University of Southern California Dissertations and Theses

The impact of demographic shifts on automobile travel in the United States: three empirical essays

(USC Thesis Other)

The impact of demographic shifts on automobile travel in the United States: three empirical essays

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

The Impact of Demographic Shifts on Automobile Travel in the United States: Three Empirical
Essays

By

Xize Wang

A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA

In Partial Fulfillment of the
Requirement for the Degree

DOCTOR OF PHILOSOPHY
URBAN PLANNING AND DEVELOPMENT

August 2017

2

3

"While his parents are alive, the son may not go abroad to a distance. If he does go abroad, he
must have a fixed place to which he goes."
–
Confucian Analects, Book IV, Chapter 19, 540 B.C ~ 400 B.C.

Dedicated to my beloved parents:
Ms. Liping Liu and Mr. Weimin Wang

4

Acknowledgements

I am indebted to many individuals for finishing this dissertation research. I am grateful to
my faculty mentor and dissertation committee chair, Dr. Marlon Boarnet. Without his enormous
support and valuable suggestions, I could have never have turned my bold ideas into established
research products. The five years working with him have been wonderful. I have learned not
only how to be a rigorous scholar, but also how to be a good man. I would also like to send my
appreciation to the professors in the rest of my dissertation committee. My discussions with Dr.
Genevieve Giuliano helped me to refine this research and explore more of the theoretical side of
demography and travel. This work has been inspired by Dr. Dowell Myers’ decades-long
insights in bringing a temporal perspective to a traditionally spatial field. My exploration in the
intersection of health, aging and travel has been greatly furthered by Dr. Julie Zissimopoulos and
her expertise in the economics of aging.
I would also like to thank many other people at USC for helping me on this dissertation.
Dr. James Moore at the Viterbi School of Engineering provided valuable suggestions when
serving on my qualifying exam committee. Dr. Jennifer Ailshire at the Davis School of
Gerontology introduced me to the Health and Retirement Study dataset for the very first time.
Dr. James Polk at the American Language Institute has provided many valuable suggestions on
how to become a better writer. For all my peer Ph.D. Students at the Price School of Public
Policy, thank you for all the years we have spent together, and the endless chats we have had
about both academics and life – you guys are the best.
5

The second essay in this dissertation used the confidential location identifiers from the
2001 National Household Travel Survey. I would like to thank Ms. Adella Santos and Ms. Jasmy
Mathipara at the US Department of Transportation for providing this dataset.
Finally, this dissertation is dedicated to my beloved family. I would like to thank my
parents, Ms. Liping Liu and Mr. Weimin Wang. Mom and Dad, thank you so much for all the
opportunities you have provided me. I could not have come this far on my long journey to
become a scholar without your intellectual, financial and emotional support. I am truly grateful. I
hope that someday I can make you both proud, just as proud as I am of you at this moment.

6

Abstract

This research quantitatively examines whether the current demographic changes in the
United States are linked to changes in automobile travel using comprehensive datasets.
Specifically, this research focuses on the impact of three major demographic shifts: immigration,
Millennials entering adulthood, and aging. The findings of the three essays suggest that these
aforementioned demographic changes significantly influence the demand of automobile travel,
controlling for socio-economic, vehicle ownership, time-specific and regional-specific factors.
Understanding the link from demographics to automobile travel creates opportunities for policy
makers to transform American cities to be more sustainable and to more effectively predict
future travel patterns based on demographic trends.

7

Table of Contents

Acknowledgements ......................................................................................................................... 3
Abstract ........................................................................................................................................... 6
Table of Contents ............................................................................................................................ 7
List of Figures ............................................................................................................................... 10
List of Tables ................................................................................................................................ 11
Chapter 1. Introduction ................................................................................................................. 13
1. Research overview ............................................................................................................. 13
2. Contributions to literature .................................................................................................. 14
3. Structure of dissertation ..................................................................................................... 17
4. References .......................................................................................................................... 19
Chapter 2. A Double-Cohort Analysis for Commute Mode Choice in Los Angeles County,
California ...................................................................................................................................... 21
1. Introduction ........................................................................................................................ 21
2. Literature ............................................................................................................................ 23
3. Data and methods ............................................................................................................... 26
3.1. Area of study and data source ..................................................................................... 26
3.2. Double cohort model .................................................................................................. 28
8

3.3. Track cohorts over time .............................................................................................. 31
4. Results ................................................................................................................................ 33
4.1. Model estimation results ............................................................................................. 33
4.2. Trajectory charts ......................................................................................................... 37
5. Discussion: impact of demographic changes on future automobile commuting ............... 41
6. Conclusion ......................................................................................................................... 43
7. References .......................................................................................................................... 45
Chapter 3. Has the Relationship between Urban and Suburban Automobile Travel Changed
across Generations? A National-Level Inquiry............................................................................. 49
1. Introduction ........................................................................................................................ 49
2. Literature ............................................................................................................................ 52
3. Methods.............................................................................................................................. 55
4. Results ................................................................................................................................ 58
4.1. Generational changes in automobility by neighborhood type .................................... 58
4.2. Regression models ...................................................................................................... 62
5. Conclusion ......................................................................................................................... 67
6. References .......................................................................................................................... 70
Chapter 4. The Impact of Health Conditions on Elderly Driving: A National-Level Longitudinal
Study Using the Health and Retirement Study ............................................................................. 74
1. Introduction ........................................................................................................................ 74
9

2. Methods.............................................................................................................................. 78
2.1. Sample of study .......................................................................................................... 78
2.2. Analytical strategy ...................................................................................................... 81
3. Results ................................................................................................................................ 85
3.1. Impact of overall health conditions on driving ........................................................... 86
3.2. Impact of overall health across socio-demographic groups ....................................... 88
3.3. Impact of specific health conditions on driving ......................................................... 89
4. Conclusion ......................................................................................................................... 93
5. Appendix – Outputs of OLS fixed effects models ............................................................. 96
6. References .......................................................................................................................... 99
Chapter 5. Conclusions and Takeaways ..................................................................................... 102
1. Conclusions ...................................................................................................................... 102
2. References ........................................................................................................................ 104

10

List of Figures

Figure 1-1 – Conceptual model of dissertation, modified from Mayer and Greenwood (1980) .. 14
Figure 2-1 – Share of commuters taking private auto to work in the US and Los Angeles County,
1970-2014 ..................................................................................................................................... 22
Figure 2-2 – Share of commuters by private auto, driving alone and carpooling in Los Angeles
County, 1970-2014 ....................................................................................................................... 27
Figure 2-3 – Illustrative trajectory chart for 25-34 in year 2000 and 35-44 in year 2000, native-
born ............................................................................................................................................... 33
Figure 2-4 – Trajectory charts for predicted probability (%) of commuting mode choice ........... 39
Figure 3-1 – Predicted personal VMT and car trips for a “typical” young adult in 1995 and 2009
....................................................................................................................................................... 67

11

List of Tables

Table 2-1 – Total population and number of the employed in Los Angeles County, 2000 and
2010............................................................................................................................................... 27
Table 2-2 – Descriptive statistics of variables used in the double-cohort modeling .................... 31
Table 2-3 – Logistic double-cohort model estimates on automobile commuting (solo driving or
carpooling) .................................................................................................................................... 35
Table 2-4 – Logistic double-cohort model estimates on commuting by driving alone ................ 36
Table 2-5 – Logistic double-cohort model estimates on commuting by carpooling .................... 37
Table 2-6 – Change of share of population by birth and immigration cohorts, 2010-2020
(projected) ..................................................................................................................................... 42
Table 3-1 – Change in automobility: 1995 - 2009 ........................................................................ 50
Table 3-2 – Descriptive statistics for the young adults (16-28) sample ....................................... 56
Table 3-3 – Change of automobile travel: 1995 – 2009 for the age group 16-28 by block-group
level residential density................................................................................................................. 60
Table 3-4 – Percentage change of automobile travel (1995 - 2009), by age groups and block-
group level residential density ...................................................................................................... 62
Table 3-5 – Regression models for automobile travel for the 16-28 age group in 1995, 2001 and
2009............................................................................................................................................... 64
Table 4-1 – Number of waves surveyed for the individuals in the sample ................................... 79
Table 4-2 – Driving, socio-economic and built environment characteristics of the study sample,
by wave ......................................................................................................................................... 80
Table 4-3 – Patterns of driving across waves ............................................................................... 80
12

Table 4-4 – Descriptive statistics on health conditions, by wave ................................................. 83
Table 4-5 – Impact of overall health conditions on senior driving ............................................... 87
Table 4-6 – Impact of overall health on driving across different socio-demographic groups ...... 89
Table 4-7 – Impact of specific health conditions on senior driving ............................................. 91
Table 4-8 – Impact of overall health conditions on senior driving (OLS models with fixed
effects)........................................................................................................................................... 96
Table 4-9 – Impact of overall health on driving across different socio-demographic groups (OLS
models with fixed effects) ............................................................................................................. 97
Table 4-10 – Impact of specific health conditions on senior driving (OLS models with fixed
effects)........................................................................................................................................... 98

13

Chapter 1. Introduction

1. Research overview
This research examines how demographic changes impact the demand for automobile
travel in the United States. The excessive reliance on private automobiles in the United States
has caused many problems including congestion, greenhouse gas emission and air pollution,
especially in major cities. However, the US has experienced declines in reliance on driving over
the past several years. For instance, the percentage of Americans taking private automobiles to
work has declined to 85.8% in 2013, the lowest since 1990 (McKenzie, 2015). Some analysts
have called this “Peak Car,” noting that driving has peaked and then declined in many
industrialized nations (Goodwin & Van Dender, 2013). Such a trend, which predates the most
recent recession, has important policy implications. To date, there has been no research that has
carefully examined whether and how current demographic trends influence automobile travel. In
this research, I quantitatively test whether demographic trends in the United States are linked to
changes in traveling by cars using comprehensive data sets. Understanding the link from
demographics to automobile travel creates opportunities for policy makers to transform
American cities to be more sustainable and to more effectively predict future travel patterns
based on demographic trends.
Specifically, I will examine the impact of three major demographic shifts on the demand
for automobile travel: immigration, Millennials entering adulthood, and aging, following the
conceptual model demonstrated in Figure 1-1.
14

Figure 1-1 – Conceptual model of dissertation, modified from Mayer and Greenwood (1980)

2. Contributions to literature
This research challenges the conventional wisdom of travel demand analysis as
influenced by derived demand theory. The derived demand theory posits that traveling is the
means rather than the ends of various activities. Based on neoclassical microeconomics, this
theory argues that the demand for travel depends solely on the price of driving and an
individual’s income, controlling for other variables (Domencich & McFadden, 1975). In other
words, this school of thought posits that the demand for travel is purely an economic decision
made by symmetric, rational economic agents. However, as argued by Mokhtarian, Salomon,
and Redmond (2001) and Anable (2005), these assumptions of symmetry and rationality are
questionable.
Studying the demands of automobile travel from a demographic perspective, this research
confirms the aforementioned arguments made by Mokhtarian et al. (2001) and Anable (2005).
15

Specifically, this research identifies two demographic effects in the demand for automobile
travel: cohort effects and aging effects. The cohort effects are supported by demographic theory,
which has argued that people from different cohorts (or demographic groups) might have
different attitudes and preferences, which have been shaped in their important stages of life
(Glenn, 1980). More specifically, birth cohort effects are shaped by the collective memories in
people’s young adult or earlier ages. Immigration cohort effects are shaped during immigrants’
early years in their new nations. In addition, aging effects are formed by a combination of health
and psychological factors associated with aging. Both cohort effects and aging effects contest the
two aforementioned assumptions in the derived demand theory: symmetry and rationality. People
in different cohorts tend to have different patterns in automobile travel; this argument violates
the symmetry assumption. The decision makers are not completely rational in the economic
sense of being influenced only or importantly by prices and income; sociological, psychological
and health factors also influence the demands for automobile travel, and this dissertation gives
evidence of birth and immigration-entry cohorts that help shape travel behavior. In short, persons
born at different times, or who entered the United States in different decades, or who have
different health conditions, travel differently when faced with the same price of travel and
personal income levels.
The research adds to the small but growing literature on the influence of demographic
factors on travel demand in the United States. First, this research contributes to the on-going
debate on the decreased travel of the Millennial young adults in the late 2000s. Blumenberg,
Ralph, Smart, and Taylor (2016) argued that the Millennials did not behave differently from
earlier generations when socio-economic and life-cycle factors were controlled. In contrast,
McDonald (2015) demonstrated that 35-50% of the decreased driving of those young adults
16

could be explained by Millennial-specific factors. This research supports the argument of
McDonald (2015) by identifying cohort effects in automobile commuting and residential density
– automobile travel associations. Second, this study addresses methodological issues in existing
studies on immigrant travel. Heavily relying on cross-sectional datasets, these existing studies
(for instance, Tal and Handy (2010)) suffer from a strong assumption that there are no
immigration cohort effects. In other words, as argued by Myers (1999), such studies estimate the
impact of length of staying in the US by comparing the new immigrants against longer-settled
immigrants at the same year. Thus, they have to assume that the immigrants entering the US in
different years will follow the same path over time. Using the double-cohort methods invented
by Myers and Lee (1996), this research uses cohort-longitudinal datasets to independently track
each immigration cohort over time. Also, this study adds to the currently small literature on the
impact of aging on automobile travel. Specifically, this is the first research to my knowledge that
examines the impact of health conditions on automobile travel in transportation policy and
planning.
From the perspectives of practitioners, this study shows the limitations of making
universal policy recommendations for an unrealistic “average” economic agent (Boarnet,
Houston, Ferguson, & Spears, 2011). Practitioners in transportation policy and planning should
be aware of the heterogeneity among different demographic groups. The lack of studies
examining such heterogeneity reflects the limitations of existing datasets: under-sampling
immigrants, having irregular time intervals between surveys, and the absence of information on
health. The results of this study should encourage future collection of datasets with better quality
to overcome these limitations.

17

3. Structure of dissertation
This dissertation has five chapters. The contents of Chapters 2 through 5 are outlined
below:
Chapter 2 (the first essay) develops double-cohort models to study the impact of
demographic change on automobile commuting in Los Angeles County, California in 2000 and
2010. The models construct demographic cohorts by years of birth and immigration to study
commuting by driving alone, carpooling and either of the previous two (private auto). Using
public-use microdata from the 2000 Census and the 2009-2011 American Community Survey, I
find statistically significant effects on commuting mode choices for most cohorts in the years
2000 and 2010. Native-born younger generations were three percentage points less likely to
commute by private auto than were the older generations when reaching the same age levels.
New immigrants residing in the US for less than ten years became increasingly likely to
commute by private auto from 2000 to 2010. A simple projection shows that these two
aforementioned demographic changes, generational shifts and immigration assimilation, have
countervailing effects on future aggregated demand for commuting by private auto.
Chapter 3 (the second essay) investigates the relationships between the built
environment and automobile travel across generations. Recent studies have found that
Millennials drive less than earlier generations. Currently, a large share of Millennials live in
central cities, and research suggests that Millennials, like earlier generations, will move to the
suburbs as they enter their prime work and child-rearing years. Will Millennials who shun
driving now continue to drive less when they move to the suburbs? To answer this question, I use
nationwide travel surveys in the United States in 1995, 2001 and 2009 to examine the association
between population density and automobile travel for young adults and test if the association is
18

growing weaker over time. Controlling for socio-economic, vehicle ownership, lifecycle, year-
specific and regional-specific factors, the impact of residential density on personal vehicle
distance traveled and number of car trips was approximately 30 percent lower for the 16-28-year-
olds in 2009 than for the young adults in the same age in 1995. This suggests that the Millennials
might still drive less than previous generations when residing in less dense suburban
neighborhoods. There might be opportunities to pursue transportation policies that foster less
reliance on private automobiles in the “New Suburbs” in the near future.
Chapter 4 (the third essay) explores a currently understudied topic related to elderly
driving: the impacts of health conditions on driving for people over 65 years of age. The mass
retirement of the Baby Boomers in the near future will bring about a more aged population in the
United States. In order to better serve the mobility needs of the growing elderly population, we
need a better understanding of the travel behavior of the elderly. I construct datasets using the
health and mobility information in the most recent five waves (2006, 2008, 2010, 2012 and
2014) of the Health and Retirement Study (HRS), a national-level longitudinal dataset. I use
fixed effects logit regressions to control for personal attitudes towards driving and show that the
impact of overall self-rated health conditions on senior driving has a larger magnitude than the
influence of poverty status, family structure and residential patterns. Such impacts of health on
driving vary by racial profile, family structure and residential living patterns. In addition to
overall health, specific conditions including physical, cognitive and vision conditions impact
senior driving. The results imply that with the current dominance of private automobiles, the
current American transportation system will not meet the mobility needs of the elderly who will
start to drive less because of declining health.
19

Chapter 5 concludes the dissertation with a brief discussion on broad implications for
the scholarship and practice of transportation policy and planning.

4. References
Anable, J. (2005). ‘Complacent car addicts’ or ‘aspiring environmentalists’? Identifying travel
behaviour segments using attitude theory. Transport Policy, 12(1), 65-78.
Blumenberg, E., Ralph, K., Smart, M., & Taylor, B. D. (2016). Who knows about kids these
days? Analyzing the determinants of youth and adult mobility in the U.S. between 1990
and 2009. Transportation Research Part A: Policy and Practice, 93, 39-54.
Boarnet, M. G., Houston, D., Ferguson, G., & Spears, S. (2011). Land use and vehicle miles of
travel in the climate change debate: getting smarter than your average bear. In Y.-H.
Hong & G. Ingram (Eds.), Climate Change and Land Policies (pp. 151-187). Cambridge,
MA: Lincoln Institute of Land Policy.
Domencich, T. A., & McFadden, D. (1975). Urban travel demand-a behavioral analysis.
Amsterdam: North-Holland Publishing Company.
Glenn, N. D. (1980). Values, attitudes, and beliefs Constancy and change in human development
(pp. 596-640). Cambridge, MA: Harvard University Press.
Goodwin, P., & Van Dender, K. (2013). ‘Peak Car’—Themes and Issues. Transport Reviews,
33(3), 243-254.
Mayer, R. R., & Greenwood, E. (1980). The design of social policy research: Prentice Hall.
McDonald, N. C. (2015). Are Millennials Really the “Go-Nowhere” Generation? Journal of the
American Planning Association, 81(2), 90-103.
20

McKenzie, B. (2015). Who Drives to Work? Commuting by Automobile in the United States:
2013 American Community Survey Reports. Washington D.C.
Mokhtarian, P. L., Salomon, I., & Redmond, L. S. (2001). Understanding the demand for travel:
It's not purely'derived'. Innovation: The European Journal of Social Science Research,
14(4), 355-380.
Myers, D. (1999). Cohort longitudinal estimation of housing careers. Housing Studies, 14(4),
473-490.
Myers, D., & Lee, S. W. (1996). Immigration cohorts and residential overcrowding in southern
California. Demography, 33(1), 51-65.
Tal, G., & Handy, S. (2010). Travel behavior of immigrants: An analysis of the 2001 National
Household Transportation Survey. Transport Policy, 17(2), 85-93.

