Ethnic neighborhoods, social networks, and inter-household carpooling: A comparison across ethnic minority groups

Journal of Transport Geography 59 (2017) 14–26

Contents lists available at ScienceDirect

Journal of Transport Geography journal homepage: www.elsevier.com/locate/jtrangeo

Ethnic neighborhoods, social networks, and inter-household carpooling: A comparison across ethnic minority groups Eun Jin Shin Sol Price School of Public Policy, University of Southern California, Los Angeles, CA 90089-0626, USA

a r t i c l e

i n f o

Article history: Received 29 January 2016 Received in revised form 26 April 2016 Accepted 8 January 2017 Available online xxxx

a b s t r a c t The implications of racial residential segregation on travel behavior have remained understudied, despite the persistent existence of segregation. Using the 2009 National Household Travel Survey, I investigate whether residence in a co-ethnic neighborhood affects the likelihood that ethnic minorities will form inter-household carpools, and if so, how such effects differ across race or ethnic groups. Inter-household carpooling requires arrangements and interactions between people living in geographical proximity, so it will likely reflect the social networks of a neighborhood. The results show that Hispanics and Asians who reside in their co-ethnic neighborhoods, regardless of immigrant status, are more likely to use inter-household carpools for non-work purposes than their counterparts living outside co-ethnic neighborhoods. In contrast, black neighborhood residency is not associated with the likelihood that African Americans will do inter-household carpooling, regardless of trip purpose. These differences across racial/ethnic groups suggest that the role of neighborhoods in promoting social ties as reflected by activities such as external carpooling is complex. Residence in a black neighborhood may be less of a choice than residence in a Hispanic or Asian neighborhood due to the long history of black segregation in the US. With less residential choice, the propensity to develop local social ties may be weaker. © 2017 Elsevier Ltd. All rights reserved.

1. Introduction Although residential segregation between whites and African Americans has decreased over time, most large U.S. metropolitan areas are still characterized by high segregation levels (Iceland et al., 2013). In 2010, the average level of black-white segregation for major U.S. metropolitan areas was about 55—that is, almost 55% of African Americans would need to relocate in order to be entirely integrated with white populations (Frey, 2011). The segregation levels between whites and Hispanics and between whites and Asians are relatively low in comparison to the figure for African Americans (44 for Hispanics, 40 for Asian), but they have increased slightly or remained similar over the last several decades with continuing immigration (Frey, 2011). As the spatial separation between whites and ethnic minorities has been a long-standing issue, the consequences of racial residential segregation have been widely researched in the academic literature: the main focus points are labor market outcomes (for a literature review, see Ihlanfeldt and Sjoquist, 1998; Preston and McLafferty, 1999), health outcomes (Williams and Collins, 2001; Acevedo-Garcia et al., 2003; Kramer and Hogue, 2009) and educational outcomes (Card and Rothstein, 2007; Quillian, 2014). However, the role of racial residential segregation in

E-mail address: [email protected].

http://dx.doi.org/10.1016/j.jtrangeo.2017.01.002 0966-6923/© 2017 Elsevier Ltd. All rights reserved.

shaping transportation-related decisions among ethnic minorities has received scant attention. Previous research on travel behavior suggests that the social environment and social networks play a crucial role in people's travel behavior (Carrasco and Miller, 2006; Roy et al., 2012; Maness et al., 2015). Ethnic neighborhoods offer a distinct social environment for those who live there; especially, those whose ethnicity is the same as the ethnic characteristics of the neighborhood. They are physically close to coethnics who share the same language and cultural background, which might facilitate the formation of social networks (Charles and Kline, 2006). This environment will also likely affect the probability of interhousehold carpooling, because such ways of traveling, especially those for non-work-trip purposes, require social networks among people who live close to one another. For instance, Liu and Painter (2012) is one study that found a positive relationship between co-ethnic neighborhood residency and the likelihood of Latino immigrant carpooling for commutes. Yet, an investigation of such a relationship among native-born ethnic minorities has been lacking. In addition, differences in the effects of co-ethnic neighborhood residency across racial/ethnic groups remain unclear. The literature on racial residential segregation has suggested that the causes behind the formation of ethnic neighborhoods and the resultant characteristics of those neighborhoods seem to differ by race/ethnicity (Charles, 2000; Emerson et al., 2001). Hence, these differences may have implications for residents' travel behavior in different ethnic neighborhoods.

E.J. Shin / Journal of Transport Geography 59 (2017) 14–26

The goal of this study is to help fill a gap in the existing literature. Using the 2009 National Household Travel Survey (NHTS), I investigate how residence in co-ethnic neighborhoods influences the likelihood of inter-household carpooling among ethnic minorities and how the effects of such co-ethnic neighborhoods differ across race/ethnicity and immigrant status. This study also aims to identify other determinants of inter-household carpooling among ethnic minority groups. Interhousehold carpooling is an important means of transportation for ethnic minorities: it accounts for nearly 15% of person trips for this population segment―more than five times the number of trips made using public transportation (2009 NHTS). Given the limited attention to inter-household carpooling in association with the social environment, the findings of this study will help provide suggestions for effective policy interventions aimed at increasing ridesharing across households and implications for future studies. The next section of this paper provides a review of the literature on race/ethnicity and travel behavior, and ethnic neighborhoods and social networks. Section 3 describes the data sets and research strategy and Section 4 presents the empirical analyses on the travel mode choices of ethnic minority groups. Section 5 discusses in detail the relationship between co-ethnic neighborhoods and inter-household carpooling. The paper concludes in Section 6 with a summary of study results and policy implications. 2. Previous research and research hypotheses 2.1. Race/ethnicity and travel behavior The relationship between race/ethnicity and travel behavior has been the focus of several empirical studies that have investigated differences between the travel patterns of white populations versus those of other racial ethnic groups (Ibipo, 1995; Giuliano, 2003; Pucher and Renne, 2003). However, despite these empirical findings of racial differences in travel behavior, efforts to understand such differences and to develop theoretical discussions have been lagging. One strand of literature on the spatial mismatch hypothesis has attempted to explain longer commutes among ethnic minorities as related to their residential location and commute modes, but the focus has been on commutes without considering non-work travel behavior (Ibipo, 1995; Taylor and Ong, 1995; Sultana, 2005). One of the most relevant theories that may help explain racial/ethnic differences in non-work travel behavior is the leisure segregation theory. Developed to explain racial variations in leisure behavior, it has revolved around the marginality and ethnicity hypotheses (Floyd, 1998). The marginal hypothesis attributes racial variations in leisure behavior to ethnic minorities' marginal socio-economic conditions, which are explained, in turn, by social structural barriers (Washburne, 1978). Previous studies have tended to operationalize marginality as socioeconomic characteristics, such as limited household income and vehicle availability (Floyd, 1998). However, the ethnicity hypothesis views the cultural differences commonly operationalized by racial variables in empirical studies as a determinant of interracial differences in leisure behavior (Allison, 1988; Floyd et al., 1994). Past studies of race/ethnicity and travel behavior have found results that are in line with the marginality and ethnicity hypotheses. The general findings of previous studies show that auto ownership and income are relatively low among ethnic minority groups when compared to whites, which partly explains the higher reliance on alternative modes of transportation for both non-work and work purposes among ethnic minorities (Blumenberg et al., 2007). Yet, after adjustments for socioeconomic status, race/ethnicity still plays a role in travel behavior for reasons that are unclear (Giuliano, 2003; Smart, 2010). However, there has been little theoretical discussion of racial/ethnic differences in travel behavior across ethnic minority groups, as previous studies concentrated on the comparison between whites and ethnic minority groups, especially between white and blacks. Thus, the current

15

study attempts to explain differences in factors that affect carpooling behavior across ethnic minority groups based on a theoretical framework of segregation and social networks. 2.2. Residential segregation, social networks, and carpooling Substantial discussions have centered on how segregated neighborhoods affect social networks and urban inequality, but most have revolved around job searching networks related to social mobility among ethnic minorities (Reingold, 1999; Elliott and Sims, 2001). This line of literature has demonstrated that African Americans living in segregated black neighborhoods suffer from lack of job networks which help their job acquisitions (Kirschenman and Neckerman, 1991; Smith, 2005). More recent findings, however, suggest that “co-ethnic neighborhood effects” on job networking might differ by racial/ethnic group; Hispanic or Asian neighborhoods, for example, as opposed to black neighborhoods, often provide ethnic-based information networks that help residents to secure employment (Wilson and Portes, 1980; Bailey and Waldinger, 1991). This mechanism seems more prevalent among immigrants, who are likely to rely heavily on localized and informal networks to find employment opportunities because of their legal status and limited English ability (Elliott, 2001). Despite the relative lack of attention paid to social support networks in the fields of geography and planning, they are an important type of social tie. Social support networks―often related to strong ties―comprise ties that aid people in handling their daily tasks (Briggs, 1998). Social support networks are distinct from job information networks, because an individuals' socio-economic status is not as relevant in social support networks as in job information networks. For example, it is possible that a low-income friend might provide better child care than a high-income friend. In addition, unlike job information networks, which may be affected by such demand factors as racial discrimination among employers, social support networks are less likely to be hampered in their formation by external factors. In their investigation of social support networks among ethnic minorities, Dominguez and Watkins (2003)—even though they did not explicitly focus on racial differences—suggested that African Americans may rely on institution-based social support networks whereas Latin Americans tend to rely on family- or friendship-based social support networks―a phenomenon similar to jobnetworking. Several other studies have found racial differences in social support networks including network size and composition, but results are mixed (Kim and McKenry, 1998; Small, 2007). Furthermore, these studies have not clarified how neighborhood conditions, especially residence in co-ethnic neighborhoods, influence social support networks among ethnic minorities. The literature on racial residential segregation suggests that the effects of co-ethnic neighborhood residency on social support networks may not be the same across racial/ethnic groups because of different causes of racial residential segregation and the resultant neighborhood characteristics. Studies have found fairly consistent evidence that African Americans continue to experience racial discrimination in the housing market, although it is more subtle than it used to be (Yinger, 1995; Massey and Lundy, 2001; Dawkins, 2004). There is little evidence to suggest that racial segregation is the result of preferences of African Americans to live in black neighborhoods. Even studies that found evidence of black self-segregation revealed that it seemed to play only a minor role in explaining black segregation (Zubrinsky and Bobo, 1996; Ihlanfeldt and Scafidi, 2002). Whereas other such racial groups as Hispanics and Asians also tend to be segregated from whites, the causes of segregation may differ fundamentally from reasons for black-white segregation. For example, many studies have demonstrated that, compared to African Americans, Hispanics and Asians are more likely to self-segregate because they prefer to live with co-ethnics (Zubrinsky and Bobo, 1996; Nguyen, 2004). The ethnic enclave hypothesis supports this argument, as it posits that

