University of Southern California Dissertations and Theses

Spatial analysis of urban built environments and vehicle transit behavior

(USC Thesis Other)

Spatial analysis of urban built environments and vehicle transit behavior

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content SPATIAL ANALYSIS OF URBAN BUILT ENVIRONMENTS

AND VEHICLE TRANSIT BEHAVIOR

by

Daniel Currie Eisman

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)

August 2012

Copyright 2012 Daniel Currie Eisman
ii

Acknowledgments
I would like to thank my Committee Chair, Dr. Robert Vos, whose feedback and
support throughout this process have been invaluable. I would also like to thank my
other committee members, Dr. Katsuhiko Oda and Dr. Daniel Warshawsky, for their time
and effort in assisting me. I am also grateful to Dr. Meredith Franklin for her help with
the processing and interpretation of data.
iii

Table of Contents
Acknowledgments ii

List of Tables v

List of Figures vi

Abstract vii

Chapter 1: Introduction 1
1.1 Motivation 2

Chapter 2: Background 5
2.1 Single Region Studies 6
2.2 Longitudinal Approaches 7
2.3 Studies at Multi-Region and Neighborhood Scales 8

Chapter 3: Methodology 12
3.1 Sampling Framework 12
3.2 Major Hypotheses and Dependent Variables 19
3.3 Measuring the Built Environment: Independent Variables 21
3.4 Confounding Variables 24
3.5 Spatial Modeling of Variables 26
3.6 Statistical Analysis 30

Chapter 4: Results 31
4.1 Analysis of Study Variables 31
4.2 Assessing Correlation of Study Variables 35
4.3 Regression Models 36
4.4 Spatial Autocorrelation 40
4.5 Modifiable Areal Unit Problem 43
iv

Chapter 5: Discussion and Conclusion 46
5.1 Future Research 48

Bibliography 52

v

List of Tables
Table 1: Dependent Variables 19

Table 2: Independent Variables 21

Table 3: Confounding Variables 25

Table 4: Distribution of Variables 34

Table 5: Breakdown of Spatial Relationship between Job and Retail Centers 35

Table 6: Best-Fit Regression Model for VMTs 37

Table 7: Best-Fit Regression Model for Automobile Ownership 38
vi

List of Figures
Figure 1: Sample Metropolitan Statistical Areas (MSAs) 13

Figure 2: Chicago, IL MSA 15

Figure 3: Miami, FL MSA 16

Figure 4: Portland, OR MSA 17

Figure 5: San Diego, CA MSA 17

Figure 6: Washington, DC MSA 18

Figure 7: Average Household VMTs 31

Figure 8: Average Automobile Ownership 33

Figure 9: Largest, Least Dense Block Group in San Diego MSA 44

vii

Abstract
In an effort to explore smart growth principles, this study offers an empirical test
of the influence of the built environment at the neighborhood scale on vehicle transit
behavior. Using U.S. Census data combined with spatial analysis techniques, the study
conducts a cross-sectional analysis of the effect of the built environment on household
automobile ownership and vehicles miles traveled (VMTs) in 75 block groups across five
metropolitan statistical areas. Variables are measured for density, job and retail access,
transit accessibility, and street connectivity. The study also considers confounding
variables including household income, regional density, extent of regional transit
network, age of neighborhood population, and individual transit expenditure. From these
data, best-fit regression models are developed for VMTs and automobile ownership.
Although there is significant unexplained variation, the regression models confirm a
statistically significant association of VMTs and automobile ownership with the built
environment. Among the implications of these findings are that (1) neighborhood density
should be encouraged in areas well-served by transit, (2) transit and smart-growth
projects will have a greater impact on VMTs in regions that have robust, existing transit
systems, and (3) new transit projects will likely be most effective in reducing vehicle
ownership if planners focus on better serving moderate and low-income neighborhoods.
Future research should examine statistical associations longitudinally, based on updated
data from the 2010 U.S. Census, and should attempt to gather primary data on VMTs at
the household and neighborhood scales.
1

Chapter 1: Introduction
This study assesses the effect of the built environment on transit behavior.
Specifically, the focus is on travel behavior in terms of vehicle transit by measuring
vehicle miles traveled (VMTs) and automobile ownership. While there are other aspects
to transit behavior, such as transit ridership, biking, and walking, this study focuses on
VMTs and automobile ownership because these variables best capture vehicle behavior.
The purpose of this study is to put the transit-oriented development principles of
smart growth to a robust empirical test by exploring the relationship between the built
environment and VMTs and automobile ownership. If built environments have an
influence on transportation behavior, it is expected that people who live in neighborhoods
with built environments that include features of transit-oriented development will drive
less than those who live in neighborhoods more characteristic of sprawl morphologies.
This is accomplished by assessing a set of built environment factors at the
neighborhood and regional scales. The following independent variables are measured in
this analysis: neighborhood density, job access, retail access, transit access, and street
connectivity. The confounding variables measured include household income, regional
density, extent of regional transit network, age of neighborhood population, and
individual transit expenditure.
Using a cross-sectional study design, this research seeks to find the spatial
relationships and variables that drive automobile ownership levels and VMTs at the
neighborhood and regional scales. Variables are measured for 75 sample block groups
across five metropolitan statistical areas (MSAs)—Chicago, IL; Miami, FL; Portland,
2

OR; San Diego, CA; and Washington, DC. The findings confirm a significant
association with some smart growth principles and lower automobile ownership and
VMTs.
1.1 Motivation
The sprawl style suburban development that became commonplace in the post-
World War II years has been blamed for several negative societal and economic impacts.
Traffic gridlock has been linked to the proliferation of far-flung, sprawling automobile-
dependent suburbs (Gordon & Richardson, 1997). Road networks lack adequate capacity
to keep up with the constantly growing demand. Even if government agencies had
adequate funding to build new roads, many see new road construction as an inefficient
and unsustainable way of meeting the transportation demands of the public. Gridlock has
both societal and economic costs in that people are spending more and more time and
money getting from one place to another (Ingram, Carbonell, Hong, & Flint, 2009).
Automobile usage has also been connected with climate change and rising energy
costs. Fuel prices have skyrocketed since 2002 and as a result so have household
transportation costs. Many people in the U.S. cannot avoid these additional costs,
because they do not have adequate alternatives to driving, such as transit.
It is important to note that household use of personal vehicle transit is not the only
undesirable aspect of sprawl development. Critics of sprawl also point to the lack of
affordable housing and housing options for moderate and low-income households, the
destruction of farmland and environmentally sensitive areas, and the negative effects on
air and water quality (Ingram et al., 2009). Still, over-reliance on driving is thought to be
3

a major negative influence of sprawl morphologies. Smart growth advocates have
advanced the notion that by changing the built environment to allow for greater density
and by creating transportation alternatives and better transit access, people will drive
substantially less (Ewing & Cervero, 2001).
Smart growth policies have now been in effect for decades in some jurisdictions.
Among the many beneficial effects of smart growth cited by its proponents is the belief
that by changing the built environment in which people live, transportation behavior can
be altered so that it is more efficient and environmentally friendly. Proponents hope that
people living in denser, more walkable neighborhoods will become less dependent on
automobiles. Thus, using the available data, it is important to try to discern the influence
of the built environment on the way we use transportation in our everyday lives.
By further studying the effects of the built environment on automobile ownership
levels and VMTs, the goal of this analysis is to better understand the implications of
density and several other spatial variables on household transportation behavior. This is
especially important as more and more local governments adopt smart growth planning
practices. Since 2001, large Sunbelt cities, such as Houston and Phoenix, which have
traditionally seen mostly sprawling automobile-oriented growth, have made substantial
public investments in regional rapid transit systems. As these regions begin to promote
more compact, transit-oriented neighborhoods, it is important to measure the effect these
neighborhoods have on the transportation habits of those who live in them.
A great deal has been written on planning theory and the benefits of smart growth.
There are also a number of empirical studies that have looked at the effects of the built
4

environment on topics related to automobile ownership and VMTs. However, most of
these studies rely on very old datasets (Zhang, 2006) or only focus on a single region
(Shay & Khattak, 2006). The studies that examined multiple regions do not analyze data
beyond the MSA-level scale (Cervero & Murakami, 2010). In studies that have looked at
VMTs or automobile ownership at the block group scale, it has typically been within a
single MSA or metropolitan area (Haas, Makarewicz, Benedict, & Bernstein, 2008).
This analysis builds on existing research by looking at a sample of 75 block
groups across five different MSAs using both neighborhood and regional variables.
Because data are examined from multiple regions, the results of this study are more
generally applicable than studies that examine a single MSA. Also, the study looks at the
influence of variables at overlapping scales (e.g., the neighborhood and the metropolitan
region).

