Table of Contents

LITERATURE REVIEWS AND CASE STUDIES

A literature review conducted by the European Commission on Transportation Research (1996) provides an extensive list of variables that should be considered in evaluating the success of transit ridership enhancement projects (see See Direct and Indirect Strategies for the Evaluation of the Successes of Transit Ridership Project in the Study by European Commission Transportation Research (1996)). The authors begin by categorizing the variables as either direct or indirect strategies. Direct strategies are those that transit agencies can pursue to increase efficiency and effectiveness of transit operations and are roughly equivalent to the internal factors discussed above. These strategies include changes in the fare levels, service quality and quantity, marketing, facilities, and technologies employed in the provision of service. Indirect strategies are generally broader public policies that influence ridership, but over which transit agencies generally do not have control.

This study concludes that most direct strategies have little impact on public transit's modal share, and thus need to be implemented in concert with indirect measures to successfully influence transit use. More specifically, the authors conclude that increasing both service frequency and transit stop density of bus stops in combination with road pricing would increase transit patronage more than any other combination of public policy actions.

An Urban Mass Transportation Administration (UMTA) report conducted by Sale (1976) examines the factors influencing transit ridership growth by analyzing the techniques used to increase ridership by more than 5 percent on transit systems in seven U.S. cities between 1971 to 1975. Sale finds that most ridership gains are in large part attributable to service expansion -- especially the route expansion in rapidly growing metropolitan areas. In addition to service expansion, Sale notes three other important factors that have a significant effect on transit mode share in the short term: strong public and political support, resulting in the availability of substantial and stable financial resources; stable or declining fare levels; and higher motor vehicle fuel prices due to the energy crisis (Sale 1976).

Cervero (1993) conducted a literature review to examine the characteristics of rail-station-adjacent housing and commercial projects thought to influence transit ridership. He finds transit use varies significantly by proximity to transit lines and stations. He cites a study of Washington, D.C., showing that the share of trips by rail and bus declines by approximately 0.65 percent for every 100-foot increase in distance of a residential site from a Metrorail station. Ridership also declines steadily as distance between stations and offices increases. These findings imply that increasing service network densities to decrease the average distance from residences and workplaces to transit stations and stops would significantly increase transit use.

SURVEYS OF AND INTERVIEWS WITH TRANSIT MANAGERS

In some studies, transit system managers were interviewed to find out what factors they thought had the greatest influence on ridership. Although perceptions are just that, managers of transit systems are in a good position to consider the relative influence of various factors on patronage.

The Transit Cooperative Research Program Research Results Digest (1995, 1998) provides results of two extensive interview studies. Each interviewed the managers of about 25 to 50 transit agencies, producing similar findings. The transit agencies were selected on the basis of increasing ridership, and the interviews were conducted by telephone. Most transit managers interviewed attribute ridership increases to a various combination of strategies, programs, and initiatives in five general categories: (1) service adjustments, (2) fare and pricing adaptations, (3) market and information initiatives, (4) new planning orientation, and (5) service coordination, consolidation, and market segmentation.

The respondents frequently mention the use of deep discount fare policies to help increase ridership as well as efforts to make passes more widely available in communities, strategies from the second category (fare and pricing adaptations) and the fifth category (service coordination, consolidation, and market segmentation). Deep discount fare policies stratify transit markets into segments based on two primary factors: frequency of use and sensitivity to cost (Fleishman 1993). Such policies generally offer a per-ride discount for the purchase of a multiple-ride pass or transit card, aiming to induce potential riders with low usage and high price sensitivity to increase overall transit patronage.

The interviews also indicate a consensus among transit managers that external factors, such as population change, new development, and regional economic conditions, probably have a greater effect on ridership than system and service design initiatives. One conclusion of the 1995 study is that because mode choice decision is strongly influenced by vehicle ownership and the private vehicle is overwhelmingly preferred by many travelers who have the choice, then strategies that target transit service alone have little chance of being very effective.

Some transit systems have found that they can increase their ridership by selling discounted transit passes in bulk to large groups. University students are a group that is willing to purchase transit, and since they are more likely to ride during off-peak periods than the general transit-riding public, transit systems do not need to increase service to accommodate university students. Brown, Hess, and Shoup (2001) report the results of a survey of university transit pass programs at 35 U.S. universities. The university typically pays the transit system an annual lump sum based on expected student ridership, and students show their university identification to board the bus. University administrators report that transit pass programs reduce parking demand, increase students' access to the campus and the community, help recruit and retain students, and reduce the cost of attending college. Transit system officials report that university transit pass programs increase ridership, fill empty seats, improve transit service, and reduce the operating cost per rider. Increases in student transit ridership ranged from 71 percent to 200 percent in the first year of university transit pass programs, and annual growth in subsequent years ranged from 2 percent to 10 percent. The universities' average cost for transit pass programs is $30 per student per year. The authors report that the 35 university transit pass programs examined during the 1997-1998 school year provide fare-free transit service for 825,000 people, but since this is only 6 percent of the 14 million students enrolled in U.S. universities, the opportunity for growth is enormous.