21

Chapter 2. A Double-Cohort Analysis for Commute Mode Choice in Los Angeles County,
California

1. Introduction
This essay develops double-cohort models to study the journey-to-work (or commute)
mode choice of different demographic groups, defined by their years of birth and immigration.
More comprehensive demographic models can help better analyze the impact of demographic
shifts on the recent trend of reduced automobility in developed economies. Data from decennial
censuses and the American Community Surveys (ACS) show that the share of American workers
commuting by private auto started to decline beginning in 2006, the first year the ACS was
available (McKenzie, 2015) (Figure 2-1). Since 1970, Los Angeles County, the “Car Capital” in
pop culture (e.g. (Haggis, 2004)), has steadily decreased its share of workers commuting by
private auto (McKenzie, 2015). Recent studies also found declines in other measures of
automobility, such as vehicle ownership and use, both in the US and in other developed
economies (Dargay, Gately, & Sommer, 2007; Millard‐Ball & Schipper, 2011). Does this signal
the end of the ever-increasing demand for private automobiles, as the “Peak Car” hypothesis
(Goodwin & Van Dender, 2013) has proposed?

22

Figure 2-1 – Share of commuters taking private auto to work in the US and Los Angeles County, 1970-2014
(Data source: Census 1970-2000 and ACS one-year sample 2006-2014)

Better analyses on the determinants of such recent trends on automobility are needed.
Economic factors have long been proved to be a substantial factor determining the demand for
driving (McFadden, 1973). In other words, the recent recession caused people to drive less
because of their reduced income. In addition, the impact of recent demographic changes on
automobile travel has drawn the attention of the recent literature on transportation policy and
planning. Studies have confirmed that Millennials in the US were less likely to drive compared
to earlier generations when reaching the same age level (Blumenberg, Ralph, Smart, & Taylor,
2016; McDonald, 2015). However, few studies have simultaneously examined the impact on
driving from various demographic changes, including the assimilation of immigrants.
Researchers and practitioners need a more comprehensive framework to analyze the
impact of demographics on travel demand. Such complicated models are difficult to build,
especially in smaller geographical scales. Models incorporating generational shifts and
immigration assimilation usually require a much larger sample size than typical travel demand
models, since they need a sufficient number of immigrants to gain enough statistical power. Most
60%
70%
80%
90%
100%
1960 1970 1980 1990 2000 2010 2020
United States
Los Angeles County
23

of the existing travel surveys in the United States, such as early rounds of the National
Household Travel Survey (NHTS), do not have enough sample size to support such models. In
contrast, the Integrated Public Use Microdata Series (IPUMS) for the censuses and ACS, popular
in activity-based travel forecasting (Tierney, 2012), can help to build such demographic models.
The IPUMS is one of the most representative public-use datasets related to transportation, with
five-percent samples for decennial censuses and three-percent samples for ACS. With such a
large sample size, IPUMS can help to build complicated demographic models at even a regional
scale. IPUMS also has better representativeness than does the NHTS since the latter oversamples
the native-born population (Chatman & Klein, 2009).
Commuting makes up a third of the total vehicle miles traveled (VMT) for households in
major metropolitan areas (Boarnet, 2011). That makes commute mode choice an important
indicator of automobility. In this essay I examine commute mode choices of different
demographic groups in Los Angeles County, California in 2000 and 2010, using double-cohort
models. I begin below by reviewing theories and empirical evidence on the impact of
demographics on travel. I introduce the area of study as well as the sources of data and methods
of analysis. I then present the results for the empirical models and trajectory charts. I further
discuss the implication of the models by showing the countervailing effects of different
demographic changes on commute by private auto in the near future. I conclude in the final
section.

2. Literature
A cohort, or demographic group, is a group of people who have experienced similar
socio-economic environments during important stages of life and who share collective memory
24

shaped by major events in history (Ryder, 1965). Sociologists believe that people in different
cohorts can have different attitudes or lifestyles (Strauss & Howe, 1997). A cohort analysis
examines such differences among different cohorts, or simply “cohort effects” on various topics.
People in different cohorts might have different patterns of travel demand because of their
differences in activity patterns, cognitive maps and attitudes and beliefs (Tal & Handy, 2010). In
this study, I follow Myers and Lee (1996) and define cohorts by both decade of birth (birth
cohort) and decade of immigrating into the US (immigration cohort) to measure the travel impact
of two major demographic changes: generational shifts and immigration assimilation.
Generational shifts, or birth-cohort effects, refer to the differences in lifestyles, attitudes
and beliefs among people in different generations when observed at the same age. Krosnick and
Alwin (1989) reviewed two hypotheses explaining generational differences. The first, the
“impressionable years hypothesis”, argues that people shape their main attitudes and lifestyles in
the age range of 18-25 and will not change over their later stages of life (Mannheim, 1952). The
second, the “increasing persistence hypothesis,” proposes that people have flexible attitudes and
lifestyles when they are young, while the flexibility gradually declines as they get older (Glenn,
1980). Both hypotheses emphasize the importance of a person’s young adulthood in shaping
lifestyles and attitudes, from both a biological perspective (decline of energy of brain tissues
after age 25) and a sociological perspective (developed and fixed social networks after age 25)
(Glenn, 1980). Recent studies in various fields support such importance of early adulthood in
generational differences in people’s political party affiliations (Lewis-Beck, 2009), views on
social policies (Alesina & Fuchs-Schündeln, 2007) and consumption patterns (Giuliano &
Spilimbergo, 2009). In the field of transportation policy, recent studies (Blumenberg et al., 2016;
Garikapati, Pendyala, Morris, Mokhtarian, & McDonald, 2016; McDonald, 2015) have found
25

that Millennials (born after 1980) have lower auto ownership and take fewer trips than did earlier
generations, both inside and outside the US.
Immigration assimilation is the process for people in recent immigration cohorts to adopt
the general lifestyles and beliefs of the target nation. The current literature discusses three major
aspects of immigration assimilation in the US (Waters & Jiménez, 2005). Socio-economic
assimilation argues that new immigrants might start their careers in the US in lower-skilled jobs
with lower income, but will achieve better job prospects with higher income when they stay in
the US longer (Chiswick, 1978). Spatial assimilation shows that immigrants are less likely to live
in ethnic enclaves as they live longer in the US and have higher socio-economic status (Massey,
1985). In addition, linguistic assimilation indicates that as immigrants from non-English
speaking backgrounds stay longer in the US, they have better fluency in English (Bean &
Stevens, 2003). These assimilation processes can all contribute to assimilation in travel behavior.
Improved socio-economic status, moving to a suburban neighborhood and having a wider social
network with better English fluency can all generate a higher demand for automobile travel. As
recent studies have shown, new immigrants tend to stay in ethnic enclaves (Massey, 1985), rely
on carpooling (Blumenberg & Smart, 2014), take public transportation (Rosenbloom & Fielding,
1998) and drive less (Tal & Handy, 2010), and all these effects are largest during their first
decade in the US.
While the impacts of generational shifts and immigrant assimilation are most well-
studied in sociology and political science, the literature on these topics has begun in the field of
transportation policy and planning. The growing body of literature utilizes various methods
including qualitative studies (e.g. Cope and Lee (2016)), cross tabulations (e.g. Delbosc (2016),
Kuhnimhof et al. (2012)) and behavioral models (e.g. McDonald (2015) and Blumenberg and
26

Smart (2014)). This study plans to contribute to this literature in the following ways. First, I
develop a comprehensive framework to incorporate the effects of both generational shifts and
immigration assimilation on travel behavior. Previous studies have focused solely on one of
these two effects in isolation from the other. Second, I innovate by using double-cohort models
to capture full cohort effects. Models based on cross-sectional data or repeated cross-sectional
data with irregular time intervals might not be able to capture the full cohort effects. For
instance, the time intervals among the most recent three NHTS (1995, 2001 and 2009) are six
and eight years, making cohort models using such datasets difficult to create age group intervals
to match them. Third, I add to the limited literature studying travel behavior using IPUMS, one
of the most representative public-use datasets related to transportation.

3. Data and methods
3.1. Area of study and data source
The area of study in this research is Los Angeles County (LA County hereafter),
California. According to the most recent census in 2010, the population of the county is
9,818,605, the most populous statewide. LA County is racially diverse, with only 27.8 percent of
its population non-Hispanic white. It is also diverse in community types and municipal-level
governance, with 88 municipalities and 53 non-incorporated areas. Among these 141
communities are well-known cities such as Los Angeles, Santa Monica and Beverly Hills. As a
major immigration gateway to the US (Singer, 2003), LA County has a large foreign-born
population. As Table 2-1 shows, from 2000 to 2010, the number of the employed in LA County
grew from 3.9 million to 4.3 million. In addition, the share of foreign-born in the workforce also
grew from 43.6% to 46.6%. Also, the share of foreign-born in the workforce is higher than the
27

share of foreign-born in the whole population, indicating that a higher percentage of foreign-born
than native-born employees.

Table 2-1 – Total population and number of the employed in Los Angeles County, 2000 and 2010
(data: 2000 Census and 2009-2011 ACS)
year
employed total population
native-born foreign-born total native-born foreign-born total
2000
n 2,183,847 1,669,266 3,853,113 5,994,794 3,529,045 9,523,839
% 56.7% 43.3% 100% 62.9% 37.1% 100%
2010
n 2,318,011 2,009,977 4,327,988 6,274,011 3,571,561 9,845,572
% 53.6% 46.4% 100% 63.7% 36.3% 100%

Figure 2-2 shows the decline in the share of commuters in LA County who take a private
auto to work. This figure can be further divided into a slightly increasing share of those who
drive alone to work and a decreasing share of those who carpool to work. Specifically, the share
of those who drive alone to work increased from 70.1% in 1990 to 70.4% in 2000 and more than
73% in 2010, while the share of those who carpool to work decreased from 15.5% in 1990 to
15.0% in 2000 and to around 10% in 2010.

Figure 2-2 – Share of commuters by private auto, driving alone and carpooling in Los Angeles County, 1970-
2014 (data source: censuses and ACS since 2006)
0%
20%
40%
60%
80%
100%
1960 1970 1980 1990 2000 2010 2020
private auto
driving alone
carpooling
28

The dataset for the double-cohort models comes from the unweighted five-percent
sample from the 2000 Census and the three-percent sample from the 2009-2011 ACS of the
IPUMS (Ruggles et al., 2010.). Here I use the three-year ACS to proxy the year 2010 since the
Census Bureau excluded questions on commuting mode choice in the 2010 Census. The dataset
includes commute mode choice of the week before the survey for each worker, as well as
information on each commuter’s socio-economic characteristics and residential density at the
public use micro data (PUMA) level. I was not able to include censuses in 1990 or earlier in my
sample since the lack residential density at a similar scale.

3.2. Double cohort model
The double-cohort models, invented by Myers and Lee (1996), aim to study three mode
choices in commuting: driving alone, carpooling and either of the previous two (in other words,
by private auto). As discussed in the previous section, “cohort” refers to a demographic group
whose lifestyles and views are shaped by similar collective memories. The double-cohort
methods construct demographic groups in two dimensions: the decade when a person is born and
the decade when a person immigrated to the United States. Lifestyles for people in different birth
cohorts (or generations) are shaped by the major events of history such as the Civil Rights
Movement, Globalization and the wide application of information and communication
technologies (ICT) (Taylor, 2014). Similarly, lifestyles for people in different immigration
cohorts (or decades immigrated to US) are reshaped by their increasing familiarity with the
society they immigrate to and by their rebuilt social network. In my double-cohort models, I
group the birth cohorts as those born in the following time frames: 1976 – 1985 (or “15-24 in
29

2000”), 1966-1975 (“25-34 in 2000”), 1956-1965 (“35-44 in 2000”) and 1946-1955 (“45-54 in
2000”). Similarly, I group the immigration cohorts as those who immigrate into the United States
in the following time frames: before 1980, in 1980s and in 1990s. The native-born are also
treated as a reference immigration cohort. In other words, each individual in my sample belongs
to both a birth cohort and an immigration cohort.
By matching the time intervals of age and immigration cohorts with the ten-year time
interval between surveys, the double-cohort models are able to fully capture the effects of both
cohorts in one model, controlling for other factors. The functional form of the models is as
below:

𝑦 = 𝛽 + 𝛽 𝑇 + 𝐁𝐂 𝐢 𝛃 𝟐 + 𝐈𝐂 𝐢 𝛃 𝟑 + 𝑇 𝐁𝐂 𝐢 𝛃 𝟒 + 𝑇 𝐈𝐂 𝐢 𝛃 𝟓 + 𝐗 𝐢𝐭 𝛃 𝟔 + 𝜀 . (2-1)

In Formula (2-1) above, yit indicates the commuting mode choice for commuter i in year
t. Since here I study three different commuting modes, I estimate three separate models for
commuting by private auto, driving alone and carpooling. The dependent variable yit will be one
if the survey respondent commutes by the mode of interest, and zero otherwise. Thus, all three
models will be logit models. Tt is a dummy variable for time effects, which equals one if the
survey year is 2010, with year 2000 as a reference term. Column vector BCi includes birth cohort
dummy variables: age 25-34 in 2000, age 35-44 in 2000 and age 45-54 in 2000, with age 15-24
in 2000 as a reference term. Similarly, column vector ICi includes immigration cohort dummy
variables: entering the US before 1980, in 1980s and in 1990s, with native born as a reference
term. Column vector TtBCi refers to the age effect specific to each birth cohort, such as change
of commuting mode for the “age 25-34 in 2000” age cohort from 2000 (at age 25-34) to 2010 (at
30

age 35-44). Similarly, column vector TtICi indicates the assimilation effect specific to each
immigration cohort (relative to the corresponding native-born cohort), such as change of
commuting mode choice for the “immigrated to US in 1980s” immigration cohort between 2000
(having stayed in the US for less than ten years) and 2010 (having stayed in the US for ten to
twenty years). Such specifications indicate two assumptions. First, these models forgo the period
effects and assume that they are the same across all cohorts. Age, period and cohort effects are
collinear, this assumption allows me to focus on the birth and immigration cohort effects (see
Myers and Lee (1996) for a detailed discussion). Second, for simplicity, I exclude three-way
interaction terms (among birth cohort, immigration cohort and year) and assume that the age
effects are the same for all immigration cohorts.
Column vector Xit includes other confounding factors associated with commuting mode
choice, including socio-economic variables such as educational attainment, household income (in
2011 US Dollars) and gender. Xit also includes information on homeownership, number of
vehicles per person and residential density at the PUMA level. As mentioned in the previous
subsection, the sample comes from the unweighted IPUMS sample for 2000 Census and 2009-
2011 ACS in Los Angeles County, with 259,015 observations. A complete list of the descriptive
statistics of the dependent and independent variables is shown in Table 2-2.

31

Table 2-2 – Descriptive statistics of variables used in the double-cohort modeling
variable mean SD variable mean SD
Commute by private auto 0.858 0.349 Education attainment

Drive alone to work 0.719 0.449 less than 9 yrs education reference

Carpool to work 0.139 0.346 9-11 yrs education 0.069 0.254
Census Year

HS graduate/some college 0.515 0.500

2000 reference 4 or more years college 0.311 0.463

2011 0.374 0.484 Household income (in 2011 USD)

Birth Cohort

less than $15,000 reference

age 15-24 in 2000 reference $15,001 - $35,000 0.122 0.328

age 25-34 in 2000 0.290 0.454 $35,001 - $55,000 0.159 0.366

age 35-44 in 2000 0.296 0.457 $55,001 - $75,000 0.148 0.355

age 45-54 in 2000 0.220 0.415 $75,001 - $100,000 0.158 0.365
Ten-year period effect

more than $100,000 0.378 0.485

25-34 to 35-44 0.103 0.304 Renter 0.461 0.498

35-44 to 45-54 0.107 0.309 Female 0.456 0.498

45-54 to 55-64 0.073 0.260 Ethnicity

Immigration cohort

Non-Latino White reference

native-born reference Non-Latino Black 0.073 0.260

pre-1980s 0.132 0.339 Non-Latino Asian/PI 0.141 0.348

1980s 0.175 0.380 Latino 0.413 0.492

1990s 0.145 0.352 Other 0.026 0.160
Ten-year duration effect for
immigrants

Number of vehicles per person 0.678 0.424

pre-1980 immigrants 0.046 0.209 Residential density (1k per sq mi) 8.588 5.691

1980s immigrants 0.064 0.245

1990s immigrants 0.059 0.236 N 259,015

3.3. Track cohorts over time
Other than the aforementioned multivariate expression, double-cohort models also have
the graphic expression using trajectory charts (Myers & Lee, 1996). Trajectory charts can
visually show how people in different cohorts behave differently over time. Most demographic
models have complicated combinations of cohort membership, time and age variables,
complicating the use of marginal effects to investigate differences among cohorts over time. For
instance, if we would like to examine differences in commuting by private auto for the age group
25-34 in 2000 and in 2010, we should sum the marginal effects of variables “year 2010” and
32

“age group 25-34 in 2000” – “year 2010” interaction. Trajectory charts, in contrast, provide only
one composite effect: expected probabilities of the mode of interest for the age group in 2000
and 2010.
After estimating the double-cohort models in Formula (2-1), I predict the logit value of
the probability of the mode choice of interest (𝑌 ) for each birth cohort (BCc) and immigration
cohort (ICc) membership combination c at time point t, with the same average 𝐗 variables across
cohorts (shown in Table 2-2), following Formula (2-2):

𝑙𝑜𝑔𝑖𝑡 [𝑃 (𝑦 )|𝑇 , 𝐁𝐂 𝐜 , 𝐈𝐂 𝐜 , 𝐗 ] = 𝑌
= 𝛽 + 𝛽 𝑇 + 𝐁𝐂 𝐜 𝛃 𝟐 + 𝐈𝐂 𝐜 𝛃 𝟑 + 𝑇 𝐁𝐂 𝐜 𝛃 𝟒 + 𝑇 𝐈𝐂 𝐜 𝛃 𝟓 + 𝐗 𝛃 𝟔 . (2-2)

Following the definition of logit shown in Formula (2-3), I then transform 𝑌 to the
corresponding probability of the mode of interest 𝑃 (𝑦 ) using Formula (2-4).

𝑌 = 𝑙𝑜𝑔𝑖𝑡 [𝑃 (𝑦 )|𝑇 , 𝐁𝐂 𝐜 , 𝐈𝐂 𝐜 , 𝐗 ] = ln 𝑃 (𝑦 )
1 − 𝑃 (𝑦 )
; (2 − 3)
𝑃 (𝑦 ) =
𝑒 1 + 𝑒 . (2 − 4)

I then put the values of 𝑃 (𝑦 ) into trajectory charts to show commuting mode choices
for each cohort over time. Figure 2-3 serves as an illustration of commuting by private auto for
the native-born, 25-34 in 2000 cohort and native-born, 35-44 in 2000 cohort. After controlling
for all the socio-economic and built environment variables as sample averages in 𝐗 , I estimate
the expected probability of commuting by private auto for these two cohorts in 2000 and 2010.
33

Then I show those dots in the trajectory chart and use lines to connect the dots belonging to the
same cohorts to indicate temporal changes for each cohort. Such trajectory charts illustrate
various effects across different age groups, cohorts and time points. They show at a specific time
point (year 2000), how expected probabilities of commuting by private auto differ across
different age groups (91.52% vs. 91.08%). They also show at a specific age level (35-44), how
expected probabilities of taking private auto to work differ across different cohorts (88.25% vs.
91.08%), and how for a specific cohort (25-34 in 2000) expected probabilities of this mode differ
at different time points (91.52% vs. 88.25%), holding the variables in vector X consistent.