5

Chapter 2: Background
For many years those in favor of smart growth principles have advocated its many
benefits. Of particular relevance to this study are the aspects of smart growth that
encourage transit use and discourage automobile dependency through changes in the built
environment.
Among the most commonly cited examples of smart growth practices for
reducing the use of personal vehicles are transit-oriented development, mixed-use zoning,
and increased transit options (Gearin, 2004). Transit-oriented development refers to the
zoning of high-density development in the immediate vicinity of transit stations. Mixed-
use zoning practices allow for the creation of buildings with multiple uses, such as an
apartment building with ground-level retail. Mixed-use zoning can also apply at the
neighborhood scale where zoning allows for a mix of different building types. The
purpose of such mixed-use development practices is to reduce overall transit demand.
Many regions increase transit options through the construction of new rail transit
systems, the addition of new bus lines, or the creation of bike trails.
Studies have found that the connection between transportation and land-use is
complex. Factors, such as household preferences, socio-demographic variables, and the
scale at which policies are enacted, determine how the connection manifests itself.
Furthermore, while urban form can influence transportation behavior, it is usually a
secondary factor to personal preferences and socio-demographics. Increased density,
mixed land-use, and transit-oriented design are thought to play only a modest role in
decreasing VMTs (Ingram et al., 2009).
6

This section provides an overview of the current body of literature on the topics
relevant to this paper. Many studies have looked into the effect of the built environment
on vehicle transit behavior; however, most of these studies only look at a single region
and measure a limited number of variables. This study takes a deeper look into the
relationship between the built environment and both vehicle miles traveled (VMTs) and
automobile ownership by (1) measuring multiple spatial variables, (2) measuring
variables at both the block group and metropolitan statistical area (MSA) scale, and (3)
collecting data from sample block groups across five MSAs.
2.1 Single Region Studies
A number of studies have looked at the factors that influence transportation
behavior in a single metropolitan area. Shay and Khattak (2006) investigated automobile
ownership in the Charlotte, NC, metropolitan area, using survey data from 2001. This
was done using a number of environmental measures, neighborhood typologies and
indices of environmental factors generated by factor and cluster analyses, and other
spatial variables. It was shown that automobile ownership is affected by socio-
demographic factors such as income and household size. However, environmental
factors, including land-use and walkability, had a greater influence on trip generation
than socio-demographic factors.
Another approach used in single region studies is to model household
transportation expenses for different types of neighborhoods. Haas et al. (2008)
developed models for predicting total out-of-pocket household annual transportation
expenditures and tested them in the Minneapolis-St. Paul, MN metropolitan area using
7

data from the 2000-2003 timeframe. These models included independent variables, such
as density, job access, neighborhood services, walkability, and transit connectivity. The
models were then used to confirm the statistically significant influence of the built
environment and transit accessibility on household transportation expenditures. In
particular, built environments that featured smaller block sizes, a greater number of
services within the neighborhood, greater residential densities, high transit connectivity,
and close proximity to major employment centers reduced the number and distance of
automobile trips.
The fact that these studies only analyzed a single region severely limits their
applicability at larger spatial scales. The likely reason that these studies only tested a
single region is that most of them relied on surveys that included data from personal
travel diaries. This allowed for the examination of transit behavior at the individual level,
but it also limits the applicability of the results to other regions.
2.2 Longitudinal Approaches
Krizek (2003) investigated the effect of neighborhood-scale, urban-form factors
on travel behavior in the Central Puget Sound region in the state of Washington. This
longitudinal study developed regression models to predict changes in travel behavior as a
function of neighborhood accessibility controlling for regional and workplace
accessibility. The study used survey data collected between 1989 and 1998. It was found
that household travel behavior changed when exposed to differing urban forms and that
higher neighborhood accessibility decreases VMTs. Using travel diaries collected in the
San Diego, CA, area in 1986, Crane and Crepeau (1998) investigated claims made by
8

smart growth advocates that urban design can influence travel behavior. The study found
that land-use played only a small role in explaining travel behavior. Additionally, the
study found no evidence to support the theory that street network patterns affect non-
work travel decisions.
2.3 Studies at Multi-Region and Neighborhood Scales
Some studies have investigated VMTs in multiple regions. Cervero and
Murakami (2010) analyzed the effects of the built environment on VMTs in 370
urbanized areas in the U.S., using data from 2003. Using structural equation models,
factors such as population density, access to employment, population of urbanized area,
and rail transit usage, the study found that population density was strongly and positively
associated with lower VMTs. Employment access, the population of the urbanized area,
and rail usage had only modest effects. This analysis took place at the metropolitan scale
as opposed to the block group scale used in this study. This analysis was likely
conducted at the MSA scale so that data could be easily compared across multiple
regions.
Other studies have investigated transit behavior at the neighborhood scale only.
Using 2003 survey data from northern California, Cao, Mokhtarian, and Handy (2007)
examined the influence of neighborhood design versus residential self-selection as the
causal factor of transportation behavior at the neighborhood level. The study relies on
data from four ―traditional’ neighborhoods and four ―suburban‖ neighborhoods with
variables related to travel behavior, neighborhood characteristics, neighborhood
preferences, travel attitudes, and socio-demographics. The study found that while
9

residential self-selection had significant impacts on travel behavior, the built environment
also had a statistically significant association with changes in travel behavior. Increased
transit accessibility was the most important factor in reducing driving.
Many studies have included only a limited number of spatial measures when
investigating vehicle transportation behavior. Zhang (2006) conducted an empirical
study of automobile dependence in the Boston area using travel survey data from 1991.
The emphasis of the study was mode choice. Spatial variables for land-use and street
connectivity were used in addition to socio-economic variables. It was found that
automobile dependence is sensitive to street network connectivity and automobile
availability. Both population and job density were also found to be important, and land-
use’s role in increasing transportation options was confirmed.
Cervero (2002) studied the influence of built environments on transportation
mode choice using 1994 survey data from Montgomery County, MD. The study
developed a normative model that weighed the influence of built environment factors,
including density, diversity, and design. It was found that density and mixed land-use
had a significant influence on transportation mode choice while urban design factors had
a more modest influence.
Kim and Brownstone (2010) looked into the impact of residential density on
vehicle usage and fuel consumption. An empirical model was developed using data from
the 2001 National Household Travel Survey. It was found that households located in
block groups in which density is greater than a 1,000 housing units per square mile will
drive less and consume fewer gallons of fuel than households in less dense areas.
10

Density in the context of the surrounding area was found to be greater than the effect of
just residential density. It was also found that moving a household from a suburban area
to an urban area reduces household VMTs by 15%. While the Kim and Brownstone
(2010) study uses the same VMT data used in this analysis, it only examined VMTs in
the context of density.
Hess and Ong (2002) used 1994 survey data from Portland, Oregon, to develop a
model to explain automobile ownership based on demographic variables and a few urban
design characteristics such as land-use mix. It was found that the presence of mixed
land-uses caused the probability of owning an automobile to decrease by 31 percent. It
was also found that non-sprawl neighborhoods are more conducive to walking and to the
use of public transit.
Some studies have used methodologies similar to the one used in this analysis.
Cervero (1996) investigated the suggestion that mixed land uses encourage non-auto
commuting. His analysis used data from the 1985 American Housing Survey that
included survey data for eleven MSAs. A regression model was created using various
land-use variables designed to capture the density and presence of mixed land-use.
Control variables measured household income, automobile ownership, location within
MSA, the presence and adequacy of transportation choices in neighborhood, and the
distance from home to work. The study found that neighborhood density had a greater
influence than mixed land-uses in influencing transportation choices, except for walking
and bicycling. The study also found significant elasticity between land-use environments
and commuting choices in the eleven MSAs.
11

Using variables that describe the built environment, this study seeks to explain
their effect on vehicle transit behavior. By measuring multiple spatial variables at both
the block group and MSA scale across five MSAs, this study provides a deeper analysis
of both VMTs and automobile ownership than previous studies.