STATISTICAL ANALYSES OF A TRANSIT AGENCY OR REGION

The studies in this group use statistical methods, such as correlation and regression analyses, to examine the relationships between transit ridership and potentially influential factors. Compared to studies in the previous two groups, these statistical analyses can not only identify the factors thought to affect ridership, but also attempt to measure the level of influence in a comparative fashion. The common approach of these studies is to use multiple regression analysis to analyze the combined effects of a variety of factors on transit use.

Using the data in Portland, Oregon, Liu (1993) constructs a variety of regression models to explain the variation in transit ridership each decade from 1960 to 1990. To test the widely held notion that declining transit use is largely a function of increasing personal income, auto ownership, and suburbanization of residence and job locations, Liu produces a model to estimate per capita transit trips as the function of the following factors:

Per capita transit capacity

Per capita passenger car registrations

Per capita transit subsidies

Per capita income

Percent of population residing in the central city

Metropolitan area population

Motor vehicle fuel prices

A time-trend variable for a period 1929-90

Annual total transit miles

Average passenger fare

Total employment in the Portland metropolitan area

The effects of World War II.

He finds that per capita income, auto ownership, and the suburbanization of both jobs and housing have significant effects on transit ridership. In a similar analysis for the period from 1949-1990, Liu finds that the size of the central city population also has a significant effect on ridership.

Using data for the period from 1971 to 1990, Liu estimates the following regression model:

Δ (Linked Trips) = -0.008
+ 0.606 Δ (Revenue Hours of Service)
- 0.285 Δ (Average Fares)
+ 0.861 Δ (Regional Employment)
+ 0.274 Δ (Real gasoline prices)
(Δ s are all annual percentage changes)

Kain and Liu (1995) produce similar results in their study of the San Diego and Houston transit systems in the early 1990s. They chose San Diego and Houston because the transit systems in these cities were adding riders during an economic downturn when most transit systems were losing riders. Using data for the period from 1968 to 1992, Kain and Liu find much of the increased ridership could be attributed to the number of revenue vehicle miles of service, average fares, regional employment levels, car ownership levels, and gasoline prices -- in other words, to a combination of internal and (primarily) external factors.

Chung (1997) estimates the effects of employment, development levels, and parking availability on Chicago Transit Authority (CTA) rapid transit ridership for the period from 1976 to 1995 in Chicago, controlling for fare policy and service levels. Chung finds that parking availability, development, and employment had greater impacts on ridership than fares, although the array of variables considered in this study was considerably less comprehensive than those used in the studies by Kain and Liu.

McLeod, Flannelly, Flannelly, and Behnke (1991) estimate multivariate time-series regression models of transit ridership based on the aggregate data for the period from 1956 to 1984 in Honolulu, Hawaii. Their models, using revenue trips and linked trips as dependent variables, include five independent variables: civilian jobs, inflation-adjusted per capita incomes, inflation-adjusted fares, the size of the transit fleet, and a variable accounting for service disruptions due to strikes. Although both internal and external factors influenced ridership, other factors thought to be important -- the number of tourists, the number of registered passenger vehicles, and gasoline prices -- were not.

Gomez-Ibanez (1996) analyzes the changes in ridership and increases in deficits for the Massachusetts Bay Transportation Authority (MBTA) in Boston in the late 20th century. He estimates the effects on ridership of both internal (fare and service policies) and external (income, demographics, and others) factors in regression models. He produces one model that predicted ridership change based on two external factors (income and employment) and three internal factors (fare, revenue vehicle miles, and a dummy variable for a 1980-81 severe budget crisis). The model predicts an 11.9 percent increase in ridership between 1970 and 1990; the actual increase was 11.8 percent. A second model, using a simple time trend for income, predicts a 9.9 percent ridership increase.

Gomez-Ibanez's models show that, at least in Boston, transit ridership is strongly affected by external factors beyond the transit agency's control. He calculates, for example, that each percentage decrease in central city jobs reduced MBTA ridership by 1.24 to 1.75 percent, and each percentage increase in real per capita income reduced MBTA ridership by 0.7 percent. The effects of fare and service policies are, by contrast, relatively small. A 1 percent increase in service increased ridership by only 0.30 to 0.36 percent, and a 1 percent reduction in fares increased ridership by 0.22 to 0.23 percent.

A subset of studies in this category examines the effects of land use and urban form on ridership using statistical methods. In general, these studies find that decentralized residential and occupational locations are difficult to serve by public transit because transit works best when a large number of people are all headed to activity nodes that contain various destinations. Dense, compact development is more conducive to efficient transit operations than dispersed and sprawling patterns of urban development.

In an analysis of transit demand in Portland, Oregon, Nelson and Nygaard (1995, cited in TCRP 1996) note that the overall housing and employment density per acre are two of the most significant determinants of transit demand among the 40 land use and demographic variables studied. These two variables alone explain 93 percent of the variance in transit demand. Similarly, Pushkarev and Zupan (1977) find that residential densities in transit corridors, together with the size of the downtown and the distance of stations from downtown, explain the level of demand for a variety of transit modes.