Figure 2-3 – Illustrative trajectory chart for 25-34 in year 2000 and 35-44 in year 2000, native-born
(Note: white and black dots indicate 2000 and 2010, respectively)

4. Results
4.1. Model estimation results
The outputs of double-cohort models on commuting by private auto, driving and
carpooling are displayed in Table 2-3, Table 2-4 and Table 2-5, respectively. Here is a summary
34

of the findings: First, most variables for cohort membership, time and the interactions of the two
are statistically significant at the 5 percent level. This shows that the differences on automobile
commuting across different cohorts does not solely come from socio-economic and built
environment factors. Second, socio-economic factors influence commuting mode choices. For
instance, compared to males, females have a lower propensity to commute by private auto and to
drive alone. Females are also more likely to carpool to work. Higher income levels are associated
with a higher propensity of commuting by private auto, driving alone and carpooling. Also,
higher educational levels are associated with a higher propensity of commuting by private auto
and driving alone but with a lower propensity of carpooling to work. In addition, there are
significant differences in commuting mode choice among ethnic groups. Third, built
environment and vehicle ownership have a small but significant effect on commute mode
choices, a result that agrees with the literature on land use and travel. Higher numbers of vehicles
per person in a household are associated with higher probabilities of commuting by private auto
and driving alone, but with lower probabilities of carpooling to work. Residential density at the
PUMA level negatively relates to the probability of commuting by all the three modes of interest.

35

Table 2-3 – Logistic double-cohort model estimates on automobile commuting (solo driving or carpooling)
variable coefficient SD variable coefficient SD
Intercept

0.355*** [0.041] Education attainment

Census Year

less than 9 yrs education reference

2000 reference

9-11 yrs education 0.170*** [0.024]

2011 0.085*** [0.028]

HS graduate/some
college
0.473*** [0.019]
Birth Cohort

4 or more years college 0.296*** [0.023]

age 15-24 in 2000 reference Household income (in 2011 USD)

age 25-34 in 2000 0.422*** [0.022]

less than $15,000 reference

age 35-44 in 2000 0.367*** [0.023]

$15,001 - $35,000 0.359*** [0.028]

age 45-54 in 2000 0.032 [0.025]

$35,001 - $55,000 0.574*** [0.028]
Ten-year period effect

$55,001 - $75,000 0.698*** [0.029]

25-34 to 35-44 -0.447*** [0.035]

$75,001 - $100,000 0.768*** [0.030]

35-44 to 45-54 -0.665*** [0.036]

more than $100,000 0.734*** [0.029]

45-54 to 55-64 -0.627*** [0.039] Renter

-0.419*** [0.014]
Immigration cohort

Female

-0.255*** [0.012]

native-born reference Ethnicity

pre-1980s 0.145*** [0.027]

Non-Latino White reference

1980s 0.014 [0.023]

Non-Latino Black 0.241*** [0.025]

1990s -0.622*** [0.022]

Non-Latino Asian/PI 0.612*** [0.024]
Ten-year duration effect for immigrants

Latino 0.335*** [0.018]

pre-1980 immigrants 0.098** [0.042]

Other 0.158*** [0.039]

1980s immigrants 0.206*** [0.034] Number of vehicles per person 1.406*** [0.020]

1990s immigrants 0.628*** [0.033] Residential density (1k per sq mi) -0.032*** [0.001]
N 259,015 p-value of likelihood ratio test

<0.001
Note: *, ** and *** indicate p < 0.1, p < 0.05 and p < 0.01, respectively.

36

Table 2-4 – Logistic double-cohort model estimates on commuting by driving alone
variable coefficient SD variable coefficient SD
Intercept -0.425*** [0.035] Education attainment

Census Year

less than 9 yrs education
reference

2000
reference

9-11 yrs education 0.152*** [0.021]

2011 0.384*** [0.022]

HS graduate/some college 0.535*** [0.016]
Birth Cohort

4 or more years college 0.575*** [0.019]

age 15-24 in 2000
reference
Household income (in 2011 USD)

age 25-34 in 2000 0.419*** [0.017]

less than $15,000
reference

age 35-44 in 2000 0.430*** [0.018]

$15,001 - $35,000 0.241*** [0.026]

age 45-54 in 2000 0.230*** [0.019]

$35,001 - $55,000 0.353*** [0.026]
Ten-year period effect

$55,001 - $75,000 0.381*** [0.026]

25-34 to 35-44 -0.390*** [0.028]

$75,001 - $100,000 0.390*** [0.026]

35-44 to 45-54 -0.611*** [0.028]

more than $100,000 0.370*** [0.026]

45-54 to 55-64 -0.597*** [0.031] Renter -0.212*** [0.011]
Immigration cohort

Female -0.251*** [0.009]

native-born
reference
Ethnicity

pre-1980s -0.013 [0.019]

Non-Latino White
reference

1980s -0.151*** [0.017]

Non-Latino Black -0.003 [0.020]

1990s -0.658*** [0.018]

Non-Latino Asian/PI 0.126*** [0.017]
Ten-year duration effect for immigrants

Latino -0.052*** [0.014]

pre-1980 immigrants 0.149*** [0.031]

Other -0.029 [0.030]

1980s immigrants 0.188*** [0.026] Number of vehicles per person 1.196*** [0.015]

1990s immigrants 0.511*** [0.027] Residential density (1k per sq mi) -0.017*** [0.001]
N 259,015 p-value of likelihood ratio test

<0.001
Note: *, ** and *** indicate p < 0.1, p < 0.05 and p < 0.01, respectively.

37

Table 2-5 – Logistic double-cohort model estimates on commuting by carpooling
variable coefficient SD variable coefficient SD
Intercept

-1.470*** [0.046] Education attainment

Census Year

less than 9 yrs education

2000

9-11 yrs education -0.020 [0.024]

2011 -0.513*** [0.029]

HS graduate/some
college
-0.303*** [0.019]
Birth Cohort

4 or more years college -0.557*** [0.023]

age 15-24 in 2000

Household income (in 2011 USD)

age 25-34 in 2000 -0.206*** [0.020]

less than $15,000

age 35-44 in 2000 -0.268*** [0.021]

$15,001 - $35,000 0.166*** [0.036]

age 45-54 in 2000 -0.267*** [0.024]

$35,001 - $55,000 0.262*** [0.035]
Ten-year period effect

$55,001 - $75,000 0.337*** [0.035]

25-34 to 35-44 0.132*** [0.036]

$75,001 - $100,000 0.372*** [0.036]

35-44 to 45-54 0.255*** [0.036]

more than $100,000 0.358*** [0.035]

45-54 to 55-64 0.233*** [0.041] Renter

-0.055*** [0.013]
Immigration cohort

Female

0.153*** [0.012]

native-born

Ethnicity

pre-1980s 0.129*** [0.023]

Non-Latino White

1980s 0.246*** [0.021]

Non-Latino Black 0.281*** [0.026]

1990s 0.301*** [0.022]

Non-Latino Asian/PI 0.383*** [0.021]
Ten-year duration effect for immigrants

Latino 0.428*** [0.018]

pre-1980 immigrants -0.026 [0.040]

Other 0.241*** [0.039]

1980s immigrants -0.013 [0.033] Number of vehicles per person -0.598*** [0.018]

1990s immigrants 0.003 [0.034] Residential density (1k per sq mi) -0.012*** [0.001]
N 259,015 p-value of likelihood ratio test

<0.001
Note: *, ** and *** indicate p < 0.1, p < 0.05 and p < 0.01, respectively.

4.2. Trajectory charts
Trajectory charts show different expected probabilities of commuting in the modes of
interest in a visually intuitive way. Figure 2-4(a-c) show the expected probability of commuting
by private auto, driving alone and carpooling for the native-born in LA County in 2000 and
2010. Figure 2-4(a) shows that except for the youngest age cohort (15-24 in 2000), all other birth
cohorts became less likely to commute by private auto from 2000 to 2010. Specifically, the
expected probabilities of commuting by private auto for the younger age cohorts are three
38

percentage points lower than for the older age cohorts at the same age levels. Figure 2-4(b) and
Figure 2-4(c) indicate that those differences for commuting by private auto derive mostly from
the three-percent differences for carpooling to work. There are no differences in expected
probabilities of driving alone to work between birth cohorts when reaching the same age levels
for the native-born. In contrast, claims based on one cross-sectional dataset may be misleading.
Take Figure 2-4(c) as an example, if we only examine the predicted probability of driving alone
to work in 2010 (i.e. the black dots), we might have an inaccurate conclusion that the likelihood
of carpooling to work for the native-born will not change much as they age.

39

Figure 2-4 – Trajectory charts for predicted probability (%) of commuting mode choice
(Note: white and black dots indicate 2000 and 2010, respectively)
40

The patterns of commuting mode choice for immigrants who have been residing in the
US for more than ten years are very similar to those of the native-born. Such patterns show the
assimilation process of new immigrants as to how they commute. Figure 2-4(d-i) are the
trajectory charts on commuting mode choices for immigrants who entered the US before 1980
and in 1980s. If we assume that all these foreign-born Angelenos have stayed in the United
States since their first entries, they have stayed in the US for more than 20 years and ten to 20
years, respectively. Comparing Figure 2-4(d-i) with Figure 2-4(a-c), we see that the expected
likelihood of commuting by private auto for each birth cohort of these longer-settled immigrants
are comparable to those of their native-born counterparts. However, these two longer-settled
immigration cohorts display slightly different patterns from the native-born in two ways. First,
compared to the native born in each birth cohort, pre-1980 immigrants are slightly more likely to
drive alone to work, while the 1980s immigrants are slightly more likely to carpool to work.
Second, the birth cohorts “25-34 in 2000” for pre-1980 and 1980s immigrants display an
increased likelihood to drive alone to work from 2000 to 2010, which contrasts with their native-
born counterpart.
New immigrants staying the US for less than ten years show strong assimilation effects in
commuting mode choice from 2000 to 2010. For their first ten years of residence in the US, new
immigrants have lower expected probabilities of commuting by private auto than do the native-
born and longer-settled immigrants. As indicated in Figure 2-4(j-l), the expected probabilities for
the 1990s immigrants to drive alone to work are around ten percentage points lower than those
for the 1980s immigrants in 2000. In addition, the expected probabilities for carpooling to work
for the 1990s immigrants are one percentage point higher than those of the 1980s immigrants in
2000. However, from 2000 to 2010, the patterns of commuting mode choice for the 1990s
41

immigrants have become similar to longer-settled immigrants and the native-born. During this
ten-year period, the expected probabilities of driving alone to work and taking private auto to
work for the 1990s immigrants increased by roughly ten percentage points and three percentage
points, respectively. Their expected probabilities of carpooling to work have decreased by
around seven percentage points. Such assimilation effects in commuting mode choices match
similar findings in studies of non-work travel ((Rosenbloom & Fielding, 1998), (Tal & Handy,
2010)).
In sum, the trajectory charts show the expected probabilities of three commuting mode
choices for different cohorts in 2000 and 2010, controlling for other confounding factors. The
charts show that in both years, longer-settled immigrants and the native-born have similar
patterns in commuting mode choices. During this ten-year period, people in younger birth
cohorts (especially 15-24 in 2000) became more likely to commute by private auto and by
driving alone, while those in older birth cohorts became less likely to do so. In contrast, new
immigrants having stayed in the US for less than ten years are less likely to commute by private
auto and by driving alone compared to longer-settled immigrants and the native born, while the
probabilities to commute by these two modes have increased significantly from 2000 to 2010. In
addition, people from all birth and immigration cohorts became less likely to carpool to work
during these ten years. Such decreased probability of carpooling has contributed to most of the
decreased likelihood of commute by private auto.

5. Discussion: impact of demographic changes on future automobile commuting
The double-cohort models presented in this study indicate the countervailing impacts of
the two major demographic changes, generational shifts and immigration assimilation, on the
42

share of automobile commuters in LA County in 2020. To examine these effects, I first project
the number of workers in each cohort in 2020 by assuming that the percentages employed for
each cohort will remain the same from 2010 to 2020. Under this assumption, I estimate the size
of the workforce for each cohort in 2020 by multiplying the 2010 share of employment by the
2020 projected population. For each cohort, the 2010 shares of employment come from the 2009-
2011 ACS and the 2020 projected population comes from Myers and Pitkin (2013). Table 2-6
shows the share of workers for each cohort in 2010, the projected share of workers for each
cohort in 2020, and the percentage point changes for these shares from 2010 to 2020.

Table 2-6 – Change of share of population by birth and immigration cohorts, 2010-2020 (projected)

Age
native-
born
foreign-born
sum
more than
20 years
10-20
years
less than
10 years
(a) share
of total
employed,
2010
15-24 9.30% 0.30% 1.10% 1.70% 12.40%
25-34 14.90% 2.60% 3.70% 4.20% 25.40%
35-44 11.10% 6.50% 4.80% 2.50% 24.90%
45-54 10.80% 8.50% 2.50% 1.40% 23.20%
55-64 7.20% 5.40% 1.00% 0.50% 14.10%
sum 53.30% 23.30% 13.10% 10.30% 100.00%
(b)
projected
share of
total
employed,
2020
15-24 7.70% 0.30% 0.70% 1.00% 9.70%
25-34 20.80% 2.10% 2.50% 2.40% 27.80%
35-44 13.50% 5.40% 4.00% 1.50% 24.40%
45-54 9.30% 9.50% 2.20% 0.90% 22.00%
55-64 7.40% 7.40% 1.00% 0.30% 16.10%
sum 58.80% 24.70% 10.40% 6.20% 100.00%
(c)
percentage
point
change:
2010-2020
15-24 -1.60% 0.00% -0.30% -0.70% -2.70%
25-34 5.90% -0.50% -1.20% -1.80% 2.30%
35-44 2.40% -1.10% -0.80% -0.90% -0.50%
45-54 -1.40% 1.00% -0.30% -0.50% -1.20%
55-64 0.30% 2.00% 0.00% -0.20% 2.00%
sum 5.40% 1.40% -2.70% -4.10% 0.00%

43

Table 2-6(c) infers different impacts from generational shifts and immigration
assimilation on the changes of the share of private auto commuters in Los Angeles County from
2010 to 2020. On one hand, there will be a higher share of young native-born workers in the age
range of 25-44 from 2010 to 2020. As shown in the trajectory chart in Figure 2-4(a), these
workers are less likely to commute by private auto than those in older age cohorts at the same
age levels. If we assume this pattern will persist in 2020, we can expect a lower share of workers
commuting by private auto due to such generational shift. On the other hand, there will be a
lower share of new immigrant workers having stayed in the US for less than ten years from 2010
to 2020. As shown in the trajectory chart in Figure 2-4(j), these new immigrant workers will be
the least likely to commute by private auto, while longer-settled immigrant workers will be more
likely to do so. If there is a more assimilated immigrant workforce in Los Angeles County in
2020, we can expect a higher share of workers commuting by private auto.
Although clearly projecting the direction of the composite of these two effects is beyond
the scope of this article, the previous discussion still has strong implications for transportation
policy and planning. Transportation policy makers and planners should be aware of these two
demographic changes in Los Angeles County and elsewhere in the near future: a growing share
of younger workers and a longer-settled, more assimilated immigrant workforce. Travel forecast
models should consider such changes and incorporate their impacts in future travel demand in
order to produce more accurate predictions.

6. Conclusion
This essay develops double-cohort models to study the commuting mode choices across
different demographic groups, defined by their years of birth and immigration, in LA County
44

from 2000 to 2010. The recent declines in automobility of the developed economies raise the
questions as to whether they have reached their “Peak Car.” This study tries to extend the
discussion by examining how demographic shifts have contributed to the current trends using
more comprehensive demographic models. It also adds to a relatively small literature on the use
IPUMS datasets in transportation policy research. Specifically, I develop the double-cohort
models using the IMPUS data for the 2000 Census and 2009-2011 ACS. The cohorts are defined
by two dimensions, birth-cohorts referring to the birth year and immigration cohorts referring to
the decade in which immigration to the US occurred. With the individual as the observational
unit, I estimated double-cohort models for commuting by driving alone, by carpooling and by
private auto (in either of the first two modes).
The outputs of these models show: first, there are statistically significant differences in
commuting mode choices across different cohorts at different time points, although some
magnitudes might be small; second, socio-economic factors and the number of vehicles are
important predictors of commuting mode choices; third, higher residential density at the PUMA
level is associated with lower probabilities of commuting by private auto, driving alone and
carpooling. In order to show commuting mode choices for different cohorts at various time
points, I use trajectory charts to show the expected probabilities on certain commuting modes,
controlling for other non-cohort factors. The trajectory charts show that longer-settled
immigrants and the native-born had similar patterns of commuting mode choices in 2000 and
2010. In contrast, new immigrants had rather low expected probabilities of commuting by private
auto and driving alone in 2000, and showed strong assimilation effects over the next ten years.
For the native-born, when reaching the same age, younger generations are roughly three percent
45

less likely to commute by private auto or by carpooling than are the older ones. In addition, there
was a universal decline in carpooling to work in all cohorts from 2000 to 2010.
These double-cohort models also indicate countervailing effects of generational shifts and
immigration assimilation on future demand of automobile commute in LA County. Using
IPUMS data in 2009-2011 ACS and population projection for 2020 by Myers and Pitkin (2013),
I projected a growing share of young native-born workers and a more assimilated immigrant
workforce in 2020. If the commuting patterns for these cohorts persist over time, a younger and
less auto-dependent native-born workforce favors a less auto-oriented LA County in the future;
and a more assimilated and more auto-dependent foreign-born workforce favors a more auto-
oriented one. In order to make smarter transportation policies for the near future, researchers and
practitioners should try to meet the potential demands for alternative commuting modes for the
younger generations and be aware of a more assimilated immigrant body.