12

Chapter 3: Methodology
This study seeks to examine how built environments at neighborhood and
regional scales relate to transportation behavior. The analysis focuses on two variables
that measure transportation behavior at the block group level: automobile ownership and
vehicle miles traveled (VMTs). The influence of the built environment on these variables
is examined across a range of cases in metropolitan statistical areas throughout the United
States, representing both neighborhoods with a dense residential population and in less
dense ―sprawl‖ neighborhoods.
The research examines vehicle transit behaviors using a cross-sectional approach
for block groups from five regions for the 2000-2001 timeframe. The study measures and
examines several independent variables related to the built environment at the scale of
neighborhoods and metropolitan regions, including density, job access, transit access,
retail access, and street connectivity. Through analysis of all of these variables, this
paper attempts to find the relationships that explain household automobile ownership
levels and VMTs.
This chapter serves as an overview of the steps taken to complete this analysis,
including the sampling framework, major hypotheses, variables, and spatial and statistical
modeling processes. All census data used in this study comes from the 2000 U.S.
Census.
3.1 Sampling Framework
This study samples seventy-five block groups, combining stratified random and
selected sampling methods. The block group is the smallest scale at which detailed
13

information is available for most of the key variables in this study. This unit of analysis
is small enough that in suburban and urban areas it covers a geographic area similar to
what might be termed a neighborhood. However, in less-populated, exurban areas the
geographic area of a block group is much larger.
The seventy-five cases are sampled from five metropolitan areas in different
regions of the United States. These metropolitan areas are Chicago, IL; Miami, FL;
Portland, OR; San Diego, CA; and Washington, DC (Figure 1). These areas were
selected because they each represent a different region of the U.S. They are also diverse
in terms of their overall population, urban form, the era in which they developed, and the
extent of their regional transit network.
Figure 1: Sample Metropolitan Statistical Areas (MSAs)

14

Throughout this study, metropolitan areas are defined by their metropolitan
statistical area (MSA). MSAs are defined by the U.S. Office of Management and Budget.
They are composed of counties or county-equivalents that cover the extent of a central
urban area or urban cluster and the surrounding area of somewhat continuous and
relatively high population density. Additional ―outlying counties‖ that have strong
economic ties to the central MSA are also included in an MSA. These outlying counties
are included in an MSA if the total in-commuting and out-commuting (i.e., the
employment interchange) exceeds 25% of the total employment in the outlying county.
The MSA is used as the regional unit of analysis in this study because it is the
only consistent definition of a metropolitan area that is available. It is also the definition
of metropolitan areas used by the census when discussing large cities and their
surrounding suburbs (United States, 2000).
The Chicago MSA (Figure 2) was the third most populated region in the 2000
U.S. Census and has an extensive transit network. It is the largest in the sample group of
MSAs and developed earlier than all but Washington, DC. While the built environment
in the city of Chicago is characterized by traditional, dense urban development, most of
its surrounding suburbs are sprawling and automobile-oriented. However, some of the
older inner-suburbs are more dense and transit-oriented.
The Miami MSA (Figure 3) was the sixth largest in the 2000 U.S. Census and has
a relatively small transit infrastructure. While development in the coastal areas is highly
dense, the vast majority of the region is characterized by low-density automobile-
dependent development typical of most Sunbelt cities.
15

Ranking twenty-fifth in the 2000 U.S. Census, the Portland MSA (Figure 4) is by
far the least populated MSA included in this study. However, the Portland area has a
relatively extensive transit system for its size. Portland was a pioneer in adopting smart
growth principles; as a result, it is denser than MSAs of comparable size.
Figure 2: Chicago, IL MSA

16

The San Diego MSA (Figure 5) ranked seventeenth in population in the 2000 U.S.
Census. It has a transit system of moderate extent. The built environment in most of the
region is less dense than other MSAs in this study.
Figure 3: Miami, FL MSA

17

Figure 4: Portland, OR MSA

Figure 5: San Diego, CA MSA

18

The Washington, DC MSA (Figure 6) was the seventh most populated MSA in
the 2000 U.S. Census. It has a relatively large transit infrastructure, and jurisdictions in
the Maryland suburbs have been subject to statewide smart growth initiatives since 1997
(Ingram et al., 2009). As a result, many suburban areas have areas of high density in the
vicinity of transit stations.
Figure 6: Washington, DC MSA

19

As shown in each of the maps above, fifteen sample block groups are taken from
each metropolitan area of which thirteen are randomly selected. Additionally, the most
dense and least dense block groups within the MSA are included in the sample.
3.2 Major Hypotheses and Dependent Variables
It is expected that the data will show that the built environment influences the
number of automobiles owned per household (Table 1). Furthermore, it is predicted that
automobile ownership will be greater in sprawl neighborhoods and lesser in denser
neighborhoods. The data used to measure automobile ownership comes from Summary
File 3 of the 2000 U.S. Census (United States Census, 2000). The census variable used
describes the total number of vehicles available in a block group. To measure automobile
ownership by household, this value is then divided by the number of households in that
block group. The result is a ratio measure that describes the average number of
automobiles owned per household within a block group. Similar studies have used
census data to measure automobile ownership (Center for Transit-Oriented Development,
2006). The number of automobiles owned by a household can be indicative of the
automobile-dependence of the individuals who live within it. Less automobile-dependent
households may only require a single car among its occupants, while the occupants of
other households may each require a car for their daily transportation needs.
Table 1: Dependent Variables
Variable Source Hypothesis
Automobile Ownership 2000 U.S. Census
The built environment influences the number of
automobiles owned per household
Vehicle Miles Traveled
2001 National Household
Travel Survey
The built environment influences household
VMTs

20

It is also predicted that the built environment influences driving such that VMTs
are greater in sprawl neighborhoods and lesser in more densely populated neighborhoods.
The data used to measure VMTs came from the 2001 National Household Travel Survey
(NHTS) that was conducted by the Federal Highway Administration. NHTS data is
reported annually at the MSA, state, and national scale. VMTs are a strong indicator of
how far individuals are traveling daily to access jobs and amenities.
While data from the NHTS is typically unavailable at any scale below the MSA
level, a model was developed to estimate the data at the census tract level using data from
the 2001 survey. The variables used to model VMTs by census tract were household
size, household income, and employment rate (Hu, Reuscher, Schmoyer, & Chin, 2007).
Unfortunately, block group scale data for VMTs do not exist nationally and would
require significant time and expense to collect. The census tract level is the smallest
scale at which VMT data have been estimated across multiple MSAs. Furthermore, the
NHTS estimation model cannot be used to downscale the survey data to the block group
level, because the employment rate variable used in the model is not available at scales
smaller than the tract level.
Therefore, sample block groups used in this study are assigned the VMT value of
their corresponding census tract. All VMT estimates are per household on an average
weekday. The tract level estimates for VMTs are given based on household size and the
number of vehicles available to a household. For the purposes of this study, the VMT
estimate for each block group is based on its mean household size and the mean number
of automobiles per household.
21

3.3 Measuring the Built Environment: Independent Variables
Dense neighborhoods are almost always less automobile friendly than less dense
areas, in part because the availability of parking is scarce. Therefore, density will likely
affect both VMTs and automobile ownership. Density of the built environment is
assessed as the number of households in a block group divided by the total acreage of the
block group. Measuring households per acre as opposed to population per acre is a more
appropriate method for this study because it is more closely related to the density of the
built environment rather than the population density of a neighborhood. That is, high
residential density likely indicates the presence of multi-family housing structures. These
buildings contain multiple units that are smaller than a typical single-family home and,
therefore, are likely to have fewer occupants per household. If measured by population
rather than by the number of households, this could lead to an underestimation of density
in such neighborhoods. A negative relationship is expected between density and both
VMTs and automobile ownership levels (Table 2).
Table 2: Independent Variables
Variable Source Hypothesis
Density 2000 U.S. Census
Expected to have negative relationship with
VMTs and automobile ownership.
Job Access
2000 Census Transportation
Planning Package
Expected to have negative relationship with
VMTs and automobile ownership.
Retail Access
2000 Census Transportation
Planning Package
Expected to have negative relationship with
VMTs and automobile ownership.
Transit Access
2011 National Transportation
Atlas Database
Expected to have negative relationship with
VMTs and automobile ownership.
Street Connectivity 2000 U.S. Census
Expected to have negative relationship with
VMTs and automobile ownership.