The Transit Cooperative Research Program (1996), using a variety of sources, analyzes the relationships between urban form and transit ridership. The authors find that residential densities have a significant influence on rail transit ridership, as does the size and density of the central business district (CBD), although the influence was found to be greater for light rail ridership than for commuter rail ridership. The study also finds that, for a 25-mile light rail line surrounded by low-density residences, increasing downtown employment from 50,000 to 300,000 for a 3-square mile CBD could increase ridership along that corridor from 18,000 to 85,000 daily boardings. Beyond a certain size, however, CBD size is not found to be important.

The TCRP study also finds that the effects of density are interrelated with employment center size, "corridor-level" urban structure, transit service characteristics, and a variety of public policies. Lastly, the types and mix of land uses influence the demand for transit as well as the use of nonmotorized modes. However, it was difficult to sort out the effects of land use mix and urban design because they are so strongly correlated with density. An analysis of travel behavior in 11 metropolitan areas surveyed in the 1985 Housing Survey suggests that both land use mix and residential densities contribute to the probability of choosing transit in mode choice decisions. The authors find that the overall level of density is more significant than the mix of land uses. Land use mix has only one-tenth as much influence on transit choice as density.

Spillar and Rutherford (1998) examine the effects of urban residential densities and income on transit ridership in five western U.S. cities -- Seattle, Portland, Salt Lake City, Denver, and San Diego -- using 1980 Census data. The data include total population counts within a given geographic area, average annual income levels in that area, and the average area in acres of each zone. Since the data were drawn from the Census, Spillar and Rutherford examine only work-related trips. They find that transit use per person grows with increasing density up to a ceiling of between 20 and 30 people per acre, which is equivalent to 0.1 to 0.2 transit trips per day per person. In terms of income, density exhibits less effect on transit use in higher-income neighborhoods (those with less than 18 percent low-income families) than in low-income areas, although the sample size analyzed was rather small.

Since car ownership, car use, and transit use are all related, a change in one variable affects other variables; however, the magnitude of effect may not be symmetrical in terms of direction. Kitamura (1989) examines the causal relationships between car ownership, car use, and transit use using surveys and trip diaries given to nearly 4,000 people in the Netherlands. He finds that a change in car ownership leads to a change in car use, which influences transit use. Conversely, he finds that significant changes in transit use are usually related to changes in car use or car ownership.

Strategies to price parking can be an effective means of increasing transit patronage for the work trip (Dueker, Strathman, and Bianco 1998). Since increasing parking costs affects relative attractiveness of traveling by transit compared to driving an automobile, it has significant effects on mode share. In 1998, in TCRP Report Number 40, a quantitative analysis of mode choice and finds (1) the probability that people pay to park is likely to influence transit share more than either transit frequency or transit accessibility, (2) transit frequency has more significant effects on transit mode share than transit accessibility, and (3) pay-to-park probability and transit frequency combined have the greatest effect on transit share. The study finds that transit share increases nearly 300 percent, from 6.5 to 24.5 percent, when transit frequency doubles from 1.0 transit revenue hours per capita to 2.0, and when the pay-to-park probability doubles from 0.05 to 0.10. The study also estimates that increasing access to a transit stop from 30 percent of the population to 60 percent increases transit use only from 8.6 to only 9.3 percent. By comparison, an increase from 10 to 15 percent of the population that expects to pay to park at work is estimated to increase the transit share from 21 to 34 percent.

The San Francisco County Transportation Authority conducted a travel study in 1995 and finds that, when parking costs exceeded transit fares by 20 to 30 percent, commuters tend to take transit rather than drive alone. The study also finds that 47 percent of the employees who drove alone report that they either park free or are provided employer-paid parking (cited in TCRP 1980).

Cervero (1990) reports that riders are generally more sensitive to changes in service than they are to changes in fares. In other words, riders are more easily attracted by service improvements than fare decreases. A study by Syed (2000) supports Cervero's findings. Syed conducts a factor analysis of the determinants of increasing transit ridership at the Ottawa-Carleton Transportation Commission (OC Transpo) using survey data on 47 variables for each of 2,000 transit riders. This analysis focuses on factors that users of the system judge the most important. Syed finds bus information is the most important factor among eight underlying factors in determining transit trips. Based on the factor analysis of the survey, Syed finds that the following factors were the most important factors in determining ridership: bus information, on-street service, station safety, customer service, safety en-route, reduced fares, cleanliness, and general transit operator attitudes. Because Syed combines the many original factors from the survey into a smaller number of categories, it may be difficult for transit agencies to implement any of the measures evaluated in the study with certainty of the probable outcome. For example, Syed lumps "on-street service" into one category that includes such aspects as on-time performance, system expansion, and frequency of service.

