Different ways to get to the same workplace: How does workplace location relate to commuting by different income groups?

Transport Policy 59 (2017) 106–115

Contents lists available at ScienceDirect

Transport Policy journal homepage: www.elsevier.com/locate/tranpol

Different ways to get to the same workplace: How does workplace location relate to commuting by different income groups? Lingqian Hu *, Robert J. Schneider University of Wisconsin-Milwaukee, School of Architecture and Urban Planning, P.O. Box 413, Milwaukee, WI 53201, United States

A R T I C L E I N F O

A B S T R A C T

Keywords: Transportation inequality Job location Commute distance Commute mode Urban structure

We examine whether commonly-observed differences in commute behavior among different income groups are associated with the location of their workplaces. Using the Chicago metropolitan area as a case study, we classify six types of workplace locations to reflect the degree of employment centralization versus decentralization and the degree of employment clustering versus dispersion. Based on the 2008 Chicago Regional Household Travel Inventory, we found that low-income workers are more likely to work in centralized but dispersed workplaces, while high-income workers are more likely to work in employment clusters. The unequal distribution of workers in different workplaces, combined with distinctive commuting patterns to certain workplaces, partly explains commonly-observed commute differences, such as shorter-distance commutes and more public transit use by lower-income workers. Regression analysis shows that the association between income and commute mode varies by workplace, and, more importantly, commute mode has a greater association with workplace locations than with income. The results suggest considering workplace locations in empirical research on commuting inequalities and when establishing transportation and housing policies.

1. Introduction Higher- and lower-income groups tend to travel to work differently in the United States: lower-income workers on average have shorter commute distances but longer commute times, and they are more likely to rely on non-automobile modes than higher-income workers (Renne and Bennett, 2014). Many factors contribute to these commuting differences, such as low-income workers’ limited budgets and low automobile ownership rates. However, a gap remains in our understanding of commute disparities: to what extent are the locations of workplaces associated with the commonly-observed differences in commuting between income groups? In answering this question, we seek to bridge literature on commuting inequality with literature on the relationship between urban spatial structure and commuting. Both streams of literature have a long history, but they have not been sufficiently connected. Investigating the association between workplaces and commutes is important because it provides a new perspective to address commuting inequalities. Policies such as discounted transit passes and other financial incentives aim to make transportation more affordable to low-income workers. Yet, if workers at various income levels are unevenly sorted into different workplace locations, baseline commute travel options for each income group may also be unequal. If this is true, then policies also * Corresponding author. E-mail addresses: [email protected] (L. Hu), [email protected] (R.J. Schneider). http://dx.doi.org/10.1016/j.tranpol.2017.07.009 Received 5 March 2017; Received in revised form 30 June 2017; Accepted 17 July 2017 0967-070X/© 2017 Elsevier Ltd. All rights reserved.

need to consider workplace-based strategies to meet the needs of lowincome workers. In fact, in the U.K., the Greater London Authority has noted this issue and published a report on the disconnection between workers at different qualification levels and their workplaces (Ennis et al., 2009). We use household travel survey data from the Chicago metropolitan area to answer three questions: How are the workplaces of different income groups distributed spatially? Do the commonly-observed differences in commute distance and commute mode between income groups vary by workplace location in a metropolitan area? And, to what extent are workplace locations associated with differences in commute modes among income groups? The next section reviews literature on how workplace locations are associated with commuting behavior of different income groups. We then describe the data and methodology, including the classification of six types of workplaces in the study area. The results section includes findings based on a descriptive analysis of how different income groups are distributed across the six types of workplaces, a descriptive analysis of commute distance and commute mode by income and by workplace, and binomial logit models that estimate the association between workplaces and commute modes of different income groups. The paper concludes with policy implications.

L. Hu, R.J. Schneider

Transport Policy 59 (2017) 106–115

2. Literature review

travel (Thompson and Matoff, 2003). Decentralization of population and employment, on the other hand, is regarded as a primary cause of the decline in transit mode share (Cervero, 1989; Schwanen et al., 2001). But in the suburbs, employment subcenters can potentially ensure the minimum employment size that supports a competitive public transit system (Susilo and Maat, 2007). Brown and Thompson (2008) found empirical support that transit services to employment subcenters can attract transit riders. Meanwhile, it is challenging to organize public transit to dispersed workplaces (Jaroszynski et al., 2017). This stream of literature identifies differences in commuting patterns to various types of workplaces, but it rarely connects workplace locations to commuting differences among population groups. For example, workers employed in the CBD tend to have long commutes, but do all CBD workers at different income levels have similarly long commutes? This is a question that we aim to answer.

2.1. Workplaces and commutes Economic functions of various workplaces in a metropolitan area naturally vary. Urban economic theory suggests that the intra-regional urban spatial structure reflects rational decisions made by various people and organizations competing for a limited supply of land (Von Thunen, 1826). In a conventional monocentric metropolitan area where jobs concentrate in the Central Business District (CBD), land users who value agglomeration economies are willing to pay high land rent to locate in the center (Alonso, 1964), particularly high-income, high-skill jobs. The high land rent encourages density. In recent decades, the monocentric urban structure dissolved in many metropolitan areas as jobs suburbanized and employment clusters emerged in the suburbs. Yet, a familiar pattern emerged around these clusters: compared with surrounding low-density suburban areas, suburban employment clusters similarly attract jobs that value agglomeration economies (Anas et al., 1998). Therefore, the contemporary intra-regional urban structure can be analyzed on two related but different dimensions (Anas et al., 1998): 1) centralization versus decentralization; and 2) clustering versus dispersion. The spatial distribution of employment opportunities can be classified in these two dimensions. The first measurement indicates the degree to which jobs are concentrated in a CBD or a central city versus suburban communities. In the U.S., decentralization has been the common trend for several decades. The second dimension denotes whether jobs are clustered in subcenters or dispersed in a low-density, relatively uniform fashion. The degree of clustering in U.S. metropolitan areas is debatable. Gordon and Richardson (1996), Lang and LeFurgy (2003), and Lee (2007) suggest that dispersion is more common than clustering, but much research still emphasizes and analyzes employment subcenters (Giuliano and Small, 1991; McDonald and Prather, 1994; Cervero and Wu, 1997; McMillen and Smith, 2003; Giuliano et al., 2007), because they affect labor markets and commuting patterns. Commuting costs are an essential factor in shaping the urban spatial form of a metropolitan area. Urban economic theory suggests that when deciding where to live, households aim to maximize utility, making tradeoffs between commuting costs, land rents, and other housing amenities (Alonso, 1964; Muth, 1969; Mills, 1970); firms aim to maximize profits, and one of the strategies is to reduce transportation costs to potential workers (Marshall, 1920; Mills, 1972). As employment locations shift from the CBD to the suburbs, theoretical models have been developed to understand commuting to multiple employment centers (Timothy and Wheaton, 2001; Wheaton, 2004). There is no consensus on the extent to which commutes to decentralized and centralized workplaces differ. The main reason is that workers could adjust their job and housing locations and eventually stabilize commute length or duration (Gordon et al., 1991; Clark and Kuijpers-Linde, 1994; Levinson, 1997; Crane and Chatman, 2004). Additionally, the existence of employment subcenters in decentralized places complicates commuting patterns (Wang, 2000). Empirical research found that commute trips to the largest employment clusters—CBDs—tend to be longer than those to employment subcenters (Cervero and Wu, 1997; Sultana, 2002), and longer to employment subcenters than to dispersed workplaces (Giuliano and Small, 1991; Manaugh et al., 2010). Even without identifying employment centers, studies found that higher density at workplaces tend to be associated with longer commute trips (Cervero, 2002; Zhang, 2004; Chen et al., 2008) and lower share of automobile commutes (Chatman, 2003). Giuliano and Small (1993) contended that employment clusters need to draw workers from larger areas, thus requiring average longer commute trips than dispersed workplaces. Transportation services available for different workplaces vary and consequently affect commutes. Transit systems tend to serve CBD-bound