22

The commute to and from work is a key aspect of people’s daily transportation
behavior. The distance between a person and their job is likely a major factor in the
decision of how to get to work Job access is measured as the distance from a block
group to an ―employment center.‖ For the purposes of this study, an employment center
is defined as a census tract that is among the top ten percent within an MSA in terms of
the total number of jobs. Similar measurements have been used by authors to establish
the location of job centers within a region in studies that attempt to estimate household
transportation costs (Haas et al., 2008). All job data are extracted from the 2000 Census
Transportation Planning Package (CTPP). A block group’s distance from the nearest
employment center could suggest how far people are traveling to get to work. The
distance between sample block groups and employment centers is measured from the
centroids of both polygons. A negative relationship is expected between job access and
both household and automobile ownership levels and vehicle miles traveled.
Similar to a workplace, retail services, such as grocery stores and shopping malls,
are accessed by many people on a frequent basis. The distance to these services likely
affects the mode of transportation used to access them. To measure retail access, the
distance between a sample block group and a retail center is measured. Those census
tracts that are among the top ten percent in an MSA in terms of the aggregate number of
retail jobs are considered ―retail centers.‖ Retail job data comes from the 2000 CTPP.
The measurement is made from the centroid of the sample block group to the centroid of
the retail center tract. A negative relationship is expected with both VMTs and
automobile ownership per household.
23

Transit is only an option for individuals if it is reasonably accessible. If it is not
easily accessible, it is less likely that a person will choose transit as their primary means
of transportation. Transit access is a measurement of the distance between a sample
block group’s centroid and the nearest transit station. This measure predicts the ease of
access to a transit system in a given census block group. Households located near transit
stations likely own fewer cars and travel shorter distances. A negative relationship is
expected with both VMTs and automobile ownership per household.
This study includes transit stations that are part of a regional light, heavy, or
commuter rail system. Unfortunately, no national dataset could be found that included
bus stop data for local or regional bus services. The spatial data for rail transit stations
comes from the 2011 National Transportation Atlas Database that uses data from the U.S.
Department of Transportation. In keeping with the timeframe of this study, only stations
that were in operation in 2000 are included.
Street connectivity is one relatively simple measure of the ease of walking, as it is
related to pedestrian connectivity in a neighborhood. The street connectivity of a
neighborhood is thought to influence transportation mode choice because people are
willing to walk greater distances more frequently in areas with high pedestrian
connectivity (Center for Neighborhood Technology, 2010).
Street connectivity is measured as the block group’s acreage divided by the
number of blocks within the block group. This measurement gives an indication of how
complete the street network is in a sample block group. A street network can be
considered as a proxy for a sidewalk network. Shorter block lengths have been shown to
24

encourage pedestrian activity (Huang, Stinchcomb, Pickle, Dill, & Berrigan, 2009). A
negative relationship is expected between street connectivity and both automobile
ownership and VMTs.
3.4 Confounding Variables
This study also tests against a series of confounding variables to explore whether
variables, not from the built environment or at spatial scales larger than the block group,
are related to block groups and vary in such a way as to confuse or suppress the observed
relationships. The confounding variables tested are household income, regional density,
extent of regional transit network, age of neighborhood population, and individual transit
expenditure (Table 3).
Income is measured as the per capita household income of a block group.
Household income has been shown to influence transportation mode choices. Higher
income households are more likely to own automobiles and use transit less. This
measure has been used in other models to explain transportation choices (Haas et al.,
2008). A positive relationship is expected between income and both automobile
ownership and VMTs.
Regional density is measured at the MSA level as the MSA population per square
mile. Dense metropolitan areas may lead to greater use of transit options than in less
dense regions. Because this variable is more oriented toward a measurement of
population density rather than strictly the density of the built environment, population per
square mile is more appropriate than using households. Regional density is expected to
relate negatively to both VMTs and automobile ownership.
25

Table 3: Confounding Variables
Variable Source Hypothesis
Household Income 2000 U.S. Census
Positive relationship expected with both
automobile ownership and VMTs.
Regional Density 2000 U.S. Census
Negative relationship expected with
both automobile ownership and VMTs.
Extent of Regional
Transit Network
2000 U.S. Census, 2011 National
Transportation Atlas Database
Negative relationship expected with
both automobile ownership and VMTs.
Age of Neighborhood
Population
2000 U.S. Census
Negative relationship expected with
both automobile ownership and VMTs.
Individual Transit
Expenditure
2000-2001 Consumer Expenditure
Survey
Negative relationship expected with
both automobile ownership and VMTs.

This study also measures the extent of a region’s transit network. Overall, people
are likely less dependent on automobiles in regions in which there is a more robust transit
network. This variable is measured at the MSA level as the number of residents per
heavy, light, or commuter rail station. This variable is obtained by dividing the MSA
population by the number of transit stations in the MSA. The extent of a transit network
in a region is expected to relate negatively to VMTs and automobile ownership.
The age of neighborhood population variable is measured at the block group level
as the median age of residents. An individual’s age may affect the transportation options
that are available to them. The median age of a neighborhood is expected to relate
significantly in the case of outliers of young or old neighborhoods. In areas with more
children or a high elderly population, people likely own fewer automobiles.
Individual transit expenditure is measured at the MSA level as the average annual
consumer expenditure on public transportation. This variable could provide an indication
of how people are using regional public transportation systems based on the average
annual dollar amount spent on transit. Data for this variable come from the Bureau of
26

Labor Statistics’ 2000-2001 Consumer Expenditure Survey. A negative relationship is
expected between individual transit expenditure in both automobile ownership and
VMTs.
3.5 Spatial Modeling of Variables
All spatial data were processed and analyzed using ArcGIS 10. The analysis uses
U.S. Census TIGER/Line shapefiles representing MSAs, block groups, tracts, and blocks
as defined in the 2000 U.S. Census. Census demographic data, in addition to NHTS and
CTPP data, were merged with their corresponding shapefiles at the tract and block group
level within ArcMap. The merge was based on the census-designated tract or block
group identification number. Throughout this study, distance was always a measurement
of distance via the road network as opposed to ―as the crow flies.‖ Additionally,
distances were always measured from the centroid of a block group or tract polygon.
The spatial modeling process began by creating the sample of fifteen block groups
for each MSA. The U.S. Census American Fact Finder website was used to create an
Excel spreadsheet of all block groups within a selected MSA. Once obtained, a
―Random‖ field was added to the spreadsheet, and a random number generator was used
to assign random values between 0 and 1. The spreadsheet was then sorted based on the
―Random‖ field, and the thirteen block groups with the lowest random number were then
used as the randomly sampled block groups from their MSA. The last two block groups
identified for the sample were the most and least dense block groups within the MSA, to
reach a total of 15 for each MSA.
27