16

E.J. Shin / Journal of Transport Geography 59 (2017) 14–26

the formation of ethnic enclaves is based on the economic benefits of ethnic agglomeration (Wilson and Portes, 1980). Empirical studies have also confirmed that the proportion of Hispanics or Asians in a neighborhood does not affect the likelihood that whites will buy a house there, whereas that of African Americans does (Emerson et al., 2001). Therefore, these studies suggest that segregated black neighborhoods stem from involuntary spatial concentrations (Emerson et al., 2001; Gans, 2008), whereas other ethnic neighborhoods are more likely to be voluntarily formed, self-defined neighborhoods (Nguyen, 2004; Peach, 2005). Given that African Americans cannot necessarily choose their neighborhoods in order to seek help from their neighbors, it may not be possible to regard black neighborhoods as indicators of potential social networks like Asian and Hispanic neighborhoods. Therefore, black neighborhoods are likely to be characterized by weaker social support networks among their own-ethnic residents compared to Hispanic and Asian neighborhoods. As argued by Charles and Kline (2006), carpooling may serve as a great proxy for social networks because (1) it is based on trust among people (punctuality, driving skills), and (2) it requires interactions among people (i.e., drivers and passengers). Transportation assistance from non-household members also greatly helps people manage mundane tasks, such as taking children to school and going shopping―in other words, it provides social support. Given that most inter-household carpools occur within a close proximity, it is likely that inter-household carpooling reflects the social support networks in a neighborhood. Based on the literature review, I hypothesize that Hispanic neighborhood residency increases the likelihood that Hispanic populations (both U.S.-born and immigrants) will form inter-household carpools because the neighborhood composition itself is more likely to be based on selfsegregation and ethnic networks. This tendency is most likely mirrored in the Asian population as well. I further hypothesize that the effects of black neighborhood residency on the likelihood of inter-household carpooling will be weaker or insignificant as compared to other ethnic groups. Meanwhile, I expect that the effects of Hispanic or Asian neighborhood residency will be larger for Hispanic/Asian immigrants than for U.S.-born Hispanics/Asians, as immigrants are more likely to rely on local-based social networks. 3. Research approach 3.1. Data and sample I use the 2009 NHTS, a cross-sectional, nation-wide survey of personal travel behavior and transportation patterns, as the main source of data. The NHTS provides information on trips made during a day (24-hour period) and includes a detailed list of socio-economic, demographic, and travel-related variables for individuals and households. The survey also has questions on the detailed, pre-categorized purposes of each trip which I use to differentiate commuting trips and non-work related trips. Out of 26 categories of non-work related trip purposes, which exclude work-related or school-related trips, I reclassify them into three categories: social/recreational (e.g., visiting friends/relatives, going out/hanging out), shopping (e.g., buying goods and groceries), and other/personal business (e.g., medical/dental services) purposes. The analyses in the present study are based on homebased trips—those originating at home for either work or nonwork purposes. The 2009 NHTS gives information for about 1,160,000 trips (including about 400,000 home-based trips) made by approximately 150,000 households and 300,000 individuals. Of the 51 Core-Based Statistical Areas (CBSA) included in the survey, I chose eight for the study. The selected CBSAs share several common features: (1) relatively large sample sizes for the five groups of ethnic minorities specified, (2) established immigrant gateways in the western or southern United

States, and (3) auto-oriented built environment characteristics. The first and second criteria are employed because CBSAs with lower proportion of a certain ethnic group are less likely to have what amounts to ethnic neighborhoods―that is, large ethnic populations that exercise significant effects on those who belong to race/ethnicity. The primary samples employed in this study are African Americans, Hispanics (both Hispanic U.S.-born and immigrants), and Asians (both Asian U.S.-born and immigrants) older than 16 and living in the study area. Non-relatives of householders are excluded from the analysis, as race information is provided among household characteristics. After excluding cases in which not all requisite information is provided, the final sample consists of about 15,000 homebased trips, made by approximately 9.200 individuals from 5800 households. Fig. 1 presents the distribution of total home-based trips across metro areas by ethnic group. 3.2. Methods and variables A multinomial logit model is used to estimate travel mode choice. The model assumes that an individual chooses a travel mode from several options in order to maximize the utility. A basic model for estimating the effects of co-ethnic neighborhood residency on travel mode choice is as follows: Y ¼ f ðE; S; T; N; MÞ where E = dummy variable indicating an ethnic neighborhood residence, S = vector of a travel decision maker's socio-demographic characteristics, T = vector of trip purpose (for the non-work trip model), N = vector of neighborhood characteristics, and M = vector of metro dummy variables. I posit that ethnic minorities' choices of travel mode are a function of ethnic neighborhood residency and other control variables (e.g., socio-economic characteristics). The dependent variable of this study is travel mode choice, which is a 5-category variable created from detailed records on trip modes in the NHTS: driving alone, inter-household carpools, household-carpools, public transit, and other means of transportation. Inter-household carpools are defined as automobile trips with non-household members, regardless of the number of household members involved. In order to separate between household-carpools and inter-household carpools, I use day trip records from NHTS, which include information on the number of household members and non-household members on each trip. Many inter-household carpools actually involve two trips wherein 1) the respondent travels from home to a middle point place to pick up a non-household member (or members) and 2) continues to the destination with this person (people). A trip is also categorized as an inter-household carpool if it starts from a respondent's home with non-household members. Although the multinomial logit model generates the results for the four categories of travel mode versus driving alone (reference category), estimation results will be reported only for inter-household carpools versus driving alone, as the focus here is on inter-household carpools. Because many people generate more than one home-based nonwork trip during a day, multiple observations (trips) generated by the same person might be correlated in some way, which goes somewhat against the assumption of the discrete choice model (i.e. the independence of observation). This does not bias the coefficients of the variables; however, standard errors can be affected, causing incorrect inferences. This problem has not been addressed in previous studies on the subject. To solve this issue, I employ the cluster option in STATA to ensure a robust standard error by making observations independent across clusters of trips made by one individual.

E.J. Shin / Journal of Transport Geography 59 (2017) 14–26

100% 90% 80% 70%

6.50% 9.70% 4.70%

5.50% 12.80%

40% 30%

15.30% 21.10%

20% 10% 0%

1.90% 18.10%

9.50%

11.10%

60% 50%

3.20% 21.10%

24.50%

10.50%

17

0.50%

1.70%

28.30%

22.10%

4.10%

4.70%

6.70%

4.60%

2.60% 5.10%

3.10% 2.00%

28.70%

30.40% 43.60%

22.20% 11.00% 11.40%

22.80% 9.00% 8.30%

16.00% 9.40% 9.00%

African Americans U.S.-born Hispanics Hispanic immigrants U.S.-born Asians

17.40% Asian immigrants

Dallas-Fort Worth-Arlington, TX

Houston-Sugar Land-Baytown, TX

Los Angeles-Long Beach-Santa Ana, CA

Miami-Fort Lauderdale-Pompano Beach, FL

Phoenix-Mesa-Scottsdale, AZ

Riverside-San Bernardino-Ontario, CA

San Diego-Carlsbad-San Marcos, CA

Tampa-St. Petersburg-Clearwater

Fig. 1. Share of home-based trips in each metro area by ethnic minority group (unit: %).

Table 1 presents the definitions and descriptive statistics for the variables used in estimating the travel mode choice. The key variable in this study is residence in co-ethnic neighborhoods; it was

measured using the 2009 American Community Survey (ACS) 5year estimates. The co-ethnic neighborhood variable is based on the residential concentration quotient (RCQ), which measures the

Table 1 Descriptive statistics for variables by race/ethnicity and immigrant status [mean (standard deviation)]. Variables

Description

African Americans Mean (S.D.)

U.S.-born Hispanics Mean (S.D.)

Hispanic immigrants Mean (S.D.)

U.S.-born Asians Mean (S.D.)

Asian immigrants Mean (S.D.)

Co-ethnic neighbor-hood

Measure 1: RCQ value of a residential census tract (continuous measure) Measure 2: living in a census tract whose RCQ is N2 (discrete measure)

2.88 (2.7)

1.15 (0.8)

1.49 (0.8)

1.67 (1.4)

2.22 (1.8)

0.50 (0.5)

0.15 (0.4)

0.29 (0.5)

0.33 (0.5)

0.45 (0.5)

Number of household vehicles Household income is less than $35,000 (dummy) Respondent's highest education completed is “less than high school graduate” or “high school graduate” (dummy) Gender of respondents (dummy) Age of respondents Age squared Employed at the time of survey (dummy) Number of household members Lived in the U.S. b10 years (dummy) Lived in the U.S. 10–30 years (reference category) (dummy) Lived in the U.S. N30 years (dummy)

1.86 (1.1) 0.35 (0.5) 0.31 (0.5)

2.37 (1.1) 0.21 (0.4) 0.33 (0.5)

1.97 (1.1) 0.49 (0.5) 0.56 (0.5)

2.75 (1.2) 0.10 (0.3) 0.13 (0.3)

2.25 (0.9) 0.12 (0.3) 0.12 (0.3)

0.59 (0.5) 51.9 (15.5) 2932 (1619) 0.54 (0.5) 2.76 (1.5) N/A

0.57 (0.5) 45.3 (16.6) 2327 (1600) 0.63 (0.5) 3.38 (1.5) N/A

0.58 (0.5) 47.7 (15.5) 2512 (1610) 0.56 (0.5) 3.76 (1.6) 0.19 (0.4) 0.46 (0.5)

0.46 (0.5) 42.6 (18.1) 2140 (1712) 0.67 (0.5) 3.32 (1.4) N/A

0.51 (0.5) 48.1 (13.2) 2482 (1337) 0.67 (0.5) 3.34 (1.3) 0.18 (0.4) 0.52 (0.5)

Socio-economic status HH veh Low-inc Low-ed

Female Age Age2 Employed HH size Recentfb Medfb Longerfb Trip purpose Commute Social Shopping Others

Trip purpose is related to commuting Trip purpose is related to social activity (dummy) Trip purpose is related to shopping activity (dummy) Trip purpose is related to other/personal activity (dummy)

Neighborhood characteristics Popdenacre Population density per acre in a residential census tract Medincome Median household income of a residential census tract ($) [Commute trips only] Job category (reference category = sales/service) Admin Clerical/admin support Manu/const Manuf, construct, maintenance, or farming Prof Professional, managerial, or technical Other Works in other industry Full time Works full-time N Total home-based trips Notes: There is no observation of U.S.-born Asians, who fall into the “Other” jobs category.

