«ON THE SOCIAL DESIRABILITY OF URBAN RAIL TRANSIT SYSTEMS Clifford Winston Vikram Maheshri Brookings Institution U.C. Berkeley December 2005 ...»
Please do not quote.
ON THE SOCIAL DESIRABILITY OF
URBAN RAIL TRANSIT SYSTEMS
Clifford Winston Vikram Maheshri
Brookings Institution U.C. Berkeley
The evolution of urban rail transit in the United States over the past twenty years
has been marked by three inescapable facts that signal an inefficient allocation of transit
resources. Rail’s share of urban travelers is declining; its deficits are rising sharply; and yet investment to build new systems and extend old ones continues.
In 1980, two million Americans got to work by rail transit. Today, in spite of an increase in urban jobs and transit coverage, fewer than one million U.S. workers use rail, causing its share of work trips to drop from 5 percent to 1 percent.1 Although rail transit’s farebox revenues have consistently failed to cover its operating and capital costs since World War II, governmental aid to cover transit deficits has been increasingly available. Since 1980, operating subsidies have climbed from $6 billion to more than $15 billion today (APTA Transit Fact Books, figures in 2001 dollars). Capital subsidies have also increased as transit agencies struggle to maintain and provide new facilities, track, and rolling stock.
These worrisome trends, however, have not curbed U.S. cities’ appetite for rail transit service. During the 1990s, Cleveland, Washington, Santa Clara, Sacramento and other cities expanded their systems, while Los Angeles, Denver, Dallas, and St. Louis built new ones. Recently, Houston and Minneapolis opened new light rail lines while small, sparsely populated cities such as Sioux City, Harrisburg, and Staunton, Virginia suggested that they want federal funds to help build their systems. And although county residents repeatedly nixed a referendum to build a $4 billion extension of Washington’s Metro out to Dulles airport, planners nevertheless circumvented popular will and diverted These figures are from the National Transit Database and the U.S. Census.
increased toll revenue from the Dulles toll road to finance a portion of the ultimate extension.
Any private firm that was losing market share and reporting increasing losses would be hard pressed to attract funds to expand. Almost certainly, it would try to determine the most efficient way to contract. Of course, a transit agency does not seek to maximize profits, but its public financing is justified only if it is raising social welfare, where social welfare can be measured as the difference between net benefits to consumers and the agency’s budget deficit, also taking into account externalities of rail transit (for instance, the reduction in roadway congestion attributable to rail.) Although the costs and benefits of public rail transit operations have been debated in the policy community (see, for example, Litman (2004)), we are not aware of a recent comprehensive empirical assessment of rail’s social desirability.2 The purpose of this paper is to estimate the contribution of each U.S. urban rail operation to social welfare based on the demand for and cost of its service. We find that with the single exception of BART in the San Francisco Bay area, every U.S. transit system actually reduces social welfare. Worse, we cannot identify an optimal pricing policy or physical restructuring of the rail network that would enhance any system’s social desirability without effectively eliminating its service.
Rail transit’s problem is its failure to attract sufficient patronage to reduce its high average costs. This problem has been complicated enormously by new patterns of urban Richmond (2001) provides an overview of U.S. rail transit systems including their ridership and financial performance. Winston and Shirley (1998) estimate the net benefits of urban rail systems as of 1990. Other researchers such as Savage (2004), Savage and Schupp(1997), Kain (1997), and Viton (1981) have assessed the welfare properties of and alternative polices to improve specific transit systems.
development. Rail operations, unfortunately, are best suited for yesterday’s concentrated central city residential developments and employment opportunities; they are decidedly not suited for today’s geographically dispersed residences and jobs. At best, urban rail service may be socially desirable in a few large U.S. cities if its operations can be adjusted to mirror successful privatization experiments conducted abroad. Ironically, however, rail transit enjoys powerful political support from planners, civic boosters, and policymakers, making it highly unlikely that rail’s social cost will abate.
An Empirical Framework for Estimating Rail Transit’s Social Benefits Urban rail transit operators do not set prices to cover operating and capital costs.
In fact, under Federal Transit Administration Section 5307, transit fares (including bus, rail, and paratransit) have to cover only some 17 percent of operating costs for most agencies to qualify for federal funds. State and municipal funding thresholds vary, but none requires even half of operating costs to be covered at the farebox. On average, the nation’s rail transit systems cover only about 40 percent of operating costs, to say nothing of their substantial capital costs (National Transit database). Like any good or service, rail’s net benefits to users are simply given by consumer surplus, but to justify continued operation, rail’s surplus must offset the difference between farebox revenues and costs.3 We develop rail transit demand and cost models to estimate users’ benefits and agencies’ budget deficits. We also account for rail’s effect on the cost of roadway congestion. A novel feature of the models is that they include variables that characterize Rail transit also directly generates a small portion of its revenues from advertisements, parking lot fees, and other auxiliary services. We explicitly exclude these revenue streams and their associated production costs when we perform our analysis.
a transit system’s network configuration and stations, enabling us to explore whether rail could enhance its social benefits by expanding or contracting its facilities in an efficient manner.4 Demand. Our empirical analysis is conducted on a panel of U.S. urban rail transit
where pit is the average fare, Z it contains exogenous network variables, X it contains D exogenous city characteristics, and uit is an error term.5 Depending on the system, transit fare is determined by a transit agency, metropolitan planning organization, or city council. It is therefore reasonable to treat fare as exogenous because it is primarily determined through a regulatory process rather than market forces. In addition, the policy bodies that set fares have little incentive to adjust them to changing market conditions because they are not subject to stringent financial performance goals.