CROSS-SECTIONAL STATISTICAL ANALYSES

Cross-sectional statistical analyses are premised on the idea that there are underlying structural relationships between factors influencing transit use. The collection of detailed transit operator data by the Federal Transit Administration (FTA) in the National Transit Database (NTD) permits comparative analysis of transit systems. Hartgen and Kinnamon (1999) develop comparative statistics for the nation's largest urban bus transit operators from nationally reported data for 1988 through 1997. Four measures of resources (vehicles, population base, fare revenue, and coverage area) are normalized and compared with seven outcome measures (operating expenses per mile, operating expenses per hour, operating costs per passenger, operating costs per passenger mile, vehicle miles of service, vehicle hours of service, and ridership). Systems are ranked according to overall performance against U.S. averages, weighting each statistic equally. Systems then are ranked within six peer groups based on population served and modes of service. Hartgen and Kinnamon find that the overall performance of bus transit systems steadily declined during their study period; only two of the 12 measures of performance improved from 1988 to 1997. They find evidence that service in general expanded: service coverage was 11 to 14 percent greater in 1997 versus 1988. However, costs per vehicle hour rose through 1997, 2 percent more than 1996 and 32 percent more than 1988. The 10 top-ranked systems for 1997 were Santa Monica, CA, Champaign-Urbana, IL, Tucson, AZ, Santa Barbara, CA, Milwaukee, WI, Long Beach, CA, Las Vegas, NV, Shreveport, LA, Durham, NC, and Newport News, VA. The study concludes that cost-effective performance depends on low unit costs, low fares, and low subsidies, with concentrated service that optimizes service utilization.

Kain and Liu (1996) conduct detailed analyses of factors that determined the level of transit ridership using the data for 184 systems over a 30-year period from 1960 to 1990. Kain and Liu essentially conduct two different econometric analyses. First, they estimate regression models for changes in ridership for the periods 1960-70, 1970-80, and 1980-90 using variables such as fare levels, revenue miles of service supplied, the rail share in revenue miles, whether the system was publicly or privately operated, and a vector of control variables (population and employment, density, area, fraction of carless households in the area). Because many of the control variables are highly correlated, only a few of them were included in each regression model. All models of ridership changes between 1980 and 1990 had R2 = 0.75 or higher.

Second, Kain and Liu estimate cross-sectional regression models for the level of ridership for four different years -- 1960, 1970, 1980, and 1990 -- using transit fares, service levels, service types, public or private ownership, and a vector of exogenous or control variables (again, only a portion of which could be included in each regression model). All models for 1990 had a high explanatory power of at least R2 = 0.95.

The results indicate that the mean fare elasticities for ridership changes during the 1980-1990 and 1970-1980 periods and the 1990 and 1980 cross section models range between -0.34 and -0.44, and that the mean revenue mile elasticities range between 0.70 and 0.89. These results imply that transit agencies will increase ridership less by reducing fares than by increasing service, although both changes are likely to reduce overall transit system performance. Since this study focuses more on the effects of four specific policy variables -- transit fares, service levels, service types, and public or private ownership -- it is not clear how explanatory variables in two groups -- both policy and control variables -- were selected from the large variety of possible variables.

Kohn (2000), in a study of 85 Canadian urban transit agencies, examines the data from 1992 and 1998 to identify significant explanatory variables to predict ridership. He concludes that the two main variables were average fares and revenue vehicle hours. Other variables he examines included demographics, hours of service, fare structure, vehicle statistics, energy consumption, employment, passenger statistics, revenues, and expenditures. However, Kohn's model includes only two main variables and does not control for other variables because two variables explain almost all variation in the ridership level (R2 = 0.97). See Statistical Results of Kohn's Model (2000) shows the results of the study. Kohn's study does not specifically account for the fact that service levels are, at least in part, a function of the level of transit demand, which calls into question the implied causality of his analysis (that is, increasing service and lowering fares is the way to increase ridership).

Hendrickson (1986) examines the significance of the share of employment in the central business district and the share of work trips by public transit using 1980 Census data for 25 large metropolitan U.S. areas, which made up 60 percent of all nationwide transit ridership. He finds that transit use was highly related to the percentage of jobs in the CBD for any given metropolitan area. He reports that the percentage of employees who worked in the CBD dropped from 8.5 percent in 1970 to 7.8 percent in 1980, while the percentage of work trips taken on public transit dropped from 12.2 percent in 1970 to 10.5 percent in 1980. His Ordinary Least Squares (OLS) regression model with only four variables -- percentage of workforce in CBD, absolute number of workers in CBD, absolute number of work transit trips, percentage of work trips taken on transit -- explains 96 percent of the variation of public transit use, signaling a strong relationship between transit use and CBD employment. This model does not consider the growth rate of an area, any other economic factors, or the land use patterns of the city (other than the CBD). For 1980, 90 percent of the variation is explained by the percentage of jobs based in the CBD rather than overall metropolitan employment. Hendrickson notes that CBD employment does not necessarily promote transit usage, but that the supply of transit to the CBD might actually bolster downtown employment. He also acknowledges that the definition of the CBD area in each city is somewhat arbitrary.

Finally, Morral and Bolger (1996) examine the effects of downtown parking supply on transit use in eight Canadian cities and 14 U.S. cities. The study finds that the number of CBD parking spaces per downtown employee had a significant influence on the percentage of CBD workers that commute to work on transit; however, their models are single, nonlinear regression models only, and do not take into account other variables.