0.35 (0.5)

0.30 (0.5)

0.17 (0.4) 0.35 (0.5) 0.30 (0.5) 0.17 (0.4)

0.22 (0.4) 0.36 (0.5) 0.26 (0.4) 0.16 (0.4)

0.20 (0.4) 0.30 (0.5) 0.29 (0.5) 0.21 (0.4)

0.26 (0.4) 0.38 (0.5) 0.22 (0.4) 0.14 (0.3)

0.27 (0.4) 0.32 (0.5) 0.25 (0.4) 0.16 (0.4)

9.29 (8.1) 51,957 (24364)

9.38 (7.9) 64,336 (26293)

13.2 (12.0) 54,366 (24266)

10.3 (11.1) 80,152 (27216)

8.66 (7.4) 81,958 (31566)

0.18 (0.4) 0.09 (0.3) 0.44 (0.5) 0.003 (0.1) 0.86 (0.3) 2848

0.15 (0.4) 0.11 (0.3) 0.41 (0.5) 0.009 (0.1) 0.80 (0.4) 4201

0.12 (0.3) 0.31 (0.5) 0.24 (0.4) 0.006 (0.1) 0.84 (0.4) 5344

0.10 (0.3) 0.04 (0.2) 0.55 (0.5) – 0.73 (0.4) 614

0.10 (0.3) 0.07 (0.3) 0.64 (0.5) 0.002 (0.0) 0.90 (0.3) 2043

18

E.J. Shin / Journal of Transport Geography 59 (2017) 14–26

concentration of each minority group in a census tract as compared to the metro area. The formula for computing RCQ is as follows: RCQ ¼

the 2009 ACS 5-year estimates. The economic status of neighborhoods is adjusted because many ethnic neighborhoods are low-income. Small (2007) demonstrated that the poverty level of a neighborhood is the most important neighborhood condition affecting the level of social networks. In addition, merely the presence of low-income residents in a neighborhood, regardless of their race/ethnicity, might be important in finding carpool partners. The median household income of a neighborhood, as derived from the 2009 ACS 5-year estimates, is included in the model to separate the income effects from the effects of co-ethnic concentrations.

Pij Pie = P j Pe

where j = (1, …, n) refers to the census tract; P ij is the number of each racial/ethnic group (African American/Hispanic/Asian) in each census tract; Pj is the total population in that census tract; Pie is the number of each ethnic group in each metro area; and P e refers to the total population in that metro area. I use two measures of co-ethnic neighborhood residency, including (1) a continuous value of RCQ and (2) a discrete measure of the co-ethnic neighborhood, and run the separate model for each case. Based on the distribution of RCQ shown in Table 1, cutoff RCQ 2 is used to define a discrete measure for a co-ethnic neighborhood, the same as Borjas (1998) has done. This measure is used to confirm whether there are any discrete effects of residence in co-ethnic neighborhoods. For RCQ calculation, Hispanic neighborhoods are measured using the total Hispanic population (i.e., including the U.S.-born and immigrants); the estimation results remain robust, even if Hispanic neighborhoods are defined separately as dominantly U.S.-born Hispanic neighborhoods for U.S.-born Hispanics and Hispanic immigrant neighborhoods for Hispanic immigrants. Asian neighborhoods are measured in the same way. A number of variables of individual socio-demographic characteristics are included in the travel mode choice model. These include level of educational attainment, gender, age, household size, household vehicle availability, and household income. In the case of immigrants, the length of U.S. residence is included in the model to capture the degree of cultural assimilation. As shown in Table 1, African Americans and Hispanic immigrants have a lower socio-economic status than do other ethnic groups, especially compared to Asians, based on diverse dimensions including employment status and educational attainment. Job characteristics are included in the commuting model because they have been identified as determinants of carpooling (Cline et al., 2009; Liu and Painter, 2012; Vanoutrive et al., 2012). This probably has to do with the fact that the type of job affects commuting time, job locations, and the composition of co-workers. Full-time status is also controlled for in the commuting model. Table 1 indicates that the distributions of occupations greatly differ by ethnic group. Whereas the regression analyses are conducted separately for commuting trips and non-work-related trips, trip purpose variables are included in the non-work travel mode choice model to differentiate between types of trips, such as those for shopping or personal business. I also control for two neighborhood characteristics, including the population density and economic status of neighborhoods. Previous studies have found that ethnic neighborhoods tend to have neo-traditional built environment characteristics (Blumenberg and Smart, 2014; Shin, in press). Population density, which has been used frequently as a proxy for the density of activity sites, is included in the model to separate its effects on carpooling. To calculate the population density, I use

4. Results 4.1. Descriptive statistics Table 2 describes the share of travel modes in home-based trips by ethnic neighborhood status (a discrete measure) and ethnic group. The share of driving alone is highest among trips made by U.S.-born Asians and lowest among trips made by Hispanic immigrants, as expected. All ethnic minority groups except Asian immigrants demonstrate lower shares of driving alone on trips generated by co-ethnic neighborhood residents than on trips generated by non-co-ethnic neighborhood residents, although these differences are statistically significant only for the two Hispanic groups. Results show that among all ethnic minority groups, the proportions of inter-household carpools are higher for trips made by co-ethnic neighborhood residents than for trips made by those living outside coethnic neighborhoods (This difference is not statistically significant for U.S.-born Asians owing in large part to small sample size). This result suggests that before controlling for other factors, such as individual socio-economic characteristics, those living in co-ethnic neighborhoods generate a higher share of inter-household carpools, regardless of race/ ethnicity and immigrant status. However, the relationship between ethnic neighborhood residency and the share of household-based carpooling shows a completely different picture: for African Americans and Asian immigrants, shares of household-based carpool trips are even lower among residents of co-ethnic neighborhoods than among those living outside such neighborhoods. Although previous studies on carpools among ethnic minorities did not distinguish household-based carpools from inter-household carpools (Liu and Painter, 2012; Blumenberg and Smart, 2014; Smart, 2015), this result implies that such a distinction is critical in associating carpools with co-ethnic neighborhood residency. 4.2. Multinomial logit regression results The variables described in Table 1 are used to construct multinomial logit regression models for each ethnic minority group, testing whether residency in co-ethnic neighborhoods affects travel mode choice after adjusting for other explanatory variables. Tables 3 and 5 exhibit the estimation results of travel mode choice for each ethnic group, using the continuous measure of co-ethnic neighborhood residency, for

Table 2 Travel mode share of home-based trips by co-ethnic neighborhood status (by ethnic group). Ethnic group

African Americans

Living in co-ethnic neighborhoods

Yes

No

Driving alone Household-carpools Interhousehold carpools N (home-based trips)

42.9 22.8 18.1 1424

44.9 26.3 14.7 1424

Hispanic immigrants

U.S. born Asians

Sig

U.S. born Hispanics Yes

No

Sig

Yes

No

Sig

Yes

No

40.7 31.3 14.9 3554

⁎⁎

26.9 33.2 17.9 1543

31.3 35.3 15.0 3801

⁎⁎

⁎ ⁎

34.5 30.0 20.2 647

47.5 28.5 16.5 200

51.7 24.2 11.8 414

⁎⁎

Note: Unit: percent; Sig column indicates statistical differences between trips by co-ethnic neighborhood status. ⁎ p b 0.05. ⁎⁎ p b 0.01.

⁎⁎

Asian immigrants Sig

Yes

No

Sig

40.5 34.5 15.6 914

37.1 40.3 11.7 1129

⁎⁎ ⁎⁎

E.J. Shin / Journal of Transport Geography 59 (2017) 14–26

19

Table 3 Multinomial logit model results for commuting trip purposes (inter-household carpools versus driving alone): continuous measure of co-ethnic neighborhoods. Variables

Co-ethnic neighborhood RCQ Locational quotient Socio-demographic characteristics HH veh Number of household vehicle Low-inc Low income household Low-ed Low level of education Female Female Age Respondent's age Age squared Age2 HH size Household size Recentfb Recent immigrant Longerfb Long-settled immigrant Job category (reference category = sales/service) Admin Clerical/admin Manu/cons Manufacture/construction Prof Professional, tech Others Other jobs Full time W Employed full-time Neighborhood characteristics Population density Popdenacre Medincome Median household income Metrodummy Constant N Pseudo R2

Metro dummy variables

African Americans

U.S.-born Hispanics

Hispanic immigrants

U.S.-born Asians

Coef.

t

Coef.

t

Coef.

t

Coef.

