Abstract
Do children suffer long-term consequences when they grow up without a car? To answer that question, this article uses propensity score matching and longitudinal data from the Panel Study of Income Dynamics. Young adults who were carless as children completed less education, worked for pay less often, experienced more unemployment, and earned less than their matched peers with consistent car access. The matching process allows me to compare like to like; it accounts for differences in income, wealth, residential location, family composition, and race. These results suggest that transportation disadvantage contributes to the intergenerational transmission of economic standing.
Introduction
Economic standing is relatively sticky in the United States (Solon 1992; Mazumder 2005; Bradbury and Triest 2016; Bradbury 2011). Children from high-income families are disproportionately more likely to earn high incomes as adults, and the same is true at the opposite end of the income spectrum. This article explores one of many potential causes of the intergenerational transmission of economic status: car ownership—or a lack thereof. I hypothesize that young people who were carless growing up complete less education, are less likely to work for pay, and earn less than those who enjoyed consistent access.
Motivation for this hypothesis comes from three sources. First, automobiles are unequally distributed across the population (Bureau of Labor Statistics 2015; Brown 2017). Second, in nearly all parts of the United States, an automobile is essential for quickly and conveniently accessing opportunities. As a result, households without cars are thought to suffer from transportation disadvantage (Lucas 2004). As I discuss below, transportation disadvantage may harm children in two ways: through general family disadvantage (e.g., lower household earnings and more expensive household items) and through constrained access to child-specific activities (e.g., organized sports and clubs). Finally, the life course perspective demonstrates that opportunities early in life shape outcomes later in life (Mortimer and Shanahan 2003).
A randomized field experiment would be the gold standard for testing this study’s hypotheses, but such a design would be both impractical and unethical. Like other scholars I rely on a second-best option: observational data. The central shortcoming of using observational data for causal inference is that we cannot observe the counterfactual—what would have happened if the treatment group had cars and the control group no longer had them. Fortunately, a suite of matching techniques offers a reasonable method for estimating the counterfactual from observational data (Rosenbaum and Rubin 1983). Adolescent development scholars have used matching to investigate the effect of participating in school sports (Van Boekel et al. 2016), smoking marijuana (Stuart and Green 2008), and experiencing a parental divorce (Frisco, Muller, and Frank 2007).
The observational data for this analysis come from the Panel Study of Income Dynamics (PSID), a nationwide, longitudinal study. I use propensity-score matching to estimate the effect of family car ownership growing up on education, employment, and earnings at ages nineteen and twenty-five. Matching accounts for observed differences but cannot account for unobservable differences between families. The analysis is restricted to respondents who had low incomes and/or low wealth at some point during childhood.
The results after matching are stark. Young adults who were carless growing up did indeed complete less education than their matched peers who always had access to a car. They also worked for pay less often and earned less. These findings are consistent with the hypothesis that transportation disadvantage contributes to the intergenerational transmission of economic standing.
Literature Review
The conceptual framework for this analysis is based on three building blocks: disparate patterns of car ownership, transportation disadvantage, and the life course perspective.
Disparate Car Ownership
Rates of car ownership are highly unequal in the United States. When it comes to race, black and Hispanic individuals are far more likely than their white counterparts to live in households without cars (20.3 percent, 12.8 percent, and 7.0 percent, respectively) (Polzin and Pisarski 2013). Households in poverty are five times more likely than those above the poverty line to be carless (25 percent vs. 5 percent in 2009) (Federal Highway Administration 2014). Many households at the bottom of the income distribution are carless (43 percent in the bottom decile and 32 percent in the second decile) (Bureau of Labor Statistics 2015). By contrast, less than 5 percent of households are carless at the top of the income distribution.