SUMMARY

This review of previous studies of transit ridership has identified several common factors that influence transit use. Among internal factors, increasing the quantity of service (in terms of service coverage and service frequency) and reducing fares are both found to have significant effects on ridership (Sale 1976; Cervero 1990; Kohn 2000). Systems with low unit costs, low fares, relatively low subsidies, and spatially concentrated service have proven the most cost-effective in increasing ridership (Hartgen and Kinnamon 1999). Kain and Liu (1996) estimate the fare elasticity of ridership with respect to fare changes to be between -0.34 and -0.44, while the elasticity of ridership with respect to changes in revenue miles of service is estimated to be between 0.70 and 0.89. A few studies found that pricing schemes, such as deep discounting, induce significant ridership increases because such schemes account for different sensitivity to price among various market segments. Some transit agencies provide discounted transit fares to students through partnerships with universities (university transit pass programs) and have been successful in increasing ridership without increasing service (Brown, Hess, and Shoup 2001). In addition to fare policies, some studies find that the quality of service -- customer and on-street service, and station and on-board safety -- is more important in attracting riders than changes in fares or the quantity of service (Cervero 1990). Syed's (2000) survey of transit users reveals that providing transit information, improving customer and on-street service, and improving station and on-board safety are generally more important to passengers than reducing fares.

Among the external factors studied, many researchers argue that residential and employment density are critical determinants of transit use, while the effects of land use mix and urban design are relatively small (Crane 2000; Cervero 1993; Pushkarev and Zupan 1977; TCRP 1996; Spillar and Rutherford 1998; Hendrickson 1986). Demographic factors, such as personal income, auto ownership, and suburbanization of residential and job locations, also have been found to significantly affect ridership (Liu 1993; Kain and Liu 1995; Gomez-Ibanez 1996). Gomez-Ibanez (1996) finds that transit ridership is strongly affected by forces beyond the transit system's control. Finally, strategies to increase parking costs or the probability drivers will have to pay for parking are found to be more effective in increasing transit mode share than increasing the level of transit service in terms of frequency and accessibility (TCRP 1980).

The studies cited here adopted a wide array of methodological approaches: literature reviews and case studies, interviews and surveys, statistical analyses of a transit agency or region, and cross-sectional statistical analyses. The more objective statistical analyses typically focus on testing the relative causal influences of internal and external factors on transit ridership. Collectively, these studies find that external factors such as population and employment growth have had more influence on ridership than internal factors such as fare and service levels. Furthermore, there are clear limits to the effectiveness of using solely internal factors to stimulate transit use (European Commission Transportation Research 1996; TCRP 1980, 1995, 1998; Gomez-Ibanez 1996). Most of the authors of these studies recommend that, to increase transit use, external measures, such as increased gasoline prices or parking costs, should be combined with internal measures, such as increasing the transit service quantity and quality, to have large effects on transit ridership. Because these indirect measures are external to the transit operator and likely to be strongly opposed by nontransit interests, combining such internal and external factors in a concerted effort to increase transit use proves difficult for transit operators.

While past studies provide valuable information for transit agencies that seek measures to increase ridership, their results are quite mixed, partly due to the variation in methodologies and data used for analysis. In general, the aggregate statistical analyses have been hampered by limited and incomplete data, particularly concerning the external influences on patronage. In contrast, the more subjective studies based on literature reviews, surveys/interviews, and case studies typically have sought to identify the factors thought by experts to affect ridership. Many of these studies, however, are relatively old, and most of them do not specifically ask about perceptions of causality or the relative influence of internal or external factors. This study does two things to update and advance the research. First, we examine a recent period in history -- the economic boom years of the late 1990s; second, we combine an array of methodological approaches used separately in previous research -- aggregate, cross-sectional data analysis, a survey of more than one hundred transit managers nationwide, and in-depth case studies of a dozen transit systems. It is to the data analysis that we turn in the next chapter.

The Big Picture: Recent Trends in Transit Patronage

At the turn of the last century, public transit systems were the centerpiece of every urban transportation system in the United States, and indeed the world. Personal travel in the cities of 1900 usually took one of two forms: walking or public transportation. At that time, 99.7 percent of all passenger miles traveled in U.S. cities were on transit (Altshuler 1981). Although transit systems of a hundred years ago operated a variety of modes -- cable cars, horse-drawn trolleys, and so forth -- the vast majority of travel was by electric streetcar.

Cities and travel in them have changed greatly in a century. Travel is now dominated by private motor vehicles, and public transit systems in the United States -- outside of New York City -- play a decidedly supplementary role. See Total Unlinked Trips (1907-1999) shows the trend in transit use during the 20th century: Transit patronage in the U.S. climbed quite steadily until the economic downturn of the Great Depression, when ridership declined steadily for almost a decade.

Total Unlinked Trips (1907-1999)

The rationing of oil, rubber, and steel during World War II, combined with a surge in war-related employment, pushed transit use to its highest-ever levels. Following the war, transit ridership plunged precipitously, and the quarter-century after the war was characterized by widespread bankruptcies among transit systems, which then were mostly private and for-profit. The advent of public subsidies for transit systems began in earnest in the mid-1960s and increased significantly into the 1980s. The effects of public subsidies were both to increase dramatically the cost of producing transit service and to stabilize transit ridership (Jones 1985; Pickrell 1988; Wachs 1989).

