THE ROLE OF PUBLIC TRANSIT

IN THE MOBILITY OF LOW INCOME HOUSEHOLDS

FINAL REPORT

May 2001

Genevieve Giuliano

Principal Investigator

Hsi-Hwa Hu

Kyoung Lee

School of Policy, Planning, and Development

University of Southern California

DISCLAIMER

The contents of this report reflect the views of the authors, who are responsible for the facts and the accuracy of the information presented herein. This document is disseminated under the sponsorship of the Department of Transportation, University Transportation Centers Program, and California Department of Transportation in the interest of information exchange. The U.S. Government and California Department of Transportation assume no liability for the contents or use thereof. The contents do not necessarily reflect the official views or policies of the State of California or the Department of Transportation. This report does not constitute a standard, specification, or regulation.

ABSTRACT

Despite substantial and increasing subsidies, public transit’s market share continues to decline; public transit’s share of person trips is less than two percent. The remaining transit market has two components: downtown commuters in the largest U.S. metropolitan areas, and transit dependents — those who are either unable or unwilling to drive, and those who do not have access to a private vehicle. Car ownership is a function of income. A fundamental justification for transit subsidies is to provide a basic level of mobility to all persons, especially the transportation disadvantaged, yet even among the disadvantaged, most travel is by private vehicle, and public transit accounts for just five percent of all person trips.

This report examines the use of public transit by low income households. Using the 1995 Nationwide Personal Transportation Survey, we analyze both stated behavior regarding usual travel and actual journey to work mode to understand the role of public transit in the mobility of low income households. We find that public transit is not a reasonable substitute for the private vehicle for most people, poor or not poor. Regular transit is associated with less trip making and less distance traveled, and the effect is more pronounced for the poor. A second major barrier to transit use is lack of access: about one-third of NPTS respondents stated that transit was not available in their town or city. Other barriers include off-peak commuting and trip patterns that are inconsistent with transit use. We conclude that transit policy should focus on retaining existing markets by improving service frequency and quality in high demand markets, by exploring more effective ways of providing transit in low demand markets, and by expanding transit to serve off-peak and off-direction commutes. We note that in most circumstances, private vehicle access is the key to improved mobility for the poor as well as the non-poor. Economic development policies to increase the supply of jobs, goods and services in low income neighborhoods are also encouraged.

TABLE OF CONTENTS

DISCLAIMER i

ABSTRACT.................................................................................................................................... ii

TABLE OF CONTENTS iii

LIST OF TABLES v

LIST OF FIGURES vi

DISCLOSURE vii

ACKNOWLEDGMENTS viii

CHAPTER ONE INTRODUCTION 1

CHAPTER TWO LITERATURE REVIEW 4

The Role of Public Transit............................................................................................................ 4

..... Who Uses Transit 5

..... The Concept of Transportation Disadvantage and Its Consequences 7

..... The Limited Market for Transit 11

CHAPTER THREE RESEARCH APPROACH, DATA AND

..... DESCRIPTIVE STATISTICS 13

..... Research Approach 13

........... Measuring Mobility 13

........... Transit Users vs. Non-Transit Users 14

........... Target Population 15

Data 16

..... Descriptive Statistics 17

..... Characteristics of Poor and Low Income Households 19

........... Travel Characteristics 22

CHAPTER FOUR DATA ANALYSIS 35

Introduction 35

..... Explaining Frequency of Transit Use 35

..... Model Form 37

Data 41

Results 44

........... Conclusions on Frequency of Transit Use 53

..... Transit Use for Commuting 54

..... Model Development 54

Data 56

Results 59

........... Conclusions on Transit Use for Commuting 63

........... Why People Don’t Use Transit 63

..... Summary from Data Analysis 64

CHAPTER FIVE CONCLUSIONS AND POLICY IMPLICATIONS 66

..... Transit and Mobility 66

........... Policy Implications 66

..... Transit Access 67

........... Policy Implications 68

..... Transit and Car Ownership 69

........... Policy Implications 69

..... Transit and Demographics 69

........... Policy Implications 70

..... Geography and Transit 70

........... Policy Implications 71

..... Transit and Complex Travel 72

........... Policy Implications 72

..... A Note on Attitudes and Transit Quality 73

APPENDIX A 74

APPENDIX B 80

BIBLIOGRAPHY 89

LIST OF TABLES

Table 2-1 Distribution of Total Boardings by Income Quintile, 1980 and 1992, Los Angeles County 6

Table 2-2 Distribution of Total Transit Costs by Income Quintile, 1980 and 1992, Los Angeles County 10

Table 3-1 Distribution of Households by Poverty Status and Size........................................... 18

Table 3-2 Distribution of Households by Low Income Status and Size................................... 18

Table 3-3 Households by Life Cycle, Poverty, Low Income Status........................................ 19

Table 3-4 Persons by Race/Ethnicity, Poverty, Low Income Status........................................ 20

Table 3-5 Total Daily Trips, Travel Distance, Travel Time..................................................... 22

Table 3-6 Travel Characteristics for Those Who Made at Least One trip............................... 24

Table 3-7 Mode Shares........................................................................................................ 25

Table 3-8 Trip Purpose........................................................................................................ 25

Table 3-9 Transit Trips by Income Group............................................................................. 26

Table 3-10 Trip Purpose by Income Group............................................................................. 27

Table 3-11 Distance to Transit Stop by MSA Size.................................................................. 28

Table 3-12 Distance to Bus Stop by Income Category and MSA............................................ 29

Table 3-13 Transit Use........................................................................................................... 30

Table 3-14 Transit Use and Travel Distance, Time.................................................................. 31

Table 4-1 Variable Description............................................................................................. 43

Table 4-2 Binary Model with Low Income Dummy............................................................... 46

Table 4-3 Binary Model with Joint Effects............................................................................. 48

Table 4-4 Ordered Logit Model with Low Income Dummy................................................... 50

Table 4-5 Ordered Logit Model with Joint Effects................................................................. 52

Table 4-6 Variable Description............................................................................................. 57

Table 4-7 Descriptive Statistics for Home to Work Trip........................................................ 59

Table 4-8 Binary Full Model................................................................................................. 60

Table 4-9 Binary Model with Joint Effects............................................................................. 62

LIST OF FIGURES

Figure 3-1.......... Poverty, Low Income Status by Number of Workers in Household 20

Figure 3-2.......... Poverty, Low Income Status by MSA Location 21

Figure 3-3.......... Trips per Day by Poverty Status 23

Figure 3-4.......... Daily Travel Time by Poverty Status 23

Figure 3-5.......... Problem When Using Transit 32

Figure 3-6.......... Measures of Transit Service Quality 34

Figure 4-1.......... Development of Transit User Analysis Sample 42

Figure 4-2.......... Development of Sample for Work Trip Analysis 58

Figure 4-3.......... Why Not Use Transit for Work Trip 64

DISCLOSURE

Project was funded in entirety under this contract to California Department of Transportation.

ACKNOWLEDGMENTS

The authors appreciate funding support from the METRANS Center, University of Southern California.

CHAPTER ONE

INTRODUCTION

The secular decline in transit demand that began in the 1930s and continues today has left the public transit industry with two major markets: downtown commuters and transit dependents (Jones, 1985). The downtown commuter market remains because of the cost and limited availability of parking in downtown areas, road congestion, and the large concentration of jobs that makes transit access relatively convenient.

Transit dependents — those who do not have access to a private vehicle — are the second major market. However, data show that this market is shrinking: as car ownership continues to increase, fewer households have no cars. Even among households without cars, more trips are made by walking and by private car than by transit or other modes (Lave and Crepeau, 1995). The transit-dependent market is increasingly an inner city, minority market. The 1995 Nationwide Personal Transportation Survey (NPTS) data show that Hispanics and Blacks account for about 60 percent of all transit riders, and most of these riders are residents of central cities (Pucher, Evans, and Wenger, 1998).

Information on how limited mobility and accessibility affects low income households, particularly those households that do not have access to a private vehicle, is limited. National survey data indicate that members of low income households make fewer daily trips and travel fewer miles than comparable members of non-low income households. Low income households that do not have cars exhibit even lower rates of travel and trip making. Low income individuals are more dependent on public transit, yet, even within the lowest income class (less than $15,000 in 1995), just 6.8 percent of all person-trips are made by transit (Pucher, Evans, and Wenger, 1998).

These statistics suggest that public transit does not play a major role in serving the needs of the transit-dependent population. Those with limited or no access to privately owned vehicles sacrifice mobility; the results of limited travel resources are shorter and fewer trips. However, public transit’s role in providing basic mobility for the transportation disadvantaged is a major justification for subsidies (Meyer and Gomez-Ibañez, 1981). Mobility is essential for access to jobs, services, and social activities, hence there is public responsibility for supplying some basic level of transportation services to those who do not or cannot drive. Concern has grown over the past decade that public transit agencies have shifted resources from basic local transit services to more costly commuter services designed to attract discretionary riders. Since local transit services are used more by low income and minority patrons, it is argued that the benefits of public subsidies are inequitably distributed (Hodge, 1995). This issue has been the basis of federal lawsuits in several large metropolitan areas, including Los Angeles.

The recent changes in welfare policy add to concerns regarding the effectiveness of public transit. Many argue that public transit must play a major role in providing access to jobs for welfare recipients, as the majority of these new workers will come from households that do not own a car. Given the large public investment in public transit and its stated purpose of providing basic mobility, it is important to understand the role of public transit in providing mobility for low income persons. Under what conditions is transit used? What are the barriers to more extensive transit use? Are these barriers the result of travel demand characteristics, or other factors?