The unequal distribution of car ownership may not be a problem if those without cars prefer it that way. Yet two pieces of evidence suggest that this is not the case. First, Brown (2017) analyzes the motivations of zero-car households in California and finds that fully 80 percent do not own a car because they are constrained in some way, chiefly because they could not afford to purchase, maintain, and/or insure a car. Second, King, Manville, and Smart (2018) find that the income gap between households with a car and those without has widened over time and is now larger than the income gap by education (college graduate vs. not) and housing tenure (owner vs. renter), leading them to conclude that being carless is “an increasingly reliable sign of poverty” (p. 10).
Of course, in some areas people with means choose not to own vehicles. Yet areas with high densities, robust transit service, and other supportive infrastructure are few and far between. Just 5 percent of the U.S. public lives in neighborhoods where the majority of trips are made without a car, and fully half of those neighborhoods are in New York City (Turley Voulgaris et al. 2017). In other words, cars enable vastly greater access to opportunities than public transit in most areas of the United States (Grengs 2010), and as a result, those who can afford a car typically own one.
Transportation Disadvantage
The second motivating fact is that automobiles enable access to opportunities—so much so that King, Manville, and Smart (2018) describe owning one as the “price of entry into [the] economic system” (p. 1). Households without cars travel fewer miles (Blumenberg and Pierce 2012) and, more troublingly, make fewer trips (Brown 2017). While lower mobility is consistent with many transportation planning goals, lower trip making is not. People who make very few trips are at risk of social exclusion (Lucas 2004).
Most work on transportation disadvantage focuses on employment (Lucas and Nicholson 2003; Ong 2002; Raphael and Rice 2002; Onésimo Sandoval, Cervero, and Landis 2011). A car enables an individual to find a job (Cervero, Sandoval, and Landis 2002; Sandoval, Cervero, and Landis 2011), helps him or her to keep a job (Gurley and Bruce 2005; Blumenberg and Pierce 2014), increases his or her work hours (Gurley and Bruce 2005), and enables higher earnings (Gurley and Bruce 2005; Raphael and Rice 2002). Most important, many (but not all) of these studies point to a causal relationship between vehicles and employment.
Beyond employment, cars are essential for living a full life in auto-oriented areas. In these settings, people without cars struggle to shop for groceries (Burns et al. 2011), visit the doctor (Wachs and Kumagai 1973; Bostock 2001), and meet up with friends and family (Delbosc and Currie 2012).
Much like their parents, children in households without vehicles almost certainly suffer from transportation disadvantage. Because adults with cars find it easier to get to work and access shopping, health care, and other opportunities, the children in those families likely benefit from higher household earnings, cheaper household goods, and greater access to destinations outside the home.
Beyond this general transportation disadvantage, which affects entire households, car ownership may affect children directly by making it easier to access structured activities like sports or clubs. Parents and educational experts frequently report that transportation constraints keep many teens from participating (Shann 2001; Dynarski, Moore, and Mullens 2003; Gardner, Roth, and Brooks-Gunn 2009). Moreover, data from the American Time Use Survey (a cross-sectional survey) indicate that teens who use cars participate more than otherwise similar teens who rely on transit, walking, or biking (citation redacted for review).
Some may wonder why a car is necessary to access children’s activities. Due to school budget cuts, many structured activities increasingly take place outside of school, far from home (Bennett, Lutz, and Jayaram 2012). Even when activities take place at school, many children still need a parent to drive them because some school districts have cut bus services (New Jersey Safe Routes to School 2012) and because most children in the United States live too far from school to walk or bike (McDonald 2008).
Life Course and Adolescent Development
The final conceptual link is the idea that experiences as a child shape life outcomes years later. This is the central premise of the life course perspective, which demonstrates that early life circumstances shape the set of choices available to an individual and one’s ability to avail oneself of those opportunities (Elder 1998; Mortimer and Shanahan 2003). In particular, experiences, behaviors, and attitudes during the transitional period of adolescence profoundly shape trajectories for the entire life (Shanahan 2000; Osgood et al. 2005; Settersten, Furstenberg, and Rumbaut 2006). In the field of travel behavior, scholars are just beginning to uncover how experience and exposure early in life can influence travel patterns later in life (Smart and Klein 2017; Thigpen 2018).