0.12

1.3

−0.06

−0.3

0.21

−0.62 −0.4 0.79 0.07 0.29 −3.8E−03 −0.02 N/A N/A

−2.0 −0.8 1.5 0.2 2.2 −2.3 −0.1

−0.26 0.32 −0.07 0.72 0.04 −7.7E−04 0.33 N/A N/A

−1.8 1.0 −0.2 2.4 0.6 −1.0 3.3

−1.31 −0.58 −0.05 −16.48 −0.73

−1.8 −0.8 −0.1 −12 −1.2

0.13 0.96 0.63 1.87 −0.28

−0.12 −2.9E−05

−2.1 −2.3

0.01 −8.0E−06

Asian immigrants t

Coef.

t

1.1

−0.06

−0.7

−0.36 0.28 0.33 0.61 −0.16 1.7E−03 0.14 0.26 −0.24

−2.7 1.1 1.2 2.5 −3.1 3.0 1.6 0.8 −0.7

−0.50 −0.11 −0.11 −0.08 −0.10 1.0E−03 0.26 0.48 −0.11

−2.1 −0.2 −0.2 −0.2 −1.1 1.0 1.6 1.2 −0.2

0.3 2.2 1.9 2.1 −0.8

−0.11 0.73 0.24 −13.7 −0.004

−0.2 2.4 0.6 −16.9 −0.01

−0.06 −0.39 −0.41 −18.6 0.10

−0.1 −0.5 −1.0 −15 0.2

0.6 −1.1

−2.0E−04 4.9E−06

−0.01 0.7

−2.0E−03 −2.0E−05

−0.1 −2.4

N/A

Controlled −3.58 504 0.35

−1.4

Controlled −3.58 919 0.13

−2.3

Controlled 0.12 1067 0.18

Controlled 2.16 545 0.16

0.1

0.9

Note: Bold denotes significance at the 90% level. The model for U.S.-born Asians has not achieved convergence, which probably has to do with the small sample size. The multinomial results for other travel modes are not reported.

commuting purposes and non-work-related purposes, respectively. Residence in co-ethnic neighborhoods might have discrete effects rather than continuous effects. Therefore, I conduct the same analyses based on a discrete measure of co-ethnic neighborhoods (RCQ N 2), shown in Table 4 (work trips) and Table 6 (non-work trips). Several studies have pointed out that the positivity (negativity) of the estimated coefficients

in the multinomial logit model does not necessarily indicate the positive (negative) effect of the corresponding variable because the rate of change in the probability of choosing a certain category is not a simple linear function of the coefficient in that category (here, inter-household carpooling) (Kim et al., 2007; Chen et al., 2015). Thus, I also calculate the elasticity of variables reported in Appendix Tables A and B. As for dummy

Table 4 Multinomial logit model results for commuting trip purposes (inter-household carpools versus driving alone): discrete measure of co-ethnic neighborhoods. Variables

African Americans

U.S.-born Hispanics

Hispanic immigrants

Coef.

Coef.

Coef.

Co-ethnic neighborhood RCQ RCQ N 2 Socio-demographic characteristics HH veh Number of household vehicles Low-inc Low income household Low-ed Low level of education Female Female Age Respondent's age Age squared Age2 HH size Household size Recentfb Recent immigrant Longerfb Long-settled immigrant Job category (reference category = sales/service) Admin Clerical/admin Manu/cons Manufacture/construct Prof Professional, tech Others Other jobs Full time Employed full-time Neighborhood characteristics Popdenacre Population density Median household income ($) Medincome Metrodummy Constant N Pseudo R2

Controlled −3.47 504 0.35

Metro dummy variable

t

t

t

U.S.-born Asians Coef.

Asian immigrants t

Coef.

t

N/A 0.45

1.0

−0.42

−1.1

0.32

1.2

0.05

0.1

−0.62 −0.47 0.81 0.08 0.29 −3.8E−03 −0.04 N/A N/A

−2.0 −0.9 1.6 0.2 2.2 −2.3 −0.2

−0.27 0.34 −0.08 0.71 0.04 −7.5E−04 0.33 N/A N/A

−1.9 1.0 −0.3 2.4 0.6 −1.0 3.3

−0.36 0.29 0.34 0.62 −0.16 1.8E−03 0.13 0.25 −0.23

−2.7 1.2 1.2 2.5 −3.2 3.0 1.6 0.8 −0.7

−0.49 −0.13 −0.08 −0.08 −0.10 0.001 0.26 0.48 −0.09

−2.1 −0.2 −0.1 −0.2 −1.1 1.0 1.6 1.2 −0.2

−1.31 −0.59 −0.09 −17.29 −0.77

−1.8 −0.8 −0.2 −14.0 −1.1

0.17 0.94 0.63 1.86 −0.26

0.4 2.1 2.0 2.1 −0.7

−0.11 0.73 0.23 −14.28 0.02

−0.2 2.4 0.6 −17.4 0.0

−0.07 −0.43 −0.41 −17.99 0.07

−0.1 −0.6 −1.0 −14.8 0.1

−0.12 −2.9E−05

−2.1 −2.2

0.01 −9.16E−06

0.6 −1.3

−9.1E−04 3.5E−06

−0.1 0.5

−0.004 −2.1E−05

−0.2 −2.5

−1.3

Controlled −3.47 919 0.13

−2.4

Controlled 0.44 1067 0.18

0.3

Controlled 2.06 545 0.16

0.9

Note: Bold denotes significance at the 90% level. The model for U.S.-born Asians has not achieved convergence, which probably has to do with the small sample size. The multinomial results for other travel modes are not reported.

20

E.J. Shin / Journal of Transport Geography 59 (2017) 14–26

Table 5 Multinomial logit model results for non-work-trip purposes (inter-household carpools versus driving alone): continuous measure of co-ethnic neighborhoods. Variables

African Americans

U.S.-born Hispanics

Hispanic immigrants

U.S.-born Asians

Asian immigrants

Coef.

t

Coef.

t

Coef.

t

Coef.

t

Coef.

t

0.01

0.2

0.37

4.0

0.22

2.3

0.21

1.9

0.03

0.5

−0.30 0.26 0.40 0.62 −0.06 3.5E−04 0.21 −0.13 N/A N/A

−3.3 1.5 2.5 4.1 −2.3 1.4 2.9 −0.8

−0.21 0.07 0.21 0.24 −4.7E−03 1.2E−05 0.23 −0.1 N/A N/A

−3.3 0.5 1.5 1.9 −0.2 0.1 4.0 −0.7

−0.41 0.06 0.39 0.66 −0.07 6.2E−04 0.29 −0.01 0.31 −0.09

−5.9 0.4 2.8 5.3 −3.1 2.9 5.0 −0.1 1.7 −0.6

0.08 0.96 −0.24 0.43 −0.03 1.3E−04 0.06 0.08 N/A N/A

0.5 2.0 −0.5 1.3 −0.6 0.3 0.4 0.2

−0.56 0.31 −0.08 0.26 −0.05 5.3E−04 0.45 −0.04 0.56 0.04

−4.7 1.1 −0.3 1.4 −1.1 1.2 4.8 −0.2 1.9 0.2

Co-ethnic neighborhood RCQ Locational quotient Socio-demographic characteristics HH veh Number of household vehicles Low-inc Low income household Low-ed Low level of education Female Female Age Respondent's age Squared age Age2 HH size Household size Employed Employed Recentfb Recent immigrant Longerfb Long-settled immigrant Trip purpose (reference category = shopping) Social Trip purpose: social Others Trip purpose: others Neighborhood characteristics Popdenacre Population density Medincome Median household income ($)

0.75 1.76

5.1 8.9

0.70 1.61

5.7 9.2

0.89 1.65

7.2 10.6

0.84 0.49

2.4 0.9

0.83 1.65

4.1 6.3

−1.8E−03 −2.7E−06

−0.1 −0.7

−0.01 7.9E−07

−1.0 0.3

0.01 3.8E−06

1.6 1.2

−0.02 −2.0E−05

−1.0 −2.5

−0.02 7.4E−07

−1.5 0.3

Metrodummy Constant N Pseudo R2

Controlled 0.61 2344 0.14

Metro dummy variable

0.8

Controlled −1.59 3282 0.08

−2.6

Controlled −0.23 4277 0.13

−0.3

Controlled −0.60 457 0.18

−0.3

Controlled −0.22 1498 0.12

−0.2

Note: Bold denotes significance at the 90% level. The multinomial results for other travel modes are not reported.

variables, average direct pseudo-elasticity indicating average percentage in probability of inter-household carpooling when a dummy variable switches from 0 to1 (or 1 to 0) is calculated instead of elasticity, following previous studies (e.g., Kim et al., 2007; Chen et al., 2015). 4.2.1. Effects of residence in co-ethnic neighborhoods The key interest of this study is the effects of co-ethnic neighborhood residency on the likelihood that ethnic minorities would carpool across

households. Holding all other factors constant, the effects of RCQ differ by trip purpose. RCQ has an insignificant effect on the commuting model, regardless of race/ethnicity and immigrant status, while it has mixed effects on the non-work trip model across ethnic groups. Specifically, RCQ has significant and positive coefficients for both U.S.-born Hispanics and Hispanic immigrants, which indicates that a higher percentage of co-ethnics in the neighborhood increases the likelihood of both Hispanic groups' inter-household carpooling

Table 6 Multinomial logit model results for non-work-trip purposes (inter-household carpools versus driving alone): discrete measure of co-ethnic neighborhoods. Variables

Co-ethnic neighborhood RCQ RCQ N 2 Socio-demographic characteristics HH veh Number of household vehicle Low-inc Low income household Low-ed Low level of education Female Female Age Respondent's age Age squared Age2 HH size Household size Employed Employed Recentfb Recent immigrant Longerfb Long-settled immigrant Trip purpose (reference category = shopping) Social Trip purpose: social Others Trip purpose: others Neighborhood characteristics Popdenacre Population density Medincome Median household income ($) Metrodummy Constant N Pseudo R2

Metro dummy variable

African Americans

U.S.-born Hispanics

Hispanic immigrants

Coef.

t

Coef.

t

Coef.

0.11

0.6

0.55

3.0

−0.31 0.26 0.40 0.62 −0.06 3.5E−04 0.21 −0.13 N/A N/A

−3.3 1.5 2.5 4.1 −2.3 1.5 3.0 −0.8

−0.21 0.09 0.26 0.24 −0.01 1.7E−05 0.24 −0.09 N/A N/A

0.75 1.75

5.1 8.9

−1.2E−03 −1.9E−06

−0.1 −0.5

Controlled 0.55 2344 0.14

0.7

U.S.-born Asians

Asian immigrants

t

Coef.

t

Coef.

t

0.39

2.6

0.72

2.1

0.33

1.7

−3.3 0.6 1.8 1.9 −0.3 0.1 4.2 −0.7

−0.41 0.07 0.39 0.66 −0.07 6.3E−04 0.29 −0.01 0.30 −0.09

−5.8 0.5 2.9 5.3 −3.2 2.9 5.1 −0.1 1.7 −0.6

0.11 0.97 −0.21 0.44 −0.03 1.2E−04 0.03 0.07 N/A N/A

0.8 2.0 −0.4 1.4 −0.6 0.2 0.2 0.2

−0.54 0.29 −0.09 0.27 −0.05 5.6E−04 0.44 −0.04 0.56 0.05

−4.6 1.0 −0.3 1.4 −1.1 1.2 4.7 −0.2 1.9 0.2

0.69 1.59

5.6 9.2

0.89 1.66

7.2 10.7

0.83 0.53

2.3 1.0

0.82 1.66

4.0 6.3

−0.01 −2.1E−06

−1.0 −0.8

0.01 2.9E−06

1.6 0.9

−0.02 −2.0E−05

−1.0 −2.6

−0.02 −1.2E−07

−1.5 0.0

Controlled −1.11 3282 0.08

−1.8

Controlled 0.04 4277 0.13

Note: Bold denotes significance at the 90% level. The multinomial results for other travel modes are not reported.