2.2. Workplaces and commute inequalities Although sufficient research has emphasized the connection between workplaces and commutes, the research that explores commuting differences among population groups tends to focus on residential locations or the simple (dis)connection between residences and workplaces, but rarely considers inherent commuting patterns associated with distinct workplace locations, suggested by the literature reviewed above. Shifting the emphasis to workplaces requires a readjustment of perspective (Shearmur, 2006). Economic, social, and other factors significantly affect commuting behavior of different population groups (Hanson and Pratt, 1988). Urban economic theory suggests that household location decisions are made based on the tradeoffs between commuting costs, land rents, and other housing amenities (Alonso, 1964; Muth, 1969; Mills, 1970). With limited budget, low-income workers are more sensitive to the tradeoff between job proximity and housing price (Adair et al., 2000), while high-income workers can put more weight on housing amenities. Assuming exogenously-given workplaces, Pinjari et al. (2011) developed a simulation model which finds that lower-income workers prefer housing closer to their workplaces. The second reason that low-income workers have different commuting patterns is associated with their low transportation mobility. For example, in the U.S., lower-income workers on average have shorter commute distances but longer commute time (Renne and Bennett, 2014), mainly because they rely on slow and inefficient public transit and nonautomobile travel modes (Taylor and Ong, 1995). The lack of automobile mobility limits lower-income workers’ job search ranges (Blumenberg and Ong, 2001), and they face more challenges to look for and acquire jobs distant from their residences. Glaeser, Kahn, and Rappaport (2008) recognized public transit services as a major reason that the poor concentrate in central cities. In a qualitative study, Boschmann (2011) even suggested that residential choices of working poor are made based on their mobility options but not their workplace locations. Third, lower-income households face more residential location constraints, including social and institutional barriers, in addition to housing unaffordability. Kain’s (1968) Spatial Mismatch Hypothesis (SMH) provides a conceptual framework to connect residential segregation with commuting differences among population groups: African Americans in the U.S. tend to be constrained in the inner cities and thus need to endure long commutes to reach decentralized jobs. The SMH was expanded to study economically disadvantaged groups, including welfare recipients (Ong and Blumenberg, 1998; Blumenberg and Manville, 2004) and the poor (Covington, 2009; Hu, 2014). Much SMH literature focuses on residential neighborhood characteristics (Jencks and Mayer, 1990; Preston and McLafferty, 2016) or residence-based job accessibility (Ihlanfeldt and Sjoquist, 1991; Shen, 1998; Cervero et al., 2002; Kawabata and Shen, 2006; 2007; Grengs, 2010; Hu, 2015). Providing circumstantial support to the SMH, Horner and Mefford (2007) found a narrower commute range for minorites than for whites, which they 107

L. Hu, R.J. Schneider

Transport Policy 59 (2017) 106–115

area has 1993 census tracts in eight counties. The census tracts are our basic geographic unit. Year 2010 employment data and travel time matrices came from the Chicago Metropolitan Agency for Planning (CMAP). Individual socioeconomic and travel data came come from the 2008 Chicago Regional Household Travel Inventory (CRHTI). This is the most recent travel survey data available for the region. Data from the American Community Survey (ACS) show only minor commute mode shifts in the region since then. Comparing the ACS 2005–2009 data with 2010–2014 data, the share of workers who commuted by private automobile decreased from 78.8% to 77.9%, bicycle increased from 0.55% to 0.67%, walk increased from 3.12% to 3.31%, and public transit remained at 12.5%. The region has experienced significant changes in its urban spatial structure. In the early 20th century, it was a typical monocentric region with a strong central city. After that, jobs have suburbanized. Based on the CMAP employment data, in 2010 the city of Chicago had 31.8% of the total regional jobs, 36.5% of service jobs—including those in FIRE (Finance, Insurance, and Real Estate) and other services, but only 20.3% of retail jobs and 18.2% of manufacturing jobs. Many studies have identified multiple employment subcenters in the region (McDonald and Prather, 1994; McMillen and McDonald, 1998). The spatial distribution of population groups also has changed. Based on the decennial Census, between 1990 and 2010, the share of the total regional population in the city of Chicago declined from 36.3% to 31.7%, and the share of the poor population declined from 70.9% to 53.6%. Such poverty deconcentration could be due partly to the region's housing policies. The Gautreaux project and later the Housing Choice Voucher (HCV) program have moved disadvantaged groups from the inner city to suburban neighborhoods. Still, racial and class segregation persists. A significant share, 78%, of HCV households stayed within Cook County, and they tended to be clustered in certain neighborhoods (Holloway, 2014).

suggested indicates less spatially diverse choices in minorities' housing locations. Although the SMH literature generally considers employment locations, the treatment of workplaces tends to be either based on simiple indicators of their locations in the inner city and the suburbs or based on their spatial proximity to residences, without further insights into spatial and transportation characteristics of workplaces in contemporary metropolitan areas. Therefore, we suggest workplace locations as the fourth factor that influences commuting inequalities. The connection is intuitive. On one hand, empirical research finds that workers at various income levels tend to be sorted into different workplace locations. Employment (sub)centers provide more job opportunities for middle-to-high income workers than for low-income workers. Shearmur and Coffey (2002) showed that highorder services and financial services tend to be strongly overrepresented in the CBD and some employment subcenters. Cervero et al. (2010) found that employment subcenters in the San Francisco Bay Area employ workers in certain middle-to-high income occupations. In other words, job opportunities for low-income workers are less likely to be clustered; rather, they tend to be dispersed (Modarres, 2011). On the other hand, existing research reviewed in the previous subsection has established that clustered workplaces are more likely to induce longer commutes and are better served by transit than dispersed workplaces. The varying baseline commuting patterns to different workplace locations, combined with the unequal distribution of income groups in different workplaces, could be associated with the differences in commutes between income groups, but we do not fully understand how workplace locations relate to the differences. Particularly, we want to answer the question: to what extent are workplace locations associated with differences in commute patterns among income groups? Of course, the constraints for low-income workers intertwine: those without automobiles cannot search for or maintain distant jobs that are not supported by good transit and therefore they tend to have short commute distances; residential segregation and unaffordable housing prices prevent low-income workers from living close to their workplaces or good transit services. Although this research does not address the joint decisions of residential location, employment location, and automobile ownership, the above review provides important background knowledge for this study. A few studies directly examine the connection between workplace locations and commuting inequality among population groups, such as male and female workers (e.g. Sang et al., 2011) and workers in different occupations (O'Kelly and Lee, 2005). Modarres (2011) visually inspected the relationship between employment centers based on zip code-level data and the average commute time in Public Use Microdata Areas (PUMA) in Southern California. The comparison, although innovative, provides limited insights for neighborhood-level commuting behavior as PUMAs are much larger than typical neighborhoods. To our best knowledge, Schleith and Horner (2014) is the only study that connects employment clusters with commutes of different income groups. They found that low-income commuters to job clusters in Leon County, FL have shorter commute distances than other income groups, but low-income commuters experienced greater increases in commute distance between 2006 and 2011. Nevertheless, Schleith and Horner did not intend to compare commutes to employment clusters with those to dispersed workplaces. In summary, the literature on commutes and urban structure rarely differentiates population groups, while the literature on commute inequality between population groups tends to use rudimentary workplace indicators—if used—that cannot sufficiently reflect the complicated urban spatial structure. This research aims to fill in the gap by examining the connection between workplace locations and differences in commutes across income groups.