When it comes to structured activities, findings from the life course literature are clear. Students who participate are more likely to complete high school (Eccles et al. 2003), have higher GPAs (Van Boekel et al. 2016), and are more likely to attend college (Eccles et al. 2003; Zaff et al. 2003). As adults, those who participated as teens are more likely to vote and to volunteer (Zaff et al. 2003). Participation also reduces negative outcomes like dropping out (Mahoney and Cairns 1997), crime, drug use (Mahoney 2000), and teen pregnancy (Newman et al. 2000). Most important, two factors suggest that the link between structured activity participation and life outcomes is causal rather than merely associative. First, many of the studies above use longitudinal research designs, and all of them control for observed differences between children who participate and those who do not. Second, there is strong theoretical and empirical evidence that explains how children benefit from participating (Eccles et al. 2003; Mahoney, Larson, and Eccles 2005).
Conceptual Framework
To summarize, some families cannot afford a car; children in those families suffer from transportation disadvantage; and as a result, the lives of those children diverge from those of their peers. In this way, being carless likely transmits economic standing from one generation to the next.
To be clear, transportation is one of many interrelated pathways through which income inequality may be transmitted from one generation to the next. A host of factors are known to influence life outcomes such as household income (Kaplan et al. 1996; Reardon 2013), race (Lu et al. 2010; Lee 2002), family composition (McLanahan and Sandefur 1994; Fomby and Bosick 2013), parental education (Meara, Richards, and Cutler 2008), and gender (Reskin and Bielby 2005; Turner and Bowen 1999; Blau and Kahn 1992; Bertrand and Hallock 2001; Probert 2005). This analysis, therefore, takes pains to account for such influences.
Method
Data for this study come from the core sample of PSID and the Transition to Adulthood Supplement (TAS). The sample includes TAS respondents who were young adults (ages nineteen to twenty-five) between 2005 and 2015. Data on childhood car access and the matching variables come from the core PSID. I use propensity scores to match respondents along six variables (income, wealth, race, family composition, residential location, and sex). With matches in place, I estimate the treatment effect of being carless as a child on education, employment, and earnings at two points: age nineteen and age twenty-five. 1
Data
PSID is a prospective, longitudinal study of American families that began in 1968 with 5,000 families (McGonagle et al. 2012). As children grew up and formed families of their own, their new families were added to the PSID sample. Over time, Latino and immigrant subsamples were added to better reflect the composition of the United States. In 2015, roughly 9,000 families (or 22,000 individuals) completed PSID. TAS was introduced in 2005 to fill a gap in the core survey methodology. TAS fills the gaps between adolescence and family formation by focusing on all young adults regardless of whether they live with their parents or in their own family unit.
Sample
Effective matching requires sufficient overlap (i.e., similarity along the covariates) between the treatment and control groups (Garrido et al. 2014; Stuart 2010). There were very few carless children from high-income/high-wealth families and to achieve sufficient overlap; I restricted the sample to individuals whose families were in the lowest two income or wealth quintiles at any point during their childhood (n = 1,692). 2
Variable Measurement
Treatment Variable
The treatment variable is family car ownership from birth to age seventeen. This study uses car ownership data from 1984 to 1986 (when the oldest members of the sample were infants) and from 1999 to 2013. 3 Car ownership is measured dichotomously in each wave (1 = family owned a car, 0 = otherwise). Because this is a longitudinal study, respondents provide family car ownership data for multiple waves (mean and median = 5 waves). Respondents were excluded if they lacked data on family car ownership in all waves (n = 1) or had car ownership data for just a single wave (n = 6).
Table 1 depicts the share of respondents who were carless once, most of the time, or all of the time during childhood for the full sample and the low-income/low-wealth subsample. Among the low-income subsample, four in ten were carless at least once, one in four were carless most of the time, and one in ten were carless in every survey wave. 4
Share of Sample Who Are Carless by Family Characteristics.