The purpose of this project was to examine the travel patterns of low income households, with a particular focus on transit use. Using data from the 1995 Nationwide Personal Transportation Survey (NPTS), we examined patterns of total daily travel and work travel. Our research had the following objectives:

1. Document the extent and frequency of transit use among low income travelers

2. Within the segment of low income households, examine the role of demographic, life cycle and location factors associated with transit use

3. Examine the role of transit in overall levels of mobility for the same population segment

4. Evaluate the market for transit among low income and minority households.

This report presents the results of our research. Our results show that barriers to transit use are fundamental. Public transit is not a reasonable substitute for the private vehicle for most people, poor or not poor, under most circumstances. Regular transit use is associated with less trip making and less distance traveled, and the effect is more pronounced for the poor. A second major barrier to transit use is lack of access: about one-third of NPTS respondents stated that transit was not available in their town or city. Other barriers include off-peak commuting and trip patterns that are inconsistent with transit use. We conclude that transit policy-makers should focus on retaining their existing markets by improving service frequency and quality, by exploring more effective ways to provide transit in low-demand markets, and by expanding transit to serve off-peak and off-direction commutes. Economic development policies to increase the supply of jobs as well as of basic goods and services in low income neighborhoods are also encouraged. Finally, we note that in most circumstances private vehicle access is the key to improved mobility for the poor as well as the non-poor.

The remainder of this report is organized as follows. Chapter Two presents a review of the literature on the role of public transit and trends in transit use, and discusses the concept of transportation disadvantage. Chapter Three presents our research approach, describes the NPTS data, and provides a descriptive analysis of travel patterns across income groups. Chapter Four presents our analysis of transit use for all travel and of transit use for the work trip. We develop and estimate models to test the effect of individual, geographic, and trip characteristics on the likelihood of using transit. The final chapter summarizes our major findings and discusses their policy implications.

CHAPTER TWO

LITERATURE REVIEW

THE ROLE OF PUBLIC TRANSIT

Despite substantial and increasing public subsidies, public transit’s market share continues to decline (Fielding, 1995). Most recently available national survey data indicates that transit’s market share is less than 2 percent of all person-trips. This tiny market share is composed of two major markets: downtown commuters in the largest US metropolitan areas and transit dependents — those who are either unable or unwilling to drive, and those who do not have access to a privately owned vehicle (Jones, 1985). Furthermore, U.S. transit ridership is heavily concentrated in a few of the largest cities: New York accounts for 40 percent of U.S. daily transit ridership; and adding Los Angeles, Chicago, Boston, San Francisco, and Washington, DC, accounts for two-thirds of the nation’s total (American Public Transit Association, 2000).

Subsidization of public transit has historically been based on two different and often conflicting objectives: 1) to provide a basic level of mobility for all persons, but especially the transportation disadvantaged, and 2) to provide an effective substitute for the private car in order to reduce automobile travel and its associated externalities, including traffic congestion, air pollution, and “urban sprawl” (Meyer and Gomez-Ibañez, 1981; Hodge, 1995; Fielding, 1995). This latter objective has emphasized the provision of rail transit, which is argued to be more attractive to choice riders and therefore more effective in achieving environmental goals. Most recently, sustainability concerns and the “smart growth” movement have placed even more emphasis on rail transit (Newman and Kenworthy, 1998).

The emphasis on rail transit is evident in transit investments. Between 1991 and 1998, total revenue vehicle miles of light rail, commuter rail, and heavy rail service increased 59, 20, and 8 percent respectively. Over the same period, bus service increased by 6 percent (U.S. Department of Transportation, Federal Transit Administration, 1998). Unfortunately, however, there is little evidence that these investments are generating the desired increases in transit ridership (Rubin, Moore, and Lee, 1999). On the contrary, new rail service generally replaces pre-existing bus service and attracts few new riders from cars (Kain, 1999; Pickrell, 1992). New rail systems are oriented to long distance, downtown commuters, who are disproportionately affluent and white (Webber, 1976; Gomez-Ibañez, 1985). In contrast, the transportation disadvantaged are concentrated in central cities, and would benefit more from increased service frequency, lower fares, and fewer transfers. In some cases, the high costs of building and operating rail systems have led to the perverse outcome of reducing transit ridership, as fares are increased and bus service is cut back in response to budget constraints. The obvious social equity consequences of these outcomes have led to a series of lawsuits against major transit operators in Los Angeles, New York, and Pennsylvania (Taylor and Garrett, 1998). In the case of Los Angeles, the Los Angeles County Metropolitan Transportation Authority (LACMTA) is currently under court order to expand and improve bus service in high-demand areas. The courts found that LACMTA’s policy of expanding rail service and not expanding bus service was discriminatory: the County’s bus passengers are on average poorer and more likely to be minorities than the County’s rail passengers.

WHO USES TRANSIT

Transit share is declining across all metropolitan areas and across all income classes. However, the loss of middle- and higher-income passengers has been greater than the loss of low income passengers, hence the poor make up an increasing share of transit users (Pucher, Evans, and Wenger, 1998). Poor and minorities now constitute the majority of transit passengers. For example, Table 2-1 gives the distribution of transit ridership by income quintile, 1980 and 1992, for Los Angeles County. By 1992, the majority of Los Angeles County transit riders were from households in the lowest income quartile; all other income categories showed a decrease in share of total boardings.[1]

Table 2-1: Distribution of Total Boardings by Income Quintile, 1980 and 1992, Los Angeles County

Income Quintile	1980 (%)	1992 (%)
Lowest	37.4	56.3
Second	27.7	21.4
Middle	17.0	11.4
Fourth	10.0	6.5
Highest	7.9	4.5

Source: Luo, 2000

Although the majority of transit riders are poor, the poor do not use transit for the majority of their trips. Using the 1995 NPTS data, Pucher, Evans, and Wenger (1998) show that average trips per day per person range from 3.4 for the lowest income category to 4.2 for the highest income category, and average miles per day per person range from 17.4 to 28.6. The greater difference in travel mileage across income categories is explained by differences in car ownership and modal use. While just 8.5 percent of all households do not have cars, one-third of the lowest-income households have no car, and almost half have one car. Limited resources leads to relatively more use of alternative modes — walking and transit — but the vast majority of all person-trips take place in private vehicles, even among the lowest-income households. While 32 percent of the poorest households (income less than $15,000) have no vehicle, 15 percent have two cars, and 5 percent have three or more cars. A car is clearly one of the first purchases that households desire to make; the car ownership rate jumps from 68 percent to 92 percent in the next-lowest income category ($15,000-$29,999) (Pucher, Evans, and Wenger, 1998).

Giuliano and Moore (1999) conducted a case study of a Los Angeles inner-city neighborhood. Interviews with local residents revealed that carpooling and paying others to get a ride were common forms of work travel for those without cars. Field observations at work sites showed a roughly even split between drive alone, carpool, and walking or transit (the two modes could not be distinguished, since transit stops were not always adjacent to work sites). Field observations at shopping centers showed a roughly even split between drive alone, carpool, and walking. Transit use was extremely low, despite the very high level of transit accessibility within the study area.

In addition to car access and income, prior research shows that transit use is related to age, sex, and race. Children and the elderly are more likely to use transit than adults under the age of 65 (Pucher, Evans, and Wenger, 1998). Gender variation on transit ridership is expected due to differences in social roles and household responsibilities. Traditional perceptions have been that women are more dependent on transit than men (Giuliano, 1979; Michelson, 1985; Pickup, 1985). However, increased female labor force participation has resulted in more complex travel patterns and consequently more demand for private vehicle use (Rosenbloom and Burns, 1993; Hayghe, 1996; Taylor and Mauch, 1996). Finally, several studies show that Blacks and Hispanics are more likely to use transit than other race/ethnic groups (Pisarski, 1996; Millar, Morrison, and Vyas, 1986; Rosenbloom, 1998). McLafferty and Preston (1997) found that Black women residing in the central city have the longest average commute times among all race and gender groups.

THE CONCEPT OF TRANSPORTATION DISADVANTAGE AND ITS CONSEQUENCES

Mobility is largely a function of resources; car ownership is correlated with income. As household income increases, car ownership also increases, and as car ownership increases, so does trip making and miles traveled. Conversely, households adapt to limited mobility resources by making fewer and shorter trips. In a case study of low income households in Austin, Texas, Clifton and Handy (1999) found that low income households have less access to a variety of goods and services due to limited mobility. Transit access is low due to limited and infrequent service. Walk access is inherently limited. Car availability is the critical factor in determining accessibility. Hence low income households engage in various forms of car sharing, from borrowing cars to taking rides.

The poor may be disadvantaged in at least three ways as a result of limited mobility. First, the poor may be “captive consumers” of goods, services, or medical care. Retail establishments may be able to charge higher prices when consumers are limited to local neighborhood stores. Households may be limited to the parks, movie theaters, and other recreational facilities close to home. Social networks may be limited to the local neighborhood. And choice of medical or dental services may be limited not simply to what one may afford, but to nearby destinations.

Research on these issues is quite limited. Studies have demonstrated the scarcity of major supermarkets and banks in inner-city areas (Cotterill and Franklin, 1995; Alwitt and Donley, 1996; Caskey, 1994). An analysis of accessibility to parks revealed lower levels of access among inner-city residents, due to both fewer local parks and limited resources to travel to more distant parks (Talen and Anselin, 1998).

The second dimension of disadvantage is what has come to be called “spatial mismatch.” The concept of spatial mismatch was developed by Kain (1968). The argument is that suburbanization has been selective — the more affluent white population has suburbanized, while the minority (and predominantly poor) population has remained in the central city. Differential rates of suburbanization are explained by many factors, including exclusionary zoning practices and discrimination in the housing market. As jobs have suburbanized (particularly low-wage jobs), central city workers have experienced a relative decline in job accessibility, which has in turn led to both higher unemployment rates and longer commutes for those who are employed. Less job accessibility implies fewer job opportunities, and hence less likelihood of finding a job, while longer commutes imply lower net wages.