3.1. Income groups We categorize the workers surveyed by the CRHTI into four groups based on their household income. Table 1 gives the descriptive statistics of the four groups. The CHRTI data include fewer workers in the two lower income groups than in the higher income groups. Commute distance is estimated as the Euclidian distance between the centroids of home and workplace census tracts, which were reported in the CRHTI. This distance measure reflects job-housing separation. We do not use network distance, which varies by travel mode. As expected, lower income workers have shorter average commute distances and are less likely to commute by automobile but more likely by transit. One exception is the high-income workers; they are more likely to use transit than higher-

Table 1 Descriptive statistics of workers by income group. Income Group (US$1000)

Workers Number of Workers Commute distance (Miles) % Automobile Mode % Transit Mode % Other Modes Households Number of Households % Rental % in Chicago % in Cook outside of Chicago % in DuPage and Lake % in Other Counties

3. Data and methodology We use the Chicago metropolitan area for our case study. The study 108

1

2

3

4

Low-income Lower-Middle Higher-Middle High-Income <20

20–49.9

50–99.9

>¼100

369 6.2 58.5 31.0 10.5

1663 7.0 72.1 19.3 8.6

4226 9.0 79.9 14.9 5.2

4301 9.6 77.7 17.2 5.1

306 59.8 52.3 25.2

1286 36.2 40.9 31.3

2748 11.6 27.6 36.5

2471 4.3 25.7 33.4

10.5 12.1

14.8 13.1

18.6 17.3

24.1 16.8

L. Hu, R.J. Schneider

Transport Policy 59 (2017) 106–115

middle income workers, also consistent with findings in empirical studies (AASHTO, 2013). Households with lower-income workers are more likely to live in rental units and in the city of Chicago but less likely in the suburbs.

are located in the city of Chicago and the inner-ring suburbs, including those commonly identified in research that used various methods (e.g. McDonald and Prather, 1994; McMillen and Smith, 2003). Table 2 quantifies the share of regional jobs by workplace location based on the two spatial dimensions. Job decentralization is significant. The city of Chicago had 31.8% of total regional employment, including 18.5% in the CBD, while the inner-ring suburbs contained 54.7% of total jobs. Clustering of jobs is also evident. In total, the CBD and the subcenters had almost half of the regional employment, but they cover just 4.3% of the total area. Because the employment subcenters in the outerring suburbs had only 1.1% of the total jobs, we treat the outer-ring suburbs as homogeneous, without differentiating subcenters and noncenter areas. As a result, we identify six types of workplaces in the

3.2. Workplace classification We classify workplace locations by two related but different dimensions (Anas et al., 1998): 1) whether the workplace is in a centralized location and 2) whether the workplace is in an employment cluster. The workplace is considered to be in a centralized location if it is in the city of Chicago and in a cluster if it is in the CBD or an employment subcenter. To identify employment centers, we apply Giuliano and Small (1991) method, which uses employment density and total employment benchmarks. Using the absolute benchmarks allows us to conduct crosssectional analysis in a large metropolitan area. We identify the CBD as the contiguous census tracts with employment density greater than ten jobs per acre in and around Loop, the elevated rail that circles Downtown Chicago. To identify employment subcenters, we use the 5-5 benchmarks—contiguous census tracts with density greater than five jobs per acre and total employment greater than 5000. This definition yields 51 employment subcenters. Fig. 1 shows that most employment subcenters

Table 2 Percentage share of total employment by workplace. (%)

Chicago

Inner-ring

Outer-ring

Total

CBD Subcenter Non-center Total

18.5 7.7 5.7 31.8

22.4 32.2 54.7

1.1 12.4 13.5

18.5 31.2 50.4 100.0

Fig. 1. Workplace location. 109

L. Hu, R.J. Schneider

Transport Policy 59 (2017) 106–115

study area: Chicago CBD, Chicago subcenters, Chicago non-center areas, inner-ring subcenters, inner-ring non-center areas, and outerring suburbs. Transportation systems serve the city of Chicago better than the suburbs and the employment centers better than non-center areas. Table 3 shows that the CBD is best served by transit: trips from 97% of the census tracts in the study area can use transit to get to the CBD. The service coverage is smaller for the suburbs than for the city of Chicago and smaller for the non-center areas than for the subcenters. Disparities also exist in travel time. Because of the CBD's central location, the average peak-hour travel time, either by automobile or transit, from all census tracts to the CBD is shorter than to the other workplaces, and the time increases as the distance between the workplaces and the CBD increases. Interestingly, private automobile travel time is longer than transit in-vehicle travel time1 for trips going into the CBD and employment subcenters in Chicago, indicated by the ratio of automobile travel time to in-vehicle transit travel time. This is reasonable considering the peak-hour congestion in the Chicago CBD and employment clusters. The relative competitiveness of transit declines in the suburbs and in the noncenter areas.

Table 3 Transit serivce coverage and average travel time by workplace. Workplace

% census tracts have transit service to the workplacea Auto Timeb (Minutes) Transit Timeb (Minutes) Auto-Transit Time Ratioc

Vm;i ¼ αm;i þ βm;i Tm;i þ λi WKi

(2)

Outer-ring Suburbs

NonCenter

97.1 96.7

96.0

78.9

67.2

38.1

56.2 57.9 45.0 53.3 1.22 1.07

56.9 56.7 0.97

63.5 68.0 0.96

66.2 76.1 0.90

74.7 99.5 0.77

include the population density of census tracts at both workplace and residence. We also include the distance to the closest METRA station and the bus route density in the residence census tract to capture accessibility to transit services. Socioeconomic characteristics include gender, age, the number of vehicles per worker, and the number of students in their households. Additionally, WKi also includes the income level, the workplace location, and the interaction terms of income and workplace dummy variables, which are the key variables that we want to test. The models do not suffer from multicollinearity based on Pearson correlation coefficients and Variance Inflation Factor.3 Based on the utility maximization theory, we use Ben-Akiva and Lerman, 1985 binomial logit model to estimate the probability of an individual worker i choosing automobile mode (m ¼ auto):

We use binomial logit models to investigate the association between workplaces and commute modes for different income groups. The samples are limited to automobile and transit commuters to maintain sufficient sample sizes for both modes. The utility of automobile and transit modes is estimated in the following ways:

(1)

Inner-ring Suburbs

a Transit service indicates whether public transit connects the census tracts to the workplace, based on whether transit network (in-vehicle) travel time is available in the CMAP dataset. b Automobile and transit travel time represent in-vehicle travel time. We did not include access and waiting times for transit because the data are missing for many origindestination pairs. c Auto-Transit Time ratio is the ratio of automobile travel time to transit travel time from the same origin to the same destination.