See Total Unlinked Trips (1980-1999) plots the trends in nationwide transit patronage over the last two decades to show recent trends in more detail: Overall transit use declined during the recession years of the late 1980s and early 1990s, but rebounded with the economy during the mid-1990s. The 9.1 billion unlinked passenger trips made in 1999 represented an 18 percent increase in just four years (APTA 1999).

Total Unlinked Trips (1980-1999)

While these recent increases in transit patronage are encouraging, they probably do not herald a return to the heyday of urban public transit seen a century ago. Although overall transit use has gradually climbed since the 1970s, and quite significantly since the mid-1990s, transit's overall share of metropolitan travel continues to fall. This is because cities continue to grow and urban travel is growing even faster. Just 1.8 percent of all person trips in the United States were made by transit in 1995, down from a 2.2 percent share in 1983, and 2.4 percent in 1977. Nationwide, 4.5 percent of all commute trips were made by transit in 1983; by 1995, this share had fallen to 3.5 percent (FHWA 1995; Pisarski 1996). Similarly, data from the U.S. Census and Nationwide Personal Transportation Survey (NPTS) indicate that transit's market share of total travel is continuing to fall despite absolute ridership increases.

Why the continuing decline in transit's market share? Researchers have attributed the decline in transit ridership in U.S. metropolitan areas since the end of World War II to factors such as suburbanization of jobs and residences, rising incomes, increasing car ownership, declining gasoline prices (in real terms), ample free parking, and the effects of changing demographics (such as the maturing of baby-boomers), and the increase in trip-chaining, particularly among women who combine both workplace and household responsibilities in their trip-making (Fleishman, et al. 1996; Pisarski 1996; Taylor and McCullough 1998).

Given transit's declining overall market share of urban travel, perhaps the most auspicious aspect of the recent upswing in transit ridership is that transit trips per capita are on the rise as well, based on projections of 1990 Census data. As shown in See Unlinked Trips per Person, Americans took an average of 31.3 trips per capita in 1999, compared to only 28.6 trips per capita in 1995 (a 9 percent increase).

Unlinked Trips per Person

To help explain the forces and factors behind the recent increases in transit ridership, we deconstruct these summary patronage trends below along two dimensions. First, we explore how changes in the factors internal to transit systems (changes in service levels, fares, etc.) have influenced ridership; then we examine how factors external to transit systems (changes in population, employment, development density, etc.) have affected ridership. We conduct this initial analysis using data derived from the National Transit Database (NTD, formerly known as Section 15 database) maintained by the Federal Transit Administration (FTA). The NTD is a system of accounts and records reported annually by the more than 500 transit systems that receive federal operating assistance. These transit systems are required to report a wide range of data to the FTA concerning the finance and operation of their system. Although the NTD is clearly the best, most comprehensive, cross-sectional transit data source, it is not without limitations. For example, not all systems report data to the NTD because systems that do not receive federal subsidies are not required to report. However, the transit systems operating the vast majority of service and carrying the vast majority of passengers in the U.S. do report to the NTD.

Effects of Internal Factors on Transit Ridership

Ridership can be affected by internal factors in two principal ways: either by changing the price charged for transit service or changing the level of service provided. We examine each of these factors below.

Changes in the Price Charged for Transit Service

During our study period, changes in average fares per unlinked trip nationwide (calculated by dividing total fare revenues by total unlinked trips) was closely related to changes in ridership. See Average Fare per Unlinked Trip shows that, controlling for the effects of inflation, average transit fares increased, although unevenly, from $0.94 per unlinked trip in 1991 to $1.04 in 1996, an 11 percent increase. Since 1996, however, average fares have declined to $0.93 per unlinked trip (all figures are in 2001 dollars).

Average Fare per Unlinked Trip

The 11 percent decrease in inflation-adjusted transit fares since 1996 is closely correlated with a 12 percent increase in total ridership and a 10 percent increase in transit trips per capita over the same period. During the 1990s, changes in average fares were closely correlated (-0.61) with changes in overall transit patronage (See Unlinked Trips vs. Average Fare per Trip). See Unlinked Trips per Person vs. Average Fare per Trip shows that changes in transit fares were even more closely correlated with changes in transit use per capita (-0.91). While such findings suggest that, during the 1990s, the demand for transit service was very sensitive to price, the causality of this relationship cannot be determined precisely without performing a more comprehensive multivariate analysis to control the wide array of factors (both internal and external to transit systems) that are thought to affect transit use (see ).

Unlinked Trips vs. Average Fare per Trip	Unlinked Trips per Person vs. Average Fare per Trip

Changes in the Level of Service Provided

While transit ridership levels were quite volatile during the 1990s, transit service levels rose steadily throughout the decade, with revenue vehicle miles increasing 24 percent between 1991 and 1999, and vehicle miles per person increasing 15 percent over the same time period (Figures See Revenue Vehicle Miles (1991-1999) and See Revenue Vehicle Miles per Person (1991-1999)).

Revenue Vehicle Miles (1991-1999)	Revenue Vehicle Miles per Person (1991-1999)

One would expect that changes in transit service levels are strongly correlated with changes in transit patronage, and in the 1990s this was the case. See Unlinked Trips vs. Revenue Vehicle Miles shows that the correlation between service levels and ridership was 0.81 during the 1990s. The correlation between service levels per capita and ridership levels per capita in See Unlinked Trips per Person vs. Revenue Vehicle Miles per Person were much lower (0.37), which suggests that factors external to transit systems (such as population and employment changes) may have influenced both service and ridership levels during the 1990s, and thus influenced some of the relationships observed here (see ). We now turn to an analysis of the external factors.