0.05

Controlled −0.44 457 0.18

−0.3

Controlled −0.22 1498 0.12

−0.2

E.J. Shin / Journal of Transport Geography 59 (2017) 14–26

21

Table 7 Robustness check results: multinomial logit model results for work and non-work-trip purposes (inter-household carpools versus driving alone): discrete measure of co-ethnic neighborhoods. Variables

Co-ethnic neighborhood RCQ N 2 RCQ N 2 Socio-demographic characteristics HH veh Number of household vehicle Low-inc Low income household Low-ed Low level of education Female Female Age Respondent's age Age squared Age2 HH size Household size Employed Employed Recentfb Recent immigrant Longerfb Long-settled immigrant Trip purpose (reference category = shopping) Social Trip purpose: social Others Trip purpose: other Occupation category (reference category = sales/service) Admin Clerical/admin Manu/cons Manufacture/construct Prof Professional, tech Other jobs Other jobs Full time Employed full-time Neighborhood characteristics Popdenacre Population density Medincome Median household income ($) Regional dummy Constant N Pseudo R2

African Americans

U.S.-born Hispanics

Work

Non-work Work

0.10

0.08

−0.49⁎⁎⁎ 0.09 0.14 0.40 −0.02 0.0001 0.06 – N/A N/A

−0.24⁎⁎⁎ 0.41⁎⁎⁎ 0.32⁎⁎⁎ 0.58⁎⁎⁎ −0.04⁎⁎

Hispanic immigrants

U.S.-born Asians

Non-work

Work

Non-work Work Non-work

−0.12

0.47⁎⁎⁎

0.25

0.39⁎⁎⁎

−0.29⁎⁎ 0.43 −0.02 0.87⁎⁎⁎

−0.19⁎⁎⁎ 0.02 0.25⁎

−0.38⁎⁎⁎ 0.27 0.44⁎ 0.54⁎⁎ −0.14⁎⁎ 0.001⁎⁎

−0.33⁎⁎⁎ 0.15 0.45⁎⁎⁎ 0.58⁎⁎⁎ −0.08⁎⁎⁎ 0.001⁎⁎⁎ 0.27⁎⁎⁎

Asian immigrants Work

Non-work

0.97⁎⁎⁎

0.02

0.25⁎

−0.01 0.41⁎⁎ −0.10

0.04 1.06⁎⁎ 0.48 0.27 −0.009 −7.3E−05 0.09 0.12 N/A N/A

−0.53⁎⁎ −0.003 −0.11 −0.43 −0.08 0.001 0.22 – 0.60 0.04

−0.40⁎⁎⁎ 0.28 0.23 0.28 −0.07 0.0006 0.35⁎⁎⁎ −0.03 0.51⁎

N/A

0.0003 0.22⁎⁎⁎ −0.04 N/A N/A

0.03 −0.0006 0.38⁎⁎⁎ – N/A N/A

0.20 −0.02 0.0002 0.20⁎⁎⁎ −0.11 N/A N/A

– –

0.83⁎⁎⁎ 1.64⁎⁎⁎

– –

0.72⁎⁎⁎ 1.51⁎⁎⁎

– –

0.91⁎⁎⁎ 1.59⁎⁎⁎

0.85⁎⁎⁎ 0.36

– –

0.83⁎⁎⁎ 1.66⁎⁎⁎

−1.47⁎⁎⁎ −0.29 −0.289 −13.7⁎⁎⁎ 0.15

– – – – –

0.15 1.26⁎⁎⁎ 0.66⁎⁎ 1.73⁎⁎ 0.26

– – – – –

−0.12 0.67⁎⁎ 0.07 −12.1⁎⁎⁎ −0.006

– – – – –

– – – – –

0.06 −0.59 −0.88⁎⁎ −16.0⁎⁎⁎ 0.12

– – – – –

1.0E−03 −6.0E−06 Controlled −0.61 978 0.22

0.007 1.0E−06

1.9E−02 −1.9E−03 −1.4E−03 3.0E−06 −2.0E−06 6.0E−06 Controlled Controlled −4.31⁎⁎⁎ −0.60 0.33 1021 3596 1182 0.16 0.09 0.20

5.8E−03 4.0E−06

−0.01 −1.6E−05⁎⁎ Controlled −0.53 515 0.16

0.002 −5.0E−06 Controlled 0.87 691 0.17

0.006 1.0E−06

−0.63 4165 0.14

0.11 – 0.19 −0.26

0.07 4687 0.13

0.12

−0.31 1867 0.12

Notes: − indicates that variables are not included in a model. Regional dummies include: West (reference category), East, South, and Midwest. The model for U.S.-born Asians has not achieved convergence, which has to do with the small sample size. The multinomial results for other travel modes are not reported. ⁎ p b 0.1. ⁎⁎ p b 0.05. ⁎⁎⁎ p b 0.01.

for non-work purposes. Based on the elasticity results, 100% increases in RCQ values lead to about a 30% and 15% increase in the probability of inter-household carpooling for non-work purposes among U.S.-born Hispanics and Hispanic immigrants, respectively. RCQ also has a significant and positive effect on the likelihood that U.S.-born Asians would form inter-household carpools for nonwork purposes, suggesting that Asian enclave effects that are significant for this group; the elasticity of RCQ is positive and larger (about 38.0%) for this group than for any other racial/ethnic groups. RCQ, however, has insignificant effects for Asian immigrants and African Americans. Tables 4 and 6 show that this discrete measure does not change the statistical significance of the co-ethnic neighborhood residence variables for the ethnic groups, except for Asian immigrants. The positive effects of co-ethnic neighborhood residency do become statistically significant for Asian immigrants in a non-work trip model when using a discrete measure of co-ethnic neighborhood residency. This result, therefore, implies that the concentration of Asian populations in the neighborhood has only a discrete and not a continuous effect on the likelihood that Asian immigrants will use inter-household carpools for non-work purposes. Asian neighborhood residency (RCQ N 2) increases the probability that Asian immigrants will use inter-household carpools for non-work purposes by 54.7% (Appendix B). Overall, these findings lend some support to the hypotheses established in Section 2.2 in that the effects of co-ethnic neighborhood residency vary across ethnic minority groups and their coefficients tend to carry expected signs in non-work models. But I do not find evidence among Hispanics and Asians that co-ethnic neighborhood effects

are stronger for immigrants than for U.S.-born ethnic minorities. Detailed discussions on each hypothesis are provided in Section 5. 4.2.2. Other determinants of inter-household carpooling Another goal of the present study is to identify the determinants of inter-household carpooling and examine how they differ across ethnic minority groups. Several individual and household characteristics are found to be associated with inter-household carpooling rather than driving alone. The coefficients of individual and household characteristics that are statistically significant in this study are similar to previous findings on the determinants of carpooling based on total population (Teal, 1987; Ferguson, 1997; Charles and Kline, 2006). However, the statistical significance of coefficients is not the same across ethnic groups and trip purposes (i.e., work versus non-work purposes), which suggests the unequal effects of these factors on inter-household carpooling across different groups in different situations. Regardless of trip purposes, variables related to the economic status are found to affect the probability of inter-household carpooling significantly; the coefficients of household vehicle availability (“HH veh”) are statistically significant and negative across all ethnic groups except for U.S.-born Asians. Instead, low household income status (“Low-inc”) has a positive coefficient for U.S.-born Asians, whereas it does not for other ethnic groups. Low levels of education (“Low-ed”), which are a proxy for individual socio-economic status, also have positive effects for African Americans and Hispanic immigrants in terms of their likelihood of interhousehold carpooling for non-work purposes. The positivity (negativity) of variables do not change based on the calculated elasticities. Thus, these consistently confirm previous findings that low socio-economic

22

E.J. Shin / Journal of Transport Geography 59 (2017) 14–26

status increases the probability of relying on non-solo driving modes of transportation, which makes ethnic groups with lower socio-economic status more likely to use inter-household carpooling. Furthermore, considering that the magnitudes of elasticities of these variables vary across ethnic minority groups, differences in socio-economic status across groups seem to contribute further to inter-group differences in nonwork travel behavior. This is in line with the marginal hypothesis. For some groups, gender (“Female”) is correlated with a choice to carpool over driving alone and their coefficients carry expected signs; previous studies have found that ethnic minority women tend to rely on alternative modes of transportation more than ethnic minority men do (e.g., McLafferty and Preston, 1997). I find that, among both Hispanic groups, females are more likely than males to use inter-household carpools irrespective of trip purpose. For instance, being a female almost doubles (increases by approximately 90%) the probability of interhousehold carpooling for commutes among U.S.-born Hispanics. Female African Americans are also more likely to use inter-household carpools for non-work trip purposes only, but there is no gender difference among Asian groups. This implies that gender effects on travel behavior differ across race/ethnicity. Age affects the likelihood of inter-household carpooling for several ethnic minority groups, although the direction of effects is mixed across groups. Larger household size (“HH Size”) also has positive and significant coefficients for all ethnic groups except U.S.-born Asians in terms of non-work trips, holding all else constant. Interestingly, the signs of coefficients are misleading for some groups; unlike the positive coefficients in Tables 4 and 6, the elasticities of variables show that larger household size will decrease the probability of inter-household carpooling for nonwork purposes, although the magnitudes of these elasticities of variables are relatively small. Meanwhile, effects are statistically significant and positive only for U.S.-born Hispanics for commuting purposes. Several studies demonstrated that the length of U.S. residency is associated with the likelihood of using alternative modes of transportation among immigrants (Chatman and Klein, 2009; Kim, 2009; Liu and Painter, 2012). Whereas the coefficients of recent immigrants (“Recentfb”) carry expected signs for both Hispanic and Asian immigrants in the non-work model, the elasticity results show a different picture. In the case of Hispanic immigrants, the direction of effects changes, whereas it does not for Asian immigrants. However, the magnitudes of elasticities (−1.0% to −2.0%) imply that the length of U.S. residency has almost no impacts on the probability of inter-household carpooling for Hispanic immigrants. The regression models indicate that some job-related characteristics are strong predictors of inter-household carpooling for commutes, and the magnitudes of elasticities are larger than those for other variables. For instance, two Hispanic groups working in manufacturing/construction/maintenance are much more likely to use inter-household carpools than are those who work in sales and services. It is likely that working hours, spatial location of jobs, and ethnic composition of jobs influence these outcomes. Employment status, however, has an insignificant effect in a non-work trip model for all ethnic groups. It appears that, among non-work trips, the trip purpose affects the propensity to use inter-household carpooling versus driving alone. Regardless of race/ethnicity and immigrant status, ethnic minorities are more likely to form inter-household carpools for social and other/personal business purposes than for shopping trips, a result which coincides with previous findings (Blumenberg and Smart, 2010). The results show that, regardless of trip purpose, population density (“Popdenacre”) barely affects the probability of inter-household carpooling versus driving alone among ethnic minorities. At the same time, the economic status of the neighborhood (“Medincome”) is found to have negative and sizable effects for some groups but not for all groups: the lower the neighborhood economic status, the higher the likelihood that residents will use inter-household carpools. Given the adjustments for population density and economic status variables, the results of the effects of co-ethnic neighborhood residency on the