3.3. Multivariate analysis of automobile commutes

Um;i ¼ Vm;i þ εm;i

Chicago

CBD SubNonSubCenter Center Center

Prðyi ¼ autoÞ ¼

where Um,i is the utility of commute mode m (m ¼ auto when the worker commutes by automobile, either as a driver or a passenger, and m ¼ transit when the worker commutes by transit, either bus or rail) for worker i. Vm,i is a vector of deterministic components of Um,i, and εm,i is a vector of unobserved errors. In Equation (2), αm, i is the intercept, βm, i and λi are vectors of coefficients. Tm, i is a vector of travel cost associated with mode m for each individual worker i. We use commute distance to reflect the spatial separation between jobs and residences and the ratio of automobile travel time to transit travel time to represent the competitiveness of automobile travel relative to transit. Because 25% of the potential origindestination pairs are not served by transit based on the CMAP data, we assume unrealistically long transit travel time—9999 min, so that the automobile to transit time ratios are not infinity. We did not include outof-pocket costs, such as gas, parking, and transit fares.2 WKi is a vector of built environment and socioeconomic variables associated with individual worker i. The built environment variables

expðVauto;i Þ expðVauto;i Þ þ expðVtran;i Þ

(3)

4. Results This section provides answers to the three research questions. 4.1. Distribution of income groups by workplace location We use descriptive statistics to examine the first research question: How are the workplaces of different income groups distributed spatially? Table 4 shows the distribution of the four income groups by workplace location. Low-income workers are more likely to work in centralized but dispersed places, while high-income workers are more likely to work in clustered workplaces of the CBD and employment subcenters. Table 4 shows that 45.9% of low-income workers are employed in the centralized workplace—the city of Chicago—and the share is the highest among all income groups. Nevertheless, only 11.4% of low-income workers are employed in the CBD, while the share of CBD workers in the high-income group is more than doubled, 24.5%. This gap in the distribution of workers in and outside of job clusters also exists in the suburbs. Only 6.6% of low-income workers work in the outer-ring suburbs, lower than the 10–13% share for the other income groups. These observations confirm the findings of existing research: low-income workers tend to work in dispersed workplaces. But, the findings also underscore that lowincome workers are more likely to work in centralized workplaces than the other income groups. This result is different from earlier literature that hypothesized suburbanizing job locations for disadvantaged groups (Kain, 1968; Ihlanfeldt, 1994).

1 Because of a data limitation, we use in-vehicle transit travel time. This introduces two possible biases. First, total transit travel time, which includes access and waiting time, is longer than reported in-vehicle travel time. Second, access and wait time is expected to be shorter in the city center where transit stops/stations are denser and transit services are more frequent than in the suburbs. Therefore, the gap between the total transit travel time and the in-vehicle time tend to be larger in the suburbs. To address these biases, we control for distance to METRA stations and bus route density in the regression analysis. 2 In the CHRTI data, only 65% of workers who commute by automobile gave valid parking information, among them 96% workers either did not pay or had their parking paid by employers. We tested the model using the subset with valid parking information. The sample size was reduced by more than a third, and the variable—whether the worker paid for parking or not—is insignificant. We also evaluated the feasibility to include transit fare. The correlation between transit fare and transit time is 0.71, based on the CMAP data. For the 25% of workers commuted between places without transit services, we assumed the transit fare to be US$99.99. The model results show that transit fare is insignificant. Therefore, we omitted the fare variable since it provides little additional information.

3 Descriptive statistics and multicollinearity diagnostics can be provided to interested readers.

110

L. Hu, R.J. Schneider

Transport Policy 59 (2017) 106–115

Table 4 Percentage distribution of workplace locations by income group. Income Group (US$1000)

1 2 3 4

Low Lower-Middle Higher-Middle High

Chicago

<20 20–49.9 50–99.9 >¼100

Inner-ring Suburbs

Outer-ring Suburbs

Total

CBD

Sub-center

Non-Center

Total

Sub-center

Non-Center

45.9 38.4 34.3 37.2

11.4 16.2 18.7 24.6

11.1 6.9 5.9 5.5

23.4 15.4 9.8 7.1

47.4 48.2 53.5 52.6

10.2 14.2 15.7 17.3

37.2 34.0 37.8 35.3

6.6 13.3 12.2 10.2

distance of low-income workers in the whole region is shorter. The general commuting patterns to different workplaces and the uneven distribution of income groups in those workplaces jointly contribute to the commonly-observed commute distance gaps between income groups. Fig. 3 shows the commute mode share by income and by workplace location. Differences in commute mode among income groups vary significantly within the study area. There are four noteworthy observations. First, lower-income workers who are employed in the CBD have similar commute mode splits as higher-income CBD workers; the transit share ranges between 59.0% and 63.2%. However, high-income workers prefer rail transit: 88% of the high-income transit commuters to the CBD use rail, while the share declines to 54% for the low-income transit commuters. One reason might be that rail transit is expensive for lowincome workers. The one-way commuter rail fare ranged between $2.15 and $8.05 in 2009 (METRA, 2010), higher than the one-way bus fare of $1.75 (Chicago Transit Authority, 2008). It is also possible that low-income workers travel to work outside the peak hours, especially those holding night shift jobs, and therefore they rely on night bus services when rail transit is unavailable. Second, workers in different income groups also have similar commute mode shares to the outer-ring suburbs: most—about 96-97%— of them rely on automobile. Nevertheless, 18% of low-income automobile commuters are passengers, while less than 1% of high-income automobile commuters are passengers. Carpooling is an important travel mode for low-income workers where transit services are limited. Third, like the commute distance gap, the largest commute mode gap occurs in the Chicago non-center areas. Specifically, 48.7% of lowincome workers who are employed in these areas commute by transit, while the share is only 8.1% for high-income workers. Fourth, transit mode shares to Chicago, including the centers and non-center areas, are higher than to the suburbs, regardless of workers’ income levels. The previous section shows that low-income workers tend to be employed in Chicago but not the suburbs. Again, the uneven distribution of income groups and the variation in commute modes to different workplaces together contribute to the average higher transit mode share of low-income workers.

4.2. Commute patterns by workplace location We also use descriptive statistics to answer the second research question: do the differences in commute distance and commute mode between income groups vary across workplaces? Fig. 2 shows the commute distance distribution by workplace location and by income. As expected, lower-income workers tend to have shorter commute distances than higher-income workers. Nevertheless, the extent of the difference varies greatly across workplaces. The distance gap is especially large in the Chicago non-center areas but not as much as in the CBD or the outer-ring suburbs. Specifically, 75.6% of low-income workers who work in Chicago non-center areas commute less than five miles, more than twice the share—35.9%—of high-income workers. We hypothesize two reasons of such short commutes: low-income workers in Chicago non-center areas may have many nearby job opportunities in their neighborhoods or low-income workers employed in these areas may easily find housing near jobs. Meanwhile, as expected, CBD jobs induce long commutes for workers at all income levels. But, the difference in the share of short-distance commuters between high- and low-income CBD workers is very small: 23.7% for low-income CBD workers and 21.7% for high-income CBD workers. Across all workplaces, commute distance for workers employed in the CBD and employment subcenters tend to be longer than that for workers in dispersed non-center areas, regardless of their income. The difference is more prominent in the city of Chicago than in the suburbs. There are two possible reasons. First, land rent is high in and around the CBD and employment subcenters. Jobs that value agglomeration economy price out housing, and therefore many workers cannot afford to live close to job centers (Sullivan, 1986). Second, employment centers draw workers from greater distances than dispersed workplaces, and therefore a larger share of workers need to travel long distance (Giuliano and Small, 1991). The previous section shows that low-income workers are less likely to work in the CBD and employment subcenters than other income groups. This section underscores that the average commute distance for workers employed in the CBD and subcenters is longer than that for workers in non-center areas, regardless of income. As a result, the average commute

Inner Suburbs

Chicago 100%

CBD

Subcenter

Non-Center

1 2 3 4

1 2 3 4

Outer Suburbs

Subcenter Non-Center N

80% 60% 40% 20% 0%

Income 1 2 3 4 Group

Walk Fig. 2. Distribution of commute distance by income group and by workplace.