Unlinked Trips vs. Revenue Vehicle Miles	Unlinked Trips per Person vs. Revenue Vehicle Miles per Person

EFFECTS OF EXTERNAL FACTORS ON TRANSIT RIDERSHIP

The data presented in the previous section show that both fare levels and service levels were closely correlated with changes in ridership during the 1990s. What can be inferred from such findings? If transit systems simply cut fares and expand service, will they attract additional riders at a rate almost proportional to the fare and service changes? Perhaps not. While the case of transit fare levels is less clear, it stands to reason that changes in transit service levels are as likely to occur in response to increasing demand for transit service as they are to be a cause of increasing demand. This raises the question of what factors outside the control of transit managers may be exerting influence on both service and demand. We examine three such factors here: unemployment levels, average wage levels, and overall economic output.

Employment Levels and Transit Ridership

Given the apparent positive relationship between transit ridership and economic cycles, we hypothesized that transit use was inversely related to unemployment rates during the 1990s for three reasons. First, journeys to and from work comprise a larger share of transit trips than auto trips (Pisarski, 1996). Second, lower-wage, less-skilled workers are more likely to lose jobs when the economy contracts. Third, transit riders, especially bus riders, are far more likely to come from low-income households than those traveling in private motor vehicles (Pucher 1995; Garrett and Taylor 1999).

Indeed, we find the unemployment rate was highly correlated (-0.70) with overall transit use during the 1990s. Nationally, the unemployment rate declined for most of the 1990s, from a high of 7.7 percent in 1992 to a low of 4.3 percent in 1999 (See Unemployment Rate (1991-1999)).

Unemployment Rate (1991-1999)	Unlinked Trips vs. Unemployment Rate

The correlation between unlinked trips and unemployment is shown in See Unlinked Trips vs. Unemployment Rate. As noted earlier, transit ridership increased every year but one from 1993 to 1999 (see ).

Gross Domestic Product and Transit Ridership Levels

A second, common measure of economic activity is the Gross Domestic Product (GDP), calculated annually by the Bureau of Economic Analysis. The GDP grew throughout the 1990s. The average annual increase during the recession years of the early 1990s was just under 2 percent per annum, while the annual rate of increase was in excess of 3 percent per year in the late 1990s (see Figures See Gross Domestic Product (1991-1999) and See Gross Domestic Product per Person (1991-1999)).

Gross Domestic Product (1991-1999)	Gross Domestic Product per Person (1991-1999)

Gross Domestic Product vs. Unlinked Trips	Average Hourly Wage (1991-1999)

We compared transit ridership trends to both the overall real (inflation-adjusted) GDP (See Gross Domestic Product vs. Unlinked Trips) and the real GDP per capita. Overall transit ridership tracked both of these measures closely -- 0.79 with the real GDP, and 0.82 with the real GDP per capita (see ).

Wage Levels and Ridership

Of all of the economic indicators tested, transit ridership tracked most closely with personal income, as measured by the average hourly wage from all industries (estimated by the Bureau of Labor Statistics). Average "real wages," measured by the BLS in $2001 using the Consumer Price Index, declined in the recession years of the early 1990s, from $13.47/hour in 1991 to $13.28/hour in 1994. For the remainder of the 1990s, average real wages increased every year, to a high of $14.13/hour in 1999 (Figure See Average Hourly Wage (1991-1999)).

While transit trips per capita were not highly correlated with either the unemployment rate (-0.16) or the real Gross Domestic Product (0.24), transit trips per capita during the 1990s were strongly correlated with changes in average real wages (0.70) (see Figure See Trips per Person vs. Average Hourly Wage ($2001)). We also found that the unemployment rate was highly negatively correlated with overall transit ridership (-0.70). In addition, the correlation between average real wages and total transit ridership during the 1990s was almost perfect (0.96) (see Figure 18 and Table B-5).

Trips per Person vs. Average Hourly Wage ($2001)	Unlinked Trips vs. Average Hourly Wage

SUMMARY OF EFFECTS OF INTERNAL AND EXTERNAL FACTORS ON RIDERSHIP

In this section, we compare national trends in transit ridership during the 1990s with a series of factors internal to transit systems (fares and service supply) and external to transit systems (unemployment, economic productivity, and wages). We expected to find a relatively high degree of correlation between transit ridership and the internal factors tested, and this was the case. However, such correlations do not necessarily imply causality; this is the "chicken or the egg" question. Increased service should increase ridership, but increased demand should also motivate transit managers to increase service. An important first step in breaking down this chicken-versus-egg question is to look for factors that may be influencing both service levels and ridership. We have analyzed three such factors here, all related to economic activity. As the summary data in See Correlation Coefficients of Internal and External Factors and Transit Ridership show, the extraordinarily strong relationships observed between an external economic measure -- unconnected to the price or supply of transit service -- suggest that many of the factors affecting changes in transit ridership may be outside of transit managers' control.