Kain’s work touched off an extended academic debate that has persisted to this day. Are the higher unemployment rates observed among central city Blacks and other minorities the result of this spatial mismatch, or the result of discrimination by employers, lack of job skills, lack of access to social networks which provide access to job opportunities, or some combination of these factors? The spatial mismatch hypothesis has been tested by comparing unemployment rates, commute distances, or net wages across otherwise similar workers living in central cities and suburbs.[2] There is some evidence of spatial mismatch in studies using average commute distance of low-wage workers, meaning that workers residing in central cities have longer commutes than workers residing in the suburbs (Ong and Blumenberg, 1998). Taylor and Ong (1995) explain observed shorter commute distances but longer commute travel times as the result of lower rates of car ownership and greater use of public transit by minority central-city residents. Evidence based on unemployment rates is mixed; lack of access to jobs explains very little of the differences in unemployment rates between central-city and suburban residents (O’Regan and Quigley, 1996; 1998). In a related study, however, Ihlandfeldt (1996) found that transit access to suburban low-wage jobs was significantly related to the probability of Black workers filling those jobs. Despite extensive research on this issue, the evidence on spatial mismatch remains mixed.

The third source of disadvantage is the cost of transport services. With regard to public transit, the poor pay relatively higher fares per unit of service than the non-poor. The poor take shorter trips and are less likely to travel during peak periods. The non-poor take longer trips and are more likely to travel during peak periods. Flat fares, or fares only loosely based on trip distance, mean that short trips have a higher price per unit (Wachs, 1989). Because transit demand is higher in poor areas, transit productivity is higher, fares contribute a higher proportion of operating costs, and subsidies per trip are lower. Shifts in transit financing have further increased the financial burden on the poor. Federal subsidies have declined, and local subsidies have increased. Federal subsidies come primarily from general revenue funds and hence are a relatively progressive income source. Local subsidies typically come from various types of use taxes, which tend to be regressive in incidence.

Results from a recent Los Angeles case study are illustrative (Luo, 2000). Table 2-2 gives the distributional incidence of total transit costs (capital and operating, calculated as three year averages) for Los Angeles County residents, for 1980 and 1992. The big changes in revenue sources between the two periods were fares, reduced federal subsidies, and new local revenues from two sales tax measures. Since relatively more poor people were using transit (see Table 2-1) and transit fares had been increased, the lowest quintile contributed a greater share of fare revenue. In addition, sales taxes are highly regressive, and the shift to a sales tax resulted in greater tax contributions from the lowest quintile. The middle quintiles were hardly affected, while the contribution of the highest income quintile declined.

Table 2-2: Distribution of Total Transit Costs by Income Quintile, 1980 and 1992, Los Angeles County

Income Quintile	1980 (%)	1992 (%)
Lowest	16.8	22.1
Second	15.6	14.7
Middle	15.7	15.8
Fourth	19.0	18.9
Highest	32.9	28.6

Source: Luo, 2000

Low income households also spend a much higher proportion of after-tax income on transportation — about one-third — than the average household, which spends about 17 percent (Deka, 2001, calculated from 1993 Consumer Expenditure Survey). Relatively high expenditures are explained by the high cost of car ownership. The poor are more likely to own older, less reliable and less fuel-efficient vehicles. Lower purchase costs are offset by higher repair and running costs.

It has been argued that one explanation for extensive car ownership even among the poorest households is the lack of high-quality transit service. Essentially, public transit is such a poor substitute for the automobile that the poor incur the expense of car ownership in order to obtain the mobility a car provides. If this is the case, then we should observe lower rates of car ownership in areas where transit service is more available. Deka (2001) conducted a Los Angeles case study to determine the relationship between transit access and car ownership. Transit access was measured with respect to census tract of residence as a gravity formulation incorporating route density and service frequency. Controlling for the dependency between the two variables (e.g., transit providers will supply more service in response to greater demand, and households without cars will locate in areas with more transit), Deka (2001) found that the relationship is small but significant. The probability of auto ownership decreases only slightly with increases in transit availability.

Another way of assessing impacts of transportation disadvantage is to look at the households who do not own cars. The share of households without cars has dropped from 21 percent in 1969 to about 9 percent in 1995. Lave and Crepeau (1995) examined households without vehicles using the 1990 NPTS data. Households residing in the New York Metropolitan Statistical Area (MSA) were excluded, because New York is so different from the rest of the US. Elderly, retired persons make up the majority of zero-vehicle households. Most of the remainder are single persons without children, and two-thirds of zero-vehicle households have no workers. As expected, persons in zero-vehicle households make an average of 1.8 trips per day, compared to the average of 3.2 trips per day. Persons in zero-vehicle households also were more likely not to have traveled at all on the survey day (40 percent vs. 21 percent for the general population).

Most zero-vehicle households are low income households, but most low income households own at least one car, as noted earlier. Therefore the question is, to what degree does no car indicate travel disadvantage vs. reduced demand for travel? Research on the elderly show that they make fewer trips and travel fewer miles than the non-elderly, whether or not they own cars (Rosenbloom, 1994a). The Lave and Crepeau (1995) analysis suggests that only a small segment of zero-vehicle households are truly disadvantaged.

THE LIMITED MARKET FOR TRANSIT

Transit’s limited market share, even among the poor and among those who do not own cars, leads to the obvious question of why. Several possible explanations have been explored. First, decentralization and the dispersion of activities make contemporary land-use patterns difficult to serve with conventional fixed-route transit. Cost-effective transit requires concentrated origins and destinations, so that transit capacity is effectively utilized. Several studies have documented the relationship between metropolitan density and transit use (e.g., Pushkarev and Zupan, 1977; Newman and Kenworthy, 1998). Dispersed origins and destinations require extensive route systems. The high costs of operating such systems leads transit agencies to offer infrequent service that cannot compete with the automobile (Meyer and Gomez-Ibañez, 1981; Fielding, 1995). Comparisons of transit travel time with auto travel time indicate that a transit trip takes 2 to 3 times as long as the same trip by car, even in areas where transit service is reasonably available (e.g., Taylor and Mauch, 1996). In suburban areas, many destinations simply cannot be reached by transit.

A second possible explanation is that even in areas where demand is adequate to support high quality transit, poor service quality, crowded buses, and fear of crime may deter transit use. There is little research on this issue. Levine and Wachs (1986) conducted an extensive study of crime in and around transit in Los Angeles, and found that fear of crime was particularly a problem for transit dependents. In a series of interviews with low income shift workers in Los Angeles, Giuliano and Moore (2000) found that long travel times, high fares, personal safety, and lack of service were the most frequent explanations given for not using transit. Crowded buses adversely affect service quality. Dwell time increases, making schedule reliability deteriorate. Heavily crowded buses may skip stops, leaving passengers stranded at bus stops. Standing on a bus is difficult for the elderly or for people carrying packages or small children.

A third explanation is spatial mismatches between where people live and where people work, as discussed in the previous section. A classic example is the reverse commute, in which central-city residents commute to suburban jobs. Transit service is oriented to the downtown commuter, and consequently reverse commuters experience a much lower level of service (Ong and Blumenberg, 1998). There is also the possibility of a temporal mismatch. Many low-wage jobs have non-traditional work hours. Office janitorial services are performed at night and on weekends. Retail jobs often require evening or weekend work. Swing and graveyard shifts still exist in manufacturing. Public transit is oriented to the traditional commute — to work (inbound) in the early morning and from work (outbound) in the late afternoon.

Finally, it is possible that contemporary lifestyles are simply incompatible with conventional transit service. Research shows that travel patterns have become more complex — people often combine a series of activities in a single travel “tour”, and many incidental stops are made in conjunction with the work trip (Hanson, 1995; Vincent, Keyes, and Reed, 1994). The extent to which these observations are true for low income households or for the transportation disadvantaged is unknown.

CHAPTER THREE

RESEARCH APPROACH, DATA AND DESCRIPTIVE STATISTICS

RESEARCH APPROACH

The purpose of this research is to evaluate the role of public transit in providing mobility for low income households.

Measuring Mobility

What is the appropriate measure of overall mobility? This is a matter of current debate (e.g., Hanson, 1995; Handy and Niemeier, 1997). On the one hand, it can be argued that the more one travels, the more benefits from travel one obtains. However, travel is costly, both in time and money, so the rational individual seeks to minimize these costs. Travel demand is an indirect demand — one travels in order to consume goods and services that are spatially dispersed.[3] Willingness to travel reflects willingness to pay for the expected benefits of the activity at the destination. Discussions of mobility often involve accessibility — to the extent that activities are more concentrated in space, less travel (mobility) is required to achieve a given level of activity benefits. However, controlling for land-use pattern, more travel should indicate more consumption of goods and services (activities), or more investment in travel in order to consume preferred bundles of goods and services.

Consider an ideal measure of mobility. Following the work of Hägerstrand (1970), mobility reflects an individual’s “activity sphere” — the geographic range of activities conducted over the course of the day. The activity sphere is determined by resources and constraints of the individual and by the spatial distribution of activity locations. Resources include such things as income, supply of transportation services, and time. Constraints may be resource related (e.g., no car, no transit available) or schedule related (e.g., fixed work hours, fixed operating hours of business establishments). The spatial distribution of activities determines the number of opportunities that may be accessed for a given quantity of travel resources. Travel outcomes are the result of the individual’s activity choices, given his/her set of resources, constraints, and spatial opportunities. An ideal measure of mobility would capture all of these factors. Unfortunately, however, the data are not available to construct such a measure.

