that there is ground work laid for a new type of smart growth and walkability joint study at a multiple region level.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Spatial analysis of urban built environments and vehicle transit behavior

PDF

Walkability study for school accessibility: case study of the San Juan, Puerto Rico elementary schools

PDF

Investigating bus route walkability: comparative case study in Orange County, California

PDF

Site location suitability analysis for a smart grid network

PDF

Residential housing code violation prediction: a study in Victorville, CA using geographically weighted logistic regression

PDF

Calculating solar photovoltaic potential on residential rooftops in Kailua Kona, Hawaii

PDF

Finding the green in greenspace: an examination of geospatial measures of greenspace for use in exposure studies

PDF

A Maxent-based model for identifying local-scale tree species richness patch boundaries in the Lake Tahoe Basin of California and Nevada

PDF

A site suitability analysis for an inland port to service the ports of Los Angeles and Long Beach

PDF

Network accessibility and population change: historical analysis of transportation in Tennessee, 1830-2010

PDF

Preparing for immigration reform: a spatial analysis of unauthorized immigrants

PDF

Historical temperature trends in Los Angeles County, California

PDF

Validating the HAZUS coastal surge model for Superstorm Sandy

PDF

Using volunteered geographic information to model blue whale foraging habitat, Southern California Bight

PDF

Obesity and healthy food accessibility: case study of Minnesota, USA

PDF

Using census data, urban-land cover classification, and dasymetric mapping to measure urban growth of the lower Rio Grande Valley, Texas

PDF

Using pattern oriented modeling to design and validate spatial models: a case study in agent-based modeling

PDF

The role of amenities in measuring park accessibility: a case study of Downey, California

PDF

Developing and implementing a GIS-based framework to identify optimal locations for clean water wells in sub-Saharan Africa

PDF

Assessing the transferability of a species distribution model for predicting the distribution of invasive cogongrass in Alabama

Asset Metadata

Creator Newland, Derek Richard (author)

Core Title Smart growth and walkability affect on vehicle use and ownership

School College of Letters, Arts and Sciences

Degree Master of Science

Degree Program Geographic Information Science and Technology

Publication Date 02/09/2015

Defense Date 01/12/2015

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag automobile ownership,OAI-PMH Harvest,smart growth,VMTs,walkability

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Vos, Robert O. (committee chair), Franklin, Meredith (committee member), Lee, Su Jin (committee member)

Creator Email dknewland@gmail.com,dnewland@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-529945

Unique identifier UC11297716

Identifier etd-NewlandDer-3166.pdf (filename),usctheses-c3-529945 (legacy record id)

Legacy Identifier etd-NewlandDer-3166-0.pdf

Dmrecord 529945

Document Type Thesis

Format application/pdf (imt)

Rights Newland, Derek Richard

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA