Do Transit-Dependent Neighborhoods Receive Inferior Bus Access? A Neighborhood Analysis in Four U.S. Cities

ABSTRACT:

Intrajurisdictional delivery of publicly provided services often results in observable service level differences that vary by spatial subunit (neighborhood). These variations are related to the sociodemographic characteristics of neighborhoods and have been hypothesized in prior literature to be the result of bias against or favoritism toward certain neighborhoods. Using path regression, this paper examines publicly provided bus service in four cities—Asheville, North Carolina; Charlotte, North Carolina; Mobile, Alabama; and Richmond, Virginia—to examine whether the socioeconomic character of a neighborhood is related to the share of municipal bus service it receives. With this analysis, we test an expanded version of Lineberry’s underclass hypothesis. Specifically, do transit-dependent neighborhoods, or those with a high percentage of non-Caucasian, low-income, elderly, or student residents receive inferior bus service? Findings confirm prior research that both standard rules and bias are present in service delivery decisions.

Notes

1 Case 1:10-cv-00055 Unites States District Court for the Northern District of Illinois Eastern Division.

2 Title VI of the Civil Rights Act of 1964 prohibits discrimination on the basis of race, color, or national origin in programs receiving Federal financial assistance.

3 Source: U.S. Census Bureau.

4 The study is limited to bus service (as opposed to bus, subway, and commuter rail) in order to limit the scope and better control the breadth of transport service made available.

5 49 CFR §25.1(a).

6 City is the appropriate geographical unit for this study as opposed to metropolitan area because the theory on service delivery is focused on intrajurisdictional rather than interjurisdictional service distributions. This point is important especially in light of bus service, which tends to be provided in multiple jurisdictions across a metropolitan area, often on a regional basis. Yet the focus of analysis is on neighborhood differences in service delivery, not regional differences.

7 Trips are reported on an “unlinked” basis, which means that each boarding is counted as a separate trip. Thus, a trip from a point of origin to a destination that requires a transfer(s) is counted as multiple trips. Passenger data are commonly reported as unlinked due to the ease in data collection.

8 In a study based on data from the 1995 Nationwide Personal Transportation Survey, Pucher, Evans, & Wenger (1998) find that African Americans make more than 10 times as high a percentage of their travel trips by bus than Caucasians and almost three times as high a percentage as Hispanics. In a more recent report issued by APTA on rider profiles (APTA, 2007), the ATPA found 35.7% of roadway transit riders are Black/African American and 13.7% are Hispanic/Latino.

9 Using 2000 Census data, Logan (2006) finds that African Americans are much more segregated from Caucasians than are Hispanics or Asians.

10 For a comprehensive overview of these access measures, see Morris, Dumble, and Wigan (1978).

11 Capacity is a proxy for comfort in that it measures crowding. Crowding is clearly only one aspect of comfort yet data on more refined measures of comfort such as the age of buses, cleanliness, safety, and climatization are not readily available by route.

12 Census data measure ethnicity in two different ways. The first is by category such as African American, American Indian, Asian, other, and two or more “races.” The second way uses the same categories but distinguishes between Hispanic and non-Hispanic. Thus, total non-Caucasian population may be calculated either way. The Pearson correlation of the two measures is 0.99. Therefore to calculate percentage non-Caucasian, we used the simpler approach of the ratio of the population in all the non-Caucasian categories to the total population.

13 The relationship between income and access may be nonlinear if very low and very high income residents do not have the means or inclination, respectively, to ride the bus. However, our data reveal a linear relationship between these two variables.

14 An alternative test for multicollinearity (e.g., Allison, 1999) uses a VIF range of 2.5–10.0 and a tolerance of under 0.10–0.40. These results for our variables are 2.9/0.33 and 3.4/0.29 for non-Caucasian and income, respectively.

15 The literature on transit service planning indicates that population density, along with the physical characteristics of a neighborhood, are key elements considered by transit planners when designing bus routes. See for example, Benn (1995).

16 For a discussion of path analysis, see, for example, Alwin and Hauser (1975).

17 The model is estimated using SPSS. Indirect effects are the product of the two coefficients leading to the dependent variable through the intervening endogenous variable. Total effect is the sum of direct and indirect values.

18 Census tracts that contain the central business district (CBD) of the city are omitted from analysis. These tracts (in the case of Richmond, the CBD is located in two tracts) contain an abundance of bus stops and routes because they are the point of origin and destination for the majority of routes. Thus, the level of access is highly skewed and not meaningful in the sense of residents’ ability to access bus service. In addition, bus routes that serve only a university or college in a city are omitted from analysis because those routes are used only to shuttle students within a university area. Express routes are omitted as well because the LITA calculation does not properly distinguish between express and non-express routes. Express routes arguably offer superior access to residents in the neighborhoods where they stop, yet the LITA methodology scores these neighborhoods as having less, not more, access because they have fewer stops. Finally, five observations are eliminated from analysis as they are identified by casewise diagnostics as outliers.

19 Dummy variables are included for each city in order to account for the distribution of data within each city. Charlotte is the reference case. Coefficients for dummy variables are unstandardized.

20 All inferential interpretations are based on the 0.05 significance level, estimates are standardized betas.

21 The dummy variables for each city are significant, indicating that the cities are statistically different from each other. Mobile, Asheville, and Richmond all have lower LITA scores on average than Charlotte. By convention, dummy variables are not shown in the path diagram. The path analysis findings hold for each city with one exception. In Asheville, commuter density is a much stronger predictor of access than carless households.

22 The other relationships are either weak or insignificant.

23 Source: http://www.Censusscope.org.

24 Bus stops that fell within ¼ mile of a census tract border were counted twice, once in their own census tract and once in the bordering tract, to reflect that walkers in one tract could catch a bus in another tract.

25 Percent no car and commuter density are transformed by base 10 log in the regression analysis to account for outliers and reduce skew in the regression standard residuals.