Bike

Auto Driver

1 2 3 4

Auto Passenger

1 2 3 4 Bus

Rail

1 2 3 4 Other

Fig. 3. Commute mode share by income group and by workplace. 111

L. Hu, R.J. Schneider

Transport Policy 59 (2017) 106–115

In Model 1, income does not affect the likelihood of automobile commuting. This result is different from the literature that finds the significance of income on travel mode (Schwanen and Mokhtarian, 2005), but it is consistent with the literature that suggests income effects on car ownership but not on the propensity to use automobile (Chen et al., 2008). Specifically, when we remove the automobile ownership variable, the low-income variable becomes significant. There is no multicollinearity between the low-income and the automobile ownership variables: the correlation coefficient is 0.147, and the VIF scores for both variables are below 1.2. Therefore, we infer that automobile ownership, rather than income, has a more direct relationship with automobile commuting. In Model 2, which includes the workplace dummy variables with the CBD as the reference, both income dummy variables are significant. Lowincome workers are less likely and high-income workers are more likely to use automobiles than middle-income workers. The workplace variables indicate that, compared with the CBD, all the other workplaces are significantly associated with higher automobile mode shares, and the association increases as the workplaces become more decentralized or more dispersed. When the automobile ownership variable is removed, the significance levels of both income variables increase, again suggesting the direct association between automobile ownership and commute mode. The differences in the income effects between Model 1 and Model 2 suggest that without considering workplace location, the income effects on commute mode could be masked. As explained in the previous section, the large share of high-income workers employed in the CBD and the small share of automobile commutes to the CBD combined lower the average automobile mode share for all high-income workers, and consequently their automobile mode share is not statistically different

4.3. Association between income and commute mode by workplace We use binomial logit regression models to answer the third research question: to what extent are workplace locations associated with the commute modes of different income groups? Table 5 shows the model results in a step-wise fashion: In addition to the control variables, Model 1 includes only the income dummy variables, Model 2 includes the income and workplace dummy variables, and finally Model 3 includes the income and workplace interaction terms. For modeling purposes, we collapsed income groups 1 and 2 to one group because the share of the original income group 1 in our sample is small (approximately 3.5% of all samples). This gave us three income groups, and we treated the middle-income group (the original income group 3) as the reference group. Across the three models, all the control variables have expected signs. Long commute distances may discourage automobile commuting: Chicago workers tend to rely on transit, particularly rail transit, for long commutes. A high ratio of automobile to transit travel time also discourages automobile commutes. Workers who live or work in highdensity places are less likely to use automobile, while those who are female, older, living in households with more cars per worker or with more children in school are more likely to commute by automobile. Note that the coefficient of workplace population density is greater than that of residence population density, suggesting the importance of the workplace built environment on commute mode choice. In particular, lower levels of automobile commuting could also be explained by high parking costs and limited parking supply, though we were unable to test this variable. Better access to transit, indicated by shorter distance to the closest METRA station and denser bus routes, is also associated with a lower likelihood to commute by automobile.

Table 5 Estimating automobile commute mode. With Income

With Workplace

Workplace and Income interaction

Coeff

p-value

Coeff

p-value

Coeff

p-value

Commute Distance Auto-Transit Time Ratio Pop Density at Workplace Pop Density at Residence House Distance to METRA (Mile) Bus Route Density in House Tract Female AGE # of Student # of Vehicle per Worker

0.070 1.052 0.044 0.019 0.228 0.014 0.194 0.010 0.097 1.589

<.0001 <.0001 <.0001 <.0001 <.0001 0.002 0.003 <.0001 0.004 <.0001

0.053 0.604 0.016 0.012 0.195 0.002 0.128 0.013 0.091 1.583

<.0001 <.0001 <.0001 <.0001 <.0001 0.656 0.085 <.0001 0.015 <.0001

0.053 0.600 0.017 0.012 0.191 0.002 0.106 0.013 0.087 1.562

<.0001 <.0001 <.0001 <.0001 <.0001 0.641 0.155 <.0001 0.022 <.0001

Low Income High Income

0.140 0.018

0.117 0.808

0.538 0.148

<.0001 0.078

0.161 0.001

0.345 0.993

1.950 2.241 3.211 2.998 4.174

<.0001 <.0001 <.0001 <.0001 <.0001

1.791 2.348 3.064 3.114 4.652

<.0001 <.0001 <.0001 <.0001 <.0001

0.368 0.434 1.059 0.434 0.803 0.894 1.122 0.266 1.503 0.388

0.255 0.152 0.000 0.154 0.024 0.024 <.0001 0.286 0.096 0.703

1.276

<.0001

Chicago Center Chicago Non-Center Inner-ring Center Inner-ring Non-Center Outer-ring Suburbs Chicago Center*Low Inc Chicago Center*High Inc Chicago Non-Center*Low Inc Chicago Non-Center*High Inc Inner-ring Center*Low Inc Inner-ring Center*High Inc Inner-ring Non-Center*Low Inc Inner-ring Non-Center*High Inc Outer-ring Sub*Low Inc Outer-ring Sub*High Inc Intercept