It is clear that an appropriate measure should capture travel for all purposes. Total travel can be measured in terms of trips, distance, and time. Trips capture the total number of activities conducted, but provide limited information. Many trips are mandatory, in the sense that household maintenance requires some amount of trip making, and most jobs require traveling to work, hence the greater regularity of trip frequency across population segments. The more interesting question is where people choose to shop or work. The spatial range of travel over the course of the day is captured by distance and time. Of these, distance is the more appropriate measure of mobility. Travel time is problematic, because it is determined both by distance and speed. In this analysis we measure mobility as total distance traveled over the course of one-day period.

From a public policy perspective, work trips are particularly important. We are concerned about whether low income households must incur higher commuting costs due to spatial mismatch, and what this may mean for employment opportunities and job retention. However, work trips are also important from a behavioral perspective, because work location and schedule are critical factors in defining daily activity and travel patterns. Therefore our analysis includes total daily travel as well as travel associated with the journey to and from work (e.g., a subset of total daily travel).

Transit Users vs. Non-Transit Users

Our literature review has shown that use of public transit has declined even among low income households, as more such households own and use private vehicles. We reviewed a series of explanations regarding why this is the case, and we noted that evidence to support some of these explanations is limited. Focusing our attention on those who use transit would provide only partial information. We would learn something about how and why these individuals use transit, but we would learn nothing about why other similar individuals do not use transit. Therefore it is appropriate to include all travel in our analysis. We are interested in such questions as,

· Under what conditions is transit used?

· Are patterns of travel and transit use different across income groups, holding relevant factors constant?

· What are the barriers to more extensive transit use? Are they the result of travel demand characteristics, or other factors?

Target Population

A third measurement issue is which population segments should be included in the analysis. Our focus is low income households, and a case could be made for restricting the analysis to such households. However, comparing travel patterns across low income and not-low income populations may provide a clearer understanding of differences in travel between these groups. Transportation disadvantage is a relative concept. Therefore all households are included in our analysis.

How do we define the low income population? After reviewing several possibilities, we selected two measures. The first measure is based on the 1995 poverty threshold, adjusted for household size, as defined by the U.S. Census Bureau. The Census definition is based on food consumption requirements. Annual costs of food consumption are used as the basis for factoring up annual income to determine the poverty definition. The poverty threshold does not vary geographically. It is based on money income before taxes, and excludes capital gains and non-cash benefits. The poverty threshold is updated annually based on the Consumer Price Index.[4] There are many problems with the U.S. Census definition (Citro and Michaels, 1995); however we decided that it was sufficiently valid for our purposes.

The second measure is based on the Department of Housing and Urban Development’s (HUD) definition of low income. HUD defines “low income” and “very low income” in order to determine eligibility for housing subsidies. HUD definitions are adjusted both for household size and geographic region, to account for especially high-cost or low-cost housing markets. The “low income” definition is approximately 80 percent of the region’s median household income.[5] HUD definitions are adjusted annually, as median income is estimated annually. The HUD definition provides a less restrictive low income category.

DATA

We use the 1995 NPTS survey for this research. The NPTS is a household-based travel survey conducted periodically by the Federal Highway Administration (FHWA). The 1995 survey included 42,000 households and 95,360 persons. The sample was drawn from a stratified random digit dial telephone sample. In addition, several metropolitan areas paid FHWA to over-sample their areas. Areas with high transit use are also over-sampled, in order to obtain as large a sample as possible of transit trips. The survey includes household, individual and vehicle information, as well as a one-day travel diary for each person 5 years old or older. The travel data were collected in a two-stage process. Households were given one-day travel diaries to complete for each eligible member of the household. The diaries were reported to the interviewer via telephone. The travel diary data includes a total of 409,025 trips. The data files also include basic geographic and demographic data drawn from the U.S. Census and updated for 1995, provided at both block and census tract level and linked to each household record. In addition to the actual one-day travel information, the survey includes information on the journey to work, transit use, and a variety of attitudinal information. NPTS is therefore an exceptionally rich dataset.

Despite its richness, however, NPTS has some serious shortcomings for this research. First, indicators of transit accessibility are very limited — access to bus or rail stops is recorded, but there is no way to measure transit network accessibility. Second, attitudinal data on transit is recorded only for those who use transit. It is therefore not possible to measure attitudes that may prevent transit use (e.g., fear of crime). Attitudes and perceptions are known to be important explanatory factors in travel behavior (Kitamura, Mokhtarian, and Laidet, 1997). Finally, job-related data are limited. There is no information on job tenure. Respondents were asked to provide the Zip Code of their place of work, but this information is not released to the public. The Zip Code information is used to generate a variable to indicate whether or not the person works in a central city. There is also information on whether the work place is fixed or variable, or at home. The occupation data were never categorized, and therefore cannot be used. There is no information on work schedule, except what can be surmised by the time the individual starts and ends his/her work trip.

A very complex weighting procedure was developed for the NPTS data, as the weights must adjust for various types of response bias as well as the over-sampling of large metropolitan areas with rail transit and of areas that contracted with NPTS for larger samples.[6] The weights also expand the sample to estimates for the US population. In order to conduct statistical tests, we adjusted the person weights to scale the sample down to its original size.[7] This is a second-best procedure, as the weighting scheme in theory requires statistical calculations that are not available in most statistics software packages. The effect of using conventional statistics is to bias downward estimates of variance, and therefore increase the probability of Type I errors (reject the null hypothesis when it should be accepted).[8] Increasing the stringency of statistical significance tests compensates for this problem.

A total daily travel data file was constructed by aggregating all travel day trips and their characteristics for each person, using the 93,560 observation NPTS person file as the working file. Travel period trips and trips longer than 75 miles were excluded from the analysis. Most of the results reported here are based on the person file, and all are based on the adjusted weights described above. Because of missing data on key variables, actual sample size varies by type of analysis.

DESCRIPTIVE STATISTICS

We begin by identifying poor and low income households. As noted earlier, poverty status is adjusted for household size. We matched the reported household income categorical data as closely as possible to the income limits defined by the U.S. Census Bureau. Table 3-1 gives the results. The highest shares of poor households are found among single-person

households (reflecting many single persons retired or unemployed), and among the largest households (reflecting households with many children). In terms of numbers, however, there are relatively few large households; single-person households make up 40 percent of all poor households.

Table 3-1: Distribution of Households by Poverty Status and Size

	Number of persons in household
	1	2	3	4	5	6	Total
Poverty HH income cut-off ($ 1995)	10,000	10,000	15,000	15,000	20,000	> 20,000
Poor (col %)	20.7	7.2	12.6	9.3	18.9	22.6	13.2
Not poor (col %)	79.3	92.8	87.4	90.7	81.1	77.4	86.8
Total (row %)	25.6	31.3	17.2	16.1	6.8	3	100

Similar information for low income households is given in Table 3-2. In this case, household income was adjusted both for geographic region (state) and household size, so income cut-off levels are given relative to the “base”. Overall, about 37 percent of all households in the sample are defined as low income. The pattern across household size is similar. Single-person households account for about 35 percent of all low income households.

Table 3-2: Distribution of Households by Low Income Status and Size

	Number of persons in household
	1	2	3	4	5	6	Total
Median family income factor (% of base)	70	80	90	base	108	116+
Low income (col %)	51.2	30.47	31.5	30.5	41.2	48.3	37.2
Not low income (col %)	48.8	69.5	68.5	69.5	58.8	51.7	62.8

Characteristics of Poor and Low Income Households

Poverty and low income status are related to life cycle, race/ethnicity, and employment. The poverty rate is highest among single-adult households with children, followed by single-adult retired households, as shown in Table 3-3. The lowest poverty rate is among households with at least two adults. The pattern is similar for low income status. Nearly two-thirds of single-adult households with children are low income, and close to three-fourths of retired single-person households fall into this category. Two-adult households, with or without children have the smallest share of low income households.

Table 3-3: Households by Life Cycle, Poverty, Low Income Status

	1 adult no kids	³2 adults no kids	1 adult + kids	³2 adults + kids	1 adult retired	³2 adults retired	Total
Poor	15.6	5.8	34.8	9.9	33.3	10.6	13.2
Not poor	84.4	94.2	65.2	90.1	66.7	89.4	86.8
Low income	42.6	22.8	65.7	20.4	72.4	42.1	37.2
Not low income	57.4	77.2	34.3	69.6	27.6	57.9	62.8
Share of total sample	18.2	23.0	5.5	34.7	7.4	11.1	100

The relationship between race/ethnicity and poverty is well documented. Table 3-4 gives shares of poor/non-poor and low income/not low income by race/ethnicity for the NPTS sample. The poverty rates for non-Hispanic Blacks and Hispanics are much higher than those for non-Hispanic Whites and Asians.

Table 3-4: Persons by Race/Ethnicity, Poverty, Low Income Status

	White	Black	Hispanic	Asian	Other
Poor	8.6	26.2	25.1	8.7	16.1
Not Poor	91.4	73.8	74.8	91.3	83.9
Low income	30.1	54.4	48.7	38.1	40.6
Not low income	69.9	45.6	51.3	61.9	59.4
Share of total sample	73.7	12.0	10.0	2.1	2.3

Poverty status is also related to employment. Figure 3-1 shows number of workers in the household by poverty status and low income status. Over half of all poor households have no workers, and an additional one-third have just one worker. Among low income households, 38 percent have no workers and 37 percent have one worker.

Figure 3-1: Poverty, Low Income Status by Number of Workers in Household

We compared residence location patterns across income categories. The poor are concentrated in the largest metropolitan areas, and in non-metropolitan areas. Figure 3-2 shows that the non-poor are relatively more concentrated in the largest metropolitan areas, but relatively less concentrated in non-metropolitan areas. Within metropolitan areas, the poor are more concentrated in the central city, and hence are more likely to reside in high-density areas, defined here as census tracts with population density of 10,000 persons per square mile or more — about 17 percent of the poor live in high-density areas, compared to 8 percent for the non-poor.