7. References
Alesina, A., & Fuchs-Schündeln, N. (2007). Good-Bye Lenin (or Not?): The Effect of
Communism on People's Preferences. The american economic review, 97(4), 1507-1528.
Bean, F. D., & Stevens, G. (2003). America's Newcomers and the Dynamics of Diversity: Russell
Sage Foundation.
Blumenberg, E., Ralph, K., Smart, M., & Taylor, B. D. (2016). Who knows about kids these
days? Analyzing the determinants of youth and adult mobility in the U.S. between 1990
and 2009. Transportation Research Part A: Policy and Practice, 93, 39-54.
Blumenberg, E., & Smart, M. (2014). Brother can you Spare a Ride? Carpooling in Immigrant
Neighbourhoods. Urban Studies, 51(9), 1871-1890.
46

Boarnet, M. G. (2011). A broader context for land use and travel behavior, and a research
agenda. Journal of the American Planning Association, 77(3), 197-213.
Chatman, D. G., & Klein, N. (2009). Immigrants and Travel Demand in the United States
Implications for Transportation Policy and Future Research. Public Works Management
& Policy, 13(4), 312-327.
Chiswick, B. R. (1978). The effect of Americanization on the earnings of foreign-born men. The
journal of political economy, 897-921.
Cope, M., & Lee, B. H. (2016). Mobility, Communication, and Place: Navigating the Landscapes
of Suburban US Teens. Annals of the American Association of Geographers, 106(2), 311-
320.
Dargay, J., Gately, D., & Sommer, M. (2007). Vehicle ownership and income growth,
worldwide: 1960-2030. The Energy Journal, 143-170.
Delbosc, A. (2016). Delay or forgo? A closer look at youth driver licensing trends in the United
States and Australia. Transportation. doi:doi:10.1007/s11116-016-9685-7
Garikapati, V. M., Pendyala, R. M., Morris, E. A., Mokhtarian, P. L., & McDonald, N. (2016).
Activity patterns, time use, and travel of millennials: a generation in transition? Transport
Reviews, 36(5), 558-584.
Giuliano, P., & Spilimbergo, A. (2009). Growing up in a Recession: Beliefs and the
Macroeconomy: National Bureau of Economic Research.
Glenn, N. D. (1980). Values, attitudes, and beliefs Constancy and change in human development
(pp. 596-640). Cambridge, MA: Harvard University Press.
Goodwin, P., & Van Dender, K. (2013). ‘Peak Car’—Themes and Issues. Transport Reviews,
33(3), 243-254.
47

Haggis, P. (Writer). (2004). Crash. In C. Schulman, D. Cheadle, B. Yari, M. R. Harris, B.
Moresco, & P. Haggis (Producer). USA: Lions Gate Entertainment Corporation.
Krosnick, J. A., & Alwin, D. F. (1989). Aging and susceptibility to attitude change. Journal of
personality and social psychology, 57(3), 416.
Kuhnimhof, T., Armoogum, J., Buehler, R., Dargay, J., Denstadli, J. M., & Yamamoto, T.
(2012). Men Shape a Downward Trend in Car Use among Young Adults—Evidence
from Six Industrialized Countries. Transport Reviews, 32(6), 761-779.
Lewis-Beck, M. S. (2009). The American voter revisited: University of Michigan Press.
Mannheim, K. (1952). The Problem of Generations. In K. Mannheim (Ed.), Essays on the
Sociology of Knowledge (pp. 276-322). London: Routledge and Kegal Paul.
Massey, D. S. (1985). Ethnic residential segregation: A theoretical synthesis and empirical
review. Sociology and social research, 69(3), 315-350.
McDonald, N. C. (2015). Are Millennials Really the “Go-Nowhere” Generation? Journal of the
American Planning Association, 81(2), 90-103.
McFadden, D. (1973). Conditional logit analysis of qualitative choice behavior. Frontiers in
Econometrics, 105-142.
McKenzie, B. (2015). Who Drives to Work? Commuting by Automobile in the United States:
2013 American Community Survey Reports. Washington D.C.
Millard ‐Ball, A., & Schipper, L. (2011). Are we reaching peak travel? Trends in passenger
transport in eight industrialized countries. Transport Reviews, 31(3), 357-378.
Myers, D., & Lee, S. W. (1996). Immigration cohorts and residential overcrowding in southern
California. Demography, 33(1), 51-65.
48

Myers, D., & Pitkin, J. (2013). The Generational Future of Los Angeles: Projections to 2030 and
Comparisons to Recent Decades.
Rosenbloom, S., & Fielding, G. J. (1998). TCRP Report 28: Transit markets of the future: the
challenge of change (0309062535).
Ruggles, S., Alexander, J. T., Genadek, K., Goeken, R., Schroeder, M. B., & Sobek., M. (2010.).
Integrated Public Use Microdata Series: Version 5.0 [Machine-readable database].
Ryder, N. B. (1965). The cohort as a concept in the study of social change. American
Sociological Review, 843-861.
Singer, A. (2003). The rise of new immigrant gateways: Center on Urban and Metropolitan
Policy, the Brookings Institution.
Strauss, W., & Howe, N. (1997). The Fourth Turning: An American Prophecy: Broadway.
Tal, G., & Handy, S. (2010). Travel behavior of immigrants: An analysis of the 2001 National
Household Transportation Survey. Transport Policy, 17(2), 85-93.
Taylor, P. (2014). The next America. Retrieved from http://www.pewresearch.org/next-
america/
Tierney, K. F. (2012). Use of the US Census Bureau's Public Use Microdata Sample (PUMS) by
State Departments of Transportation and Metropolitan Planning Organizations (Vol.
434): Transportation Research Board.
Waters, M. C., & Jiménez, T. R. (2005). Assessing immigrant assimilation: New empirical and
theoretical challenges. Annual Review of Sociology, 105-125.

49

Chapter 3. Has the Relationship between Urban and Suburban Automobile Travel
Changed across Generations? A National-Level Inquiry

1. Introduction
This essay investigates the relationships between the built environment and automobile
travel across generations. Recent studies found that Millennials in various developed economies
are less likely to travel by automobile compared to early generations (e.g. (Kuhnimhof,
Armoogum, et al., 2012)). For instance, in the United States, the Millennials who were 16-28
years old in 2009 took 24.3 percent fewer car trips and had 10.2 percent fewer personal vehicle
miles traveled (VMT) compared to those at the same age in 1995 (Table 3-1). Comparing travel
diaries from the National Household Travel Survey (NHTS) in 2009 and the Nationwide
Personal Transportation Survey (NPTS) in 1995, Table 3-1 shows that although reductions in
driving also happened in other age groups, the drop for the 16-28-year-olds (Millennials vs.
Generation X) is the largest. One interesting question intriguing the researchers and practitioners
in transportation planning is whether the existing patterns of automobile travel for the
Millennials will persist over time. To answer this question, scholars identified factors influencing
the travel behaviors of Millennials in the early 2010s, including the recession, delayed lifecycles
and residence in urban neighborhoods (Blumenberg, Ralph, Smart, & Taylor, 2016; Brown,
Blumenberg, Taylor, Ralph, & Voulgaris, 2016; Garikapati, Pendyala, Morris, Mokhtarian, &
McDonald, 2016; McDonald, 2015; Ralph, Voulgaris, Taylor, Blumenberg, & Brown, 2016).
Since Millennials have started to move from urban neighborhoods to suburban neighborhoods
50

because of higher income and plans for having children (Casselman, 2015), it is likely that they
will increase their distances and frequencies of automobile use.

Table 3-1 – Change in automobility: 1995 - 2009
Age group 1995 2009 Change
Change
in %
p-value
personal VMT
16-28 33.3 29.9 -3.4 -10.2% <0.001
29-41 38.4 36.3 -2.1 -5.5% <0.001
42-54 38.2 36.7 -1.4 -3.8% <0.001
55-67 33.7 33.8 0.1 0.2% 0.848
68- 24.9 25.4 0.4 1.6% 0.318
car trips
16-28 4.2 3.2 -1.0 -24.3% <0.001
29-41 4.7 4.1 -0.6 -13.7% <0.001
42-54 4.7 4.1 -0.6 -12.6% <0.001
55-67 4.5 3.9 -0.6 -12.5% <0.001
68- 4.2 3.6 -0.6 -13.3% <0.001
Note: Subjects not residing in MSAs and with personal VMT larger than 214 were excluded

However, being more reliant on private automobiles does not guarantee that Millennials
will drive as much as the previous generations did even if they live in similar suburban
neighborhoods. Theories in demography highlight the importance of life experiences in young
adulthood in shaping a generation’s long-term attitudes and lifestyles (Glenn, 1980; Mannheim,
1952). Because of either the recession or their delayed lifecycles, the Millennial young adults
have stayed in urban neighborhoods for an extended period of time. Taking numerous short trips
by public transportation or ridesharing services in young adulthood, the Millennials might
develop special activity patterns, cognitive maps and attitudes (Tal & Handy, 2010). Whenever
possible, suburban Millennials might be more willing to try “non-traditional” options of traveling
such as commuter rail, express buses and carpooling than previous generations were. Thus, when
making similar moves from urban to suburban neighborhoods, the Millennials might not increase
51

their personal VMTs and car trips as much as the previous generations of similar socio-economic
conditions and in the same lifecycle stages did.
In order to test this hypothesis in the context of the United States, I compared automobile
travel among the Millennials and the Generation-Xers across different built environment patterns
in metropolitan areas as both groups reached the same age (young adulthood). This study uses
the travel diary data from the three most recent nationwide travel surveys in the US: the 1995
NPTS, the 2001 NHTS and the 2009 NHTS. I first compared the changes in patterns of
automobile travel from 1995 to 2009 in different residential density categories for each age
group. For most neighborhood types, the Millennial young adults had significantly lower
personal VMTs and car trips than the Generation-Xers had at the same age. The decline in
automobility for the 16-28-year-olds is the largest and the most common across all age groups.
Controlling for socio-economic, vehicle ownership, lifecycle, year-specific and regional-specific
factors, I also proposed two regression models and showed that the marginal effects of
residential density on personal VMT and car trips are both significantly different between the
Millennial and the Generation-X young adults. Specifically, the influence of residential density
on personal VMT for the 16-28-year-olds in 2009 is 27 percent lower than those in 1995.
Similarly, the influence of residential density on number of car trips for the 16-28-year-olds is 38
percent lower than those in 1995. In addition, I predicted their personal VMT and car trips using
these models, controlling for the aforementioned factors. Compared to the “average Generation-
Xer” (in 1995), an “average Millennial” (in 2009) with the same controlling factors had lower
numbers of predicted personal VMTs and car trips in the same type of suburban neighborhoods.
Even if these findings cannot guarantee that the travel patterns of the Millennial young adults
52

will certainly persist over time, they imply that transportation planners should test various travel
demand management policies on suburban Millennials.
The second section reviews the previous literature covering Millennial travel, theories on
generations and the relationship between built environment and automobile travel. The third
section introduces the datasets and the methods of my analysis. The fourth section reports and
interprets the findings and the final section concludes with some discussions.

2. Literature
Since the early 2010s, researchers in transportation planning have noticed that, in
developed economies such as the United States (Blumenberg et al., 2016; McDonald, 2015),
Canada (Marzoughi, 2011) and Germany (Kuhnimhof, Buehler, Wirtz, & Kalinowska, 2012),
Millennials drive considerably less than the young adults 15-20 years ago. In addition, they were
less likely to hold a driver’s license (Delbosc & Currie, 2013; Kuhnimhof, Armoogum, et al.,
2012). Researchers started to wonder if this trend had contributed to the stagnant growth of
automobile travel (or “Peak Car”) at that time (Goodwin & Van Dender, 2013). Recently, the
distance of vehicle traveled in the US has rebounded to an all-time high (Federal Highway
Administration, 2017), largely due to the resurgence of the economy and the decrease in fuel
prices (Bastian, Börjesson, & Eliasson, 2016). However, whether the Millennials will maintain
such “multi-modalism” (Circella et al., 2017) in the future continue to intrigue researchers and
practitioners in transportation planning.
Recent studies in the United States found that most of the decline in automobile travel
among Millennials are caused by the recession and the delayed lifecycles of these young adults.
Using data from nationwide travel surveys conducted in 1990, 2001 and 2009, Blumenberg et al.
53

(2016) found that being unemployed was the largest factor contributing to the reduced mobility
of these young adults. Using data from the American Time Use Survey, Garikapati et al. (2016)
argued that the Millennials’ delayed lifecycle caused them to postpone their adoption of the
activity patterns of previous generations. Using a nationwide travel survey in 2009, Ralph et al.
(2016) showed that the built environment patterns have a small, yet significant effect on
Millennial driving. Using nationwide travel survey data in 1995, 2001 and 2009, McDonald
(2015) quantitatively decomposed the factors contributing to the decline of automobile travel of
the young adults. The relative contribution of lifestyle-related demographic shifts, Millennial-
specific factors and time fixed effects are 10-25%, 35-50% and 40%, respectively.
Cohort theory, originally developed by demographers, argues that people in different
generations tend to hold unique attitudes and lifestyles (Strauss & Howe, 1997). People in a
generation, or birth cohort, have experienced similar socio-economic conditions and major
historical events at similar lifecycle stages (Ryder, 1965). Such collective memory, especially in
the ages of 18-25, is especially powerful in shaping the life-long attitudes and lifestyles for
people in different generations (Glenn, 1980; Mannheim, 1952). Recent studies have
demonstrated the long-term impacts of early-adulthood experiences on people’s political
attitudes (Lewis-Beck, 2009), views on social policies (Alesina & Fuchs-Schündeln, 2007) and
consumption patterns (Giuliano & Spilimbergo, 2009). Many Millennials have stayed in
neighborhoods with good transit supply during their young adulthood for an extended period of
time (Brown et al., 2016). They have also been exposed to a wide application of information and
communication technology from their early ages on (Circella & Mokhtarian, 2017). Such
experiences, associated with the economic hardship and the delayed lifecycles that the
Millennials have experienced, might be able to influence their behaviors of automobile travel
54

when they move to suburban neighborhoods by impacting their activity patterns, cognitive maps
and attitudes and beliefs (Tal & Handy, 2010). This suggests that the relationship between the
built environment and automobile travel among Millennials might be different from that of
previous generations.
The decades-long literature on the built environment and automobile travel has shown
that people residing in urban neighborhoods drive less than those residing in suburban
neighborhoods (Ewing & Cervero, 2010). In addition, recent studies have confirmed that built
environment patterns can affect driving patterns after controlling for the effects of “self-
selection” (Cao, Mokhtarian, & Handy, 2009). From a microeconomic perspective, compared to
their suburban counterparts, urban neighborhoods close trip origins and destinations. Thus,
driving is less attractive in urban neighborhoods because the time and monetary costs of
alternative modes are lower (Boarnet, 2011). Among various built environment measurements,
density is probably the most widely-used proxy to other built environment patterns (e.g.
(Brownstone & Golob, 2009; Chatman, 2008; Chen, Gong, & Paaswell, 2008). In the context of
the United States, residential density often correlates with factors such as land use mix and
patterns of street design (Ewing & Cervero, 2010). The current literature shows that a 100
percent increase in neighborhood-level residential density is associated with a 4-12 percent
reduction in household-level vehicle miles traveled, controlling for other built environment
factors (Ewing & Cervero, 2010). However, the reduction could be as high as 34% if no other
factors are controlled for (Chatman, 2008).
Most of the discussions on the travel behavior of young adults in the United States have
focused on the determinants of driving, rather than built environment – driving dynamics. The
generational theory developed by demographers suggests that relationships between urban and
55

suburban automobile travel among young adults might differ across generations. Examining such
potential differences can help us better understand the current debate on the future patterns of
automobile travel among Millennials, especially after they move to less dense neighborhoods.

3. Methods
The dataset for this study comes from the public sample of the most recent three
nationwide travel surveys in the United States, the 1995 NPTS, 2001 NHTS and 2009 NHTS
(Federal Highway Administration, 1997, 2004, 2011). I define the Millennials as people born
after 1980 (Taylor, 2014); thus, they were 16-28 years old in 2009. The young adults at the same
age in 1995 were born between 1965 and 1980 and belong to the previous generation, namely
Generation X. This study proposes two variables in automobile travel: personal VMT and
personal car trips during the survey day. I have excluded the subjects whose personal VMT were
larger than 214 (98
th
percentiles of total personal VMT) and who were residing outside
metropolitan statistical areas (MSAs). Therefore, the total number of eligible young adults (16-
28 years’ old) across these three survey years is 36,407 (Table 3-2). This dataset in repeated
cross-sectional. In other words, the Federal Highway Administration (FHWA) surveyed different
respondents in different survey years. According to Table 3-2, the share of the young adults
surveyed in 1995, 2001 and 2009 are 18%, 36% and 46%, respectively, indicating that FHWA
have increased the sample size of these nationwide travel surveys over time. The average age of
these young adults was 21.7, and 51% of the survey respondents were female.

56

Table 3-2 – Descriptive statistics for the young adults (16-28) sample
Mean Std. Dev.
Personal VMT for the survey day 32.38 (34.69)
Number of car trips in the survey day 3.63 (2.57)
Population density (1,000 persons per square mile) 5.40 (6.81)
Survey year

1995 0.18 (0.38)
2001 0.36 (0.48)
2009 0.46 (0.50)
Household income (in 1,000 2009 US Dollars) 67.21 (39.22)
Driver status 0.87 (0.34)
Household vehicles per person 0.79 (0.40)
Age 21.74 (3.97)
Female 0.51 (0.50)
Having children 0.17 (0.37)
Worker 0.70 (0.46)
Education

Less than high school 0.14 (0.34)
High school/Associate degree 0.54 (0.50)
Bachelor's degree 0.15 (0.36)
Graduate degree 0.03 (0.18)
Not available 0.13 (0.34)
Race of household head

Non-Hispanic White 0.76 (0.43)
Non-Hispanic Black 0.07 (0.25)
Non-Hispanic Asian/Pacific Islander 0.04 (0.20)
Hispanic 0.11 (0.31)
Other races 0.02 (0.15)
Size of metropolitan areas

In an MSA of less than 250,000 0.16 (0.37)
In an MSA of 250,000 - 499,999 0.14 (0.34)
In an MSA of 500,000 - 999,999 0.13 (0.33)
In an MSA or CMSA of 1,000,000 - 2,999,999 0.22 (0.41)
In an MSA or CMSA of 3 million or more 0.36 (0.48)
N 36,407
Note: Subjects not residing in MSAs and with personal VMT larger than 214 were excluded

I used the eight residential density categories provided by these surveys to categorize
neighborhood-level built environmental patterns. Although the residential density categories are
57

unfortunately the only built environment patterns available in the public samples of NPTS and
NHTS, they are usually indicators of other neighborhood design patterns in the context of the
United States (Ewing & Cervero, 2010). These categories are, in unit of persons per square mile,
less than 100; 100-500; 500-1,000; 1,000 to 2,000; 2,000 to 4,000; 4,000 to 10,000; 10,000 to
25,000 and more than 25,000
1
. I first compare the changes in personal VMT and personal car
trips of the 16-28-year-old in 1995 and in 2009 in each of these neighborhood categories. That
enables me to determine whether the reductions of automobility of the young adults from 1995 to
2009 solely came from their different residential patterns. Then I repeat the same analysis for
other age groups: 29-41, 42-54, 55-67 and 68+ to see if the patterns were similar across age
groups. If the patterns for the 16-28-year-old group differ from those for other age groups, the
changes in automobile travel did not solely derive from temporal factors impacting everyone in
the same way.
I also propose two regression models to compare the residential density – automobile
travel dynamics for young adults across generations, controlling for socio-economic, vehicle
ownership, lifecycle, year-specific and regional-specific factors. The models follow the equation
below:

𝑇𝐵 = 𝑓 (𝛽 + 𝛽 𝐵𝐸 + 𝐲𝐞𝐚𝐫 𝐢 ∙ 𝛃 𝟐 + 𝐵𝐸 𝐲𝐞𝐚𝐫 𝐢 ∙ 𝛃 𝟑 + 𝐗 𝐢 ∙ 𝛃 𝟒 + 𝜀 ), (3-1)

where TBi is the travel behavior variable of respondent i, including personal VMT and number of
car trips on the survey day. Notation f(∙) indicates that neither model is an ordinary least square

1
Examples of neighborhood types (unit: persons per square mile): 0-100, Antelope Valley, Lancaster, CA; 100-500,
City of Industry, CA; 500-1,000 Friendly Hills, Mendota Heights, MN; 1,000-2,000, Arden Hills, MN; 2,000-4,000,
Rancho Palos Verdes, CA; 4,000-10,000, Wilburton, Bellevue, WA; 10,000-25,000, downtown Bellevue, WA;
25,000 or above, Koreatown, Los Angeles, CA.
58

regression. Personal VMT is a left-censored variable; thus it is estimated by a Tobit regression.
The number of car trips is a count variable, and is thus estimated by a negative binomial
regression. The independent variables include an intercept term β0, a variable BEi measuring the
built environment patterns of the residence, a vector yeari indicating the year of the survey (with
1995 as the reference term), and a vector interacting BEi and yeari to separately estimate the
coefficients of BEi on TBi for 1995, 2001 and 2009, and a vector Xi including the control
variables. Since the public samples of these travel surveys only provide residential density
categories, I construct a continuous BEi by using the midpoint of each category
2
. The control
variables in Xi include household income (in 2009 US Dollars), driver status, household vehicle
per person, age, age squared, gender, worker status, gender interacted with worker status, having
children, gender interacted with having children, level of education, race of household head and
the size of metropolitan areas. Xi also includes 50 state fixed effects dummy variables to control
for inter-state variations for the 50 states and Washington D.C.
3
Table 3-2 shows the means and
standard deviations of the dependent and independent variables in the sample of the models.