1.599

AIC SC 2 Log L Pseudo R2 Max-rescaled Pseudo R2

6276 6369 6250 0.269 0.437

<.0001

1.334 4981 5110 4945 0.361 0.586

112

<.0001

4954 5156 4898 0.364 0.591

L. Hu, R.J. Schneider

Transport Policy 59 (2017) 106–115

commuting differences are directly associated with the uneven workplace distribution of each income group as well as the distinct commute patterns to different workplaces. Third, the association between income and commute mode varies by workplace location. Lower income is generally associated with a lower share of automobile commutes, except for commuters to the CBD and Chicago subcenters. More importantly, the association between income and commute mode is relatively weaker than the association between workplace location and commute mode, resulting in higher automobile use among low-income workers who are employed in non-CBD workplaces than among middle-income workers who are employed in the CBD. The variation in the relative efficiency of transit services could partly explain how workplace locations are associated with commute modes of different income groups. The Chicago CBD is well connected to the rest of the region through public transit, while the transportation cost to the CBD by automobile is high. Table 3 shows that automobile travel to the CBD takes longer than transit in-vehicle travel time during the peak hours, which may deter private automobile usage, even for high-income workers. On the other hand, the outer-ring suburbs have very slow transit services, if available, and thus almost require private automobile travel. Meanwhile, jobs in the Chicago non-CBD areas and the inner-ring suburbs can be reached by transit from most places in the study area, and transit travel time is not significantly longer than automobile travel time. Therefore, low-income workers, particularly those who have difficulty affording automobile travel can rely on transit to commute to these workplaces. These empirical results echo the long-standing concerns of conventional CBD-centered transit investment with respect to the distribution of benefits (Garrett and Taylor, 1999), although we recognize the compelling economic rationales for investing in places where there are robust transit markets. We detect workplaces where transit services are in high demand by low-income workers: Chicago non-center areas, and we suggest innovative transit service improvements in these areas. It is challenging to provide conventional fix-route transit to serve lowincome commuters employed in dispersed workplaces. It is often argued that paratransit could provide affordable and politically acceptable transit services (Mulley and Nelson, 2009; Campbell et al., 2015), particularly for workers who commute during off-peak hours to dispersed workplaces (Ambrosino et al., 2004; Finn, 2012). Nevertheless, successful paratransit systems that serves a wide range of riders in the U.S. are rare. Such paratransit services require better integrating different types of transit service and encouraging stakeholder collaboration (Davison et al., 2012). Our results from Chicago show that the integration and collaboration need to be coordinated across local jurisdiction boundaries in the metropolitan area. In fact, the Chicago Transit Authority and the Regional Transit Authority provide required ADA (Americans with Disabilities Act) paratransit services for the disabled and the elderly. These services can be expanded to serve low-income workers. Advancing technology can enhance transportation mobility for lowincome workers. Shared-mobility, including ride-sharing and carsharing, has experienced significant growth (Chan and Shaheen, 2012; Kodransky and Lewenstein, 2014). Using simulation models, Agatz et al. (2011) found that dynamic ride-sharing could work in sprawling areas. Although shared-mobility is still relatively unaffordable for low-income workers’ regular commute trips, the cost of shared-mobility could be significantly reduced with the advancement of autonomous vehicle technologies (Fagnant and Kockelman, 2016), because of the saving from labor costs. Another advantage of shared autonomous vehicles is that the operation and the revenue could be optimized at the system level, rather than at the individual vehicle level as in current Taxi or ridesharing services (Fagnant et al., 2015). Therefore, region-wide shared autonomous vehicle services can better serve commuters to dispersed workplaces if designed so. One concern is that lower-income workers might be slow to adopt new technology and therefore might not gain mobility benefits as soon as other income groups. For example, the share of adults who own smart phones is much lower in low-income households than in

from the average automobile share of middle-income workers. Results from Model 2 suggest the importance of considering workplace locations when testing the relationship between income and commuting. The results of Model 3 show the complicated joint relationship between income, workplace location, and commute mode. With the interaction terms, interpretation of the income dummy variables needs to be based on the workplace reference group. Both income dummy variables are insignificant, indicating that, all else being equal, income is not associated with the commute mode of CBD workers—the reference group. This finding is consistent with the descriptive statistics. Similarly, interpretation of the workplace dummy variables needs to be based on the income reference group. All workplace dummy variables are significant, indicating that middle-income workers—the income reference group—who commute to non-CBD workplaces are more likely to drive than middle-income workers who are employed in the CBD. The interaction terms in Model 3 indicate that the association between income and commute mode varies by workplace location. The interaction terms of Chicago subcenters and the two income dummy variables are insignificant, suggesting that income is not associated with the commute mode of Chicago subcenter workers. Meanwhile, the interaction term of Chicago non-center and low-income workers is significant, suggesting that, compared with the middle-income workers who commute to the Chicago non-center workplaces, low-income workers who commute to the same workplaces are less likely to use automobile (log odds-ratio of 1.059). Because the workplace dummy variable of “Chicago Non-Center” indicates that middle-income workers who commute to these workplaces are more likely to use automobile than the middle-income workers who commute to the CBD (with the log oddsratio of 2.348), we further infer that low-income workers who commute to Chicago non-center workplaces, despite being less likely to drive than middle-income workers travelling to the same workplace locations, are more likely to use automobile than the middle-income workers who commute to the CBD, with the log odds-ratio of 1.289 (¼2.348–1.059). By the same token, low-income workers who commute to the inner-ring suburbs—including the subcenters and non-center areas—and the outer-ring suburbs are less likely to use automobile than the middle-income workers who commute to these respective workplaces. Nevertheless, combining the effects of income and workplace location, we find that low-income workers who are employed in the suburbs are still more likely to use automobile than the middle-income workers who commute to the CBD. In other words, the association between workplace location and commute mode is greater than that between income and commute mode. This is also shown by the larger magnitude of the workplace dummy variables than the respective interaction terms. Further, high-income workers who commute to employment subcenters in the inner-ring suburbs are more likely to use automobile than the middle-income workers who commute to the same workplaces, but these two income groups are similarly likely to commute by automobile to the other workplaces. 4.4. Conclusion and policy implications This research investigates the relationship between workplace locations and differences in commute behavior among income groups in the Chicago metropolitan region. It investigates commuting inequalities from a different perspective, emphasizing the need to differentiate workplace locations in contemporary metropolitan areas. We found answers to our three research questions. First, workplace locations vary by income. Lower-income workers are more likely to work in non-center areas (dispersed locations) in the city of Chicago (a centralized location), while higher-income workers are more likely to work in the CBD and employment subcenters (clustered locations). Second, the size of the gaps in commute distance and commute mode between income groups vary across the study area. Workers in the CBD and the outer-ring suburbs have similar commuting behavior, while the gaps are especially large in Chicago non-center areas. The commonly observed 113

L. Hu, R.J. Schneider

Transport Policy 59 (2017) 106–115

high-income households (Smith, 2015). Bridging the technology and information gaps for low-income workers to fully enjoy shared mobility is important. Further, we found that three-quarters of low-income workers who work in Chicago non-center areas commute less than five miles (Fig. 2). Some of these people walk to work, underscoring the need to invest in well-maintained sidewalks and safe street crossings. In addition, many of these short commutes are currently being made by private automobile or bus (Fig. 3) but could be comfortably travelled by bicycle, especially if low-stress infrastructure (e.g., separated bicycle lanes, multi-use trails, neighborhood greenways) was available (Lusk et al., 2011; Minikel, 2012). As bike sharing services have extended to neighborhoods with more socioeconomically disadvantaged residents, these residents have bicycled more (Goodman and Cheshire, 2014). Still, it can be costly to increase bike share station network density in the suburbs. Dockless bike sharing systems (i.e., bikes with embedded GPS and no physical station hardware) (Fishman, 2016) could be a future solution to serve commuters to dispersed workplaces. Affordable housing policies should also consider workplace locations, while continuing to address housing segregation. This research shows that low-income workers are much more likely to work in Chicago noncenter areas and that they have particularly shorter commute distance in these areas, compared with other income groups. On one hand, persistent housing market segregation can limit low-income workers’ residential location options. Zax and Kain (1990) found that segregation can deter African Americans from moving further from their workplaces. Lowincome minorities might face similar location constraints and thus have to live in proximity to jobs in these places. On the other hand, we suspect that Chicago non-CBD areas could be preferred residential locations for low-income workers for two reasons. First, the affordable housing supply could be sufficient: the city of Chicago, mainly its nonCBD areas, had 69% of the total regional rental units with the monthly rent below US$700, based on the 2006–2010 ACS data. Second, transit service coverage may be relatively good in these areas for low-income workers. Therefore, it is important to ensure and expand housing opportunities in similar areas where job opportunities for low-income workers are abundant. In fact, the dispersed workplaces for lowincome workers offer plenty of options for conveniently-located affordable housing, some of which can be close to transit stops and stations and some of which can be within short distance of workplaces. This research is exploratory in nature by investigating an oftenneglected factor—workplace locations—in explaining commute inequalities. Further studies on commuting should consider the distribution of workers in different workplaces and commute options available for different workplace locations. Additionally, the spatial and socioeconomic structure of the Chicago metropolitan area is different from many other regions. Case studies of different metropolitan areas can improve our understanding on how workplace locations relate to commute differences between population groups.

