Rapid Bus and Rapid Rail:
Peak Period Service Supply vs. Observed Passenger Utilization
A publictransit.us Monograph
 
Leroy W. Demery, Jr. • J. Wallace Higgins • Updated June 19, 2005
 
Copyright 2003–2007, Leroy W. Demery, Jr., and J. Wallace Higgins
 
History of Updates
 
Original Document: November 14, 2002
Table of Contents added March 19, 2003
Linear regression models updated, reference added, minor corrections May 25, 2003
Minor additions September 23, 2003.
Figures 1 through 7 replaced; section 2G added December 1, 2003
Converted to this document format, June 19, 2005
 
Abstract
Travel demand and service supply interact to create transit service consumption, e.g. ridership. Demand analysis receives greater attention during planning, but supply issues are often overlooked. This paper examines empirical data regarding the relationship between transit service supply and consumption for both bus and rail modes. Regression analysis is used to identify a strong, systematic relationship between offered and utilized capacity, using data from U.S. and Canadian cities on service supply, consumption and temporal distribution of travel on rapid bus and rapid rail systems. This analysis demonstrates strong correlation between peak service supply and consumption, permitting prediction of peak service consumption from supply levels within a fairly broad demand range. Peak consumption levels assumed by some previous studies are found to have been unrealistically high.
Consequently, costs for various bus and rail projects have proven higher, and ridership lower, than predicted during planning. If the peak period supply of transit service is inadequate, some potential consumers will choose not to ride, leading to higher-than-predicted costs per passenger. Crowding often discourages patronage in markets where consumers have competitive alternatives to public transit service. This occurs at crowding levels significantly lower than the "capacity" figures assumed by transit planners. Peak-period service consumption patterns suggest a significant and observable consumer preference for rail. These findings have important implications for planning and cost analysis, particularly when bus and rail modes are compared. A stronger case for rail transit might be made than some previous studies have found, but bus modes have significant advantages in some situations.
About the Authors
J. Wallace Higgins is a consultant to the East Japan Railway Co., Ltd., formerly with the University of Tsukuba (Japan) and The RAND Corporation.
Leroy W. Demery, Jr., Senior Associate, Carquinez Assocates, is a transport research specialist, formerly a mathematics teacher at Roosevelt High School, Seattle. Contact: Carquinez Associates, P.O. Box 6076, Vallejo, CA 94591-6076, ldemery@publictransit.us.
 