Note: Categories describe childhood circumstances, not current circumstances. Income and wealth quintiles are mean values.
Potential Matching Variables
Effective matching should include all observable variables that could plausibly be linked to one’s chances of receiving the treatment (i.e., family car ownership) or to the outcomes (i.e., education, employment, and earnings) (Stuart 2010). The following seven variables served as candidates for matching 5 :
Income quintile during childhood (minimum)
Wealth quintile during childhood (minimum)
Race of the household head during childhood (1 = ever had a black head, 0 = otherwise)
Family composition during childhood (1 = had a single mother most of the time, 0 = otherwise) 6
Residential location during childhood (1 = counties in the center of a large city [>1 million], 2 = counties on the fringe of a large city [>1 million], 3 = counties in a medium-sized city [250,000 to 1 million], and 4 = all others) 7
Maternal education (no high school, high school graduate, and college graduate)
Sex
Table 1 depicts how being carless varied along each of these candidate matching variables. Children were more likely to be carless if they came from families with few financial resources or from families headed by black parents, single mothers, or mothers with limited education. As expected, children from the center of large urban areas were the most likely to be carless. Nevertheless, many children outside of urban centers were carless at some point during childhood.
Outcome Variables
Table 2 details the eleven outcome variables examined in this study. Members of the low-income/low-wealth sample (the focus of this analysis) completed less education, were unemployed more, 8 and earned less than those in the full sample.
Outcome Variables at Ages Nineteen and Twenty-five.
Source: Panel Study of Income Dynamics 1989–2015.
Note: The sample sizes differ because (1) some respondents are not yet twenty-five years old; (2) some are missing data, particularly for earnings; and (3) earnings among earners includes a subsample of respondents.
Matching and Estimating the Treatment Effect
Using matching to estimate treatment effects requires two distinct steps: develop suitable matches, and then estimate the treatment effect (Stuart 2010). Crucially, these steps should be independent; the analyst should develop matches without regard to the final outcomes. Analysts develop suitable matches iteratively by assessing match quality and refining matches. Important to note, the treatment and control groups must balance on all of the variables used to calculate the propensity score (Rosenbaum and Rubin 1983).
In the simplest form of matching, analysts compare outcomes for respondents who match perfectly along all of the observed covariates. While it is intuitive, exact matching poses problems when there are many covariates or the covariates are measured continuously (Stuart 2010). Propensity scores offer a useful alternative because respondents are matched along a single scalar, the propensity score, which is the probability of being treated (e.g., carless) given the covariates (Rosenbaum and Rubin 1983).
Propensity scores were estimated using two logistic regression models with treatment status as the dependent variable ([1] lacked a car at least once vs. always had a car and [2] lacked a car most of the time vs. always had a car). The seven matching variables described above serve as the explanatory variables and members of the treatment group (i.e., children who were carless) were matched with multiple members of the control group (i.e., children with consistent car access) to improve efficiency (Stuart 2010). These models fit relatively well, with McFadden R2 values of .210 and .292, respectively.
However, the primary criterion for assessing propensity score matches is not the fit of the regression model but rather the extent to which the covariates balance after matching (Stuart 2010). The Stata commands “pscore,” “psmatch2,” and “pstest” assess balance. 9 Unfortunately, matches that included maternal education would not balance, and therefore this variable was dropped from the models. 10 Matching along the remaining variables reduced bias (by 68 to 100 percent) and resulted in balanced matches.
The next step is to estimate the treatment effect, which is done by comparing the outcome variables (e.g., employment status or earnings) along matched pairs and then averaging the differences over the sample. 11 For example, consider rates of college attendance among matched respondents. If 60 percent of the control group and 70 percent of the treatment group attend college, the estimated treatment effect is 10 percentage points. Throughout the article, differences in outcomes between the matched pairs are reported using percentage point differences (for education and employment) and dollar values (for earnings).
I present the average treatment effect on the treated, which is the effect of being carless for those who actually experienced carlessness. Results were qualitatively similar when estimating the average treatment effect (not pictured).