4. Results
4.1. Generational changes in automobility by neighborhood type
The Millennial young adults had significantly lower personal VMT and numbers of car
trips than did their counterparts in Generation X for almost every block-group level residential
density category. The upper panel in Table 3-3 shows the differences in personal VMTs among
the 16-28-year-old between 1995 (Generation-Xers) and 2009 (Millennials). For the residential

2
For the densest category (>25,000 persons per square mile), I use 30,000 persons per square mile.
3
The information on the state of residence for the respondents from the 2001 NHTS comes from the geo-coded
sample, since it is not available for all respondents in the public sample.
59

density categories ranging from “less than 100 persons per square mile” to “4,000 to 10,000
persons per square mile,” the average personal VMT of the Millennials is significantly lower
than that of the Generation-Xers, with differences from 8 to 21 percent. The difference is not
significant for the “10,000 persons per square mile” category. In contrast, for the young adults
residing in the densest neighborhood category, the average personal VMT for the survey day
increased from 11.6 in 1995 to 16.0 in 2009. This counterintuitive fact might imply that more
Millennial young adults in a higher socio economic status chose to live in dense neighborhoods
than did their Generation X counterparts. The lower panel shows that the Millennial young adults
took on average one fewer car trip per day than did their Generation X counterparts in all but the
densest neighborhood categories. For the densest neighborhood category, the average numbers of
car trips of those young adults in 1995 and 2009 are both 1.5.

60

Table 3-3 – Change of automobile travel: 1995 – 2009 for the age group 16-28 by block-group level residential
density
Residential density (persons
per square mile)
1995 2009 Change
Change
in %
p-value
Personal VMT

0 to 100 45.1 41.5 -3.6 -8.0% 0.046
100 to 500 42.6 38.2 -4.4 -10.2% <0.001
500 to 1,000 39.8 32.6 -7.2 -18.0% <0.001
1,000 to 2,000 35.3 31.1 -4.1 -11.7% <0.001
2,000 to 4,000 35.1 27.7 -7.4 -21.0% <0.001
4,000 to 10,000 31.1 25.9 -5.2 -16.7% <0.001
10,000 to 25,000 24.3 23.4 -1.0 -4.0% 0.429
25,000 - 11.6 16.0 4.4 37.4% 0.016
All 33.3 29.9 -3.4 -10.2% <0.001
Car trips

0 to 100 4.4 3.1 -1.3 -28.5% <0.001
100 to 500 4.5 3.2 -1.3 -28.4% <0.001
500 to 1,000 4.6 3.3 -1.3 -28.8% <0.001
1,000 to 2,000 4.5 3.4 -1.1 -25.1% <0.001
2,000 to 4,000 4.5 3.3 -1.2 -26.5% <0.001
4,000 to 10,000 4.4 3.1 -1.2 -28.4% <0.001
10,000 to 25,000 3.7 2.8 -0.9 -24.6% <0.001
25,000 - 1.5 1.5 0.0 -0.2% 0.9859
All 4.2 3.2 -1.0 -24.3% <0.001
Note: Subjects not residing in MSAs and with personal VMT larger than 214 were excluded

As Table 3-4 shows, compared to 16-28 age group, the reductions in the average personal
VMTs and average numbers of car trips over these 14 years for other age groups are either not as
common or are in a smaller magnitude. As the upper panel demonstrates, among the eight
neighborhood categories, six have experienced a significant reduction in personal VMT for the
16-28 age group. The numbers of density categories for other age groups are all smaller: five for
29-41, four for 42-54, zero for 55-67 and one for 68+. In addition, the percentage reductions of
personal VMT in each neighborhood category for those older age groups are generally smaller
61

than those for the 16-28 age group. The lower panel shows that the reductions in the average
numbers of car trips exist in all but one neighborhood category across all age groups. However,
the reductions for the 16-28-year-old are the only ones exceeding 20%. These findings show that
reduced automobile travel for young adults happens across various neighborhood types. In
contrast to our intuitions, reduced automobility is not an urban phenomenon.

62

Table 3-4 – Percentage change of automobile travel (1995 - 2009), by age groups and block-group level
residential density
Residential density (persons per
square mile)
16-28 29-41 42-54 55-67 68-
Personal VMT

0 to 100 -8.0% - - - -
100 to 500 -10.2% -4.7% - - -
500 to 1,000 -18.0% -5.4% -6.1% - -
1,000 to 2,000 -11.7% -14.6% -5.3% - -7.8%
2,000 to 4,000 -21.0% -6.2% -9.1% - -
4,000 to 10,000 -16.7% -10.8% -5.6% - -
10,000 to 25,000 - - - - -
25,000 - 37.4% 36.9% - - -
Car trips

0 to 100 -28.5% -19.8% -16.4% -15.9% -12.9%
100 to 500 -28.4% -16.0% -15.1% -13.4% -15.2%
500 to 1,000 -28.8% -16.3% -15.8% -13.8% -20.4%
1,000 to 2,000 -25.1% -16.0% -12.4% -13.4% -14.6%
2,000 to 4,000 -26.5% -16.9% -14.6% -15.3% -15.6%
4,000 to 10,000 -28.4% -15.7% -15.0% -15.0% -15.2%
10,000 to 25,000 -24.6% -13.3% -10.9% -10.1% -8.0%
25,000 - - - - - -
Note: “-” indicates differences not significant at a five-percent level

4.2. Regression models
The two regression models shown in Table 3-5 reveal that the personal VMTs and car
trip frequencies of the young adults increase when the residential density decreases. However,
the slope of the increase for the young adults in 2009 is significantly lower compared to their
counterparts in 1995 and 2001. In each model the interaction term between residential density
and year 2009 dummy variable is significant at a five percent level. For personal VMT, the
coefficients of density for the 1995 and 2001 young adults are both -1.155, while the coefficient
for the 2009 group is -0.845, the summation of the coefficients of the variables “population
density” and “population density X 2009.” Similarly, for the number of car trips, the coefficient
63

of density for the 1995 and 2001 young adults are both -0.013, while the coefficient for the 2009
group is -0.008. In other words, holding other factors constant, the magnitude of the association
between residential density and personal VMT for the young adults in 2009 was 27 percent
lower than for those in 1995. In addition, the magnitude of the association between residential
density and the number of car trips was 38 percent lower for young adults in 2009 than for those
in 1995.

64

Table 3-5 – Regression models for automobile travel for the 16-28 age group in 1995, 2001 and 2009

Tobit model for personal
VMT
Negative binomial model
for number of car trips
Coef. Std. Err. Coef. Std. Err.
Population density (1,000 people per square mile) -1.155*** (0.063) -0.013*** (0.001)
Survey year

1995 ref.

ref.

2001 -2.205*** (0.837) -0.127*** (0.015)
2009 -7.202*** (0.825) -0.342*** (0.015)
Density-year interactions

Population density X 2001 0.085 (0.077) <0.001 (0.001)
Population density X 2009 0.310*** (0.080) 0.005*** (0.002)
Household income (in 1,000 2009 US Dollars) 0.034*** (0.005) 0.001*** (0.000)
Driver status 13.556*** (0.651) 0.568*** (0.014)
Household vehicles per person 9.459*** (0.570) 0.147*** (0.010)
Age 3.968*** (0.874) -0.073*** (0.016)
Age squared -0.080*** (0.019) 0.001*** (0.000)
Female 2.540*** (0.711) 0.089*** (0.014)
Having children 7.963*** (0.888) 0.103*** (0.016)
Female X having children -6.064*** (1.053) 0.056*** (0.019)
Worker 5.146*** (0.654) 0.110*** (0.013)
Female X worker -2.207*** (0.833) -0.014 (0.016)
Education

Less than high school ref.

ref.

High school/Associate degree 4.107*** (0.744) 0.028** (0.014)
Bachelor's degree 4.859*** (0.944) 0.011 (0.017)
Graduate degree 3.677*** (1.319) -0.004 (0.024)
Not available 1.425* (0.840) -0.065*** (0.016)
Race of household head

Non-Hispanic white ref.

ref.

Non-Hispanic black -3.987*** (0.822) -0.070*** (0.016)
Non-Hispanic Asian/Pacific Islander -0.550 (1.017) -0.054*** (0.020)
Hispanic -0.933 (0.681) -0.018 (0.013)
Other races -1.718 (1.271) -0.034 (0.024)
Size of metropolitan areas

In an MSA of less than 250,000 ref.

ref.

In an MSA of 250,000 - 499,999 1.611** (0.727) -0.035*** (0.013)
In an MSA of 500,000 - 999,999 2.103** (0.841) -0.037** (0.015)
In an MSA or CMSA of 1,000,000 - 2,999,999 0.956 (0.716) -0.047*** (0.013)
In an MSA or CMSA of 3 million or more 2.314*** (0.718) -0.088*** (0.013)
State fixed effects

Alabama ref.

ref.

Alaska -0.141 (6.101) 0.088 (0.110)
65

…

Wyoming -11.303 (9.642) 0.283* (0.161)
Constant -35.447*** (10.948) 1.740*** (0.201)
N 36,407 36,407
p-value of likelihood ratio test <0.001 <0.001
Note: Standard errors in parentheses, *, **, *** indicate p<0.1, p<0.05 and p<0.01, respectively

In addition, most of the control variables are also significantly associated with the two
dependent variables. Being a driver and living in a household with more cars per person
positively correlate with the distance and frequency of automobile travel among the young
adults. Age positively associates with personal VMT but negatively associates with the number
of car trips; the strength of association decreases as age increases. The influences of gender,
worker status and having children on automobile travel are not intuitive because of the
interaction terms. Specifically, female young adults with no work or no children have higher
personal VMT and personal car trips than do their male counterparts. For both male and female
young adults, being employed and having children positively associate with both personal VMT
and personal car trips. Being better educated positively associates with personal VMT but does
not correlate with the number of personal car trips. In addition, the race of the household head
correlates with automobile travel for young adults. Specifically, compared to the young adults
with a non-Hispanic white household head, those with a non-Hispanic African American
household head tend to travel by car less frequently and for shorter distances; those with a non-
Hispanic Asian or Pacific Islander household head tend to have lower frequencies of car trips;
and those with a Hispanic household head do not have significant differences in automobile
travel. Also, young adults staying in larger MSAs tend to have higher personal VMTs and lower
numbers of car trips. This suggests that, in larger MSAs, each automobile trip made by young
adults is longer. Finally, the negative coefficients of the two year dummy variables on year fixed
66

effects indicate the general dampening of automobile travel that occurs on young adults across
all different characteristics mentioned above.
Since the marginal effects in non-linear regressions are not intuitive, I predicted the
personal VMT and number of car trips for a “typical” young adult residing in neighborhoods
with different residential densities in both 1995 and 2009 (Figure 3-1). “Typical” indicates that
all other control variables shown in Table 3-5 were set as their mean values, as demonstrated in
Table 3-2
4
. As indicated by Figure 3-1, holding socio-economic, vehicle ownership, lifecycle,
year-specific and regional-specific factors constant, a “typical” young adult increases personal
VMT and number of car trips after moving from an urban to a suburban neighborhood. In
addition, the increase of both measures of automobility for the young adults in 2009
(Millennials) is flatter than those in 1995 (Generation X) when making the same move. The
predicted personal VMT for the 2009 “typical” young adult was higher than that for the 1995 one
when residing in neighborhoods with densities higher than 2,300 per square mile. In contrast,
when residing in neighborhoods with densities lower than 2,300 per square mile, the “typical”
young adult in 2009 had lower predicted personal VMT (Figure 3-1(a)). The predicted numbers
of car trips for the “typical” young adult in 2009 was lower than those for the ones in 1995
across all values of residential density (Figure 3-1(b)).

4
The mean values of the 50 state fixed effect variables, not shown in Table 2 due to space limit, are available upon
request.
67

Figure 3-1 – Predicted personal VMT and car trips for a “typical” young adult in 1995 and 2009
(Note: refer to Section 3 for examples of each neighborhood type)

In sum, the analyses in this section show that at the same young adult age (16-28), the
Millennials (1) traveled shorter and less frequently in automobiles than did the Generation-Xers
in most neighborhood types; and (2) increased their distance and frequencies of automobile
travel less than did the Generation-Xers when making similar city-to-suburb moves. Such
patterns still hold even if we control for the socio-economic, vehicle ownership, lifecycle, year-
specific and regional-specific factors.

5. Conclusion
Using nationwide travel diaries in the United States for 1995, 2001 and 2009, I found that
young adults aged 16-28 in 2009 generally traveled in automobiles in shorter distances and less
frequently compared to those in 1995 when they were residing in neighborhoods with similar
levels of residential density. I also found that, when making similar city-to-suburb moves, the
increases in the distance and frequencies of automobile travel of the young adults in 2009 are
68

around 30 percent lower than those for their counterparts in 1995, controlling for socio-
economic, vehicle ownership, lifecycle, year-specific and regional-specific factors. Note that in
this case, the 16-28-year-old in 2009 and 1995 are Millennials and Generation-Xers,
respectively.
Admittedly, Millennials will likely drive more when they obtain better job opportunities,
enter later stages of their lifecycle, and move to less dense neighborhoods as a result. However, it
is also possible that people in this generation will not drive as much as those in the previous
generation even after they moving to the suburbs. Even if it is not clear whether the relationship
between urban and suburban travel for the Millennial young adults as I found will certainly
persist in their later ages, the demographic theory suggests that we should be more careful before
simply answering “no.” Staying in urban neighborhoods for an extended period of time during
these Millennials’ young adulthood might impact their travel behavior in the long run. As a
recent survey in California shows, the attitudes towards lifestyles and travel of the Millennials
are different from those of the Generation X (Circella et al., 2017). It is likely that staying in
neighborhoods with a more sufficient supply of transit, a better walking environment and more
opportunities for carpooling during young adulthood will make the Millennials more open to a
variety of travel demand management policies.
Although this study cannot directly test these implications using existing data, it suggests
potential values for transportation planners to test the effectiveness of pilot projects targeting the
new Millennial dwellers in the suburbs. Projects such as bike share, park and ride, commuter rail
and vanpool programs
5
might be more effective for the Millennials than for those in previous
generations. Many people moved to the suburban neighborhoods because of reasons other than

5
For instance, Microsoft partners with various local government agencies to provide more sustainable commuting
options for their suburban employees, see: https://www.microsoft.com/about/csr/environment/carbon/operations/
69

their desired ways of traveling, and scholars (e.g. (Levine, 2006)) argued that neighborhoods
fostering multimodalism might be undersupplied. Even if the total distance of automobile travel
has bounced back recently in the US, largely because of the reduced gasoline prices, adopting
these program might still be popular and successful.
From a researcher’s perspective, there are two important questions that will be
particularly helpful in extending this study: first, whether the attitudes about and preferences for
automobile travel differ across generations; and second, whether these generational differences,
if any, will persist over time. Answering these questions can provide further evidence as to
whether the built environment – automobile travel dynamics for each generation will persist over
time. This would require longitudinal datasets that might be expensive and time-consuming to
obtain, but it would be especially helpful for researchers in the field of transportation planning.
In addition, researchers should examine the generational differences of the impact of
metropolitan-level built environment characteristics on the distance and frequencies of
automobile travel.

70

6. References
Alesina, A., & Fuchs-Schündeln, N. (2007). Good-Bye Lenin (or Not?): The Effect of
Communism on People's Preferences. The american economic review, 97(4), 1507-1528.
Bastian, A., Börjesson, M., & Eliasson, J. (2016). Explaining “peak car” with economic
variables. Transportation Research Part A: Policy and Practice, 88, 236-250.
Blumenberg, E., Ralph, K., Smart, M., & Taylor, B. D. (2016). Who knows about kids these
days? Analyzing the determinants of youth and adult mobility in the US between 1990
and 2009. Transportation Research Part A: Policy and Practice, 93, 39-54.
Boarnet, M. G. (2011). A broader context for land use and travel behavior, and a research
agenda. Journal of the American Planning Association, 77(3), 197-213.
Brown, A. E., Blumenberg, E., Taylor, B. D., Ralph, K., & Voulgaris, C. T. (2016). A taste for
transit? Analyzing public transit use trends among youth. Journal of Public
Transportation, 19(1), 49-67.
Brownstone, D., & Golob, T. F. (2009). The impact of residential density on vehicle usage and
energy consumption. Journal of Urban Economics, 65(1), 91-98.
Cao, X., Mokhtarian, P. L., & Handy, S. L. (2009). Examining the impacts of residential self ‐
selection on travel behaviour: a focus on empirical findings. Transport Reviews, 29(3),
359-395.
Casselman, B. (2015). Think Millennials Prefer The City? Think Again. Retrieved from
https://fivethirtyeight.com/datalab/think-millennials-prefer-the-city-think-again/
Chatman, D. G. (2008). Deconstructing development density: Quality, quantity and price effects
on household non-work travel. Transportation Research Part A: Policy and Practice,
42(7), 1008-1030.
71

Chen, C., Gong, H., & Paaswell, R. (2008). Role of the built environment on mode choice
decisions: additional evidence on the impact of density. Transportation, 35(3), 285-299.
Circella, G., Alemi, F., Berliner, R., Tiedeman, K., Lee, Y., Fulton, L., . . . Mokhtarian, P. L.
(2017). Multimodal Behavior of Millennials: Exploring Differences in Travel Choices
Between Young Adults and Gen-Xers in California. Paper presented at the 96th
Transportation Research Board Annual Meeting.
Circella, G., & Mokhtarian, P. L. (2017). Impacts of Information and Communication
Technology. In G. Giuliano & S. Hanson (Eds.), The geography of urban transportation
(pp. 86-111). New York, NY: Guilford Press.
Delbosc, A., & Currie, G. (2013). Causes of youth licensing decline: A synthesis of evidence.
Transport Reviews, 33(3), 271-290.
Ewing, R., & Cervero, R. (2010). Travel and the Built Environment. Journal of the American
Planning Association, 76(3), 265-294.
Federal Highway Administration. (1997). 1995 Nationwide Personal Transportation Survey
user's guide. Retrieved from http://nhts.ornl.gov/1995/Doc/UserGuide.pdf
Federal Highway Administration. (2004). 2001 National Household Travel Survey user's guide.
Retrieved from http://nhts.ornl.gov/2001/usersguide/UsersGuide.pdf
Federal Highway Administration. (2011). 2009 National Household Travel Survey user's guide.
Retrieved from http://nhts.ornl.gov/2009/pub/UsersGuideV2.pdf
Federal Highway Administration. (2017). Traffic Volume Trends.
https://www.fhwa.dot.gov/policyinformation/travel_monitoring/tvt.cfm
72