Figure 3-2: Poverty, Low Income Status by MSA Location

It is well known that persons from low income households have less access to private vehicles. Just 5 percent of all households do not have any drivers (defined as person having a valid driver’s license), but 12 percent of low income households and 22 percent of poor households have no drivers. This is in part a function of older, retired persons (more likely female) making up a large portion of poverty households. Similarly, our sample has 7.7 percent of households having no private vehicle, but 17 percent of low income households and 30 percent of poor households have no private vehicle. Fully three-fourths of poor households have one or zero vehicles, indicating limited car access. In contrast, almost two-thirds of non-poor households have two or more private vehicles.

Travel Characteristics

We turn now to a description of basic travel characteristics. We describe total daily travel, travel by mode, and travel by purpose. Also included is a description of transit use, access to transit, and attitudes regarding transit.

Total Daily Travel

Table 3-5 gives mean and median values for total daily trips, travel distance and travel time. The averages include zero trips, e.g., persons who did not travel on the diary day. Table 3-5 shows clearly that poor or low income persons travel less by any measure than non-poor or non-low income persons. About one-fourth of the poor made no trips on the travel day. Since many more poor or low income persons did not travel, average travel distance and travel time are significantly lower as well. Differences between poor and non-poor are shown graphically in Figures 3-3 and 3-4, which give cumulative distributions for total trips and total daily travel time respectively.

Table 3-5: Total Daily Trips, Travel Distance, Travel Time

	Trips			Distance (miles)		Time (minutes)
	% no trips	mean	Median	mean	median	mean	Median
Poor	25.4	3.1	2	18.1	6	47.3	30
Not poor	13.2	4.0	4	30.9	20	61.1	50
Low income	20.0	3.5	3	23.3	12	52.3	40
Not low income	11.9	4.2	4	32.6	22	63.2	52

When the poor or low income do travel, they travel shorter distances and spend less time traveling, as shown in Table 3-6. Less time spent traveling is the result of fewer trips — note that average trip time is virtually the same for all groups. However, average trip distance is shorter for the poor and low income groups, indicating that lower-income travelers take more trips on slower modes, e.g., transit and non-motorized modes.

Table 3-6: Travel Characteristics for Those Who Made at Least One Trip

	Daily Distance (miles)		Daily Time (minutes)		Average Trip Distance		Average Trip Time
	mean	median	Mean	median	mean	Median	mean	median
Poor	24.3	12.0	63.4	49	6.2	3.7	16.3	13.3
Not poor	35.6	25.0	70.4	60	8.4	5.7	16.5	13.3
Low income	29.0	17.9	65.3	50	7.2	4.5	16.2	13.2
Not low income	37.0	26.4	71.8	60	8.6	6	16.6	13.4

Trip shares by mode are given in Table 3-7. Limited car access is evident for the poor. Although the vast majority of all trips are taken by POV even among the poor, less than half of these trips are made as the POV driver. About two-thirds of all trips by non-poor or non-low income are made as the POV driver, and only about one-quarter of trips are made as a POV passenger. In contrast, close to 90 percent of all trips are made by POV for the non-poor. The poor also make a large share of trips by non-motorized modes; note that the walk/bike share is more than twice as large as the transit share. The poor are the heaviest users of transit, yet transit trips account for only about 5 percent of all their trips. Tables 3-5 through 3-7 suggest that the poor compensate with limited travel resources by traveling less overall (fewer, shorter trips), and by using alternative modes.

Table 3-7: Mode Shares

Mode	Poor	Not poor	Low income	Not-low income
POV - driver	47.0	64.7	56.3	65.9
POV - pass	30.6	25.9	28.1	25.6
Bus/rail	5.2	1.4	3.1	1.1
Walk/bike	13.9	5.7	9.6	5.1
Other	3.3	2.3	2.9	2.2

Differences in trip purpose across income groups are mainly for work and school/church (Table 3-8). Persons from poor households are less likely to be employed, hence the poor make fewer work trips. The difference in work trips is partially offset by a relatively greater share of school/church trips among the poor

.

Table 3-8: Trip Purpose

Purpose	Poor	Not poor	Low income	Not low income
Work/work related	13.9	21.9	17.4	22.8
Shop	20.9	19.7	20.9	19.4
Personal business	27.5	25.5	26.2	25.4
School/church	12.6	8.4	10.2	8.2
Social/recreational	25.0	24.3	25.1	24.0
Other	0.1	0.2	0.1	0.1

Transit Use: Actual Trips

Given the small share of transit trips that were taken on the travel diary day (about 1.2 percent of all trips for the entire sample), it is difficult to learn very much about transit use within any population segment. However, for completeness, we present some basic information on transit use from the day trip file. Of the trips taken by transit on the travel day, the shares by mode and income group are given in Table 3-9. Sample shares of the income groups are given in the last row of the table for comparison purposes. The poor account for about one-quarter of all transit trips, which implies about twice the rate of transit use as the non-poor. The low income group accounts for over half of all transit trips, also proportionately greater than their sample share. More interesting is the split across modes, with the poor making over 80 percent of their transit trips by bus, and the non-poor making 60 percent of their trips by bus. The greater use of bus by the poor has been documented in prior research, and it is at the center of social justice controversies.

Table 3-9: Transit Trips by Income Group

	Poor	Not poor	Low income	Not low income
Share of transit trips	26.7	73.3	52.9	47.1
Share bus (col %)	82.0	60.0	77.0	53.4
Share rail (col %)	18.0	40.0	23.0	46.4
Share persons	13.2	86.8	37.2	62.8

We examined trip purpose for those who used transit on the travel day. The largest category is work or work related (35.7 percent), followed by social or recreational activities (19 percent), family or personal business (14 percent), shopping (12 percent), and school (10 percent). The remainder of trips is spread across six additional trip purpose categories. Transit trip purpose is different from trips by all modes (see Table 3-8). Transit is more likely to be used for work travel and less likely to be used for other purposes. Table 3-10 gives trip purpose divided into work and non-work by income group. Since the poor are less likely to be employed, they are less likely to use transit for a work trip. As income goes up, so does the use of transit for the work trip.

Table 3-10: Trip Purpose by Income Group

Trip purpose	Poor	Not poor	Low income	Not low income
Work & related	16.4	41.1	22.2	47.1
Non work	83.6	58.9	77.8	52.9

The NPTS asks respondents about access to a transit stop (for those who stated that they had access to public transit). They are asked the distance to the nearest bus stop and rail transit stop or station from their residence.[9] Average reported distances by MSA size are given in Table 3-11. We also computed average reported distance to a transit stop, which was the average distance to the closest stop, whether bus or rail. It turned out that even in the largest MSAs, the average to the closest stop is equivalent to the average to a bus stop, meaning that even in the few metropolitan areas that have extensive rail transit systems, the bus system is more ubiquitous. The data in Table 3-11 is as expected, in that distance to a transit stop declines with increasing metropolitan size, and access to rail transit is quite limited for all but the largest MSAs. However, Table 3-11 also shows surprisingly long average distances. It turns out that the distribution is skewed towards a few large values, and the median for each category is much shorter than the mean.

Table 3-11: Distance to Transit Stop by MSA Size

MSA Size	Distance to bus (miles)	Distance to rail (miles)
Not in MSA	2.5	4.0
< 250K	1.8	5.5
250K - 500K	1.2	4.0
500K - 1 M	1.3	6.8
1 M - 3 M	0.8	6.5
> 3 M	0.7	2.6

Table 3-12 gives average and median reported distance to a bus stop, by MSA size category, and by income status. We use three categories for MSA size, as the middle categories are quite similar to one another. Table 3-12 shows: 1) that the poor or low income groups live closer to a bus stop than the not poor or not low income groups, regardless of MSA size, 2) that most people live closer to a bus stop than the average would indicate, 3) that more than half of the entire sample live within 1/2 mile of a bus stop (meaning more than half of those for whom transit is available in their town or city). Those who are dependent on transit locate near stops, as would be expected. We also computed average distance to a stop for those who actually used transit, and that average was 1/2 mile or less for all but the smallest MSAs and those living outside MSAs.

Table 3-12: Distance to Bus Stop by Income Category and MSA

	Poor		Not poor
	Average	Median	Average	Median
MSA > 3 M	0.35	0.10	0.71	0.30
Other MSAs	0.87	0.20	1.12	0.50
Not in MSA	1.85	0.30	2.46	0.50
	Low income		Not low income
MSA > 3 M	0.52	0.20	0.74	0.30
Other MSAs	0.84	0.20	1.22	0.50
Not in MSA	2.29	0.50	2.44	0.50

Transit Use: Usual Behavior

The travel diaries were recorded on varying days of the week, so the daily trip patterns reflect both weekday and weekend activity patterns. As was indicated in Table 3-5, about 25 percent of poor persons did not travel at all on the survey day. Others may use transit irregularly, and therefore would be unlikely to have taken a transit trip on the survey day. We therefore chose another approach. The NPTS also included questions on usual travel behavior. Among these were questions on transit use and availability. We expected that many more people were at least occasional users of transit than took a transit trip on the survey day. We use these questions to examine transit use.