To assess whether the consequences of being carless vary by age, I reran the analysis for those who were carless as children only (ages 1–12; n = 218), as teens only (ages 13–17; n = 146), and as both (n = 286). These analyses are presented in Table 3.
Does the Effect of Car Access Vary by Age? Matched Results.
Source: Panel Study of Income Dynamics 1989–2015.
Note: Low-income/low-wealth subsample only. All analyses match on family income, wealth, race of the family head, single parent, residential location, and sex. Reference group = always had a car. Treatment effect reflects percentage point differences or differences in earnings in dollars. Child = ages 1 to 12; teen = ages 13 to 17.
p < .10. **p < .05. ***p < .01.
I conducted sensitivity analysis by reestimating the treatment effect using alternative specifications of the matching variables and found that the signs and magnitudes of the results were broadly consistent across various specifications (not pictured). Moreover, the results were similar when excluding respondents who ever lived in New York City (n = 28).
Finally, because I do not aim to estimate a population-level treatment effect or to determine the prevalence of childhood carlessness in the United States, I use unweighted data.
Caveats
Before discussing the results, three important caveats deserve mention. Matching can account for bias only from observed variables (Stuart 2010). If intangibles like motivation, intelligence, “grit” (Duckworth 2016), or substance abuse are associated with vehicle ownership, failure to include these factors may lead to overestimates of the vehicle effect. Second, given the ephemeral nature of car ownership for families at the margin (Klein and Smart 2017) and the infrequency of PSID, I may omit some instances when a family lacked a car. This may lead me to underestimate the treatment effect because some members of the control group (i.e., those who always had a car) were actually carless at some point.
Finally, and most important, the measure of residential location employed here is a very coarse proxy for public transit service, walkability, and car dependence. While many small counties are car dependent, some—like Suffolk County, Massachusetts (Boston)—are quite amenable to travel by other modes. Similarly, car dependence varies considerably between large counties such as New York County (Manhattan) and Maricopa County (Phoenix). Finally, car dependence can vary considerably even within a single county—like Los Angeles County. Despite this limitation, the effect of being carless is likely larger in car-dependent locations with few alternatives, and—notwithstanding some exceptions—much of the United States matches that description (Turley Voulgaris et al. 2017). As a result, I argue that the results presented here hold for most young people and will continue to hold as poverty suburbanizes.
Results
Descriptive Results
Table 4 compares the outcome variables by childhood car access for the low-income/low-wealth sample. From the descriptive comparisons we can see that young people who lacked consistent car access as children completed less education, were less likely to work for pay, and earned less than those who were classified as always having cars. We turn next to the matched results, which account for differences between these groups and offer a better estimate of the effect of being carless as a child.
Education, Employment, and Earnings by Childhood Car Access, Descriptive Results.
Source: Panel Study of Income Dynamics 1989–2015.
Note: Low-income/low-wealth subsample only. These comparisons do not account for matching. Differences represent percentage point differences or difference in earnings in dollars. Differences are relative to those who were classified as always having cars.
p < .10. **p < .05. ***p < .01.
Matched Results
Table 5 presents the results of the propensity score matching for the low-income/low-wealth subsample. The descriptive trends hold after matching along observable covariates. More specifically, when it comes to education, young people who lacked access to cars growing up were less likely than matched peers to graduate from high school. In turn, they were also less likely to attend college and less likely to eventually graduate from college. 12
Effect of Childhood Car Access on Education, Employment, and Earnings, Matched Results.
Source: Panel Study of Income Dynamics 1989–2015.
Note: Low-income/low-wealth subsample only. All analyses match on family income, wealth, race of the family head, single parent, residential location, and sex. Reference group = always had a car. Treatment effect reflects percentage point differences or differences in earnings in dollars.
p < .10. **p < .05. ***p < .01.