To build the sampling frame, a shapefile with all U.S. block groups was added in
ArcMap along with a shapefile of all MSAs. A clip operation of the block groups layer
using the MSA layer was performed to create a shapefile of just the block groups that fall
within the MSA.
The clipped block groups layer was then projected using the appropriate state
plane projection for the region. Using the calculate geometry tool, the acreage of each
block group was calculated and added to the table. A table containing general block
group level demographic data was then merged with the block group data. Among these
new data was a variable containing the total number of households within each block
group. The field calculator tool in ArcMap was used to calculate the variable for density
by dividing the total number of households in a block group by its acreage. This new
attribute was the measure of density used in this study. The block group with the highest
value was added to the sample as the most dense block group within the MSA. The block
group with the lowest value that had at least 100 households was added to the sample as
the least dense block group within the MSA.
For the dependent variables, automobile ownership per household was obtained
by merging the census data table containing the total automobiles owned in each block
group. A field was then added to the table for automobile ownership per block group,
and the field calculator tool was used to divide total automobiles by total households for
each block group.
For block group VMTs, NHTS estimates for the corresponding census tracts were
used. The NHTS estimates for each tract varied depending on the size of a household
28

and the number of vehicles owned in that household. For the purposes of this study, the
mean household size and automobile ownership value of a sample block group was used
to determine the NHTS estimate of VMTs in its corresponding census tract. The census
tract level estimate of VMTs was then applied to its corresponding block group.
The process of determining job and retail access began by identifying the job and
retail centers within each MSA. The CTPP data for the locations of jobs were presented
at the tract level. The data were sorted by the field representing total jobs within each
tract. All tracts except those within the top 10% of total jobs were deleted. The table was
then brought into ArcMap and merged with the tract shapefile for the MSA, creating a
job centers shapefile. The same process was used for identifying retail centers. Retail
and job access were measured as the distance between the centroid of a sample block
group and the centroid of a retail and job center, respectively. All centroids were created
in ArcMap using the feature-to-point tool.
The measure of distance for retail access, job access, and transit access was
calculated on the road network. To do this, a road network was created within ArcMap.
ESRI’s detailed roads shapefile available for download through ArcGIS Online was used
to create the street network. Once downloaded, the street layer was clipped using the
MSA shapefile in order to make the file size more manageable. In ArcCatalog, a new
street network was created using the detailed roads shapefile. The new network took into
account roadway elevations (i.e., grade-separated intersections) and used length as a
constraint. The default options were used for all other settings.
29

Once the street network was added to ArcMap, ―a new route‖ was created. Using
the network location tool, a location was created on top of a centroid. A second location
was placed on the centroid of the nearest job center, retail center, or transit station. A
new route was then created on the street network between the two network locations.
The distance was then recorded. If there were several candidates for the closest job
center, retail center, or transit station, multiple measurements were taken to determine the
closest. This process was repeated for all selected block groups in order to determine its
closest job center, retail center, and transit station.
Street connectivity was measured as a block group’s total acreage divided by the
number of census blocks in the block group. To measure street connectivity, a shapefile
containing the blocks within an MSA was added into ArcMap. An SQL query was
created using the Select by Attribute tool to select all blocks that share the same block
group identification as the selected block groups. The total number of blocks in each of
the selected block groups was counted and added to the selected block groups attribute
table. A new street connectivity field was then created in the selected block groups table.
Data were populated by using the field calculator to divide the total acreage of the block
group by the total number of blocks.
For the confounding variables, the data for block group per capita household
income were taken directly from the census. To measure regional density, the MSA
shapefile was projected into the appropriate state plane coordinate system. The square
mileage of the MSA was then measured using the calculate geography tool. Total MSA
population data were taken from the census and added to the selected block groups data
30

table. The Field Calculator was used to calculate the MSA total population divided by the
MSA square mileage.
3.6 Statistical Analysis
Using both the SPSS and R statistics packages for processing, the best fitting
multivariate regression models were developed for each dependent variable. To
accomplish this, descriptive statistics, such as mean, standard deviation, and histograms
were drawn for each variable. Skewness of variables was observed. A full correlation
matrix of variables was also drawn to check for multi-colinearity. As detailed in the
Results section, the data were further explored for both spatial autocorrelation and the
modifiable areal unit problem (MAUP). Various hypotheses were confirmed or rejected
as related in the Results and Discussion and Conclusion sections.

31

Chapter 4: Results
This chapter examines the results of the analysis. It covers the following topics:
(1) analysis of study variables, (2) assessment of correlation in study variables, (3)
regression models, (4) spatial autocorrelation, and (5) the modifiable areal unit problem.
4.1 Analysis of Study Variables
Households from the sample block groups have a mean VMT value of 50 miles
on an average weekday (Figure 7) with a standard deviation of 22. The means for all of
the MSAs fall within the standard deviation. The most dense block groups have a mean
VMT value of 21 miles while the least dense block groups have a mean of 73 miles.
Figure 7: Average Household VMTs

0
10
20
30
40
50
60
70
80
All
Samples
Chicago Miami Portland San Diego DC Random
Samples
Most
Dense
Least
Dense
Average Household VMTs
32

The mean number of automobiles owned per household in the sample block
groups is 1.68 (Figure 8) with a standard deviation of 0.5. The mean automobile
ownership does not vary greatly in any of the MSAs. The most dense block groups own
a mean of .79 automobiles while the least dense own a mean of 2.23. For both dependent
variables, the means are similar across all five MSAs, and the most and least dense block
groups vary as expected.
Both dependent variables are normally distributed (Table 4). On average, the
sample block groups are located 5.72 miles from job centers, 5.66 miles from retail
centers, and 7.43 miles from the nearest transit station. The values for street connectivity
range from 1 to 2,920 acres per block with the lowest values signifying block groups
estimated to have the greatest pedestrian connectivity. The average street connectivity
value is 129 acres per block. Natural log transformations are used for all of the
independent variables when they are included in a model. This is a common practice in
regression modeling to standardize skewed variables and improve model fit (Allison,
1999).
In the independent variable histograms, the most and least dense block groups are
significant outliers that cause the histograms to skew right. For the job access and retail
access variables, over 85% of the block groups are less than ten miles from a job center
and retail center. The remaining block groups (including all five least dense block
groups) are between 10 to 61.3 miles from the closest job or retail center. The variable
for transit access is similar to the job and retail access variables in terms of distribution
with over 82% of the block groups located within ten miles of a transit station. The same
33

block groups that were outliers for the job and retail access variables are among the 13
that are more than 10 miles from a transit station. The highest value for transit access is
65 miles.
Both the density and street connectivity variables are highly clustered. Only
seven of the 75 sample block groups had a density value of more than twenty households
per acre. Of the seven, five were the most dense block groups selected from each MSA.
The other two block groups are in the San Diego MSA. For street connectivity only 6
block groups had a value of more than 250 acres per block. Of these six, five are the
least dense block groups in each MSA. The sixth block group is in the Washington, DC
MSA.
Figure 8: Average Automobile Ownership

0
0.5
1
1.5
2
2.5
All
Samples
Chicago Miami Portland San Diego DC Random
Samples
Most
Dense
Least
Dense
Average Household Automobile Ownership
34

The average block group density is 13.36 households per acre. The average per
capita household income is $24,390. The average of the median age of the block group’s
population is 36.52.
The variables for individual transit expenditure, regional density, and extent of a
regional transit network are all MSA-level variables. Therefore, there are only five
unique values for these variables. The MSA averages for annual transit expenditure
range from $385-$861 per person. The MSA averages for regional density range from
373.67-1,610.66. The average number of citizens per transit station ranges from 21,212-
117,466.
Table 4: Distribution of Variables
Variable Distribution
Natural Log
Transformation Used in
Models
Dependent

Automobile Ownership Normal No
Vehicle Miles Traveled Normal No
Independent

Density Skewed Right Yes
Job Access Skewed Right Yes
Retail Access Skewed Right Yes
Transit Access Skewed Right Yes
Street Connectivity Skewed Right Yes
Confounding

Household Income Normal No
Regional Density Normal No
Extent of Regional
Transit Network
Normal No
Individual Transit
Expenditure
Normal No
Age of Neighborhood
Population
Normal No

35

4.2 Assessing Correlation of Study Variables
A significant challenge in building the best-fit models for VMTs and automobile
ownership is that many of the independent variables are significantly correlated with one
another. In particular, distance to job center and distance to retail center were found to
correlate at Pearson’s R=.797 (p<0.01 level of significance). The strong correlation
between job access and retail access is likely due to the difficulty of distinguishing
between the two variables spatially. Tracts that were designated as a job center or a retail
center were often both a job and a retail center. In every MSA, combined job and retail
center tracts were more common than separate tracts for job centers or retail centers
(Table 5).
Table 5: Breakdown of Spatial Relationship between Job and Retail Centers
MSA Job Centers Retail Centers Job and Retail Centers
Chicago 68 68 120
Miami 26 26 36
Portland 17 17 26
San Diego 25 25 36
D.C. 47 47 57
Total 183 183 275