Table 3-13 gives transit use by frequency of use. We define “regular user” as a person who uses transit at least once per week, and “occasional user” as using transit at least once per month. The other categories are self-explanatory. As expected, a much larger proportion of respondents are transit users to some degree; about 14 percent of the entire sample uses transit at least occasionally. Table 3-12 shows that the share of respondents who state that transit is not available is about the same for all groups.[10] This may seem surprising, given that we would expect that those with limited car access would locate in areas where transit service is available. However, a large portion of the poor live in non-urban areas, hence it is not unreasonable that many poor persons do not have access to transit. Among those for whom transit is available, poor or low income persons are more likely to be regular transit users.

Table 3-13: Transit Use

Transit Use	Poor	Not poor	Low income	Not low income
Regular user	16.7	7.0	11.7	6.3
Occasional user	6.6	5.8	5.5	6.1
Not a user	38.8	48.7	44.5	49.1
Transit not available	37.9	38.5	38.3	38.5

As expected, regular use of transit is associated with shorter travel distance, but longer travel time, as shown in Table 3-14 for poor and non-poor only. The relative difference in travel distance between regular transit use and no transit use is slightly greater for poor, but the relative difference in travel time is much greater for the poor. This may be result of less POV use by the poor. For those with limited access to a car, most trips are taken by transit or walking. In contrast, the regular transit users among the non-poor are likely making other trips by POV. Note that persons may or may not have used transit on the survey day, hence these numbers are indicative of general levels of mobility associated with transit use.

Table 3-14: Transit Use and Travel Distance, Time

		Regular user	Occasional user	Not a user
Daily travel Distance	Poor	14.0	15.3	18.9
	Not poor	23.9	31.9	31.2
Daily Travel Time	Poor	65.6	50.7	45.7
	Not poor	70.8	71.9	62.3

It is well known that attitudes are an important explanatory factor in travel behavior. The NPTS includes a series of questions on attitudes regarding the use of public transit. Respondents are asked how big a problem it is to get a seat, transfer, etc. There is one question on being worried about crime. Unfortunately, however, these attitudinal questions were asked only of those who stated they used transit at least once per month, and they were asked in alternating blocks. The survey gives us information on people who use transit, but not on people who do not use transit. We therefore cannot use the attitudinal information in our later analysis. Even for people who use transit, there are relatively few responses for any given question. The information provided is therefore only suggestive. We provide descriptive information for illustration.

Figures 3-5A through 3-5C show results on measures of the cost of using public transit. In each case, the respondent is asked, “Thinking about your use of public transit, please tell me whether this is a large problem, small problem, or no problem at all for you….” In all cases the poor view the measure as a large problem more than the non-poor, with the biggest difference on cost. Transferring is not a problem for most transit users, and the time it takes to use public transit is slightly more of a problem for the poor.

C. Time it takes

to Use Transit

B. Difficulty of

Transferring

A. Cost of Using

Public Transit

Figure 3-5: Problem When Using Transit

Figures 3-6A through 3-6D show results on measures of service quality. Somewhat more of the poor find crowding or getting a seat a large problem. The biggest difference between poor and non-poor is on the cleanliness of stations and vehicles. This may be the result of the poor being more likely to use inner-city services. Interestingly, there is little difference in perceptions about the availability of transit. This is counter to the hypothesis that the poor have more difficulty commuting to work via transit, but note that the poor (as we have defined them) are less likely to be employed. A major concern for planners and policy makers is fear of crime. Although a larger proportion of the poor consider fear of crime a large problem, about the same proportion of poor and non-poor do not consider crime a problem. All this tells us is that those who use transit do not worry a lot about crime. Presumably, those who do worry about transit crime would not use transit if at all possible. We might speculate that while crowded or dirty buses are an inconvenience, they are not a deterrent to transit use to the extent that fear of crime might be a deterrent. The greater propensity of the poor to perceive these factors as problems may be the result of more regular use of transit, or may reflect transit dependency. The poor may use transit even if they do not like to use it, because they have no other choice, whereas the non-poor have other choices and will not use transit if it is perceived to be inconvenient, uncomfortable or dangerous.

Finally, respondents were asked about having access to a car when needed. Not surprisingly, almost 40 percent of the poor viewed access to a car as a large problem, compared to less than 20 percent of the non-poor — an expected result consistent with the greater likelihood of the poor being transit dependent.

The descriptive analysis has presented basic information on travel patterns and transit use among poor and low income persons. We now turn to a more formal analysis of transit use.

Figure 3-6: Measures of Transit Service Quality

CHAPTER FOUR

DATA ANALYSIS

INTRODUCTION

This chapter presents our data analysis of transit use. We conducted two types of analyses. As noted previously, only a very small share of trips were made by transit on the survey day. Restricting our analysis to those who actually used transit on the survey day would provide limited insight on why people use transit. We therefore used the data on frequency of transit use for one type of analysis. This allows us to compare those who use transit regularly, those who use transit occasionally, and those who do not use transit at all. Although transit accounts for less than 2 percent of all trips, it accounts for about 6 percent of work trips. Our second type of analysis examines actual use of transit for the journey to work.

EXPLAINING FREQUENCY OF TRANSIT USE

The first part of our analysis estimates a series of models to explain frequency of transit use. We begin by developing a set of hypotheses regarding factors that may be related to transit use, based on the literature. We identify five groups of factors:

1. Demographic characteristics

2. Economic factors

3. Geographic factors

4. Travel characteristics

5. Attitudes

The three “TOO groups” (too old, too young, too handicapped) are often captive transit riders. Thus we expect children and the elderly to use transit more frequently than non-elderly adults. Based on prior literature, we also expect Blacks, Asians, and Hispanics to be more likely transit users. The role of gender is not clear. On the one hand, some population segments (single mothers) are likely to have low incomes and hence be more dependent on transit. On the other hand, women’s increasingly demanding social and household roles increases demand for auto travel. Many travel behavior theorists argue that travel choices are joint choices made at the household level. Household members decide who gets the car, who takes the children to school, etc., and allocate travel resources and responsibilities accordingly. The composition of the household therefore may affect transit use.

The major economic factors known to affect transit use are household income and car ownership. Household income is related to employment status and number of workers in the household. Employment status is also important, since transit is more likely used for the work trip, and transit service is oriented to serve the traditional peak period work trip. As household income increases, so does car ownership. The key factor for the individual is car access, or the availability of cars for household drivers. If there are fewer cars than drivers, there is more likelihood of using transit.

It was noted earlier that the largest U.S. metropolitan areas account for most U.S. transit ridership. This is due to the higher cost of using private vehicles (congested roads, limited and costly parking), particularly in downtown areas. In addition, the central parts of the largest MSAs have relatively high development densities and more extensive transit service, making transit more competitive with the private auto. Finally, a large share of poor and minority households live in the central parts of the largest MSAs. Therefore we include measures of metropolitan size and density. Access to transit service is of course a necessary condition for using transit. Distance to stops and transit headways are typical measures of transit availability. Since we have no information on headways, we consider only distance to the nearest transit stop.

It is argued that complex travel patterns — making several trips per day and combining trips into multi-stop tours — are dependent upon the private vehicle. Therefore multiple stop should reduce the likelihood of transit use. Strathman and Dueker (1995) found that commuters were reluctant to use transit because of the stops they made on the way to or from work. Trip scheduling may affect transit use — travel late at night or on weekends is unlikely to be made by transit.

Prior literature shows that attitudes are important predictors of travel behavior. As we noted earlier, fear of crime or other negative perceptions of transit may prevent transit use. It is unfortunate that the NPTS data preclude our consideration of attitudes.

Model Form

The dependent variable constructed from the survey responses is categorical (see Table 3-12 in previous chapter), hence OLS regression is not appropriate. We estimated three different model forms.

Model 1: Binary Logistic

The first model is a simple binary logistic regression model, where the dependent variable is simplified to “transit user” (anyone who uses transit bimonthly or more frequently) and “not a transit user” (everyone else). An individual is able to use transit only if it is available. In one sense we are modeling a conditional choice (using transit given that it is available). We do not consider a conditional choice model, because transit availability is determined by residential location — a long-term decision. We are more interested in the choice of using transit when it is available, so we restrict our model to the transit use choice. We therefore exclude from our sample those who do not have access to transit.

The binary model estimates the probability that an individual is a transit user as a function of the four groups of factors discussed above. It has the following form:

(1)

or equivalently

(2)

where

estimated probability of being a transit user

independent variables

estimated coefficients

This functional form guarantees that will always be a number between 0 and 1. The functional form assumes independence among the observations and extreme value distributed error terms. The logistic model is estimated via maximum likelihood, and model significance is tested via the likelihood ratio test. In logistic regression, the estimated coefficients can be interpreted as the change in the log odds associated with a change in the independent variable of one unit.

Model 2: Multiple Category Logistic

Although the binary logistic model is relatively easy to estimate and interpret, it does not take advantage of the three categories of transit users. It seems quite reasonable that people who use transit regularly may differ in some significant ways from people who use transit occasionally. In this case we have what we may define as an ordered categorical dependent variable, since the categories can be rank ordered from highest (regular user) to lowest (not a user). There are two possibilities for estimating models with ordered categorical dependent variables. The first is a multiple category logistic model. This is an extension of the binary logistic model that takes advantage of the IIA (Independence of Irrelevant Alternatives) property of the logit model, e.g., that the ratio of choice probabilities for any two alternatives is independent of the probabilities of all other alternatives in the choice set.

Taking equation (1) and expanding to R categories,

= r = 1, 2, ¼ R (3)

We estimate binary models for each category pair, using one category as the reference category in each case. For three categories, two non-redundant logits can be formed:

(4)

(5)

where,

intercept

to = independent variables

to regression coefficients of regular user

to regression coefficients of occasional user

The resulting estimated coefficients are interpreted with respect to the log odds of the pairwise comparison. For example, if never using transit is the reference category, the coefficients of the frequent user equation tell us the effect of the given independent variable on the probability of being a frequent user relative to the probability of never using transit.