likelihood of inter-household carpooling are likely because of ethnic networks rather than the other characteristics of neighborhoods.

4.3. Robustness checks The analyses in this study are restricted to eight metropolitan areas based on the assumption that only metropolitan areas with sizable populations of each ethnic group have ethnic neighborhoods that exert significant effects on those who belong to a given race/ ethnicity. In addition, the study area is also limited to southern and western regions of the United States, which have relatively similar urban development patterns. However, the representativeness of the study area might be questionable due to the small area covered in our investigation. Thus, robustness checks are attempted by conducting the analyses based on larger samples that include six additional large U.S. metropolitan areas: Chicago-NapervilleJoliet, New York-Northern New Jersey-Long Island, Virginia Beach-Norfolk-Newport News, Washington-Arlington-Alexandria, Richmond, and Boston-Cambridge-Quincy CBSAs. These metro areas are excluded in the analyses shown in Section 4.2 because they have fewer proportions of Hispanic or Asian groups, or because of their geographical locations. As a result of adding these regions to the study, approximately three-fourths of the total ethnic minority NHTS samples living in any CBSAs are included, which suggests the generalizability of the findings. The results of the robustness checks for the multinomial logit regressions are shown in Table 7 (discrete measures of co-ethnic neighborhoods). The results of the elasticities of the variables are reported in Appendix C. Overall, the coefficients and elasticities of variables remain largely similar to the analyses in Section 4.2. The results of co-ethnic neighborhoods' effects on inter-household carpooling for commute purposes are found to be insignificant for all ethnic groups. Whereas Asian and Hispanic co-ethnic neighborhood effects on the probability of inter-household carpooling for non-work purposes exist among Hispanics and Asians, black neighborhoods show no significant effects on the probability that African Americans would use inter-household carpools. This is notable, because some of the metropolitan areas added for the robustness checks, including Virginia and Washington D.C. areas, have relatively large black populations compared to other metro areas. This suggests that regardless of the size of the black population, black neighborhood effects do not play significant roles in the formation of inter-household carpools among black residents. The magnitude of the effects of co-ethnic

Fig. 2. Predicted probability of inter-household carpooling for non-work-trip purposes by concentration level of co-ethnics in one's residential neighborhood.

E.J. Shin / Journal of Transport Geography 59 (2017) 14–26

23

Fig. 3. Predicted probability of inter-household carpooling for non-work-trip purposes by co-ethnic neighborhood status panel a. (left): total sample; panel b. (right): those with low socioeconomic status (i.e., low education level, low household income, and one household vehicle) but who are otherwise average people.

neighborhood variables does not greatly differ from those shown in Section 4.2, which is a clear indication of the robustness of the findings.

5. Discussion For a better understanding of the extent to which co-ethnic neighborhood residency affects the likelihood of inter-household carpools for the average person in each ethnic group, Figs. 2 and 3a show the predicted probability of inter-household carpooling for non-work purposes by a continuous and discrete measure of co-ethnic neighborhoods, respectively. These probabilities are calculated by ethnic group with all other explanatory variables being held at their group mean. These two figures demonstrate that an increase in the concentration of co-ethnics positively affects the probability of non-work related inter-household carpooling for all ethnic groups but African Americans. For instance, an average U.S.born Hispanic who lives in Hispanic-dominant neighborhoods (RCQ = 3) has a 30% probability of inter-household carpooling for non-work purposes whereas the probability would be 18% for such a person in a neighborhood with an average regional concentration level of Hispanics (RCQ = 1). As the regression results confirm, although the continuous measure of co-ethnic neighborhood residency has insignificant effects among Asian immigrants, there is a fairly large difference in the predicted probability of inter-household carpooling between co-ethnic neighborhood residents and non-co-ethnic neighborhood residents among this group, based on a discrete measure in this group. Specifically, Asian immigrants living in Asian neighborhoods have about a 21% probability of interhousehold carpooling, which is much higher than the probability among those living outside Asian neighborhoods (about 14%). This probably has to do with the heterogeneity of Asian immigrants. Whereas a large proportion of Hispanic immigrants originated in Mexico and many South American countries use the same language, Asian immigrants are heterogeneous according to their country of origin, leading to relatively large within-group variances of languages and cultures. It is therefore possible that the share of total Asians should reach a certain level (here, RCQ N 2) in order to have a sizable number of a certain Asian subgroup (e.g., Korean, Vietnamese) in a neighborhood. The findings of this study imply that the effects of co-ethnic neighborhood residency across ethnic groups are uneven. The results confirm the hypothesis that Hispanic and Asian dominant neighborhoods foster the formation of social networks and the resultant use of inter-household carpools, whereas black neighborhoods do not. Some might point out that the predicted probability of inter-household carpooling is, on average, higher among African Americans than among other groups, which might suggest social network effects in their neighborhoods. However,

this higher probability seems to result from the generally relatively low socioeconomic status of African Americans rather than from any social networks in black neighborhoods. This interpretation is based on Fig. 3b, which presents the predicted probabilities of inter-household carpooling based only on those with low economic status across ethnic groups. Fig. 3b shows that in circumstances of low economic status, the probability of using inter-household carpools is not especially higher for African Americans than for other ethnic groups. In addition, Fig. 3a and b shows the findings supporting the marginality hypothesis. The lower probability of inter-household carpooling among Asian groups seems largely explained by their higher economic status; the predicted probabilities of inter-household carpooling are even higher among Asian groups once samples are restricted to those with low economic status (Fig. 3b). On the other hand, the findings of this study do not support the hypothesis that the effects of co-ethnic neighborhood residency would be stronger for immigrants than for U.S.-born ethnic minorities. Figs. 2 and 3 demonstrate that, regardless of which co-ethnic neighborhood residency measures are used, co-ethnic neighborhood effects are larger for U.S.-born populations than for immigrants among both Hispanics and Asians. However, these differences in the effects of co-ethnic neighborhood residency between U.S.-born Hispanics and Hispanic immigrants, and between U.S.-born Asians and Asian immigrants are not statistically significant. Smart (2015) demonstrated that the effects of residence in an immigrant neighborhood are stronger for immigrants than for the U.S.born population, but my finding implies that this is not true for those whose race/ethnicity is the same as the ethnic characteristics of their neighborhood. That is, U.S.-born ethnic minorities seem to experience similar effects from co-ethnic neighborhood residency compared to their immigrant counterparts.

6. Conclusion The purpose of this paper is to investigate the travel behavior implications of racial residential segregation. Ethnic neighborhoods provide a unique social environment for those who belong to the same ethnic group. Geographical proximity among coethnics may encourage group-oriented activities and transport assistance, based on cultural affinity and a common language. Residence in co-ethnic neighborhoods might also slow down adaptation to the U.S. automobile culture among immigrants owing to the social influence. However, the results of this study show that, after having adjusted for socio-economic characteristics and other variables, the effects of co-ethnic neighborhood residency on the probability of carpooling across households differ across ethnic groups. While co-neighborhood residency is positively associated with the

24

E.J. Shin / Journal of Transport Geography 59 (2017) 14–26

probability of inter-household carpooling for non-work purposes among both Hispanics and Asians, such ethnic neighborhood effects are found to be insignificant for African Americans. In addition, the effects of co-ethnic neighborhood residency are not particularly stronger for Hispanic immigrants and Asian immigrants compared to their U.S.born counterparts. These findings imply that the role of neighborhoods in fostering social ties, as reflected by travel behavior, is complex. One potential explanation for such complexity is that the constraints related to residential location choice vary by ethnic group: this may result in differences in local social networks across ethnic neighborhood types. However, as the findings of this study rely only on quantitative evidence, other potential explanations for this phenomenon should be further explored. For example, it is possible that African Americans have different channels of ridesharing, other than neighborhood ethnic ties, such as institutions. Hence, qualitative studies on how and why each ethnic group forms inter-household carpools are needed to reveal the mechanisms behind the neighborhood effects. Moreover, in order to reach a conclusion on the consequences of racial residential segregation with regard to social support networks, it would be interesting in future research to investigate the diverse measures of social support networks among ethnic minorities in association with their residential location. In addition to investigating the different roles of co-ethnic neighborhood residency in promoting inter-household carpools across ethnic minority groups, the present study also contributes to the literature by revealing other important determinants of inter-household carpooling and the different effects of these determinants across racial/ethnic groups. In terms of commute models, household vehicle availability and occupation types appear to be the chief predictors for the probability that ethnic minority groups will use inter-household carpools. A lower economic status neighborhood is also found to be a strong motivator for forming interhousehold carpools for commutes among African Americans and Asian immigrants. Variables related to socio-economic status also play a significant role in the non-work models, although the effects are not as