Job access and retail access were not significant in either the VMT or automobile
ownership regression models. An additional variable was created for the mean of the
distance to job center and retail center variables for each block group. This variable was
also not significant in either model. Although job access and retail access were also
highly correlated with transit access, the variable for distance to transit was consistently
more significant in test models and therefore was selected over the other distance
36

measures. Because the three distance variables were highly correlated, including more
than one in a regression model did not increase the model’s explanatory power.
All independent variables were found to have significant bi-variate correlations
with both dependent variables (p≤0.01). Of the confounding variables, transit spending
and transit stations per capita showed a bi-variate correlation with VMTs at the p≤0.05
level of significance. For automobile ownership, only MSA density was correlated at the
p≤0.01 level of significance.
4.3 Regression Models
A best-fit linear regression model was developed for both dependent variables.
The VMT model includes two variables–the natural log transformation of block group
density and the extent of a region’s transit network (Table 6). Adjusted R-squared is used
in this study to compare the predictive power of different models. For models that have a
low number of samples and many predictor variables, adjusted R-squared minimizes bias
(Agresti, 2009). The adjusted R-squared for the model is .404 meaning that about 40% of
the variation is explained by density and the extent of the regional transit network. The
model shows that as density increased, VMTs decreased. Additionally, as the number of
residents per transit station increased, VMTs also increased.
The transit access variable was not included in this model but was very close to
having a statistically significant association with VMTs. The job access, retail access,
and street connectivity variables showed no significant association with VMTs when
included in models. Household income could not be included in the model because
37

income was a variable used to estimate tract-level VMTs from the 2001 National
Highway Travel Survey (NHTS) data.
A model was also developed for VMTs using only the randomly selected block
groups, thereby removing induced bias from the sample. In addition to the block group
density and the extent of a region’s transit network, distance to the nearest transit station
was found to be statistically significant and therefore is included in this model. The
adjusted R-squared of the second model was .459.
Table 6: Best-Fit Regression Model for VMTs
Vehicle Miles Traveled
Adjusted R-Squared 0.404
Model Intercept Significance
Constant 17.396 0
Block Group Density -6.524 0
Extent of Regions Transit Network -2.85 0.006
N=75

In the best-fit model for automobile ownership, three variables were found to be
significant: (1) block group density, (2) distance to nearest transit station, and (3) per
capita household income. The adjusted R-squared of this model was .445 (Table 7). The
results of the model showed that higher block group density was related to a decrease in
automobile ownership. As the distance to the nearest transit station increased, so did
automobile ownership. Similarly, as household income increased, automobile ownership
increased.
The removal of induced bias in the sample does not significantly change the
model. A second model was developed using only the random samples. This model used
the same variables as the first automobile ownership model, and the direction of each
38

variable was the same as in the previous model. The second model’s adjusted R-squared
was .479.
As with the VMT model, access to job centers and retail centers has no significant
association with automobile ownership. Street connectivity also shows no significance in
the automobile ownership model. This would seem to confirm an association between
the smart growth principles of transit access and density with automobile use. Not
surprisingly, the model results also indicate an association between wealth and
automobile ownership. This indicates that wealthier households are more likely to own
more than one car and presumably drive more even when controlling for built
environment factors.
The best-fit regression model for vehicle miles traveled (VMTs) shows that the
variables for density and the extent of a region’s transit network explain over 40% of
average weekday VMTs per household. When the induced bias introduced by including
the most and least dense block groups in each MSA is removed, the model explains
nearly 46% of average weekday VMTs per household. When working with just the
random samples, the distance of a block group to the nearest transit station becomes
statistically significant and is therefore included in the model.
Table 7: Best-Fit Regression Model for Automobile Ownership
Automobile Ownership
Adjusted R-Squared 0.445
Model Intercept Significance
Constant 12.061 0
Block Group Density -4.437 0
Distance to Transit 1.999 0.049
Household Income 2.105 0.039
N=75
39

Additionally, the hypothesis that the built environment influences the number of
automobiles owned per household is supported by the data results. For automobile
ownership, the best-fit regression model indicates that density, distance to transit, and
household income explain over 44% of household automobile ownership. With the
removal of the most and least dense block groups that create induced bias, the same
variables explain nearly 48% of household automobile ownership rates.
Several of the independent and confounding variables do not increase the
explanatory power of either model and, therefore, are not included. As mentioned above,
the variables for job and retail access are not included because of their high correlation
with transit access. Street connectivity is correlated with both VMTs and automobile
ownership at the p≤0.05 level of significance. However, it is not significant when
included in either model.
For the confounding variables, the variables for regional density, age of
neighborhood population, and individual transit expenditure are omitted from both
models. Regional density does not have a bi-variate correlation with VMTs, but it does
with automobile ownership at the p≤0.05 level of significance. The age of neighborhood
population variable does not have a bi-variate correlation with either VMTs or
automobile ownership. Individual transit expenditure has a bi-variate correlation with
VMTs at the p≤0.05 level of significance. It does not correlate with automobile
ownership. When included in either model, these variables were not found to be
significant.
40

4.4 Spatial Autocorrelation
Spatial autocorrelation measures the relationship of a variable among multiple
occurrences in space (O’Sullivan & Unwin, 2010). In some cases, data from locations
within close proximity may be more likely to be similar than data from further locations.
Thus, adjacent location, rather than the study variables, may explain or enhance statistical
associations between block groups and the dependent variables, either within or between
regions.
To test for autocorrelation of the sample block groups within each MSA, two
approaches have been used. First, contrast models were developed for all the MSAs to
test for the potential significance of variables not yet measured at the regional level.
Second, comparisons of fixed effect and mixed effect models were developed to test the
potential of the MSAs and the regional level variables to explain residual variance.
Third, the X and Y coordinates of the centroids for each of the sample block groups were
measured and entered into the regression models to test for association of proximity of
the block groups with the dependent variables within each region.
For the contrast model, a dummy variable has been created for each MSA to test
for the significance of a particular MSA compared to all the others (i.e., the test region=1,
all other regions=0). VMTs have been tested by running the model with the variables for
density, the extent of a region’s transit network, and an MSA dummy variable. The
model was run five times using each MSA dummy variable. None of the regions showed
a significant association. Thus, no single region stands out as requiring further
investigation to develop variables not yet measured.
41

The contrast model for automobile ownership was run using the log
transformation of density, per capita household income, the distance to nearest transit
station, and an MSA dummy variable. The results of the contrast model for automobile
ownership only show a significant association for the San Diego MSA. The San Diego
automobile ownership contrast model has an adjusted R-squared of 0.466 as opposed to
the best-fit model that has an adjusted R-squared of 0.445. This indicates that there is
unexplained variation occurring in the San Diego MSA. The fact that a block group is
located within the San Diego MSA increases the predictive power of the automobile
ownership regression model. Households in the same neighborhood conditions own
more automobiles in the San Diego MSA than in the other MSAs evaluated. This occurs
for reasons at the regional level in San Diego that remain unexplained.
Another technique to explore the influence of the MSA’s regional level variables
is to compare fixed effect and mixed effect models. In this approach, variables
representing the regions are input into the regression models all at once, rather than
region-by-region as with the contrast models.
Further investigation comparing fixed effect and mixed effect regression models
for the two dependent variables gives a different indication of the role of regional level
variables than the contrast model approach. In these models, the regression for
automobile ownership is not statistically better when accounting for regional variation.
However, when comparing models for the VMT model, regional effects are found to
improve the fit of the regression.
42