Blumenberg, E., Manville, M., 2004. Beyond the spatial mismatch: welfare recipients and transportation policy. J. Plan. Lit. 19 (2), 182–205. Blumenberg, E., Ong, P., 2001. Cars, buses, and jobs: welfare participants and employment access in Los Angeles. Transp. Res. Rec. 1756, 22–31. Boschmann, E.E., 2011. Job access, location decision, and the working poor: a qualitative study in the Columbus, Ohio metropolitan area. Geoforum 42 (6), 671–682. Brown, J.R., Thompson, G.L., 2008. The relationship between transit ridership and urban decentralisation: insights from Atlanta. Urban Stud. 45 (5–6), 1119–1139. Campbell, M., Farwell, R., Weinberger, R., 2015. Stretching transportation anatomy: FlexBus services in central Florida. Transp. Res. Board 94th Annu. Meet. 15-5570. Cervero, R., 1989. America's Suburban Centers: the Land Use-transportation Link. Routledge. Cervero, R., 2002. Built environments and mode choice: toward a normative framework. Transp. Res. Part D 7, 265–284. Cervero, R., Komada, Y., Krueger, A., 2010. Suburban Transformations: from Employment Centers to Mixed-use Activity Centers. University of California Transportation Center. Cervero, R., Sandoval, O., Landis, J., 2002. Transportation as a stimulus of welfare-towork: private versus public mobility. J. Plan. Educ. Res. 22 (1), 50–63. Cervero, R., Wu, K.L., 1997. Polycentrism, commuting, and residential location in the San Francisco Bay area. Environ. Plan. A 29, 865–886. Chan, N.D., Shaheen, S.A., 2012. Ridesharing in north America: past, present, and future. Transp. Rev. 32 (1), 93–112. Chatman, D.G., 2003. How density and mixed uses at the workplace affect personal commercial travel and commute mode choice. Transp. Res. Rec. J. Transp. Res. Board 1830, 193–201. Chen, C., Gong, H., Paaswell, R., 2008. Role of the built environment on mode choice decisions: additional evidence on the impact of density. Transportation 35 (3), 285–299. Chicago Transit Authority, 2008. President's Report. Retrieved 3/12, 2016, from http:// www.transitchicago.com/assets/1/board_presentations/0810presidentsreport.txt. Clark, W.A., Kuijpers-Linde, M., 1994. Commuting in restructuring urban regions. Urban Stud. 31 (3), 465–483. Covington, K.L., 2009. Spatial mismatch of the poor: an explanation of recent declines in job isolation. J. Urban Aff. 31 (5), 559–587. Crane, R., Chatman, D.G., 2004. Traffic and sprawl: evidence from us commuting from 1985-1997. Plan. Mark. 6 (1), 14–22. Davison, L., Enoch, M., Ryley, T., Quddus, M., Wang, C., 2012. Identifying potential market niches for demand responsive transport. Res. Transp. Bus. Manag. 3, 50–61. Ennis, N., Theseira, M., McMullan, A., 2009. Commuting Patterns in London by Qualification Level and Employment Location. Greater London Authority, London, UK. Fagnant, D.J., Kockelman, K.M., 2016. Dynamic ride-sharing and fleet sizing for a system of shared autonomous vehicles in Austin, Texas. Transportation. http://dx.doi.org/ 10.1007/s11116-016-9729-z. Fagnant, D.J., Kockelman, K.M., Bansal, P., 2015. Operations of shared autonomous vehicle Fleet for Austin, Texas, market. Transp. Res. Rec. J. Transp. Res. Board (2536), 98–106. Finn, B., 2012. Towards large-scale flexible transport services: a practical perspective from the domain of paratransit. Res. Transp. Bus. Manag. 3, 39–49. Fishman, E., 2016. Bikeshare: a review of recent literature. Transp. Rev. 36 (1), 92–113. Garrett, M., Taylor, B., 1999. Reconsidering social equity in public transit. Berkeley Plan. J. 13 (1). Giuliano, G., Redfearn, C., Agarwal, A., Li, C., Zhuang, D., 2007. Employment concentrations in los angeles, 1980-2000. Environ. Plan. A 39 (12), 2935–2957. Giuliano, G., Small, K.A., 1991. Subcenters in the los angeles region. Reg. Sci. Urban Econ. 21 (2), 163–182. Giuliano, G., Small, K.A., 1993. Is the journey to work explained by urban structure? Urban Stud. 30 (9), 1485–1500. Glaeser, E.L., Kahn, M.E., Rappaport, J., 2008. Why do the poor live in cities? The role of public transportation. J. Urban Econ. 63 (1), 1–24. Goodman, A., Cheshire, J., 2014. Inequalities in the London bicycle sharing system revisited: impacts of extending the scheme to poorer areas but then doubling prices. J. Transp. Geogr. 41, 272–279. Gordon, P., Richardson, H.W., 1996. Beyond polycentricity: the dispersed metropolis, los angeles, 1970-1990. J. Am. Plan. Assoc. 62 (3). Gordon, P., Richardson, H.W., Jun, M.-J., 1991. The commuting paradox: evidence from the top twenty. J. Am. Plan. Assoc. 57 (4), 416–420. Grengs, J., 2010. Job accessibility and the modal mismatch in Detroit. J. Transp. Geogr. 18 (1), 42–54. Hanson, S., Pratt, G., 1988. Reconceptualizing the links between home and work in urban geography. Econ. Geogr. 64 (4), 299–321. Holloway, A.M., 2014. From the City to the Suburbs: Characteristics of Suburban Neighborhoods where Chicago Housing Choice Voucher Households Relocated. Urban Studies Research 2014. Horner, M.W., Mefford, J.N., 2007. Investigating urban spatial mismatch using jobhousing indicators to model home-work separation. Environ. Plan. A 39 (6), 1420–1440. Hu, L., 2014. Changing job access of the poor: effects of spatial and socioeconomic transformations in Chicago, 1990–2010. Urban Stud. 51 (4), 675–692. Hu, L., 2015. Job accessibility of the poor in los angeles: has suburbanization affected spatial mismatch? J. Am. Plan. Assoc. 81 (1), 30–45. Ihlanfeldt, K.R., 1994. The spatial mismatch between jobs and residential locations within urban areas. Cityscape 1 (1), 219–244. Ihlanfeldt, K.R., Sjoquist, D.L., 1991. The effect of job access on black and white youth employment: a cross-sectional analysis. Urban Stud. 28 (2), 255–265.

References AASHTO (American Association of State Highway and Transportation Officials), 2013. Commuting in America 2013: the National Report on Commuting Patterns and Trends. Adair, A., McGreal, S., Smyth, A., Cooper, J., Ryley, T., 2000. House prices and accessibility: the testing of relationships within the Belfast urban area. Hous. Stud. 15 (5), 699–716. Agatz, N., Erera, A.L., Savelsbergh, M.W., Wang, X., 2011. Dynamic ride-sharing: a simulation study in metro Atlanta. Procedia-Social Behav. Sci. 17, 532–550. Alonso, W., 1964. Location and Land Use. Toward a General Theory of Land Rent. Harvard Univ. Press, Cambridge, MA. Ambrosino, G., Nelson, J., Romanazzo, M., 2004. Demand Responsive Transport Services: towards the Flexible Mobility Agency, ENEA. Italian National Agency for New Technologies, Energy and the Environment. Anas, A., Arnott, R., Small, K.A., 1998. Urban spatial structure. J. Econ. Lit. 36 (3), 1426–1464. Ben-Akiva, M.E., Lerman, S.R., 1985. Discrete Choice Analysis: Theory and Application to Travel Demand. MIT Press, Cambridge, MA.