The results for employment at age nineteen are relatively muted (and statistically insignificant). While employment at age nineteen does not necessarily indicate a problem, unemployment does because unemployed teens would like to work but cannot get jobs. Teens who were carless growing up were 5.9 percentage points more likely to be unemployed at age nineteen even when matching along the covariates.
Employment at age twenty-five is an important indicator of adolescent development, and the matched results reveal that twenty-five-year-olds who grew up without consistent access to cars were less likely to work and more likely to be unemployed than those with consistent car access.
When it comes to earnings, young people who were carless at least once earned $1,800 less at age nineteen and $3,900 less at age twenty-five. These are substantial reductions in earnings: carless kids earn 17 percent less at age nineteen and 25 percent less at age twenty-five than matched peers. For those who were carless most of the time, average earnings were $1,300 lower (12 percent) at age nineteen and $4,800 lower (31 percent) at age twenty-five. When excluding young people with zero earnings, the results were generally insignificant, indicating that childhood car access likely influences earnings via employment status rather than the wage rate.
Variation by Age
Table 3 compares the effect of being carless as a child and as a teen. Due to small sample sizes, the results should be interpreted with caution. Respondents who lacked cars as children did not differ meaningfully at age nineteen or twenty-five from matched peers who always had cars. By contrast, outcomes differed for respondents who lacked cars as teens. This provides tentative support for the hypothesis that car access is especially beneficial for teens, perhaps by providing access to structured activities. Finally, outcomes differed most for respondents who lacked cars both as children and as teens.
Conclusion
This study demonstrates that low-income young adults who were carless growing up went on to complete less education, were unemployed more, worked less, and earned less than matched peers who enjoyed consistent access to cars. The use of propensity score matching supports a causal interpretation of these findings. The matching process accounts for differences in six observable variables (household income, wealth, family composition, race, and residential location). Most important, this work demonstrates that the consequences of being carless as a child persist into young adulthood and—given the importance of early outcomes on life trajectories (Elder 1998; Mortimer and Shanahan 2003)—these effects likely last a lifetime. These results are generalizable only to low-income respondents. Young people from families who are carless by choice—and who can continue to access opportunities without a car—would almost certainly not suffer the same long-term consequences of being carless. Ultimately, car ownership likely helps transmit low-economic standing from one generation to the next; children whose parents’ could not afford cars growing up earn less than their peers.
One option to reduce the long-term consequences of being carless growing up is to increase ownership of reliable cars for families at the margin by raising vehicle asset limits for welfare recipients, offering government-backed automobile loans for low-income families, and expanding programs that donate cars directly to those in need (Lucas and Nicholson 2003). If implemented, these efforts would reduce the number of children who grow up without cars, and as a result, children in those households would be better able to access structured activities and would be less likely to suffer from general transportation disadvantage because their parents would enjoy better employment prospects and greater access to basic services.
Some may counter that this approach is not only impractical but counterproductive. After all, increasing car ownership is at odds with efforts to ease congestion, reduce air pollution, encourage active travel, and reduce crashes. Rather than increase car ownership, we could reduce the consequences of being carless by expanding transit service, improving bicycle and pedestrian infrastructure, facilitating access to shared mobility services (e.g., via vouchers and requirements to serve low-income areas), building more densely, and increasing the mix of housing and employment. If these efforts are successful, children in carless households would be able to get to structured activities safely and conveniently without cars, and their parents would enjoy similar car-free access to work and basic services.
Both policy options would require considerable monetary and political investment. Choosing the optimal blend of strategies will require careful consideration of feasibility and efficacy, which is beyond the scope of this study. Instead, what this study offers is a call to arms. Existing transportation conditions leave many poor children behind, setting them up for a lifetime of constrained employment options and limited earnings. This problem is urgent, and it demands our attention. Helping children and their families access opportunities—either with cars or via other means—should be central to our transportation planning efforts.
Footnotes
Acknowledgements
I would like to thank the amazing staff at the Panel Study of Income Dynamics Data Users Workshop, as well as Nick Klein, Marty Wachs, and the anonymous reviewers for their thoughtful comments.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