Model 3: Ordered Logit

Another way of modeling an ordered dependent variable is to consider the choice process as

(6)

As in the previous choice models, is unobserved. Rather, we observe

if

if

if (7)

The are unknown parameters to be estimated from the . Depending upon our assumptions on error terms, we obtain a probability model with the general form of

(8)

where, = cumulative probability for the j^th category

= threshold for the j^th category

= predictor variables

= regression coefficient of the predictor variable

The thresholds or constants in the model, (corresponding to the intercept in linear regression models) depend only on which category’s probability is being predicted. Rather than predicting the actual cumulative probabilities, the model predicts a function of those values. This function is called the link function, which is a transformation of the cumulative probabilities that allows estimation of the model. We tried both the logit and probit link functions. The probit function assumes normally distributed errors with mean of 0, a strong assumption for non-linear data. We therefore decided to use the logit function. With the logit function, the probabilities are,

Logit () = log (9)

Testing for Income Effects

How should income be incorporated into the model, given that we are interested in transit use among the low income population? For the binary model, we can simply include an income dummy variable. Including such a variable implies that income only has a scale effect (increasing or decreasing the probability of being a transit user). However, it is also possible that interaction effects may exist. For example, low income may have more effect on using transit for women than for men. Results of prior studies do not preclude such effects, hence a correctly specified model should consider them. There are two ways to test for joint effects. One way is to estimate separate models for each income group and test for differences between coefficients. From equation (1) we specify the exponential of as

for low income,

for not low income, (10)

and we test whether for each .

The second way is to estimate a single model,

(11)

where independent variables

low income dummy variables

independent low income interaction variables

tests the independent effect of low income, and the tests the joint effect of low income with each of the independent variables. The two methods are equivalent. Estimating equation (10) separately for each group generates the same coefficients as equation (11) for the base group (in our case the not low income group). For the binary model, we use the second method.

For ordered logit, there is a third alternative. It is possible to test for differences in variability of the independent variables, as for example if there is more variation in transit use among low income households. Since we have no reason to expect such differences in variation, we restrict our tests to the independent and joint income effects, using a model of the form of equation (11).

Data

The data for this analysis were drawn from the original sample of 93,560 persons. This sample yielded 48,546 valid cases for the transit use analysis. Figure 4-1 illustrates the process of filtering data. The first filter was based on the household survey question, “is transit available in your town or city?” The second filter was based on the question regarding usual behavior.

This question was asked only of persons 16 years old or older, and only of persons completing their own questionnaires. This left a sample of 50,035 observations. Additional missing data on key variables further reduced the sample, ultimately yielding 48,546 observations distributed across the three transit use categories as shown in the bottom panel of Figure 4-1. Variable descriptions and definitions are given in Table 4-1.

Figure 4-1: Development of Transit User Analysis Sample

Table 4-1: Variable Description

Variable Name	Description
Dependent variables
TUSER	1= transit user, 0= non user
PTUSED	1= non user, 2= occasional user, 3= regular user
Demographic characteristics
AGO	1= age 65 and older
BLACK _M	1= black male
BLACK_F	1= black female
HISPANIC_M	1= Hispanic male
HISPANIC_F	1= Hispanic female
HHSIZE1	1= single person household
Economic characteristics
LOWINC1	1= Low income household member
NOEMP	1= retired or not employed
NOCAR	1= no cars and no drivers in household
MORECAR	1= more cars than drivers in household
Geographic characteristics
LARGMSA	1= Living in MSA size with more than 3 million population
LOWDENS	1= Living in a census tract with population density < 500 persons/mi^2
HIGHDENS	1= Living in a census tract with population density >10,000 persons/mi^2
S_DIST	1= Access to transit within 0.5 mile of home
Trip characteristics
FRQ	Total trips/day

Results

We present results for Model 1 and Model 3. Model 2 results are summarized in Appendix 4B.

Model 1 Results

Table 4-2 gives results for the binary model with the full sample and low income dummy variable. The table gives coefficient values (first column of numbers) and their standard errors (second column of numbers). Coefficients in bold are significant at p <.05. The overall model is significant. We provide the Cox & Snell (Psuedo-R) R-square, and percent of observations correctly predicted. However, these goodness-of-fit measures should be used with caution because of the large sample size and because we are using weighted data.

The coefficient of the low income dummy is significant and of the wrong sign, though of relatively small magnitude. Older age is significant and negative, also contrary to expectations. The result suggests that controlling for all other factors (including car ownership), the elderly are less inclined to use transit. These issues are further discussed in a later section. Also contrary to expectations, single person household status is not associated with the likelihood of being a transit user. We used several different combinations of household composition variables in other models (not shown here), and none of them were consistently significant. Apparently household composition is not a significant factor, once other demographic, economic and demographic factors are controlled.

The coefficients for Blacks, and Hispanics, both male and female, are significant. The value of the two Black coefficients suggests that sex is not significant, while the opposite is the case for Hispanics. We used joint race/sex variables in order to test for such differences. When we estimate models with separate variables for sex and race, the coefficient for sex is not significant (results not shown).

The coefficient for employment status is significant and positive, suggesting that those who are not employed are more likely to be transit users. The coefficients for the car access variables are strongly significant. Having no car in the household is the single greatest predictor of being a transit user, as expected. Having more cars than drivers in the household is associated with less likelihood of being a transit user, also as expected.

All of the coefficients for our geography measures are significant and of the expected sign. The largest MSAs have the most transit service, and the relative attractiveness of transit is increased by the scarcity and price of parking and by the availability of commuter express services. Note that our sample contains only those people who stated that transit is available to them. Hence these results do not reflect differences in the availability of transit across MSAs or neighborhood density.

Our transit access variable coefficient also has the expected sign. People who live close to a transit stop are more likely to be transit users. In contrast, our measure of complex travel, trips per day, has a significant coefficient with unexpected sign. This issue is further discussed below.

Table 4-2: Binary Model with Low Income Dummy

Variable	B^a	S.E.
Constant	-2.466	.049
Demographic
Age  65	-0.526	.056
Single person HH	0.002	.042
Black female	0.482	.051
Black male	0.425	.058
Hisp female	0.229	.060
Hisp male	-0.157	.062
Economic
Low income	-0.073	.034
Not employed	0.087	.038
No cars	2.246	.061
Cars > drivers	-0.347	.050
Geography
Large MSA	0.745	.032
High density	0.974	.039
Low density	-0.474	.063
Transit access
Stop within .5 mi	0.518	.038
Travel
Trips/day	0.021	.005
N	34442
- 2 Log Likelihood	29450.89
Psuedo-R	0.177
Percent correct	81.5

^aBold = sig. at p < .05

Table 4-3 gives results for the binary model with joint interaction terms. This model was estimated using a 60 percent random sample of the data, as computer memory limits precluded using the entire sample. Even with the addition of the joint interaction terms, the coefficient of the low income dummy remains negative and significant. Ten of the 14 interaction term coefficients are significant, indicating that there are many differences between the two groups. As noted in the previous section, the independent effect variable coefficients are equivalent to the coefficients of the “not low income” group, had we estimated the model separately for each group. The interaction term coefficients are therefore the difference between the effects for the two groups. The sum of the two coefficients (the independent plus the interaction for a given variable) corresponds to low income group coefficients. For example, older age has a significantly more negative effect on the likelihood of being a transit user for the low income group than for the not low income group. This may reflect the low rates of trip making among the low income elderly. Not only are they less likely to be transit users, they are less likely to travel at all. Continuing with the race/sex variables, we note that the coefficients are different between the income groups in every case. It is the interaction of poverty and race that is associated with higher likelihood of transit use for Blacks of both sexes and for female Hispanics.

Low income persons who are not employed are more likely to be transit users than others, suggesting greater transit dependency among those not employed. There is no difference in the effect of car ownership between the two groups. Turning to geography, the effect of living in the largest MSAs has a weaker effect for the low income group, but residential density has a greater effect. The density results suggest that low income persons are more sensitive to whatever factors density measures. However, the causality could be in the opposite direction, e.g., transit dependents are more likely to live in high-density neighborhoods and less likely to live in low-density neighborhoods. The same argument could be made for the result on transit access. In Chapter Three we noted that low income persons live closer on average to a transit stop than not low income persons. Finally, the trip measure is significant and negative for the low income group, as expected.

Table 4-3: Binary Model with Joint Effects

Base = Not low income

	Independent effects		Low income
Variable	B^a	S.E.	B	S.E.
Constant	-2.414	.074
Low income dummy	-0.402	.150
Demographic
Age  65	-0.283	.093	-0.605	.147
Single person HH	-.067	.068	-.155	.117
Black female	0.209	.091	0.750	.138
Black male	0.156	.099	0.660	.155
Hisp female	-.486	.116	1.050	.170
Hisp male	-.226	.105	.262	.163
Economic
Not employed	-.047	.063	.514	.101
No cars	2.038	.140	.201	.171
Cars > drivers	-0.331	.071	-0.004	.164
Geography
Large MSA	0.901	.050	-.471	.091
High density	.834	.064	0.446	.107
Low density	-.206	.093	-0.403	.184
Transit access
Stop within .5 mi	0.424	.055	0.510	.120
Travel
Trips/day	.021	.008	-.044	.015
N	20219
- 2 Log Likelihood	17450.83
Psuedo-R	.183
Percent correct	81.9

^aBold = sig. at p < .05

Model 3 Results

As with Model 1, we estimate two versions of Model 3, a model with low income dummy, and a model with joint interaction terms. Results for the first ordered logit model are given in Table 4-4, and results for the second are given in Table 4-5. Overall the results are consistent with the binary model results. In both cases, the ordered model does a good job of distinguishing between levels of transit use, as indicated by the significance of the threshold coefficients and their different values with respect to one another. The model in Table 4-4 shows the low income dummy coefficient as significant and again of the wrong sign. We suspect that this result may be due to the correlation between car ownership and income. As before, older age is associated with less likelihood of being a regular transit user. The coefficients for Blacks of both sexes and for Hispanic females are significant and positive. Note that the coefficient values indicate that for Blacks, race is the key factor, not sex. As in the binary models, not having a car is the most powerful predictor of transit use; the negative effect of having more cars than drivers is much smaller than the positive effect of having no cars. All the geography variable results are as expected and are consistent with the binary model. It bears noting that we have already controlled for transit being available, hence these results indicate that among those for whom transit is available, living in a large MSA and/or in high-density residential areas is associated with a higher probability of being a regular transit user. And, all else equal, having a transit stop nearby increases the probability of transit use. As in the binary model, our trip frequency variable coefficient is significant and positive, contrary to expectations.