For the VMT regression model, a random effects model with a random slope and
intercept for each MSA is better than a fixed effects only model, suggesting both that
VMTs differ for each MSA and that each MSA has a different association between VMT
and population density at the block group level. In the better fitting models, transit
stations per capita are entered as a categorical variable rather than a continuous variable
(i.e., only five unique values). This creates a regional identifier similar to using a dummy
variable.
The best-fit model for VMTs occurs when an interaction term between the
variables for density and extent of regional transit network is used. This gives an
adjusted R-squared of 0.504, explaining variance not explained in the initial model that
gives an adjusted R-squared of 0.404 (see Table 6). The model shows that an increase in
density at the block group scale and an increase in the extent of the regional transit
network at the regional scale enhance each other’s effects in decreasing VMTs.
The third approach to testing for spatial autocorrelation was to test whether
proximity of block groups within regions could explain residual variation. For the VMT
regression model, adding a spatial correlation term does not explain any of the residual
variance. However, for automobile ownership, proximity of block groups within a region
was found to have a significant association.
Spatial relationships of block groups explain a good share of residual variance in
the automobile ownership regression model. For the automobile ownership model,
including correlation of spatial coordinates, the adjusted R-squared is 0.567 compared
with 0.445 in the initial model (see Table 7). This means that the model can best predict
43

automobile ownership in non-sampled block groups within a given MSA when the
distance between these block groups and the sampled block groups is known and taken
into account.
The proximity measure in the best fit, spatial model for automobile ownership
also competes with the linear hypothesis for the distance to transit variable, rendering it
non-significant in the model. Further investigation demonstrated a non-linear
relationship between distance to transit and automobile ownership. The spatial model
indicates that automobile ownership increases as a function of distance to transit but then
gradually levels off and decreases at large distances to stations. There is little theoretical
support for this finding, and it deserves further exploration in future studies.
Overall, the tests performed to measure autocorrelation in the block group
samples help explain residual variation in both regression models. For VMTs,
autocorrelation has no effect within the MSAs. However, it does explain variance in
VMTs between the MSAs. Conversely, autocorrelation explains variance in automobile
ownership within the MSAs.
4.5 Modifiable Areal Unit Problem
The Modifiable Areal Unit Problem (MAUP) describes the effect that arbitrary
areal geometric units have on geographic analyses (Montello & Sutton, 2006). All of the
areal units examined in this study are designated by the U.S. Census Bureau and therefore
cannot be controlled in study design. Because the block groups used in the sample data
vary significantly in size, the MAUP is a potential issue. The potential for the MAUP is
44

greatest with the block group density variable because it is directly related to the acreage
of a block group.
To test for these issues, the largest and least dense block groups in the sample
were examined. The three least dense block groups were in the Miami, Portland, and San
Diego MSAs. Areal examination revealed that development was generally ex-urban in
all three block groups. The vast majority of land was undeveloped in 2000. Small areas
of development existed but were mostly spread out (Figure 9).
Figure 9: Largest, Least Dense Block Group in San Diego MSA

45

If any of these large block groups were to be broken up into smaller units,
equivalent to the mean block group size of their MSA, the density variable of the smaller
units would most likely have a value of 0. If randomly selected from the smaller units, it
would be very unlikely that a unit with any development in it would be selected. As a
result, the areal unit has little effect on the density variable throughout most of these
block groups. The density value is already extremely low in these block groups and
would be only slightly lower if broken up. There are no areas of significant density
within any of these block groups.

46

Chapter 5: Discussion and Conclusion
In general, evidence from the regression models developed in this study support
an association between key factors in the built environment and automobile transit
behaviors at the neighborhood level. The hypothesis that the built environment
influences people to drive less is supported by the research results, however, with a lot of
unexplained variation.
The data show that the built environment and the extent of a region’s transit
network are significantly associated with the daily transportation habits of households.
Proximity to a transit station is also significantly associated with the transportation habits
of households in urban and suburban areas. Additionally, the number of automobiles
owned by a household is significantly associated with the density of the built
environment, proximity to a transit station, and household income.
This study put the core principles of smart growth to a robust empirical test. In
particular, the study examined the claims that density, street connectivity, and access to
jobs, services, and transit would lead to less dependency on automobiles. The findings
confirm that density and the extent of a region’s transit network are significantly
associated with vehicle miles traveled (VMTs). Additionally, automobile ownership is
significantly associated with density, transit access, and household income.
These findings have many public policy implications. At the neighborhood level,
the findings suggest that density should be encouraged through zoning legislation in areas
well served by transit. This echoes the findings of past research that suggests that density
is associated with vehicle transportation behavior (Kim & Brownstone, 2010). This is
47

already a common practice in many jurisdictions through the use of transit-oriented
development (TOD). This practice helps to maximize benefits from the large public
investment required to build transit lines.
At the regional level, these findings suggest that investments in transit and smart
growth projects will have a greater impact on VMTs in areas that have existing transit
systems with dense service networks. Because VMTs have a significant association with
the extent of a region’s transit network, the impact of an expansion to an existing transit
system, or organized development around transit, could potentially be greater than that of
a new transit line in a region that does not have an existing public transit system. The
findings also suggest that the continued use of TOD projects in jurisdictions well served
by transit could lead to fewer VMTs. Lastly, the study results suggest that in areas with
existing TOD projects, or in areas that are already relatively dense, more density should
be considered as part of the master planning process.
Because of the association between household income and automobile ownership,
new transit projects could be most effective in low-income areas. Individuals in more
affluent neighborhoods own more automobiles and therefore are less likely to use public
transit even when the built environment is conducive to driving less. Expansion of transit
lines and improved transit access in middle and lower-income neighborhoods could lead
to reduced vehicle ownership and VMTs than similar expansions in wealthier
neighborhoods. Past studies have also found a significant association between income
and vehicle transportation behavior (Shay & Khattak, 2006).
48

5.1 Future Research
Future research should include additional data to better capture the factors that
drive VMTs and automobile ownership. At the time of this study, spatial data for bus
routes and stops was inconsistent and therefore not included. Few jurisdictions had
publicly available bus stop data. Bus route data were available in some areas but not all.
The inclusion of bus data could paint a clearer picture of transit access in future studies,
particularly in areas with a limited regional rail transit system.
Almost all of the block group level census data used in the study came from
Summary File 3 of the 2000 U.S. Census. At the time of this study, Summary File 3 data
from the 2010 U.S. Census had not been released. When these data become available,
this study should be reexamined to evaluate the changes in transportation behavior over
the ten years from 2000-2010. Longitudinal studies have previously been conducted in
an attempt to establish a causal link between the built environment and travel behavior
(Cao et al., 2007). A future study could examine the same block groups as this study and
report the updated variable data. If the 2010 U.S. Census block groups have changed
significantly since the 2000 U.S. Census, a new sample of block groups from the same
MSAs could be used to examine changes.
The smallest aggregation at which VMT data were nationally available was the
census tract level. Even these data were a one-time estimate based on the MSA level
National Household Travel Survey (NHTS) data. For the most part the VMT data that
are available at the block group or neighborhood level come from household surveys
sponsored by metropolitan planning organizations or universities. These surveys are
49

almost always for a single MSA. Future studies with the time and funding to do so could
gather survey data for VMTs at the block group or neighborhood level across multiple
MSAs.
An additional confounding variable that could be used in future research is retail
fuel price data. Data for the average retail price of fuel at the MSA level could not be
obtained for the year 2000. The cost of fuel could be an important factor in the driving
habits of households. These data are available at the MSA level from the mid-2000s to
the present from the Oil Price Information Service. Future studies, using 2010 U.S.
Census data, could take advantage of MSA-level fuel price data. An additional
component of the cost of automobile transportation that should be examined in future
studies is the cost of parking. In highly dense areas, such as the downtown area of a
major city, parking can be a significant monthly expense for commuters.
Future research should also use more sophisticated and complete measures of
walkability. Street connectivity in this study was measured a block group’s acreage
divided by the number of blocks within the block group. A more sophisticated measure
of overall walkability that takes into account infrastructure factors such as sidewalks,
elevation change, shade, safety, and quality of walking paths could yield a fuller picture
of the walking conditions in a block group. Street connectivity could also be measured
using variables related to street connectivity such as intersection density, street network
density, and average street block length (Huang et al., 2009).
In future attempts to examine and explain the phenomena explored in this study,
additional regions and block groups should be included. This may increase the predictive
50