Introduction
Urban planners seeking effective and affordable alternatives to more highways and more automobiles must face the question of whether bus or rail rapid transit provides superior performance and lower costs, and whether either mode will attract enough passengers to justify the public investment. Through regression analysis, this paper demonstrates a strong correlation between peak period service supply and observed passenger utilization. Service supply does not "cause" consumption as such, but the observed correlation permits reasonable predictions of peak service consumption for a given supply level within a fairly broad range of demand levels.
Transit service consumption requires both travel demand and service supply. Demand and supply are necessary conditions for consumption, but neither, by itself, is sufficient. Only the combination of demand and supply may be described as "necessary and sufficient" for consumption to occur. Consumption levels may be constrained by inadequate service supply levels. The validity of this is suggested by the following observations: 1.) the number of peak-period passengers per meter of transit vehicle length is remarkably consistent for various recent fixed-guideway lines, and 2.) significant increases in service consumption have followed increases in peak-period service supply.
This paper also demonstrates that the peak service consumption levels assumed by many previous studies are often unrealistically high, given the levels of peak service supply that have been stated or assumed. As a result, actual costs per passenger for various bus and rail projects have often proven higher, and ridership lower, than predicted during the project planning phase. If peak service supply is not adequate to meet projected demand, patronage will not reach predictions. This in turn will lead to higher costs per passenger than predicted. Peak-period crowding levels will discourage additional patronage, in markets where consumers have competitive alternatives to public transit service, falling significantly short of the "capacity" levels often assumed by planners and vehicle designers.
The consistent disparity in the U.S. and Canada between bus and rail modes in terms of peak-period service consumption patterns suggests a significant -- and observable -- consumer preference for rail transit, at least during peak travel times. These findings have important implications for cost analysis, particularly when comparing the costs of bus and rail alternatives. In many locations, a stronger case for rail transit might exist than some previous studies have found. However, bus modes retain significant advantages in some situations, particularly in travel corridors with lower demand or where there are opportunities for phased introduction of unconnected shorter, less costly route segments.
This analysis of urban fixed-guideway modal capacities, presented below, is based on empirical data from existing U.S. and Canadian bus and rail systems. The paper outlines a range of feasible service-supply levels, and determines commensurate levels of service consumption, for the major modal families: rapid bus (BRT) and rapid rail (RRT). These are expressed both in terms of peak-hour and all-day (average weekday) ridership. The results are specific to the U.S. and Canada, but the analytical framework itself need not be limited to this area. The paper is organized into five parts, the first of which outlines the relevance of peak service supply and vehicle occupancy to supply-side analysis of modal choice. The second part compares the service-supply characteristics of the major modal families and contrasts these with assumptions often made in alternatives analysis studies. The third section outlines differences between BRT and RRT peak vehicle occupancy levels that suggest a consumer preference for the latter. The fourth section outlines related analytical issues, and the fifth presents our conclusions.
1)  Service-Supply and Consumption Parameters: Relevance to Modal Choice
Supply-side analysis of transit service seeks to determine what service-supply levels are appropriate given predictions of service consumption, and to consider which combination of mode, fixed facility design, operating practices and other project aspects can meet projected patronage and project objectives. Such analysis can shed much light on appropriate modal choice and effectiveness, based upon empirical data about the travel decisions consumers actually make.
This analysis focuses on the relationship between service supply and observed consumption patterns rather than on preceding travel-demand predictions. Individual demand factors are not specifically treated. Instead, various determinants of demand are postulated to exist, in combinations and at levels sufficient to produce -- in concert with service supply -- the predicted levels of service consumption. In other words, service supply and service consumption levels are compared against assuming a fixed "background" of travel demand in a given travel market. This analytical framework is pertinent provided that: 1) service consumption levels reflect a balance between capacity supplied and demand parameters, and are not established by predicted "demand" factors alone, and 2) travel demand in the corridor of interest are is sufficiently high that increasing service supply parameters will result in increased service consumption.
The need for this analysis stems from the nature of the relationship among demand, supply and consumption for public transit service in the U.S. and Canada. In this sector, the usual market relationship among quantities demanded, supply and price exists tenuously, if at all. The political decision mechanisms that set prices (fares) are almost never influenced by conditions in local travel markets. Once set, fares are usually held stable, until transit budget conditions require the next fare increase. Thus, fares and consumer willingness to pay (WTP) exert little influence over the supply of transit service. Price is an important determinant of demand for transit service, but this price is not free to vary with demand and supply. The price of transit service in most U.S. and Canadian urban travel markets, is generally constant over the short term, given the political nature of the fare-setting process.