Garikapati, V. M., Pendyala, R. M., Morris, E. A., Mokhtarian, P. L., & McDonald, N. (2016).
Activity patterns, time use, and travel of millennials: a generation in transition? Transport
Reviews, 36(5), 558-584.
Giuliano, P., & Spilimbergo, A. (2009). Growing up in a Recession: Beliefs and the
Macroeconomy: National Bureau of Economic Research.
Glenn, N. D. (1980). Values, attitudes, and beliefs Constancy and change in human development
(pp. 596-640). Cambridge, MA: Harvard University Press.
Goodwin, P., & Van Dender, K. (2013). ‘Peak Car’—Themes and Issues. Transport Reviews,
33(3), 243-254.
Kuhnimhof, T., Armoogum, J., Buehler, R., Dargay, J., Denstadli, J. M., & Yamamoto, T.
(2012). Men Shape a Downward Trend in Car Use among Young Adults—Evidence
from Six Industrialized Countries. Transport Reviews, 32(6), 761-779.
Kuhnimhof, T., Buehler, R., Wirtz, M., & Kalinowska, D. (2012). Travel trends among young
adults in Germany: increasing multimodality and declining car use for men. Journal of
Transport Geography, 24, 443-450.
Levine, J. C. (2006). Zoned out: Regulation, markets, and choices in transportation and
metropolitan land-use. Washington D.C.: RFF Press.
Lewis-Beck, M. S. (2009). The American voter revisited: University of Michigan Press.
Mannheim, K. (1952). The Problem of Generations. In K. Mannheim (Ed.), Essays on the
Sociology of Knowledge (pp. 276-322). London: Routledge and Kegal Paul.
Marzoughi, R. (2011). Teen travel in the Greater Toronto Area: A descriptive analysis of trends
from 1986 to 2006 and the policy implications. Transport Policy, 18(4), 623-630.
73

McDonald, N. C. (2015). Are Millennials Really the “Go-Nowhere” Generation? Journal of the
American Planning Association, 81(2), 90-103.
Ralph, K., Voulgaris, C. T., Taylor, B. D., Blumenberg, E., & Brown, A. E. (2016). Millennials,
built form, and travel insights from a nationwide typology of US neighborhoods. Journal
of Transport Geography, 57, 218-226.
Ryder, N. B. (1965). The cohort as a concept in the study of social change. American
Sociological Review, 843-861.
Strauss, W., & Howe, N. (1997). The Fourth Turning: An American Prophecy: Broadway.
Tal, G., & Handy, S. (2010). Travel behavior of immigrants: An analysis of the 2001 National
Household Transportation Survey. Transport Policy, 17(2), 85-93.
Taylor, P. (2014). The next America. Retrieved from http://www.pewresearch.org/next-
america/

74

Chapter 4. The Impact of Health Conditions on Elderly Driving: A National-Level
Longitudinal Study Using the Health and Retirement Study

1. Introduction
The mass retirement of the Baby Boomers in the near future will accelerate the aging of
the American population. The Census Bureau’s projections show that by 2030, one out of every
five Americans will be 65 or older (Ortman, Velkoff, & Hogan, 2014). Such a rapid aging trend
will considerably change the future landscape of passenger transportation in the United States.
Generally speaking, people drive less when they age, especially after the age of 75 (Alsnih &
Hensher, 2003). For instance, the most recent National Household Travel Survey, conducted in
2009, showed that 93.7% of the respondents 50-59 years old were drivers, but the percentages
for those who were 60-69, 70-79 and 80+ were 91.4%, 83% and 61.7%, respectively (Santos,
McGuckin, Nakamoto, Gray, & Liss, 2011). As Giuliano (2004) and Rosenbloom (2001) have
demonstrated, even if seniors drive less when they age, they do not take more public
transportation to compensate for reduced driving. In other words, most seniors who reduce their
driving also suffer from reduced mobility. Considering that the now-retiring Baby-Boomers are
the most car-dependent generation to date (Bush, 2003), the challenge to fulfill seniors’ mobility
needs will be even greater over the next 20 years. Loss of mobility among the elderly has been
shown to decrease out-of-home activity (Marottoli et al., 2000), weaken networks with friends
(Mezuk & Rebok, 2008) and even increase mortality risks (Edwards, Perkins, Ross, & Reynolds,
2009). To better prepare for such an oncoming challenge, researchers in transportation policy
75

and planning need to have a deeper understanding of the factors associated with seniors’
decisions to drive.
The literature in transportation policy and planning has demonstrated that socio-economic
characteristics and built environment patterns influence people’s decisions to drive. Socio-
economic characteristics include gender, ethnicity, income, family structure and employment,
while built environment patterns include density, diversity and design patterns at the
neighborhood and metropolitan levels (Ewing & Cervero, 2010). From a microeconomic
perspective, socio-economic and built environment factors impact a person’s decision to drive by
influencing the relative time and monetary costs among different travel modes (Boarnet, 2011).
Most studies in this line of research focus on the general population across all age groups. Using
behavioral models on local or national samples, the few existing studies of senior driving showed
that seniors living in neighborhoods with higher population densities and better accessibility
were less car-dependent and more likely to use public transportation (Giuliano, 2004; Hu, 2006;
Lee, Zegras, Ben-Joseph, & Park, 2014).
In addition to socio-economic and built environment factors, health conditions also
impact people’s decision to drive, especially among seniors. The health impacts on driving can
be explained by a capability theory adopted from psychology (Fuller, 2005). The capability
theory suggests that people make their decisions to drive by comparing their capability against
the task demands. Thus, people drive if they believe that their capability exceeds the task
demands of driving. In contrast, if their capability does not exceed the task demands, they will
see potential driving trips as too risky and decide not to drive. Based on this theory, health
conditions can impact the decision to drive for the elderly either through functional limitations
because of their health conditions, or through self-regulation due to the expected effects of the
76

corresponding medications (Rosenbloom & Santos, 2014). In other words, seniors estimate their
capability to drive based on their health conditions, and if they believe they have health problems
rendering them less able to meet the task demands, they will choose not to drive. From a
microeconomic perspective adopted by the random utility theory suggested by McFadden
(1973), capability theory suggests that poor health conditions will decrease the utility of driving
for the seniors. The decreased utility will make driving less appealing to those seniors. The
California Department of Motor Vehicles (2017) suggested three types of health conditions
which could impact senior driving: physical conditions, cognitive conditions and vision
conditions.
Except for Rosenbloom and Santos (2014), researchers in transportation policy and
planning have not widely explored the impacts of health conditions on senior driving.
Gerontologists have studied a related but not identical concept, “driving cessation,” referring to
the complete abandonment of driving. The literature in gerontology shows that self-rated health,
cognitive functions (Anstey, Windsor, Luszcz, & Andrews, 2006), slower speed of processing,
congestive heart failure and poorer physical performance (Edwards et al., 2008) are related to a
higher likelihood of driving cessation. In addition, driving cessation is more common among
people who are minorities, female (Choi & Mezuk, 2013) and 75 or older (Alsnih & Hensher,
2003). However, these findings cannot be directly applied to transportation policy and planning
for the following reasons. First, “driving cessation” is not identical to travel mode choice, and the
latter concept has a higher policy significance in transportation policy and planning. Second,
studies in gerontology tend to omit the built environment and policy variables which are known
to have an impact on travel mode choice. Third, most of the US-based gerontology work did not
use a geographically representative sample; thus, their results might be robust enough to draw
77

conclusions in public health, but not in more geographic-sensitive fields such as transportation
policy and planning.
Another gap in the literature on transportation policy and planning is the dominance of
cross-sectional models. Travel behavior studies using cross-sectional datasets might suffer from
biased estimates if preferences for and attitudes pertaining to different travel modes have not
been controlled (Cao, Mokhtarian, & Handy, 2009). Specifically, attitudes towards driving are
often correlated with both the factors of interest (such as built environment factors) and travel
behavior (such as driving) (Ewing & Cervero, 2010). Studies have shown that omitting
attitudinal factors is more likely to inflate than deflate the real impacts of the factors of interest
on driving (Cao et al., 2009), even if such endogeneity issues could theoretically impact the
estimations in both directions (Cao & Chatman, 2016). Assuming those unobservable factors are
constant over time, longitudinal studies, controlling for personal fixed effects are theoretically
the best way to eliminate such “self-selection biases” (Mokhtarian & Cao, 2008).
This study tries to fill these gaps by analyzing the health impact of driving using a
nationally representative dataset – Health and Retirement Study (HRS) – controlling for socio-
economic, built environment and policy variables. The health conditions include both overall
health conditions and specific health conditions such as physical, cognitive and vision
conditions. The study also examines whether the impact of overall health conditions on driving
differs across seniors with different socio-economic characteristics such as gender, race and
poverty status. In addition, this study also contributes to a small but growing body of research
that applies longitudinal studies in transportation policy and planning (e.g. (Boarnet, Wang, &
Houston, 2016)). Using data from the HRS, I apply a series of fixed effects logit regressions to
examine the impacts of the overall and specific health conditions on driving for Americans who
78

are 65 or older. In the next session, I introduce the sample and the methods employed in this
study. Then I demonstrate the results of the models and discuss their implications for
practitioners and researchers in transportation policy and planning.

2. Methods
2.1. Sample of study
This study uses a longitudinal sample containing information on driving, socio-economic,
built environment, policy, and health characteristics of Americans 65 or older in 2006, 2008,
2010, 2012 and 2014. I constructed this sample using the HRS, a nationally representative survey
of Americans who are over 50 years old. The HRS is sponsored by the National Institute on
Aging and is conducted by the University of Michigan. The HRS surveys the same respondents
to collect information on demographics, family structure, personal income and wealth, pensions
and insurance policies, retirement plans and health conditions every year from 1992 to 1996, and
every two years since then. Currently the most recent wave of HRS data available to the public is
for 2014 (University of Michigan, 2017). I constructed the dataset for this study using the most
recent five waves of the HRS. I am not able to include earlier waves because the HRS did not
collect information on driving before 2006. In the HRS survey, all respondents who were 65 or
older were asked “Are you able to drive?” and if the answer was “yes” then another follow-up
question was posed “Have you driven in the past month?” I used the answers to these questions
to construct the information on driving as whether the person had driven in the past month, with
“no” indicating the person had not. In other words, the respondents who “had not driven in the
past month” include both those who were not able to drive and those who were able to drive but
had chosen not to. The dataset comes from both the RAND file and the FAT file of the public-
79

use sample of the HRS (University of Michigan, 2017). As Table 4-1 shows, there are 16,103
individuals in the sample. Of these individuals, 37.1% were surveyed in all five waves.

Table 4-1 – Number of waves surveyed for the individuals in the sample
Number of waves being surveyed Number of
individuals
Share
Five waves 5,974 37.1%
Four waves 2,471 15.3%
Three waves 2,310 14.3%
Two waves 2,607 16.2%
One wave 2,741 17.0%
Sum 16,103 100.0%

Table 4-2 shows the driving, socio-economic and built environment characteristics of the
study sample by survey waves. The study sample is not weighted. The table shows that roughly
73% of the survey respondents had driven in the past month. Around half of the individuals
surveyed had driven across all the waves surveyed, and 16% of the individuals surveyed had not
driven in any of the waves surveyed (Table 4-3). In addition, the respondents were on average in
their mid-70s; around 10% were below the poverty line; 75% of the respondents were non-
Hispanic white; around 9% were Hispanic and around 14% were non-Hispanic Black; 57% of
the respondents were female; 30% of the respondents were living alone; and 18% of the
respondents were employed at the time, either full-time or part-time
6
. In addition, 72% of the
respondents were living in single-family houses, which is unfortunately the only available built
environment variable because of the confidentiality concerns of the public HRS sample. These
average statistics, except for the ages, are constant across the five waves. However, the

6
According to the 2010 Census, 56.9% of the 65-or-older were female. According to the 2009-2011 American
Community Survey, 2.5% of the 65-or-older were employed full time. Other socio-economic variables for the 65-or-
order, including being employed part-time, are not available.
80

magnitude of increases (half a year) across waves are smaller than the time intervals (two years),
indicating that the HRS constantly includes new and younger respondents in each wave. Thus,
this sample is not a balanced panel dataset. Specifically, the average number of waves per
respondent is 3.39. Also, for each wave, the data in this sample constitute more than 99% of the
respondents being asked questions on driving (in other words, 65 or older). This indicates that
the sample of this study is representative of the HRS sample.

Table 4-2 – Driving, socio-economic and built environment characteristics of the study sample, by wave
Variable Mean S.D. Mean S.D. Mean S.D. Mean S.D. Mean S.D.
Wave 8 9 10 11 12
Respondent had driven in the past month (=1 if yes) 0.73 (0.45) 0.73 (0.44) 0.73 (0.44) 0.73 (0.44) 0.74 (0.44)
Respondent was able to drive (=1 if yes) 0.79 (0.42) 0.79 (0.41) 0.79 (0.41) 0.79 (0.41) 0.80 (0.40)
Age of respondent at the end of survey 74.91 (7.64) 75.26 (7.50) 75.84 (7.37) 76.10 (7.39) 76.17 (7.57)
Family is below poverty line (=1 if yes) 0.09 (0.29) 0.10 (0.30) 0.11 (0.31) 0.11 (0.31) 0.10 (0.31)
Living in a single-family house (=1 if yes) 0.71 (0.45) 0.72 (0.45) 0.72 (0.45) 0.72 (0.45) 0.71 (0.45)
Non-Hispanic white (=1 if yes) 0.77 (0.42) 0.76 (0.42) 0.76 (0.43) 0.75 (0.43) 0.73 (0.44)
Non-Hispanic black (=1 if yes) 0.13 (0.34) 0.13 (0.34) 0.14 (0.35) 0.14 (0.34) 0.14 (0.35)
Hispanic (=1 if yes) 0.08 (0.27) 0.08 (0.27) 0.09 (0.28) 0.09 (0.29) 0.10 (0.30)
Female (=1 if yes) 0.57 (0.49) 0.58 (0.49) 0.58 (0.49) 0.58 (0.49) 0.59 (0.49)
Living alone (=1 if yes) 0.30 (0.46) 0.30 (0.46) 0.30 (0.46) 0.30 (0.46) 0.30 (0.46)
Employed (=1 if yes) 0.18 (0.39) 0.19 (0.39) 0.17 (0.38) 0.17 (0.38) 0.17 (0.38)
Year 2006 2008 2010 2012 2014
Number of respondents 11,342 11,329 10,917 10,716 10,335

Table 4-3 – Patterns of driving across waves
Patterns of driving across waves
Number of
individuals
Share
Drove in all waves surveyed 8,180 50.8%
Did not drive in any waves surveyed 2,549 15.8%
Changed from driving to not driving 2,114 13.1%
Changed from not driving to driving 132 0.8%
Changed driving status at least twice 387 2.4%
Only surveyed in one wave 2,741 17.0%
Sum 16,103 100.0%

81

2.2. Analytical strategy
This study tries to answer the following three questions: (1) do health conditions
influence senior driving, and if they do, in what magnitude? (2) do health conditions influences
seniors with different socio-economic characteristics differently? And (3) what specific health
conditions can impact senior driving? In order to answer these questions using this longitudinal
dataset, I use a group of fixed effects logit regression models, as indicated in the formula below:

𝑙𝑜𝑔𝑖𝑡 (𝑑𝑟𝑖𝑣𝑒 ) = β
+ β
ℎ𝑒𝑎𝑙𝑡 ℎ
+ 𝐗 𝐢𝐭 ∙ β
𝟐 + 𝛍 𝐭 + 𝛎 𝒊 + ε
. (4-1)

In this formula, 𝑑𝑟𝑖𝑣𝑒 is the dependent variable which equals one if the respondent i
had driven in the past month in year t (it equals zero otherwise); ℎ𝑒𝑎𝑙𝑡 ℎ
indicates the health
conditions of respondent i in year t; vector 𝐗 𝐢𝐭 includes the socio-economic and built
environment factors that are theoretically related to driving, including poverty, being employed,
living in a single-family house, living in a single-person household, age and age squared (refer to
Table 4-2 for descriptive statistics); vector 𝐗 𝐢𝐭 also includes dummy variables for the U.S. census
divisions of the respondents, in order to control for regional-specific factors. In addition, these
census division fixed-effect variables could also help to control for different licensing policies
for senior drivers across different geographical areas. Ideally, dummy variables for states should
be used here, but unfortunately the public HRS sample does not contain states of residence for
confidentiality concerns. Vectors 𝛍 𝐭 and 𝛎 𝒊 contain the wave (time) and individual fixed effects,
respectively. Specifically, 𝛎 𝒊 includes gender and racial profiles of respondent i since they
generally do not change over time. In other words, the impact of gender and race on senior
driving cannot be directly estimated in those models. In these fixed effects logistic regressions,
82

the local average treatment effects used in the estimations come from the variations of the
dependent and independent variables for the same individual i across different waves (time
points) t.
The health variables, shown in Table 4-4, include both overall and specific health
conditions. Here I propose two variables on overall health conditions: a factor variable on self-
rated health (“excellent”, “very good”, “good”, “fair” and “poor”), and a dummy variable for
whether respondent i was in “fair” or “poor” health in year t. According to Table 4-4, in each
wave, around a third of the respondents were in “fair” or “poor” self-rated health. In addition, to
investigate whether the impact of the overall health condition on driving can change for seniors
in different socio-economic circumstances, I introduce a series of models containing dummy
variable “‘fair’ or ‘poor’ self-rated health” and its interaction with various socio-economic and
built environment characteristics, including being in poverty, being female, living alone, living in
single-family housing and being in a certain race.