Rail demand is also influenced by the configuration of a transit system’s network.
Travelers are more likely to use rail if it provides more comprehensive coverage of a An alternative (indirect) measure of rail transit’s benefits is rail’s impact on housing prices in the surrounding residential area near stations. Diaz (1999) summarizes studies of the effect that some systems have had on residential property values. Baum-Snow and Kahn (2000) provide a recent study.
We will interchange the term city with urbanized area and metropolitan (statistical) area (MSA). Urbanized areas and MSAs are determined by U.S. Census demographic criteria; nonetheless, they are often associated with a distinct city. Data on rail transit systems tend to pertain to a metropolitan statistical area.
given area, offers more conveniently located stations, offers more connectivity, and accommodates travel in both north-south and east-west directions (i.e., its network is nonlinear). The following components of a transit network enable us to derive four metrics from graph theory to capture the effect of a rail system’s entire network on
actual number of links and the maximal number of links, given a set of stations. Greater connectivity can improve access to different points on the network.
Hagget and Chorley (1969) provide a complete discussion of the measures.
on a minimum value of one, characterizing a perfectly linear network. Larger values characterize less linearity and thus broader coverage of a given geographical area.
We also characterize rail’s service with the ratio of its peak service frequency to base service frequency.7 Because rail and bus systems are often designed to complement each other, we also control for bus transit’s peak-to-base ratio. Larger rail and bus frequency peak-to-base ratios should increase demand.
Turning to the characteristics of a city that may influence rail demand, we control for regional gasoline prices, the average number of days below 32 degrees Fahrenheit, and the average commute time. Higher gasoline prices may induce commuters to switch from driving, which would increase transit use, but they may also depress overall economic activity, which would tend to decrease transit use. Thus, the a priori effect of gasoline price is ambiguous. Cold temperatures tend to increase transit ridership because they discourage commuters, in particular, from walking and biking. A high average commute time in a city is likely to be caused by lengthy work-trip distances and factors that contribute to road congestion. The former may reduce travelers’ accessibility to rail and decrease the demand for transit, while the latter may increase the demand for rail transit as travelers try to avoid congestion. Thus, the a priori effect of average commute The peak service frequency is defined as the maximum average frequency of the 7-10am morning and 4-7pm evening rush hours. The base service frequency is defined as the average frequency of the 10am-4pm off-peak period. It is reasonable to treat the frequency ratio as exogenous because the timing and duration of the peak are fixed and not influenced by demand, while peak and off-peak schedules are largely determined by labor contracts. We also tried to capture service frequency by specifying a system’s vehicle miles (seat miles are not available), but the peak-to-base frequency ratio produced more plausible and reliable estimates.
time is also ambiguous.8 Finally, we include the metropolitan area population and resident households’ median annual income. Demand for rail transit should be positively related to an area’s population. Although transit is often regarded as an inferior good, the average income of travelers who use rail is much higher than the average income of travelers who use bus (Winston and Shirley (1998)), so rail transit may be a normal good.9 Cost. Regulations constrain rail transit agencies from abandoning or adding track and stations to optimize their operations; hence, it is inappropriate to assume that they are in long-run equilibrium. We therefore specify a short-run total cost function where we include the network variables discussed previously to control for the capital inputs of track and stations that are fixed in the short-run. Formally, short run total costs, Cit, for
transit system i during year t are expressed as:
It is possible that average commute times may be higher in cities that have a measurable share of rail transit users, which would suggest that average commute time may be endogenous. We consider this possibility in our estimations.
We explored some additional and alternative variables to capture service quality and city characteristics. Because heavy rail systems can operate at higher speeds than light rail systems, we included a dummy variable for systems with predominantly heavy rail service, but it was statistically insignificant. For climate effects, we also specified average snowfall and a light rail dummy variable interacted with snowfall or cold temperatures because light rail stations tend to be above ground. These variables were insignificant. We also specified average commute distance instead of average travel time to work, but it was poorly measured and had virtually no effect on demand. Finally, we specified a dummy variable for new systems—those that had operated for five years or less—to explore whether some travelers might be attracted to rail because of its novelty.
However, the dummy was statistically insignificant.
where Qit is output in passenger miles, wit contains factor prices, Z it contains exogenous network variables, Yit contains other exogenous influences on cost, and ν it is an error term.10 Given that fares can be assumed to be exogenous, it is reasonable to treat output as exogenous in the cost function.11 We include factor prices for labor and fuel. The price of labor is given by the hourly wage for transit workers, including fringe benefits.
Because different rail transit systems use different combinations of fuel types (gas, electricity, kerosene, ethanol, and so on), we computed a standardized price per kilowatthour of energy using the appropriate physical constants (i.e., KWH equivalents) for each one.12 We capture the effect of capital on short-run costs by including the network variables that we used in the demand model. Holding output constant, we expect networks with greater density, connectivity, or more closely situated stations to have higher costs. The a priori effect of network linearity on costs is not clear; greater linearity requires a system to use less rolling stock to serve its stations, while greater nonlinearity may enable a system to achieve higher load factors that would reduce costs.
Urban rail transit cost functions have been estimated previously by Pozdena and Merewitz (1978), Viton (1980), and Savage (1997).