These issues will be further discussed in the next chapter, where we take an in-depth statistical look at the agencies across the country that have increased ridership since the mid-1990s.

The Bright Picture: Analyzing Transit Systems With Significant Ridership Gains During the 1990s

While overall transit ridership was up during the mid-and late-1990s, not all transit systems increased transit ridership. Some posted dramatic ridership gains, some tracked national trends, and some lost riders. Our focus here is on transit systems that added riders between the end of the economic recession in 1994 and the end of the economic boom in 1999 when, as noted in the previous chapter, transit use began to rise.

As with the data analyzed in the analysis in this chapter is drawn primarily from the Federal Transit Administration's National Transit Database (NTD). While the NTD data presented in the previous chapter were drawn from the entire sample of 587 reporting transit agencies, this chapter narrows this sample, and our analysis, in several ways. First, we eliminated all systems that do not operate some form of fixed route transit -- bus, trolleybus, light rail, heavy rail, commuter rail, ferryboat, cable car, inclined plane, monorail, jitney, or automated guideway. In other words, we excluded all agencies that operate only demand-response or taxi services. For the many agencies that provide both fixed-route and demand response or taxi services, we included data only on the fixed-route modes (so the data analyzed here may differ slightly from NTD published "totals" for each agency).

In all, 414 agencies offered some form of fixed-route service and reported data to the NTD during the late 1990s. Of these, 367 agencies submitted complete data for both 1995 and 1999. Of those 367 agencies, 227 (or 62 percent of the entire sample) increased ridership (measured as unlinked trips) during a four-year period between 1995 and 1999. Those 227 agencies carried more than 86 percent of the total unlinked trips reported to the FTA in 1999; each of those 227 systems and their patronage during the study period is listed in .

The ridership data reported in this chapter, and throughout this document, are for unlinked trips. Most transit researchers would agree that linked trips (trips that include transfers) and passenger miles data (total trips' average trip length) are more telling and less biased measures of transit use. However, reliable, comparable cross-sectional data for those measures of transit service consumption are not available. Lacking data on those measures, we (and nearly all previous research on transit ridership) use unlinked trip data.

SUMMARY OF AGENCIES THAT INCREASED RIDERSHIP

Perhaps the most distinguishing characteristic of transit systems that added riders during the late 1990s is that they have no distinguishing characteristics. They come in all shapes and sizes, from all areas of the country; some operate one way and others operate another; and they operate in a wide variety of settings.

Transit Modes Operated

Transit operators of all kinds increased ridership, including those with just one mode of operation and those that operated many forms of transit. Nine different modes -- bus, light rail, ferryboat, heavy rail, commuter rail, trolleybus, cable car, automated guideway, and monorail -- were represented among the agencies that increased ridership.

With trips on buses composing 62 percent of transit trips nationwide (APTA 1999), it makes sense that buses were the most represented mode. Two hundred eleven of the agencies (93 percent) had at least some bus service, while 82 percent of the agencies operated only buses. The second-most represented mode was light rail, which only 13 agencies operated. See Mode Combinations of Agencies with Increased Ridership (1995-1999) details how many of the agencies operated each mode; See Frequency of Modes in Agencies with Increased Ridership (1995-1999) shows all the combinations of modes featured in the agencies that increased ridership (note in See Frequency of Modes in Agencies with Increased Ridership (1995-1999) that the "# of agencies" adds up to more than 100 percent, because some agencies operate more than one mode).

Twenty-five agencies operated some combination of modes, including Boston's Massachusetts Bay Transportation Authority, which has six different modes of operation -- bus, trolleybus, heavy rail, light rail, commuter rail, and ferryboat. Regardless, there is no dominant mode other than buses.

Agency Size

As agencies operated in a variety of different modes, the agencies that reported ridership increases also came in all different sizes. Of the 227 agencies, the smallest reported increase was by the Huntsville Department of Transportation (AL), with 217 annual trips; the largest reported increase was by the New York Metropolitan Transit Authority (New York MTA), with 536,000,000 annual trips. The New York MTA is, by far, the largest transit system in the United States. The New York MTA subway and bus system report more than 2.4 billion unlinked trips in 1999, an increase of 28 percent in just four years, and a significant recovery from several years of precipitous losses in the early 1990s (Taylor and McCullough 1998). Most of the other largest transit agencies also experienced patronage increases: the Chicago Transit Authority (CTA), the Los Angeles County Metropolitan Transportation Authority (LACMTA), the Washington Metropolitan Area Transportation Authority (WMATA), Boston's Massachusetts Bay Transportation Authority (MBTA), and the San Francisco Municipal Railway. Of the 10 largest transit agencies in the United States in 1995, only two lost ridership -- SEPTA in Philadelphia and Baltimore's MTA.

In fact, 38 of the 49 U.S. transit systems (78 percent) that carry 20 million passengers per year or more increased ridership during the late 1990s, and accounted for 91 percent of the total growth in patronage nationwide (see See Agencies with Increased Ridership, by Size). By comparison, about three-fifths of the "large" (61 percent), "medium" (62 percent), "small" (56 percent), and "very small" (59 percent) transit agencies added riders during the late 1990s.