pronounced as in the commute model. The magnitudes of these effects were found to differ across ethnic minority groups, suggesting that differences in socio-economic status across groups directly and indirectly influence inter-group variations in travel behavior. This also confirms that marginal socio-economic conditions affect reliance on ridesharing across households. In addition to trip purposes, gender is relatively a strong predictor of inter-household carpooling for non-work purposes among African Americans and Hispanic immigrants. Finally, the findings of the present study have several important transportation policy implications, especially related to the forecast of future travel demand. Historical trends show that two types of segregation—Hispanic–white and Asian–white—have remained steady or have increased, whereas black–white segregation has decreased over the past several decades. Considering that co-ethnic neighborhood effects for U.S.-born Asians and Hispanics are as pronounced as for their immigrant counterparts, such neighborhood effects on travel outcomes are likely to endure, even after immigration slows down. The findings of this study can therefore inform policy making for the provision of infrastructure and systems appropriate to the needs of these population segments. In addition, as the present study confirms that a network system promotes inter-household carpools in Hispanic and Asian neighborhoods, policy interventions that facilitate such a system, thereby helping more residents of those neighborhoods find ridesharing opportunities with their neighbors, should be provided. Acknowledgements I would like to thank Dr. Genevieve Giuliano for her valuable advice and support. Constructive feedback from Dr. Marlon Boarnet, Dr. Manuel Pastor, Dr. Lisa Schweitzer, and two anonymous reviewers is very much appreciated. Special thanks to Dr. Joon-ki Kim for helpful suggestions. Thanks to the U.S. Department of Transportation for sharing data. All errors and omissions are my responsibility.

Appendix A. Elasticity (continuous variables) and average direct pseudo-elasticity (dummy variables) of variables: continuous measures of co-ethnic neighborhoods

Variables

Co-ethnic neighborhood RCQ Locational quotient Socio-demographic characteristics HH veh Number of household vehicle Low-inc Low household income Lowedu Low educational level Female Female = 1 Age Respondent's age Squared age Age2 HH size Household size Employed Employed = 1 Recentfb Recent immigrant Longerfb Long-settled immigrant Trip purpose (reference category = shopping) Social Trip purpose: social Others Trip purpose: others Occupation category (reference category = sales/service) Admin Clerical/admin Manu/cons Manufacture/construction Prof Professional/tech Other jobs Other jobs Working full-time Full time Neighborhood characteristics Popdenacre Population density Medincome Median household income ($)

African Americans

U.S.-born Hispanics

Hispanic immigrants

U.S.-born Asians

Asian immigrants

Work

Work

Work

Work

Non-work

Work

Non-work

Non-work

Non-work

Non-work

N/A 23.9%

1.1%

−5.7%

29.8%

30.7%

15.0%

38.0%

−10.6%

9.3%

−65.6% −25.7% 120.0% 9.3% 1607% −873.1% −151.2% – N/A N/A

−4.2% 6.3% 17.0% 55.7% −176.2% 67.7% −9.7% −3.4%

−56.6% 39.1% −8.5% 91.7% 121.6% −111.3% 97.7% – N/A N/A

−1.8% 17.6% 2.8% 12.9% −113.2% 56.6% −6.9% 4.2%

−80.2% 21.9% 34.6% 35.5% −703.8% 361% 51.5% – 16.1% −21.5%

−11.9% 9.4% 20.6% 32.0% −175.2% 82.2% −11.7% 22.8% −1.4% 11.4%

12.0% 163.4% −10.4% 15.3% −328.9% 131.2% −26.4% 82.2% N/A N/A

−97.1% −3.9% −16.7% −12.8% −494.2% 242.8% 64.0% – 43.6% −1.9%

−35.5% 30.4% −8.7% 16.6% 38.1% −29.2% 18.5% −8.9% 5.4% 16.0%

– –

27.0% 86.9%

– –

18.0% 37.7%

– –

16.2% 30.3%

48.3% −29.1%

– –

22.0% 27.3%

−63.4% 108.4% −2.6% −98.8% −44.9%

– – – – –

9.2% 143.5% 80.3% 357.4% −23.9%

– – – – –

−8.2% 102.3% 19.6% −100% 3.6%

– – – – –

– – – – –

−11.9% −27.0% −37.7% −100% 51.2%

– – – – –

−69.9% −108.3%

−3.1% −13.9%

8.5% −53.7%

−6.0% 2.8%

−0.3% 27.1%

5.3% 2.0%

−13.9% −105.8%

−3.2% −152%

−1.4% −21.9%

Note: – indicates that the variables are not included in the model.

E.J. Shin / Journal of Transport Geography 59 (2017) 14–26

25

Appendix B. Elasticity (continuous variables) and average direct pseudo-elasticity (dummy variables) of variables: discrete measures of co-ethnic neighborhoods Variables

Co-ethnic neighborhood RCQ N 2 RCQ N 2 Socio-demographic characteristics HH veh Number of household vehicle Low-inc Low household income Lowedu Low educational level Female Female Age Respondent's age Squared age Age2 HH size Household size Employed Employed Recentfb Recent immigrant Longerfb Long-settled immigrant Trip purpose (reference category = shopping) Social Trip purpose: social Others Trip purpose: others Occupation category (reference category = sales/service) Admin Clerical/admin Manu/cons Manufacture/construction Prof Professional/tech Other jobs Other jobs Working full-time Full time Neighborhood characteristics Popdenacre Population density Medincome Median household income ($)

African Americans

U.S.-born Hispanics

Hispanic immigrants

U.S.-born Asians

Asian immigrants

Work

Work

Work

Work

Non-work

Work

Non-work

Non-work

Non-work

Non-work

N/A 57.2%

6.2%

−32.5%

35.4%

43.1%

16.9%

71.2%

6.9%

54.7%

−68.9% −30.3% 123.2% 9.0% 1624% −881% −154.5% – N/A N/A

−4.4% 6.2% 16.8% 55.6% −177.4% 67.9% −9.7% −3.5%

−57.2% 41.7% −8.8% 89.0% 117.1% −109% 96.1% – N/A N/A

−0.9% 19.8% 6.1% 13.0% −116.5% 58.5% −3.7% 5.1%

−79.2% 23.7% 35.2% 36.7% −708.4% 363% 50.9% – 15.1% −20.4%

−11.7% 1.8% 21.0% 31.7% −176.6% 82.7% −10.8% 22.6% −2.0% 11.6%

17.9% 151.1% −9.1% 16.2% −354.4% 141.5% −33.6% 56.4%

−95.3% −6.0% −13.8% −12.4% −493.4% 243.2% 62.2% – −2.0% 11.6%

−30.7% 27.8% −9.1% 17.7% 24.2% −22.0% 15.7% −9.6% 44.0% −0.5%

– –

26.9% 86.9%

– –

23.6% 40.7*%

– –

16.3% 30.6%

46.6% −27.1%

– –

21.1% 28.2%

−64.1% 220% −7.0% −98.9% −46.1%

– – – – –

12.6% 138.5% 80.1% 357.6% −22.5%

– – – – –

−8.0% 102.3% 17.9% −100% 6.4%

– – – – –

– – – – –

−13.3% −28.9% −37.3% −100% 51.1%

– – – – –

−68.9% −95.1%

−2.8% −11.5%

9.2% −61.6%

−5.1% −12.3%

−1.2% 19.4%

5.9% −2.3%

−16.3% −105%

−4.3% −159%

−2.5% −30.6%

Note: – indicates that the variables are not included in the model.

Appendix C. Elasticity (continuous variables) and average direct pseudo-elasticity (dummy variables) of variables: discrete measures of co-ethnic neighborhoods

Variables

Co-ethnic neighborhood RCQ N 2 RCQ N 2 Socio-demographic characteristics HH veh Number of household vehicle Low-inc Low household income Lowedu Low educational level Female Female Age Respondent's age Squared age Age2 HH size Household size Employed Employed Recentfb Recent immigrant Longerfb Long-settled immigrant Trip purpose Social Trip purpose: social Others Trip purpose: others Occupation category (reference category = sales/service) Admin Clerical/admin Manu/cons Manufacture/construction Prof Professional/tech Other jobs Other jobs Full time Working full-time Neighborhood characteristics Population density Popdenacre Medincome Median household income ($)

African Americans

U.S.-born Hispanics

Hispanic immigrants

U.S.-born Asians

Asian immigrants

Work

Non-work

Work

Non-work

Work

Non-work

Work

Non-work

Work

Non-work

11.4%

3.3%

−11.2%

30.0%

37.0%

22.4%

97.8%

2.5%

38.2%

−85.8% 10.8% 19.9% 35.3% −99.7% 27.0% 7.4% – N/A N/A

−0.7% 25.8% 15.5% 49.0% −130.3% 45.9% −4.6% 1.4%

−52.0% 56.2% −7.0% 124.5% 51.4% −79.8% 104.8% – N/A N/A

3.8% 16.3% 6.3% 7.0% −140.2% 63.7% −11.6% 3.2%

−42.4% 18.9% 48.4% 25.0% −641.3% 333.4% 24.1% – 9.1% −20.1%

10.4% 4.0% 22.3% 22.4% −199.5% 84.8% −16.3% 21.6% 7.1% 9.4%

23.2% 169.3% 46.1% 5.9% −109.6% 24.9% −24.7% 13.9%

−97.0% 5.3% −8.4% −37.2% −384.2% 201.4% 44.5% – 60.4% 5.7%

−11.0% 21.6% 12.2% 14.8% −45.8% 4.9% 10.1% −5.1% 4.5% 22.9%

– –

37.4% 82.2%

– –

17.4% 37.9%

– –

17.9% 30.8%

47.9% −39.1%

– –

21.1% 26.5%

−74.9% −21.0% −22.2% −100.0% 16.8%

– – – – –

11.4% 216.6% 85.2% 389.4% 25.4%

– – – – –

−9.3% 89.4% 7.0% −100.0% 2.7%

– – – – –

– – – – –

−8.5% −39.7% −56.9% −100.0% 16.9%

– – – – –

0.6% −27.8%

−0.4% 0.5%

19.8% 17.3%

−4.4% −11.7%

−4.9% 36.6%

−1.8% 0.6%

−6.0% −90.5%

1.4% −32.5%

16.4% −16.4%

N/A

Note: – indicates that the variables are not included in the model.