83

Table 4-4 – Descriptive statistics on health conditions, by wave
Variable Mean S.D. Mean S.D. Mean S.D. Mean S.D. Mean S.D.
Wave 8 9 10 11 12
Overall health
Self-rated health (categorized)

excellent 0.08 (0.28) 0.07 (0.26) 0.08 (0.27) 0.07 (0.26) 0.07 (0.25)

very good 0.26 (0.44) 0.27 (0.44) 0.29 (0.45) 0.29 (0.45) 0.28 (0.45)

good 0.32 (0.46) 0.33 (0.47) 0.33 (0.47) 0.33 (0.47) 0.34 (0.47)

fair 0.23 (0.42) 0.23 (0.42) 0.21 (0.41) 0.22 (0.41) 0.23 (0.42)

poor 0.10 (0.30) 0.10 (0.31) 0.09 (0.29) 0.09 (0.29) 0.09 (0.28)
With "fair" or "poor" self-rated health (=1 if yes) 0.34 (0.47) 0.33 (0.47) 0.30 (0.46) 0.31 (0.46) 0.32 (0.47)
Physical conditions
Body-mass index (categorized)

normal 0.33 (0.47) 0.32 (0.47) 0.31 (0.46) 0.31 (0.46) 0.30 (0.46)

underweight 0.02 (0.15) 0.02 (0.15) 0.02 (0.14) 0.02 (0.15) 0.02 (0.15)

overweight 0.38 (0.48) 0.38 (0.48) 0.37 (0.48) 0.37 (0.48) 0.37 (0.48)

obese 0.27 (0.44) 0.28 (0.45) 0.29 (0.46) 0.30 (0.46) 0.31 (0.46)
Difficulties in activities of daily living (=1 if yes) 0.22 (0.42) 0.22 (0.41) 0.23 (0.42) 0.22 (0.42) 0.23 (0.42)
Difficulties in activities using large muscles (=1 if
yes)
0.68 (0.47) 0.67 (0.47) 0.69 (0.46) 0.68 (0.47) 0.68 (0.47)
Diagnosed high blood pressure (=1 if yes) 0.64 (0.48) 0.67 (0.47) 0.69 (0.46) 0.70 (0.46) 0.70 (0.46)
Diagnosed heart diseases (=1 if yes) 0.32 (0.47) 0.32 (0.47) 0.33 (0.47) 0.33 (0.47) 0.33 (0.47)
Diagnosed arthritis (=1 if yes) 0.68 (0.47) 0.69 (0.46) 0.70 (0.46) 0.70 (0.46) 0.70 (0.46)
Cognitive conditions

Diagnosed stroke (=1 if yes) 0.10 (0.30) 0.10 (0.30) 0.11 (0.31) 0.11 (0.32) 0.11 (0.31)
Having depression (=1 if CESD score > 0) 0.60 (0.49) 0.58 (0.49) 0.57 (0.49) 0.58 (0.49) 0.57 (0.50)
Fair or poor self-rated memory (=1 if yes) 0.33 (0.47) 0.32 (0.47) 0.32 (0.47) 0.33 (0.47) 0.34 (0.47)
Not often feeling relaxed after sleep (=1 if yes) 0.36 (0.48) 0.40 (0.49) 0.40 (0.49) 0.41 (0.49) 0.43 (0.50)
Diagnosed psychiatric diseases (=1 if yes) 0.16 (0.36) 0.16 (0.37) 0.17 (0.37) 0.17 (0.38) 0.18 (0.38)
Vision conditions

Fair or poor self-rated distant vision (=1 if yes) 0.16 (0.37) 0.15 (0.36) 0.15 (0.36) 0.16 (0.37) 0.16 (0.37)
Fair or poor self-rated near vision (=1 if yes) 0.19 (0.39) 0.19 (0.39) 0.20 (0.40) 0.21 (0.41) 0.21 (0.41)
Year 2006 2008 2010 2012 2014
Number of respondents 11,342 11,329 10,917 10,716 10,335

Following the categorization of the California Department of Motor Vehicles (2017), I
also propose 13 specific health variables in this study based on physical, cognitive and vision
conditions. There are six variables on physical health conditions: a factor variable on whether a
person is underweight, overweight or obese (with normal weight as reference), a dummy variable
84

on difficulties in activities of daily living (either in bathing, dressing, eating, bedding or
walking), a dummy variable on difficulties in activities using large muscles (sitting for two
hours, getting up from a chair, kneeling, or pushing large objects), and three dummy variables
for being diagnosed with high blood pressure, heart disease, or arthritis. There are also five
dummy variables for cognitive conditions, including whether the respondent had a stroke,
experienced depression, had fair or poor self-rated memory, insomnia, or psychiatric diseases.
Finally, there are two dummy variables for vision conditions, one for near and one for distant
vision. As Table 4-4 shows, roughly 30% of the total respondents were obese; and about 40%
had at least some symptoms of depression. Besides the fixed effects logistic model including all
these 13 physical, cognitive and vision conditions, I also propose 13 fixed effects models, each
of which includes only one health variable. For these 13 models, the estimated size of
associations between a specific health condition and driving will include both the “direct effects”
from the condition of interest, and the “indirect effects” from other conditions correlated with
this condition.
The application of the fixed effects models using a longitudinal dataset is helpful to
reduce biases in estimation, which often occurs in cross-sectional studies in transportation policy
and planning. In my models, I include individual fixed effects to control for individual
characteristics which do not change over time. Here I assume that all factors not included in 𝐗 𝐢𝐭 ,
which might correlate with both driving and health conditions, are constant across these five
waves. Among these factors, preferences and attitudes towards driving are the biggest source of
biased estimates in cross-sectional studies in transportation policy and planning (Boarnet, 2011).
To my knowledge, no previous studies have addressed whether preferences and attitudes towards
driving correlate with both driving and health conditions simultaneously. Assuming those
85

attitudes to be constant over time, I can reduce the possibility of biased estimates in these
models.

3. Results
The results of the fixed effects logit models show that, controlling for socio-economic
and built environment variables, deteriorating health conditions discourage seniors from driving.
Table 4-5, Table 4-6 and Table 4-7 show the outputs of the models with overall health variables,
overall health interacted with socio-economic variables and specific health variables. Note that a
large share of the respondents did not change driving status across all the waves they were
surveyed in. Observations with constant dependent variables across all waves cannot provide the
variations needed in the estimation of the fixed effects logistic models. Thus, those observations
cannot be included in the estimations. As shown in Table 4-5, Table 4-6 and Table 4-7, around
20% of the sample are used to estimate the coefficients. However, the estimations are not biased
even if some observations do not contribute to the local average treatment effects used to identify
the impacts of health on senior driving. As a robustness check I re-ran these 22 models using
ordinary least squared (OLS) regressions with fixed effects (in which 100% of the sample are
used). The outputs of these regular fixed effects models support the conclusions made from those
fixed effects logit models
7
.
Since the coefficients in logit regressions are challenging to interpret directly, I used odds
ratios in interpretations. The odds ratio is defined following the equation below:

7
The outputs of these OLS models with fixed effects are available in the appendix (Section 4-5).
86

𝑜𝑑𝑑𝑠 𝑟𝑎𝑡𝚤𝑜 =
𝑃 (𝑦 = 1|𝑥 = 1)
𝑃 (𝑦 = 0|𝑥 = 1)
𝑃 (𝑦 = 1|𝑥 = 0)
𝑃 (𝑦 = 0|𝑥 = 0)
= exp 𝛽 , (4 − 2)

where y is the dependent dummy variable, x is the independent dummy variable of interest, and
𝛽 is the estimated coefficient of x on y. According to this equation, the odds ratio is the ratio of
the odds of y occurring given x equals 1, to the odds of y occurring given x equals 0; and the
definition of odds is the ratio of the probability of y happening to that of y not happening, given a
specific value of x. The odds ratio for the logistic regressions is equal to the exponential of the
coefficients of the variable of interest. An odds ratio equal to one means there is no effect of x on
y; an odds ratio larger than one means x has a positive effect on y; and an odds ratio smaller than
one means x has a negative effect on y.

3.1. Impact of overall health conditions on driving
As Models 1-2 show, the overall self-rated health conditions have significant impacts on
driving for the elderly (Table 4-5). A senior citizen in “fair” or “poor” health is less likely to
drive compared to those in “excellent” health, with an odds ratio of 0.543 or 0.308, respectively.
In contrast, neither of the coefficients of “very good” and “good” is significant at the 5% level,
indicating that neither of the odds for a senior in these two conditions to drive is different from a
person in “excellent” health. Model 2 groups the health conditions “fair” and “poor” and
compared them with the grouped health conditions “excellent,” “very good” and “good.” It
shows that seniors in “fair” or “poor” health are less likely to drive than those with “excellent,”
“very good” or “good” health, with an odds ratio of 0.544.

87

Table 4-5 – Impact of overall health conditions on senior driving
Dependent Variables
Model 1 Model 2
Coef.
p-
value
Odds
Ratio
Coef.
p-
value
Odds
Ratio
Self-rated health (categorized)

excellent reference

very good
-0.101
0.593

(0.189)

good
-0.161
0.410

(0.195)

fair
-0.611***
0.003 0.543

(0.202)

poor
-1.177***
<0.001 0.308

(0.218)

With "fair" or "poor" self-rated health (=1
if yes)

-0.608***
<0.001 0.544

(0.088)
Family is below the poverty line (=1 if
yes)
-0.220*
0.081 0.803
-0.215*
0.088 0.807
(0.126) (0.126)
Living in a single-family house (=1 if
yes)
0.235*
0.057 1.265
0.251**
0.041 1.286
(0.124) (0.123)
Living alone (=1 if yes)
0.371***
0.005 1.449
0.390***
0.003 1.478
(0.133) (0.133)
Employed (=1 if yes)
1.261***
<0.001 3.530
1.305***
<0.001 3.689
(0.207) (0.207)
Age
1.880***
<0.001 N/A
1.882***
<0.001 N/A
(0.196) (0.196)
Age squared
-0.016***
<0.001 N/A
-0.016***
<0.001 N/A
(0.001) (0.001)
N 10,591 10,591
Note: conditional fixed effects logit models estimated by within-individual variations, *, ** and *** indicate
significant at 0.1, 0.05 and 0.01 level, respectively; N indicates number of observations used after dropping
individuals with constant driving choices across the five waves; standard errors in parentheses; controlled for
census division fixed effects, poverty, employment status, type of house, living alone, age, age squared and wave
fixed effects. N/A indicates no odds ratio because the corresponding independent variable is continuous.

Models 1-2 also show that the magnitudes of the impacts from overall health conditions
on senior driving are larger than those from poverty status, residential patterns and family
structure (Table 4-5). In both models, being in poverty and being older are both negatively
associated with driving for the seniors. In addition, a positive coefficient of age and a negative
88

coefficient of age squared indicates that the impact of age on driving increases as age increases.
In contrast, living in single-family houses, living alone and being employed are all positively
associated with senior driving. However, the former two variables are significant only at a 10%
level. The signs of the coefficients of these variables all conform with the empirical findings in
the literature on transportation policy and planning, as summarized by Ewing and Cervero (2010)
and Boarnet (2011). In addition, the absolute values of the coefficients of the overall health
variables are larger than those of being in poverty, living in single-family houses and living
alone.

3.2. Impact of overall health across socio-demographic groups
Models 3-8, presented in Table 4-6, show that the impact of overall health on senior
driving does not vary with poverty status, gender or employment status. These models include
the overall health condition variable interacted with six socio-economic factors that could
theoretically impact senior driving. The overall health variable is the dummy variable “with
‘fair’ or ‘poor’ health condition,” and the six socio-economic factors are being in poverty, being
female, living alone, being employed, living in single-family houses and being in a certain race.
As Table 4-6 shows, the interaction terms between the dummy variable on overall health
condition and being in poverty, being female and being employed are not significant. The results
indicate that having “fair” or “poor” self-rated health has the same impact for the seniors with
different income levels, genders and employment status. In contrast, the interaction term between
overall health and being in a certain race is significant at a 5% level (Model 8). Such results
indicate that the impact of overall health condition on the non-Hispanic African American
elderly is significantly smaller than that on the non-Hispanic white elderly. The impact of health
89

on driving for the Hispanic elderly or the elderly in other races is not different from that of non-
Hispanic white elderly. In addition, the interaction terms between overall health and living alone
(Model 5) and overall health and living in single-family houses (Model 7) are significant at the
10% level. That indicates that the overall health condition has a larger impact on driving for the
elderly not living alone or living in single family houses, although the statistical significance of
these differences is smaller.

Table 4-6 – Impact of overall health on driving across different socio-demographic groups
Model Health Dependent Variable Coef. Std. Err. p-value N
3 With "fair" or "poor" self-rated health
-0.627*** (0.092) <0.001
10,591
Health interacted with poverty
0.150 (0.213) 0.482
4 With "fair" or "poor" self-rated health
-0.646*** (0.138) <0.001
10,591
Health interacted with female
0.063 (0.179) 0.723
5 With "fair" or "poor" self-rated health
-0.702*** (0.105) <0.001
10,591
Health interacted with living alone
0.287* (0.170) 0.092
6 With "fair" or "poor" self-rated health
-0.612*** (0.090) <0.001
10,591
Health interacted with employed
0.075 (0.351) 0.831
7 With "fair" or "poor" self-rated health
-0.411*** (0.147) 0.005
10,591
Health interacted with living in single-
family housing
-0.281* (0.168) 0.095
8 With "fair" or "poor" self-rated health
-0.745*** (0.104) <0.001
10,591
Health interacted with race

non-Hispanic white
reference
non-Hispanic black
0.557** (0.234) 0.017
Hispanic
0.450 (0.320) 0.159
other races
0.158 (0.651) 0.809
Note: conditional fixed effects logit models estimated by within-individual variations, *, ** and *** indicate
significant at 0.1, 0.05 and 0.01 level, respectively; N indicates number of observations used after dropping
individuals with constant driving choices across the five waves; standard errors in parentheses; controlled for
census division fixed effects, poverty, employment status, type of house, living alone, age, age squared and wave
fixed effects.

3.3. Impact of specific health conditions on driving
Models 9-22, shown in Table 4-7, identify specific physical, cognitive and vision
conditions significantly correlates to senior driving. Model 9 includes all the 14 health condition
90

variables and identifies seven health variables associated with senior driving, controlling for
socio-economic, built environment and policy characteristics, and wave fixed effects. Compared
to the “normal weight” category, senior citizens who are obese are more likely to drive with an
odds ratio of 1.618. In addition, factor “overweight” is also positively associated with senior
driving but at a 10% significance level. Senior citizens with limitations in physical strength are
less likely to drive compared to those without them. Specifically, seniors with difficulties
involving any activities in daily living (bathing, dressing, eating, bedding or walking) are less
likely to drive with an odds ratio of 0.453, while seniors with difficulties involving any activities
requiring large muscles (sitting for two hours, getting up from a chair, kneeling or pushing large
objects) are also less likely to drive with an odds ratio of 0.745. Seniors diagnosed with high
blood pressure are more likely to drive with an odds ratio of 1.411. Heart disease or arthritis have
no significant effects on senior driving at a 5% level.

91

Table 4-7 – Impact of specific health conditions on senior driving
Health Dependent Variable
Model 9 Models 10-22
Coef.
p-
value
Odds
Ratio
N Coef.
p-
value
Odds
Ratio
N
Physical conditions

8,458

Body-mass index (categorized)

10,613
normal
reference Reference
underweight -0.114 0.694

-0.488** 0.055

(0.289) (0.254)
overweight 0.261* 0.052

0.330*** 0.005 1.392
(0.134) (0.118)
obese 0.481** 0.012 1.618 0.432*** 0.009 1.540
(0.191) (0.165)
Difficulties in activities of
daily living (=1 if yes)
-0.792*** <0.001 0.453 -1.096*** <0.001 0.334 10,613
(0.104) (0.092)
Difficulties in activities using
large muscle (=1 if yes)
-0.294*** 0.008 0.745 -0.566*** <0.001 0.568 10,613
(0.111) (0.098)
Diagnosed high blood pressure
(=1 if yes)
0.344** 0.046 1.411 0.097 0.504

10,559
(0.172) (0.145)
Diagnosed heart diseases (=1 if
yes)
-0.211 0.194

-0.286** 0.041 0.751 10,572
(0.162) (0.140)
Diagnosed arthritis (=1 if yes) -0.001 0.996

0.125 0.367

10,585
(0.171) (0.138)
Cognitive conditions

Diagnosed stroke (=1 if yes) -0.578*** 0.002 0.561 -0.875*** <0.001 0.417 10,582
(0.190) (0.160)
Having depression (=1 if
CESD score > 0)
-0.269*** 0.004 0.764 -0.538*** <0.001 0.584 10,613
(0.093) (0.085)
Fair of poor self-rated memory
(=1 if yes)
-0.158 0.101

-0.220** 0.017 0.803 8,690
(0.096) (0.092)
Not often feeling relaxed after
sleep (=1 if yes)
0.016 0.853

-0.194** 0.013 0.823 10,483
(0.088) (0.078)
Diagnosed psychiatric diseases
(=1 if yes)
-0.091 0.633

-0.298** 0.042 0.742 10,562
(0.191) (0.146)
Vision conditions

Fair or poor self-rated distant
vision (=1 if yes)
-0.266** 0.022 0.766 -0.426*** <0.001 0.653 10,501
(0.116) (0.098)
Fair or poor self-rated near
vision (=1 if yes)
-0.051 0.627

-0.256*** 0.004 0.774 10,517
(0.105) (0.090)
92

Note: conditional fixed effects logit models estimated by within-individual variations, *, ** and *** indicate
significant at 0.1, 0.05 and 0.01 level, respectively; N indicates number of observations used after dropping
individuals with constant driving choices across the five waves; standard errors in parentheses; controlled for
census division fixed effects, poverty, employment status, type of house, living alone, age, age squared and
wave fixed effects.

Model 9 also shows that cognitive and vision conditions significantly impact senior
driving (Table 4-7). Specifically, seniors who had suffered from a stroke or who had been
diagnosed depression were less likely to drive, with odds ratios of 0.561 and 0.764, respectively.
In addition, seniors with “fair” or “poor” self-rated distant vision were less likely to drive than
those with “excellent,” “very good” or “good” distant vision, with an odds ratio of 0.766. Other
variables for cognitive and vision conditions, including suffering from memory loss, reporting
frequent episodes of insomnia, being diagnosed with psychiatric diseases and with “fair” or
“poor” self-rated near vision, have no statistically significant effects on senior driving.
Health conditions often correlate with each other. Even if some health conditions do not
directly impact senior driving, they might be associated with senior driving by being correlated
with other health conditions impacting driving. From a practical perspective, policy makers and
planners might be interested in using health conditions to identify potential reductions in demand
for driving. In this case, they might be more interested in the simple associations between health
factors and senior driving, disregarding any confounding effects from other health conditions
which might correlate with the health conditions of interest.
Hence, here I also propose Models 10-22, each of which only includes one specific health
variable (Table 4-7). The findings of these models can be summarized as follows: First, five
specific health conditions, which do not have a statistically significant impact on senior driving
in Model 9, become significantly associated with senior driving. These five conditions are:
diagnosed heart diseases, arthritis, episodes of insomnia, memory loss, and “fair” or “poor” near
93

vision. Second, with one exception, all specific health conditions impacting senior driving in
Model 9 are also statistically significant in Models 10-22, the only exception is diagnosed high
blood pressure. The coefficients of the conditions that are significant in these two groups of
models is smaller in Model 9 than those in Models 10-22. Both of these two findings infer that
health conditions positively correlate with each other. Models 10-22 will be useful in identifying
health conditions associated with driving for the seniors, including both the direct effects
estimated in Model 9, and the indirect effects from other conditions correlated with the
conditions of interest.