114

L. Hu, R.J. Schneider

Transport Policy 59 (2017) 106–115 Pinjari, A.R., Pendyala, R.M., Bhat, C.R., Waddell, P.A., 2011. Modeling the choice continuum: an integrated model of residential location, auto ownership, bicycle ownership, and commute tour mode choice decisions. Transportation 38, 933–958. Preston, V., McLafferty, S., 2016. Revisiting gender, race, and commuting in New York. Ann. Am. Assoc. Geogr. 106 (2), 300–310. Renne, J.L., Bennett, P., 2014. Socioeconomics of urban travel: evidence from the 2009 national household travel survey with implications for sustainability. World Transp. Policy & Pract. 20 (4). Sang, S., O'Kelly, M., Kwan, M.-P., 2011. Examining commuting patterns : results from a journey-to-work model disaggregated by gender and occupation. Urban Stud. 48 (5), 891–909. Schleith, D., Horner, M.W., 2014. Commuting, job clusters, and travel burdens. Transp. Res. Rec. 2452, 19–27. Schwanen, T., Dieleman, F.M., Dijst, M., 2001. Travel behaviour in Dutch monocentric and policentric urban systems. J. Transp. Geogr. 9, 173–186. Schwanen, T., Mokhtarian, P.L., 2005. What affects commute mode choice: neighborhood physical structure or preferences toward neighborhoods? J. Transp. Geogr. 13 (1), 83–99. Shearmur, R., 2006. Travel from home: an economic geography of commuting distances in montreal. Urban Geogr. 27 (4), 330–359. Shearmur, R., Coffey, W.J., 2002. Urban employment subcenters and sectoral clustering in montreal: complementary approaches to the study of urban form. Urban Geogr. 23 (2), 103–130. Shen, Q., 1998. Location characteristics of inner-city neighborhood and employment accessibility of low-wage workers. Environ. Plan. B 25, 345–365. Smith, A., 2015. U.S. Smartphone Use in 2015. Pew Research Center, Washington, DC. Sullivan, A.M., 1986. A general equilibrium model with agglomerative economies and decentralized employment. J. Urban Econ. 20 (1), 55–74. Sultana, S., 2002. Job/housing imbalance and commuting time in the Atlanta metropolitan area: exploration of causes of longer commuting time. Urban Geogr. 23 (8), 728–749. Susilo, Y.O., Maat, K., 2007. The influence of built environment to the trends in commuting journeys in The Netherlands. Transportation 34 (5), 589–609. 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. Thompson, G.L., Matoff, T.G., 2003. Keeping up with the Joneses: radial vs. multidestinational transit in decentralizing regions. J. Am. Plan. Assoc. 69 (3), 296–312. Timothy, D., Wheaton, W.C., 2001. Intra-urban wage variation, employment location, and commuting times. J. Urban Econ. 50, 338–366. Von Thunen, J.H., 1826. The Isolated State, Translated by Carla M. Pergamon Press, Wartenburg. New York. Wang, F., 2000. Modeling commuting patterns in Chicago in a GIS environment: a job accessibility perspective. Prof. Geogr. 52 (1), 120–133. Wheaton, W.C., 2004. Commuting, congestion and employment dispersal in cities with mixed land-use. J. Urban Econ. 55, 417–438. Zax, J., Kain, J.F., 1990. Race and commutes. J. Urban Econ. 28, 336–348. Zhang, M., 2004. The role of land use in travel mode choice. J. Am. Plan. Assoc. 70 (3), 344–360.

Jaroszynski, M., Brown, J., Bhattacharya, T., 2017. An examination of the relationship between urban decentralisation and transit decentralisation in a small-sized US metropolitan area. Urban Stud. 54 (6), 1500–1518. Jencks, C., Mayer, S.E., 1990. Residential Segregation, Job Proximity, and Black Job Opportunities. Inner-city poverty in the United States, pp. 187–222. Kain, J.F., 1968. Housing segregation, negro employment, and metropolitan decentralization. Q. J. Econ. 82 (2), 175–197. Kawabata, M., Shen, Q., 2006. Job accessibility as an indicator of auto-oriented urban structure: a comparison of Boston and Los Angeles with Tokyo. Environ. Plan. B Plan. Des. 33, 115–130. Kawabata, M., Shen, Q., 2007. Commuting inequality between cars and public transit: the case of the San Francisco Bay Area, 1990-2000. Urban Stud. 44 (9), 1759–1780. Kodransky, M., Lewenstein, G., 2014. Connecting Low-income People to Opportunity with Shared Mobility. Institute for Transportation and Development Policy and Living Cities. Lang, R.E., LeFurgy, J., 2003. Edgeless cities: examining the noncentered metropolis. Hous. Policy Debate 14 (3), 427–460. Lee, B., 2007. “Edge” or “edgeless” Cities? Urban spatial structure in us metropolitan areas, 1980 to 2000. J. Reg. Sci. 47 (3), 479–515. Levinson, D.M., 1997. Job and housing tenure and the journey to work. Ann. Reg. Sci. 31 (4), 451–471. Lusk, A., Furth, P.G., Morency, P., Miranda-Moreno, L.F., Willett, W.C., Dennerlein, J.T., 2011. Risk of injury for bicycling on cycle tracks versus in the street. Inj. Prev. 17, 131–135. Manaugh, K., Miranda-Moreno, L.F., El-Geneidy, A.M., 2010. The effect of neighbourhood characteristics, accessibility, home–work location, and demographics on commuting distances. Transportation 37, 627–646. Marshall, A., 1920. Principles of Economics: an Introductory Volume. Macmillan. McDonald, J.F., Prather, P.J., 1994. Suburban employment centres: the case of Chicago. Urban Stud. 31 (2), 201–218. McMillen, D.P., McDonald, J.F., 1998. Suburban subcenters and employment density in metropolitan Chicago. J. Urban Econ. 43 (2), 157–180. McMillen, D.P., Smith, S.C., 2003. The number of subcenters in large urban areas. J. Urban Econ. 53 (3), 321–338. METRA, 2010. Metra's New Fare Structure Begins Feb. 1. Retrieved 3/12, 2016, from https://metrarail.com/metra/en/home/utility_landing/newsroom/newsroom_ archive/2010NewsroomArchive/metra_s_new_farestructurebeginsfeb1.html. Mills, E.S., 1970. Urban density functions. Urban Stud. 7 (1), 5–20. Mills, E.S., 1972. Studies in the Structure of the Urban Economy. Johns Hopkins Press, Baltimore, MD. Minikel, E., 2012. Cyclist safety on bicycle boulevards and parallel arterial routes in berkeley, California. Accid. Anal. Prev. 45, 241–247. Modarres, A., 2011. Polycentricity, commuting pattern, urban form: the case of southern California. Int. J. Urban Reg. Res. 35.6, 1193–1211. Mulley, C., Nelson, J.D., 2009. Flexible transport services: a new market opportunity for public transport. Res. Transp. Econ. 25 (1), 39–45. Muth, R.F., 1969. Cities and Housing: the Spatial Pattern of Urban Residential Land Use. University of Chicago Press, Chicago, IL. O'Kelly, M.E., Lee, W., 2005. Disaggregate journey-to-work data: implications for excess commuting and jobs-housing balance. Environ. Plan. A 37, 2233–2252. Ong, P., Blumenberg, E., 1998. Job access, commute and travel burden among welfare recipients. Urban Stud. 35 (1), 77–93.

115