power of the models developed. Additionally, it would potentially allow for better
analysis of the differences in VMTs and automobile ownership between the MSAs and
between regions within MSAs.
This study found that job and retail access are not significantly associated with
either VMTs or automobile ownership. These findings are in contrast with some of the
tenets of planning theory that suggest that improved access to jobs and shopping will
reduce automobile use. At the same time, the methods used in this study to define job
and retail centers only capture concentrations of employment and not necessarily areas of
mixed-use development. While mixed-use development is a key element of smart growth
theory, past studies have shown that mixed land use only slightly decreases overall VMTs
(Ingram et al., 2009). Future studies should develop methods for measuring the presence
of mixed-use development to see how it affects the significance of job and retail access as
they relate to VMTs and automobile ownership.
This analysis could be taken a step further by investigating the link between
transportation behaviors and household transportation costs. Once this link is
established, it could be used to investigate housing affordability when transportation costs
are factored in.
The spatial autocorrelation results suggest that the region a block group is in has
an effect on VMTs and automobile ownership. This study is not able to fully explain the
variance between regions, and therefore further research into the history and culture of a
region could be important. Within regions, the spatial autocorrelation results for
automobile ownership indicate that future studies should consider the inclusion of data at
51

scales between the neighborhood and MSA level. These scales could include sub-regions
within MSAs like major road and transit corridors.
Finally, future research should examine the factors that drive VMTs and
automobile ownership that were not captured in this study. The best-fit model developed
in this study explains approximately 40% of VMTs and 45% of automobile ownership.
Additional studies should seek to explain more difficult to understand factors, such as the
personal transportation preferences of individuals. Recent research indicates that the
built environment may only have a differential impact on walking trips and that an
individual’s attitude towards walking is more important in shaping walking habits (Joh,
Nguyen, & Boarnet, 2011).
A study on the effects of individual preference on transit behavior could be
accomplished by examining these variables at an individual scale as opposed to the
neighborhood or block group level. Studies could also examine the neighborhood level
environment using nested scales. Individual-level data would likely have to be obtained
through surveying, and the research design of such a study would need to be altered to
ensure that the block group samples collected within each MSA are diverse in the income
levels they represent.

52

Bibliography
Agresti, A., & Finlay, B. (2009). Statistical methods for the social sciences.
Upper Saddle River, NJ: Prentice Hall.

Allison, P. D. (1999). Multiple regression. Thousand Oaks, CA: Pine Forge Press.

Cao, X., Mokhtarian, P., & Handy, S. (2007). Do changes in neighborhood
characteristics lead to changes in travel behavior? A structural equations modeling
approach. Transportation, 34, 535-556.

Center for Neighborhood Technology. (2010, April 2). Penny wise pound fuelish: New
measures of housing + transportation affordability.

Center for Transit-Oriented Development and Center for Neighborhood Technology.
(2006, January). The affordability index: A new tool for measuring the true
affordability of a housing choice.‖ Washington, DC: Brookings Institution.

Cervero, R. (1996). Mixed land-uses and commuting: Evidence from the
American Housing Survey. Transportation Research, 30(5), 361-377.

Cervero, R. (2002). Built environments and mode choice: Toward a normative
framework. Transportation Research Part D, 7, 265-284.

Cervero, R., & Murakami, J. (2010). Effects of built environments on vehicle
miles traveled: Evidence from 370 urbanized areas. Environment and Planning,
42, 400-418.

Crane, R. & Crepeau, R. (1998). Does neighborhood design influence travel? A
behavioral analysis of travel diary and GIS data. Transportation Research-D,
3(4), 225-238.

Gearin, E. (2004). Up against the sprawl: Public policy and the making of Southern
California. Minneapolis: University of Minnesota Press.

Ewing, R., & Cervero, R. (2001). Travel and the built environment: A synthesis.
Transportation Research Record, 1780, 87-114.

Gordon, P., & Richardson, H. (1997). Are compact cities a desirable planning goal?
Journal of the American Planning Association, 63(1), 95-106.

53

Haas, P., Makarewicz, C., Benedict, A., & Bernstein, S. (2008). Estimating
transportation costs by characteristics of neighborhood and household. Journal of
the Transportation Research Board, No. 2007, 62-70.

Hess, D. B., & Ong, P. M. (2002). Traditional neighborhoods and automobile
ownership. Transportation Research Record, Paper No. 02-2594, 35+.

Hu, P., Reuscher, T., Schmoyer, R., Chin, S. (2007, May 2). Transferring 2001
National Household Travel Survey. Oak Ridge National Laboratory.

Huang, L., Stinchcomb, D., Pickle, L., Dill, J., & Berrigan, D.. (2009). Identifying
clusters of active transportation using spatial scan statistics. American Journal of
Preventive Medicine 37(2).

Ingram, G., Carbonell, A., Hong,Y., & Flint, A., eds. (2009). Smart growth policies: An
evaluation of programs and outcomes. Lincoln Institute of Land Policy.

Kim, J., & Brownstone, D. (2010). The impact of residential density on vehicle
usage and fuel consumption. University of California: Department of
Economics.

Joh, K., Nguyen, N., & Boarnet, M. (2011). Can built and social
environmental factors encourage walking among individuals with negative walking
attitudes? Journal of Planning Education and Research from
http://jpe.sagepub.com/content/early/2011/11/29/0739456X11427914.

Krizek, Kevin J. (2003 Summer). Residential relocation and changes in urban travel:
Does neighborhood-scale urban form matter? Journal of the American Planning
Association, 69(3), 265.

Montello, D., & Sutton, P. (2006). An introduction to scientific research methods
in geography. Thousand Oaks: Sage Publications.

O'Sullivan, D., & Unwin, D. (2010). Geographic information analysis. Hoboken, NJ:
John Wiley & Sons.

Shay, E., & Khattak, A. (2006). Autos, trips and neighborhood type: Comparing
environmental measures. Transportation Research Record, Paper 07-1119, 1-18.

United States. Office of Management and Budget. (2000, December 27). Standards for
defining metropolitan and micropolitan statistical areas. Federal Register, 65, 249.

54

United States Census 2000. Summary File 3 (SF 3), from
http://www.census.gov/census2000/sumfile3.html.

Zhang, M. (2006). Travel choice with no alternative: Can land use reduce
automobile dependence? Journal of Planning Education and Research, 25,
311-326.

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Smart growth and walkability affect on vehicle use and ownership

PDF

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

PDF

Installing public electric vehicle charging stations: a site suitability analysis in Los Angeles County, California

PDF

Estimating populations at risk in data-poor environments: a geographically disaggregated analysis of Boko Haram terrorism 2009-2014

PDF

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

PDF

The built environment, tour complexity, and active travel

PDF

Integrating spatial visualization to improve public health understanding and communication

PDF

Investigating the association of historical preservation and neighborhood status in Detroit, 1970-2015

PDF

A spatial narrative of alternative fueled vehicles in California: a GIS story map

PDF

Finding food deserts: a study of food access measures in the Phoenix-Mesa urban area

PDF

Relocation bay: identifying a suitable site for the Tampa Bay Rays

PDF

Evaluating transit and driving disaggregated commutes through GTFS in ArcGIS

PDF

Site location suitability analysis for a smart grid network

PDF

Selection of bridge location over the Merrimack River in southern New Hampshire: a comparison of site suitability assessments

PDF

Preparing for immigration reform: a spatial analysis of unauthorized immigrants

PDF

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

PDF

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

PDF

Population disaggregation for trade area delineation in retail real estate site analysis

PDF

Out-of-school suspensions by home neighborhood: a spatial analysis of student suspensions in the San Bernardino City Unified School District

PDF

Healthy mobility: untangling the relationships between the built environment, travel behavior, and environmental health

Asset Metadata

Creator Eisman, Daniel Currie (author)

Core Title Spatial analysis of urban built environments and vehicle transit behavior

School College of Letters, Arts and Sciences

Degree Master of Science

Degree Program Geographic Information Science and Technology

Publication Date 07/06/2012

Defense Date 05/18/2012

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

Tag automobile ownership,built environments,GIS,OAI-PMH Harvest,spatial analysis,vehicle miles traveled,vehicle transit

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Vos, Robert O. (committee chair), Oda, Katsuhiko (Kirk) (committee member), Warshawsky, Daniel N. (committee member)

Creator Email deisman1@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-52335

Unique identifier UC11289419

Identifier usctheses-c3-52335 (legacy record id)

Legacy Identifier etd-EismanDani-918.pdf

Dmrecord 52335

Document Type Thesis

Rights Eisman, Daniel Currie

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