References Acevedo-Garcia, D., Lochner, K.A., Osypuk, T.L., Subramanian, S.V., 2003. Future directions in residential segregation and health research: a multilevel approach. Am. J. Public Health 93 (2), 215–221. Allison, M.T., 1988. Breaking boundaries and barriers: future directions in cross-cultural research. Leis. Sci. 10 (4), 247–259.

Bailey, T., Waldinger, R., 1991. Primary, secondary, and enclave labor markets: a training systems approach. Am. Sociol. Rev. 56 (4), 432–445. Blumenberg, E., Smart, M., 2010. Getting by with a little help from my friends… and family: immigrants and carpooling. Transportation 37 (3), 429–446. Blumenberg, E., Smart, M., 2014. Brother can you spare a ride? Carpooling in immigrant neighbourhoods. Urban Stud. 51 (9), 1871–1890.

26

E.J. Shin / Journal of Transport Geography 59 (2017) 14–26

Blumenberg, E., Donahue, M., Handy, S.L., Lovejoy, K., Rodier, C.J., Shaheen, S., Volker, J., 2007. Travel of Diverse Populations: Literature Review. Institute of Transportation Studies. Borjas, G.J., 1998. To ghetto or not to ghetto: ethnicity and residential segregation. J. Urban Econ. 44 (2), 228–253. Briggs, X., 1998. Brown kids in white suburbs: housing mobility and the many faces of social capital. Hous. Policy Debate 9 (1), 177–221. Card, D., Rothstein, J., 2007. Racial segregation and the black–white test score gap. J. Public Econ. 91 (11), 2158–2184. Carrasco, J.A., Miller, E.J., 2006. Exploring the propensity to perform social activities: a social network approach. Transportation 33 (5), 463–480. Charles, C.Z., 2000. Neighborhood racial-composition preferences: evidence from a multiethnic metropolis. Soc. Probl. 47 (3), 379–407. Charles, K.K., Kline, P., 2006. Relational costs and the production of social capital: evidence from carpooling. Econ. J. 116 (511), 581–604. Chatman, D.G., Klein, N., 2009. Immigrants and travel demand in the United States implications for transportation policy and future research. Public Works Manag. Policy 13 (4), 312–327. Chen, C., Zhang, G., Tian, Z., Bogus, S.M., Yang, Y., 2015. Hierarchical Bayesian random intercept model-based cross-level interaction decomposition for truck driver injury severity investigations. Accid. Anal. Prev. 85, 186–198. Cline, M., Sparks, C., Eschbach, K., 2009. Understanding carpool use by Hispanics in Texas. Transp. Res. Rec. 2118, 39–46. Dawkins, C.J., 2004. Recent evidence on the continuing causes of black-white residential segregation. J. Urban Aff. 26 (3), 379–400. Dominguez, S., Watkins, C., 2003. Creating networks for survival and mobility: social capital among African-American and Latin-American low-income mothers. Soc. Probl. 50 (1), 111–135. Elliott, J.R., 2001. Referral hiring and ethnically homogeneous jobs: how prevalent is the connection and for whom? Soc. Sci. Res. 30 (3), 401–425. Elliott, J.R., Sims, M., 2001. Ghettos and barrios: the impact of neighborhood poverty and race on job matching among blacks and Latinos. Soc. Probl. 48 (3), 341–361. Emerson, M.O., Chai, K.J., Yancey, G., 2001. Does race matter in residential segregation? Exploring the preferences of white Americans. Am. Sociol. Rev. 66 (6), 922–935. Ferguson, E., 1997. The rise and fall of the American carpool: 1970–1990. Transportation 24 (4), 349–376. Floyd, M.F., 1998. Getting beyond marginality and ethnicity: the challenge for race and ethnic studies in leisure research. J. Leis. Res. 30 (1), 3. Floyd, M.R., Shinew, K.J., McGuire, F.A., Noe, F.P., 1994. Race, class, and leisure activity preferences: marginality and ethnicity revisited. J. Leis. Res. 26 (2), 158. Frey, W.H., 2011. The New Metro Minority Map: Regional Shifts in Hispanics, Asians, and Blacks from Census 2010. Metropolitan Policy Program, Brookings Institution, Washington, DC. Gans, H.J., 2008. Involuntary segregation and the ghetto: disconnecting process and place. City & Community 7 (4), 353–357. Giuliano, G., 2003. Travel, location and race/ethnicity. Transp. Res. A Policy Pract. 37 (4), 351–372. Ibipo, J.A., 1995. Racial differences in the commuting behavior of women in Buffalo, 1980– 1990. Urban Geogr. 16 (1), 23–45. Iceland, J., Sharp, G., Timberlake, J.M., 2013. Sun belt rising: regional population change and the decline in Black residential segregation, 1970–2009. Demography 50 (1), 97–123. Ihlanfeldt, K.R., Scafidi, B., 2002. Black self-segregation as a cause of housing segregation: evidence from the multi-city study of urban inequality. J. Urban Econ. 51 (2), 366–390. Ihlanfeldt, K.R., Sjoquist, D.L., 1998. The spatial mismatch hypothesis: a review of recent studies and their implications for welfare reform. Hous. Policy Debate 9 (4), 849–892. Kim, S., 2009. Immigrants and transportation: an analysis of immigrant workers' work trips. Cityscape 155–169. Kim, H.K., McKenry, P.C., 1998. Social networks and support: a comparison of African Americans, Asian Americans, Caucasians, and Hispanics. J. Comp. Fam. Stud. 313–334. Kim, J.K., Kim, S., Ulfarsson, G.F., Porrello, L.A., 2007. Bicyclist injury severities in bicycle– motor vehicle accidents. Accid. Anal. Prev. 39 (2), 238–251.

Kirschenman, J., Neckerman, K.M., 1991. In: Jencks, C., Peterson, P. (Eds.), ‘We'd Love to Hire Them, But…’: The Meaning of Race for Employers. The Urban Underclass. Brookings Institution, Washington, DC, pp. 203–232. Kramer, M.R., Hogue, C.R., 2009. Is segregation bad for your health? Epidemiol. Rev. 31 (1), 178–194. Liu, C.Y., Painter, G., 2012. Travel behavior among Latino immigrants: the role of ethnic concentration and ethnic employment. J. Plan. Educ. Res. 32 (1), 62–80. Maness, M., Cirillo, C., Dugundji, E.R., 2015. Generalized behavioral framework for choice models of social influence: behavioral and data concerns in travel behavior. J. Transp. Geogr. 46, 137–150. Massey, D.S., Lundy, G., 2001. Use of Black English and racial discrimination in urban housing markets new methods and findings. Urban Aff. Rev. 36 (4), 452–469. McLafferty, S., Preston, V., 1997. Gender, race, and the determinants of commuting: New York in 1990. Urban Geogr. 18 (3), 192–212. Nguyen, M.T., 2004. The self-segregation of Asians and Hispanics: the role of assimilation and racial prejudice. Race Soc. 7 (2), 131–151. Peach, C., 2005. The ghetto and the ethnic enclave. Desegregating the City: Ghettos, Enclaves, and Inequality, pp. 31–48. Preston, V., McLafferty, S., 1999. Spatial mismatch research in the 1990s: progress and potential. Pap. Reg. Sci. 78 (4), 387–402. Pucher, J., Renne, J.L., 2003. Socioeconomics of urban travel: evidence from the 2001 NHTS. Transp. Q. 57 (3), 49–77. Quillian, L., 2014. Does segregation create winners and losers? Residential segregation and inequality in educational attainment. Soc. Probl. 61 (3), 402–426. Reingold, D.A., 1999. Social networks and the employment problem of the urban poor. Urban Stud. 36 (11), 1907–1932. Roy, P., Martínez, A.J., Miscione, G., Zuidgeest, M.H.P., van Maarseveen, M.F.A.M., 2012. Using social network analysis to profile people based on their e-communication and travel balance. J. Transp. Geogr. 24, 111–122. Shin, E.J., 2016. Unraveling the effects of residence in an ethnic enclave on immigrants’ travel mode choices. J. Plan. Educ. Res. http://dx.doi.org/10.1177/ 0739456X16663309 (in press). Small, M.L., 2007. Racial differences in networks: do neighborhood conditions matter? Soc. Sci. Q. 88 (2), 320–343. Smart, M., 2010. US immigrants and bicycling: two-wheeled in Autopia. Transp. Policy 17 (3), 153–159. Smart, M.J., 2015. A nationwide look at the immigrant neighborhood effect on travel mode choice. Transportation 42 (1), 189–209. Smith, S.S., 2005. “Don't put my name on it”: social capital activation and job-finding assistance among the Black urban poor. Am. J. Sociol. 111 (1), 1–57. Sultana, S., 2005. Racial variations in males' commuting times in Atlanta: what does the evidence suggest? Prof. Geogr. 57 (1), 66–82. Taylor, B.D., Ong, P.M., 1995. Spatial mismatch or automobile mismatch? An examination of race, residence and commuting in US metropolitan areas. Urban Stud. 32 (9), 1453–1473. Teal, R.F., 1987. Carpooling: who, how and why. Transp. Res. A 21 (3), 203–214. U.S. Department of Transportation, Federal Highway Administration, 2009. National Household Travel Survey. http://nhts.ornl.gov. Vanoutrive, T., Van De Vijver, E., Van Malderen, L., Jourquin, B., Thomas, I., Verhetsel, A., Witlox, F., 2012. What determines carpooling to workplaces in Belgium: location, organisation, or promotion? J. Transp. Geogr. 22, 77–86. Washburne, R.F., 1978. Black under-participation in wildland recreation: alternative explanations. Leis. Sci. 1 (2), 175–189. Williams, D.R., Collins, C., 2001. Racial residential segregation: a fundamental cause of racial disparities in health. Public Health Rep. 116 (5), 404. Wilson, K.L., Portes, A., 1980. Immigrant enclaves: an analysis of the labor market experiences of Cubans in Miami. Am. J. Sociol. 295–319. Yinger, J., 1995. Closed Doors, Opportunities Lost: The Continuing Costs of Housing Discrimination. Russell Sage Foundation, New York. Zubrinsky, C.L., Bobo, L., 1996. Prismatic metropolis: race and residential segregation in the city of the angels. Soc. Sci. Res. 25 (4), 335–374.