4. Conclusion
Using a nationally-representative longitudinal dataset from the Health and Retirement
Study (HRS), this study shows that deteriorating health conditions discourage elderly Americans
from driving. Controlling for socio-economic, built-environment and regional-specific factors,
having “fair” or “poor” self-rated overall health makes a senior citizen less likely to drive
compared having “excellent”, “very good” or “good” health, with an odds ratio of 0.561. This
impact of overall self-rated health on driving is larger than those from poverty status, residential
patterns and family structure. In addition, the impact from overall self-rated health on driving is
larger for seniors living with other household members and in single-family houses; and smaller
for the non-Hispanic African American seniors relative to non-Hispanic white ones. The models
also show that specific health conditions, including being obese, having difficulties in activities
of daily living and using large muscles, being diagnosed with high blood pressure, having
suffered from a stroke, having depression and reporting “fair” and “poor” distant vision
94

significantly impact senior driving. I also identified other physical, cognitive and vision
conditions associated with senior driving.
The models proposed in this study imply that the existing passenger transportation system
in the U.S. will likely face enormous challenges in the future. With the current dominance of
private automobiles and the popularity of driving among the Baby Boomers, the current
American transportation system cannot fulfill the mobility needs of the elderly who start to drive
less because of declining health. On the one hand, most suburban areas where the majority of the
seniors are residing do not have an adequate supply of public transportation; on the other hand,
public transportation is not popular among seniors because of its inconvenience, safety concerns
and undesirable quality of service. Thus, seniors with deteriorating health conditions will have to
limit their mobility. As the US population ages, this problem will grow more severe.
Transportation policy makers and planners should be aware of this challenge and be
proactive about potential solutions. To draw more senior costumers, policy makers and planners
need to reshape the current public transportation system to making it more flexible, more
convenient and more senior-friendly. They could also use the findings of this study to work with
health providers to identify the seniors’ potential mobility needs based on changes in their health
conditions. In addition, the findings of this study imply a large market potential for self-driving
automobiles for the elderly. Driverless cars could provide the flexibility and convenience the
elderly need, without their worrying about their ability to drive due to health limitations.
This study explores an understudied topic in transportation policy and planning and
shows that health conditions can impact a senior’s decision to drive in addition to factors which
have been widely discussed in the literature: socio-economic, built environment and policy
factors. Applying a longitudinal dataset, this study also contributes to a currently small but
95

growing body of literature on longitudinal analyses in travel behavior. Controlling for individual
fixed effects is theoretically the most robust way to eliminate potential biases from personal
beliefs about and attitudes toward different travel modes.
Admittedly, confidentiality concerns prevent the public HRS sample from providing
more built-environment variables other than housing type. Even if some studies have argued that
housing type is more effective in predicting travel patterns compared to traditional
neighborhood-level built-environment variables such as density or access to public transportation
(Chatman, 2013), including these latter variables will be helpful as a robustness check. In
addition, including metropolitan-level urban form variables, requiring less confidential data, will
also be useful. As discussed by Bento, Cropper, Mobarak, and Vinha (2005), metropolitan-level
urban form characteristics such as population centrality or public transportation density are even
more effective in predicting travel mode choice compared to neighborhood-level urban design
patterns.
Hence, researchers and practitioners in transportation policy and planning need travel
survey data that are longitudinal and that include more detailed health information about the
survey respondents. One limitation of this study is that I could explore only one travel behavior,
since the HRS does not have information on travel behaviors for other modes. Travel surveys
generally have detailed information on the mobility patterns of the respondents, but most travel
surveys are cross-sectional and do not contain detailed health information. Admittedly,
longitudinal travel surveys with both comprehensive mobility patterns and detailed health
conditions will be very costly. However, they will also be highly beneficial for researchers and
practitioners to better understand the mobility patterns of the elderly and to better prepare for the
oncoming mobility challenges among the growing elderly population.
96

5. Appendix – Outputs of OLS fixed effects models

Table 4-8 – Impact of overall health conditions on senior driving (OLS models with fixed effects)
Dependent Variables
Model 1 Model 2
Coef. p-value Coef. p-value
Self-rated health (categorized)

excellent
reference

very good 0.004 0.478

(0.006)

good 0.005 0.418

(0.006)

fair -0.026*** <0.001

(0.007)

poor -0.105*** <0.001

(0.008)

With "fair" or "poor" self-rated health
(=1 if yes)
-0.045*** <0.001
(0.004)
Family is below the poverty line (=1 if
yes)
-0.024*** <0.001 -0.024*** <0.001
(0.005) (0.005)
Living in a single-family house (=1 if
yes)
0.037*** <0.001 0.038*** <0.001
(0.005) (0.005)
Living alone (=1 if yes) 0.035*** <0.001 0.035*** <0.001
(0.005) (0.005)
Employed (=1 if yes) 0.026*** <0.001 0.028*** <0.001
(0.005) (0.005)
Age 0.131*** <0.001 0.134*** <0.001
(0.006) (0.006)
Age squared -0.001*** <0.001 -0.001*** <0.001
(0.000) (0.000)
N 54,182
Note: OLS fixed effects models estimated by within-individual variations, *, ** and *** indicate
significant at 0.1, 0.05 and 0.01 level, respectively; N indicates number of observations used;
standard errors in parentheses; controlled for census division fixed effects, poverty, employment
status, type of house, living alone, age, age squared and wave fixed effects.

97

Table 4-9 – Impact of overall health on driving across different socio-demographic groups (OLS models with
fixed effects)
Model Health Dependent Variable Coef. Std. Err. p-value N
3 With "fair" or "poor" self-rated health
-0.047*** (0.004) 0.000
54,182
Health interacted with poverty
0.018** (0.009) 0.038
4 With "fair" or "poor" self-rated health
-0.041*** (0.005) 0.000
54,182
Health interacted with female
-0.006 (0.007) 0.391
5 With "fair" or "poor" self-rated health
-0.051*** (0.004) 0.000
54,182
Health interacted with living alone
0.020*** (0.007) 0.003
6 With "fair" or "poor" self-rated health
-0.052*** (0.004) 0.000
54,182
Health interacted with employed
0.071*** (0.010) 0.000
7 With "fair" or "poor" self-rated health
-0.031*** (0.006) 0.000
54,182
Health interacted with living in single-
family housing
-0.019*** (0.007) 0.007
8 With "fair" or "poor" self-rated health
-0.054*** (0.004) 0.000
54,182
Health interacted with race

non-Hispanic white

non-Hispanic black
0.031*** (0.010) 0.001
Hispanic
0.031*** (0.012) 0.010
other races
0.040* (0.024) 0.099
Note: OLS fixed effects models estimated by within-individual variations, *, ** and *** indicate significant at
0.1, 0.05 and 0.01 level, respectively; N indicates number of observations used; standard errors in parentheses;
controlled for census division fixed effects, poverty, employment status, type of house, living alone, age, age
squared and wave fixed effects.

98

Table 4-10 – Impact of specific health conditions on senior driving (OLS models with fixed effects)
Health Dependent Variable
Models 10-22
Coef. p-value N Coef. p-value N
Physical conditions
49,116

Body-mass index
(categorized)

54,241
normal Reference

Reference
underweight
-0.023*
0.063

-0.048***
<0.001

(0.012) (0.011)
overweight
0.023***
<0.001
0.026***
<0.001
(0.005) (0.005)

obese
0.032***
<0.001
0.031***
<0.001
(0.007) (0.006)
Difficulties in activities of
daily living (=1 if yes)
-0.084***
<0.001
-0.123***
<0.001 54,241
(0.004) (0.004)
Difficulties in activities using
large muscle (=1 if yes)
-0.006*
0.092
-0.024***
<0.001 54,241
(0.003) (0.003)
Diagnosed high blood
pressure (=1 if yes)
0.020***
<0.001
0.016***
0.003 54,105
(0.006) (0.005)
Diagnosed heart diseases (=1
if yes)
-0.010*
0.071
-0.015***
0.005 54,160
(0.005) (0.005)
Diagnosed arthritis (=1 if
yes)
0.013**
0.022
0.031***
<0.001 54,134
(0.006) (0.005)
Cognitive conditions

Diagnosed stroke (=1 if yes)
-0.089***
<0.001
-0.113***
<0.001 54,183
(0.008) (0.007)
Having depression (=1 if
CESD score > 0)
-0.010***
0.001
-0.032***
<0.001 54,241
(0.003) (0.003)
Fair of poor self-rated
memory (=1 if yes)
-0.009***
0.006
-0.015***
<0.001 49,752
(0.003) (0.003)
Not often feeling relaxed
after sleep (=1 if yes)
0.004
0.205
-0.011***
<0.001 53,832
(0.003) (0.003)
Diagnosed psychiatric
diseases (=1 if yes)
-0.026***
0.001
-0.043***
<0.001 54,149
(0.008) (0.007)
Vision conditions

Fair or poor self-rated distant
vision (=1 if yes)
-0.031***
<0.001
-0.048***
<0.001 53,812
(0.005) (0.004)
Fair or poor self-rated near
vision (=1 if yes)
-0.006
0.138
-0.023***
<0.001 53,807
(0.004) (0.004)
Note: OLS fixed effects models estimated by within-individual variations, *, ** and *** indicate
significant at 0.1, 0.05 and 0.01 level, respectively; N indicates number of observations used; standard
errors in parentheses; controlled for census division fixed effects, poverty, employment status, type of
house, living alone, age, age squared and wave fixed effects.
99

6. References
Alsnih, R., & Hensher, D. A. (2003). The mobility and accessibility expectations of seniors in an
aging population. Transportation Research Part A: Policy and Practice, 37(10), 903-916.
Anstey, K. J., Windsor, T. D., Luszcz, M. A., & Andrews, G. R. (2006). Predicting Driving
Cessation over 5 Years in Older Adults: Psychological Well ‐Being and Cognitive
Competence Are Stronger Predictors than Physical Health. Journal of the American
Geriatrics Society, 54(1), 121-126.
Bento, A. M., Cropper, M. L., Mobarak, A. M., & Vinha, K. (2005). The effects of urban spatial
structure on travel demand in the United States. Review of Economics and Statistics,
87(3), 466-478.
Boarnet, M. G. (2011). A broader context for land use and travel behavior, and a research
agenda. Journal of the American Planning Association, 77(3), 197-213.
Boarnet, M. G., Wang, X., & Houston, D. (2016). Can new light rail reduce personal vehicle
carbon emissions? A before-after, experimental-control evaluation in Los Angeles.
Journal of Regional Science. doi:10.1111/jors.12275
Bush, S. (2003). Forecasting 65+ travel: An integration of cohort analysis and travel demand
modeling. (Ph.D. dissertation), Massachusetts Institute of Technology.
California Department of Motor Vehicles. (2017). Retrieved from
https://www.dmv.ca.gov/portal/dmv/detail/about/senior/senior_top
Cao, X., & Chatman, D. (2016). How will smart growth land-use policies affect travel? A
theoretical discussion on the importance of residential sorting. Environment and Planning
B: Planning and Design, 43(1), 58-73.
100

Cao, X., Mokhtarian, P. L., & Handy, S. L. (2009). Examining the impacts of residential self ‐
selection on travel behaviour: a focus on empirical findings. Transport Reviews, 29(3),
359-395.
Chatman, D. G. (2013). Does TOD Need the T? Journal of the American Planning Association,
79(1), 17-31.
Choi, M., & Mezuk, B. (2013). Aging without driving evidence from the health and retirement
study, 1993 to 2008. Journal of Applied Gerontology, 32(7), 902-912.
Edwards, J. D., Perkins, M., Ross, L. A., & Reynolds, S. L. (2009). Driving status and three-year
mortality among community-dwelling older adults. The Journals of Gerontology Series
A: Biological Sciences and Medical Sciences, 300–305.
Edwards, J. D., Ross, L. A., Ackerman, M. L., Small, B. J., Ball, K. K., Bradley, S., & Dodson,
J. E. (2008). Longitudinal predictors of driving cessation among older adults from the
ACTIVE clinical trial. The Journals of Gerontology Series B: Psychological Sciences
and Social Sciences, 63(1), P6-P12.
Ewing, R., & Cervero, R. (2010). Travel and the Built Environment--A Meta-Analysis. Journal
of the American Planning Association, 76(3), 265-294.
Fuller, R. (2005). Towards a general theory of driver behaviour. Accident Analysis & Prevention,
37(3), 461-472.
Giuliano, G. (2004). Land use and travel patterns among the elderly. Transportation in an aging
society, 27, 192-210.
Hu, H.-H. (2006). Travel Patterns, Land Use, and the Elderly. University of Southern California.
Lee, J. S., Zegras, P. C., Ben-Joseph, E., & Park, S. (2014). Does urban living influence baby
boomers’ travel behavior? Journal of Transport Geography, 35, 21-29.
101

Marottoli, R. A., de Leon, C. F. M., Glass, T. A., Williams, C. S., Cooney, L. M., & Berkman, L.
F. (2000). Consequences of driving cessation decreased out-of-home activity levels. The
Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 55(6),
S334-S340.
McFadden, D. (1973). Conditional logit analysis of qualitative choice behavior. Frontiers in
Econometrics, 105-142.
Mezuk, B., & Rebok, G. W. (2008). Social integration and social support among older adults
following driving cessation. The Journals of Gerontology Series B: Psychological
Sciences and Social Sciences, 63(5), S298-S303.
Mokhtarian, P. L., & Cao, X. (2008). Examining the impacts of residential self-selection on
travel behavior: A focus on methodologies. Transportation Research Part B:
Methodological, 42(3), 204-228.
Ortman, J. M., Velkoff, V. A., & Hogan, H. (2014). An aging nation: the older population in the
United States. Retrieved from Washington, DC.:
Rosenbloom, S. (2001). Sustainability and automobility among the elderly: An international
assessment. Transportation, 28(4), 375-408.
Rosenbloom, S., & Santos, R. (2014). Understanding older drivers: an examination of medical
conditions, medication use, and travel behavior. AARP Report.
Santos, A., McGuckin, N., Nakamoto, H., Gray, D., & Liss, S. (2011). Summary of Travel
Trends: 2009 National Household Travel Survey. Retrieved from
University of Michigan. (2017). Health and retirement study user guides. Retrieved from
https://hrs.isr.umich.edu/documentation/user-guides

102

Chapter 5. Conclusions and Takeaways

1. Conclusions
Using comprehensive datasets for automobile travel, this research provides a closer look
at the influence of demographic changes on automobile travel in the United States. Using census
and ACS data for 2000 and 2010, the double-cohort models reveal the countervailing effects of
generational shifts and immigration assimilation on the future demand for commuting by
automobiles in Los Angeles County, California. A younger and less-auto-dependent native-born
workforce favors a less auto-oriented Los Angeles in the future, while a more assimilated and
more auto-dependent foreign-born workforce favors a more auto-oriented Los Angeles. Using
nationwide travel diary data for 1995, 2001 and 2009, I found that controlling for socio-
economic, vehicle ownership, lifecyle, year-specific and regional-specific factors, the young
adults in 2009 (Millennials) have a weaker association between residential density and
automobile travel than their 1995 counterparts (Generation-Xers). That indicates that the
Millennials might still travel less by automobiles than the previous generation when residing in
less dense suburban neighborhoods. Using the most recent five waves of HRS (2006, 2008,
2010, 2012 and 2014) to control individual-specific factors, my fixed-effects models show that
deteriorating health conditions make the elderly less likely to drive. Hence, the current
transportation system in the US cannot fulfill the mobility needs of a more aged population in the
future.
This research does not try to provide a simple answer to the question “will the current
demographic shifts increase or decrease the overall demand of automobile travel?” Nevertheless,
103

the findings of this research do hint at the potential impact of each of these aforementioned
demographic change on automobile travel in the US. Generational shifts, especially the
Millennials entering adulthood, favor a less auto-dependent future. Besides, aging is also likely
to favor a less auto-dependent future. In contrast, a more assimilated immigration population will
be more auto-dependent. Recently, travel mileage by automobiles has reached an all-time high
(Federal Highway Administration, 2017). This is likely because of the recovery of the economy
and the reduction in fuel prices (Bastian, Börjesson, & Eliasson, 2016). This research suggests
that controlling for the economic conditions and fuel prices, the demographic changes still
significantly influence the demand for automobile travel.
This research has three takeaways for the practitioners in transportation policy and
planning. First of all, the practitioners should be aware of the ongoing demographic changes and
their implications as to automobile travel. The assimilation of immigrants, Millennials entering
adulthood and aging are likely to bring about a future landscape of the demand for automobile
travel that differs quite dramatically from the current one. Second, the practitioners hence need to
incorporate demographic factors when forecasting travel demands. Thirdly, these planners and
policy makers should test the pilot projects targeting specific demographic sub-groups in the
population. The findings of this research indicate the potential for travel demand management
policies for the suburban Millennials, and for providing flexible public transportation services to
the elderly.
From the perspective of researchers, this research examines a currently understudied
topic in transportation policy and planning: the impacts of demographic changes on travel. The
generational theory proposed by demographers suggests that people in different demographic
groups might have different patterns as to automobile travel (Ryder, 1965). In addition, the third
104

essay in this dissertation is among the first to examine the impact of health on elderly driving in
the field of transportation policy and planning. These studies cannot be finished without datasets
covering different time points. Analyzing repeated cross-sectional datasets enables us to track
demographic groups over time; longitudinal datasets even allow us to track individuals over
time. None of these are possible by merely using cross-sectional data. This research encourages
future studies to collect and analyze longitudinal datasets in order to provide more robust policy
implications.

2. References
Bastian, A., Börjesson, M., & Eliasson, J. (2016). Explaining “peak car” with economic
variables. Transportation Research Part A: Policy and Practice, 88, 236-250.
Federal Highway Administration. (2017). Traffic Volume Trends.
https://www.fhwa.dot.gov/policyinformation/travel_monitoring/tvt.cfm
Ryder, N. B. (1965). The cohort as a concept in the study of social change. American
Sociological Review, 843-861.

Abstract (if available)

Abstract This research quantitatively examines whether the current demographic changes in the United States are linked to changes in automobile travel using comprehensive datasets. Specifically, this research focuses on the impact of three major demographic shifts: immigration, Millennials entering adulthood, and aging. The findings of the three essays suggest that these aforementioned demographic changes significantly influence the demand of automobile travel, controlling for socio-economic, vehicle ownership, time-specific and regional-specific factors. Understanding the link from demographics to automobile travel creates opportunities for policy makers to transform American cities to be more sustainable and to more effectively predict future travel patterns based on demographic trends.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Who learns where: understanding the equity implications of charter school reform in the District of Columbia

PDF

Unraveling decentralization of warehousing and distribution centers: three essays

PDF

Location of warehouses and environmental justice: Three essays

PDF

Essays on the economics of cities

PDF

The demand for reliable travel: evidence from Los Angeles, and implications for public transit policy

PDF

The long-term impact of COVID-19 on commute, employment, housing, and environment in the post-pandemic era

PDF

Three essays on housing demographics: depressed housing access amid crisis of housing shortage

PDF

Urban air pollution and environmental justice: three essays

PDF

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

PDF

The built environment, tour complexity, and active travel

PDF

Essays on congestion, agglomeration, and urban spatial structure

PDF

Active travel, outdoor leisure, and neighborhood environment: path analysis, Los Angeles County

PDF

Environmental justice in real estate, public services, and policy

PDF

Productive frictions and urbanism in transition: planning lessons from traffic flows and urban street life in Ho Chi Minh City, Vietnam

PDF

The flexible workplace: regional tendencies and daily travel implications

PDF

Property and labor formalization in the age of the sharing economy: Airbnb, housing affordability, and entrepreneurship in Havana

PDF

Spatial and temporal expenditure-pricing equity of rail transit fare policies

PDF

The role of public policy in the decisions of parents and caregivers: an examination of work, fertility, and informal caregiving

PDF

The impact of mobility and government rental subsidies on the welfare of households and affordability of markets

PDF

The interactions between housing and business

Asset Metadata

Creator Wang, Xize (author)

Core Title The impact of demographic shifts on automobile travel in the United States: three empirical essays

School School of Policy, Planning and Development

Degree Doctor of Philosophy

Degree Program Urban Planning and Development

Publication Date 07/17/2017

Defense Date 06/23/2017

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

Tag demography,health,longitudinal studies,OAI-PMH Harvest,Planning,Transportation

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Boarnet, Marlon (committee chair), Giuliano, Genevieve (committee member), Myers, Dowell (committee member), Zissimopoulos, Julie (committee member)

Creator Email xizewang@usc.edu,xwang0624@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-399661

Unique identifier UC11265693

Identifier etd-WangXize-5517.pdf (filename),usctheses-c40-399661 (legacy record id)

Legacy Identifier etd-WangXize-5517.pdf

Dmrecord 399661

Document Type Dissertation

Rights Wang, Xize

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