Table 4-4: Ordered Logit Model with Low Income Dummy

Variable	B^a	S.E.
Threshold 1	2.446	.049
Threshold 2	3.346	.051
Demographic
Age  65	-0.531	.054
Single person HH	-0.036	.041
Black female	0.521	.048
Black male	0.444	.055
Hisp female	0.243	.057
Hisp male	-0.123	.060
Economic
Low income	-.073	.033
Not employed	0.052	.037
No cars	2.380	.054
Cars > drivers	-0.371	.049
Geography
Large MSA	0.749	.032
High density	1.033	.037
Low density	-0.476	.063
Transit access
Stop within .5 mi	0.539	.038
Travel
Trips/day	0.015	.005
N	33651
- 2 Log Likelihood	15870.39
Psuedo-R	0.205

Results for the ordered logit joint interaction model (Table 4-5) are similar to that of the comparable binary model. There is no low income dummy variable; independent low income effects are captured in the threshold variables. Most of the interaction coefficients are significant, again indicating that there are many differences between the two income groups with regard to using transit. Many of the coefficients of the demographic variables are different. As with the binary model, the negative effect (e.g., reducing likelihood of using transit) of age is greater for low income persons, as is single-person household status. The effect of race is more positive for low income Blacks and Hispanics. Again, it is the combination of low income and race that matters. Not being employed increases the probability of using transit for low income persons, but not for others. The effects of car ownership are the same across both groups.

The results on the geography variable coefficients are also similar to the binary model results. Living in the largest MSAs has a less positive effect, while living in a low-density neighborhood has a more negative effect. As before, access to a transit stop has a more positive effect for the low income group. We noted above that these results may be indicative of more transit dependency and hence greater likelihood to live near a transit stop, regardless of MSA size. In contrast, the not low income group, who we presume are largely choice riders, are more sensitive to service quality, and therefore more likely to use transit where it is most convenient. Also as before, the trip frequency variable coefficient is significant and negative for the low income group, but significant and positive for the higher-income group.

Table 4-5: Ordered Logit Model with Joint Effects

Base = Not low income

	Independent effects		Low income
Variable	B^a	S.E.	B	S.E.
Threshold vars
Threshold 1	2.531	.049
Threshold 2	3.427	.051
Demographic
Age  65	-0.248	.072	-0.535	.109
Single person HH	.082	.051	-.266	.085
Black female	0.324	.069	0.461	.098
Black male	0.206	.076	0.610	.113
Hisp female	-.160	.082	.790	.117
Hisp male	-.233	.081	.284	.121
Economic
Not employed	-.075	.049	.261	.074
No cars	2.374	.100	-.037	.119
Cars > drivers	-0.384	.055	.049	.122
Geography
Large MSA	0.920	.038	-.501	.064
High density	1.009	.047	.069	.076
Low density	-0.257	.073	-0.684	.141
Transit access
Stop within .5 mi	0.443	.042	0.389	.069
Travel
Trips/day	.029	.006	-.044	.009
N	33651
- 2 Log Likelihood	15633.81
Psuedo-R	.211

^aBold = sig. at p < .05

Conclusions on Frequency of Transit Use

The results here are mostly consistent with the literature. First, for the entire sample, demographic, economic, and geographic factors all affect the likelihood of being a transit user. Contrary to the literature, older age is associated with a lower probability of being a transit user, although we know that the elderly are often transit dependent. We think our results are a function of the dependent variable — how often people use transit — and likely reflect the lower propensity to travel by any mode among the elderly. Other model specifications not shown in this report revealed that sex by itself was not significant. We noted in our literature review that recent research has indicated mixed results on women’s use of transit. Race/ethnicity is positively associated with the likelihood of using transit, even when economic status and geography are taken into account. As expected, car availability is a powerful predictor of transit use. We noted that there is an element of interdependency here, since those who prefer to use cars are more likely to have them, and those who prefer to use transit are less likely to have them. However, given overwhelming preferences for auto travel, this effect is likely to be rather small. The geography and transit access variables performed as expected.

The joint interaction models showed that there are differences in the relationships between demographic factors and probability of transit use between the two income groups. Race/ethnicity effects are more pronounced within the low income group, suggesting that it is the intersection of poverty and race that is associated with difference transit use patterns. There are also differences in the effect of geography, with residence in large MSAs associated with greater likelihood of being a transit user for those with higher incomes. We noted that this is likely a choice rider effect, with transit a relatively attractive option for commuters in the largest MSAs. Living in low-density residential areas has a more negative effect for the low income group. This is difficult to interpret. We might speculate that this is a function of the greater propensity of the low income group to be regular transit users. This is consistent with the strong effect of access to a transit stop for the low income group.

Our efforts to capture the effect of complex travel behavior were not effective. The trips per day variable was either not significant or had the wrong sign. Part of the problem is the small number of people who are regular transit users, and, among them, the lack of variability in trip making. We thought that the problem was that the measure was too gross; we wanted something to capture chained or sequenced trips. However, measures of chained trips or the number of tours made per day did not perform any better. Perhaps trip chaining is simply not a significant factor in usual transit use.

TRANSIT USE FOR COMMUTING

In this section we use data from the day trip file to examine transit use for the work trip. Our analysis is based on home-to-work and work-to-home trips which include all stops made between home and work and between work and home.

MODEL DEVELOPMENT

We follow a similar process to develop and estimate models of transit use for the journey to work. In this case, the modeling task is straightforward, as we wish to model the choice of using transit for the trip to/from work. We are concerned only with the choice of whether or not transit is used. We have no information on the alternatives available to each individual, so we cannot estimate a modal choice model. This is a simple binary choice, and the logistic model presented in the previous section is appropriate.

What are the factors that may influence transit use for the work trip? The extensive literature on this topic suggests four groups of factors:

1. Demographic characteristics

2. Level of service and availability of modes

3. Residence and work location

4. Travel and schedule characteristics

Demographic characteristics include sex, race, age, and household composition. Women are more likely to work closer to home, to work part-time, and have more household responsibilities than men. In addition, in households where the number of drivers exceeds the number of cars, the male is more likely to have access to the car. These considerations lead to mixed expectations. To the extent that women have more binding schedule constraints, we expect lower probability of transit use. To the extent that women have lower wages (associated with part-time work) and less access to cars, we expect higher probability of transit use.

Prior research shows that Blacks use transit at higher rates than any other race/ethnic group. Hispanics and Asians also use transit more than Whites. We also expect transit use to be associated with age: younger workers are more likely to have lower wages and hence may be more inclined to use transit. Older workers are likely at the peak of their earning years, and therefore may be less inclined to use transit.

We noted in the previous section that household composition is important, because travel decisions are made at the household level. For the journey to work, the circumstances and responsibilities of each worker may affect modal choice. Single persons with children generally have the lowest household income and therefore are likely to be transit dependent. Households with more than one worker and with children have higher incomes, more complex family schedules, and more access to cars. These households are less likely to use transit for the journey to work.

Level of service and availability of modes includes car availability, transit access, and transit availability. Our previous analysis showed that not having a car was the most powerful predictor of being a regular transit user. Few employed people live in households with no cars, but for those few, we expect transit use. Conversely, those in households with high car access are not likely to use transit. Our previous analysis also showed that access to a transit stop was a significant predictor of transit use. We expect the same here.

As noted in our literature review, the availability of transit for low-wage workers has become a major policy issue. The argument is that low-wage workers often have work schedules that require off-peak commuting, and their commutes are often in the reverse direction. We use the start work time as an indicator of commuting schedule. If a person starts work outside of the traditional AM peak period, we expect less likelihood of using transit.

Our previous analysis also showed that living in a high-density neighborhood, or living in the largest metropolitan area, is positively related to being a regular transit user. We expect the same results for commuting. In addition, we know from previous research that commutes to jobs in the central city are more likely to be made by transit. Transit systems are oriented to central city commutes, and the often-high price of parking in central cities provides a disincentive for car commuting.

Finally, there is the issue of complex travel patterns and household schedule constraints. If an individual has many responsibilities and schedule constraints, we hypothesize that it becomes more difficult to use transit. In the previous analysis, we used the simple measure of total trips per day, and our results were unsatisfactory. In this case we have more choices, since we have data on the actual trip made, and on the sequence of trips included in both the trip from home to work, and the trip from work to home. We therefore use the total number of stops in the home to work and work to home chains as our indicator of complex travel patterns. In addition, we use part-time work as an indicator of a possibly irregular work schedule. The list of variables used in our model is given in Table 4-6.

DATA

The data for the logistic model estimations were also drawn from the original sample of 93,560 persons. This sample yielded 11,709 valid cases for analysis, after screening for those who made a work trip on the travel day, had transit available, and who had information on whether or not their job was located in the central city. Figure 4-2 illustrates the process of screening the data. About