The relative independence of transit service demand and supply from price has several important ramifications. Demand and consumption must be treated as separate, although related, parameters. Since the "invisible hand" does not bring supply into line with demand, observed consumption may not represent total demand. Latent or hidden demand may exist. Prospective consumption may be suppressed by factors other than price changes to levels that exceed WTP. Since the supply/demand equilibrium, expressed as observed consumption, is not established with respect to price alone, this must also be shaped by other factors influencing consumer choice. One of the most important of these factors is service supply, particularly during peak commute hours.
This paper's focus on service-supply issues stems from two key assumptions:
1.  A transit service will serve its potential demand only if potential users find that service supply is adequate. If service supply is insufficient, then some potential consumers will not ride. It will then appear to many observers that project demand was overestimated and unit costs per passenger underestimated, when in fact the problem is with supply, not demand.
2.  Realistic patronage projections rely on sufficient levels of service supply that reflect occupancy levels acceptable to consumers. Inadequate treatment of service supply, understating these relationships, can lead to poor investment and engineering decisions, generating controversy and requiring costly, difficult and time-consuming corrections.
The focus on peak-period service supply stems from the temporal distribution of urban travel demand typical of large U.S. and Canadian cities. Between 85 and 95 percent of annual transit service consumption occurs on weekdays. In most cases, about 50 percent of weekday service consumption occurs during the four to five busiest "peak period" hours. Thus, service levels provided on a fixed-guideway transit facility during the peak commute period have a substantial impact on total potential weekday ridership. This service supply may be expressed in terms of vehicles operated, or seats/places provided per hour. Regression analysis demonstrates a strong correlation between peak-period service supply and service consumption (Section 2F).
Peak vehicle occupancy or PVO (peak-hour passengers per vehicle, or p/v) refers to the average passenger load carried aboard each vehicle, past the maximum-load point, during the busiest hour, in the busier direction. This consumption parameter reflects a complex relationship among several demand and supply factors. It is also an indicator of consumer choice, reflecting decisions of when to travel and by what mode. Among the factors influencing this choice include willingness to travel as a standing rather than a seated passenger and willingness to tolerate the existing level of crowding aboard transit vehicles (whether standing or seated). The ancillary effects of high crowding levels include increased loading and unloading times at stops, irregular intervals between vehicles and lack of space aboard them. These impacts of overcrowding also undoubtedly influence consumer choice. PVO is likely to level off at the point where, on average, prospective transit users will choose to travel at a different time, or by a different mode. The service-consumption level at which this occurs is likely to fall short of theoretical maxima.
"Capacity" ratings used for transit vehicles typically reflect arbitrary standards of floor space per passenger. These are highly inconsistent between different transit operators. Observed PVO in most metropolitan areas is much less than the vehicle capacity (Cv) usually assumed by planners. The authors have not found any cases where such "loading standards" were based on the wishes of consumers, demonstrated consumer choice or feedback from surveys of potential passengers or focus groups.
Figure 1 illustrates the relationship between peak-period service supply and consumption, with demand held constant, on a hypothetical fixed-guideway facility where service supply may be increased without limit. In this example, changes in service levels do not affect demand parameters; for example, vehicle size or train length may be expanded without limit but service frequency remains constant. Following introduction of service, consumption grows as service supply is increased, while PVO remains constant, at a "threshold" level of crowding (i.e. number of standing passengers). At this threshold, on average, additional prospective passengers would choose to travel at a different time or by a different mode. The maximum consumption level reflects equilibrium between the totality of demand and service supply. No further increase in ridership would occur if service levels were increased beyond this point. PVO would simply decrease with each increment of added service.
Figure 2 reflects relationships more typical of actual transit facilities -- which cannot increase service without limit. Overall weekday service consumption (ridership) falls below the hypothetical ceiling established by demand factors. This occurs because of the threshold effect described above. Once peak service levels reach the practical maximum, ridership ceases to grow unless the facility attracts additional off-peak and reverse-peak traffic, or changes in demand parameters raise the vehicle-occupancy threshold.
Figure 3 illustrates the changes which follow introduction of a new fixed-guideway service into an urban travel market. The increased consumption generated by the more-attractive service is not uniformly distributed. Instead, the new ridership is concentrated into peak commute hours, reflecting the principal market for such service: CBD work trips. This changes the temporal distribution of service consumption. This effect, anticipated four decades ago by San Francisco Bay Area Rapid Transit District (BART) planners (Proctor 1960), is demonstrated by data collected by San Diego Trolley, Inc.: during the first three years of LRT operation, the majority of ridership growth occurred during peak periods (San Diego Association of Governments, 1984). Similar results have been reported elsewhere. These may be quantified by changes in "peak traffic share," (PTS), a measure of the "peaking effect" which is characteristic of urban transit and highway travel.
PTS, an indicator variable, reflects factors, including consumer choice that shapes the temporal distribution of weekday travel demand in urban areas. These are not fixed, unchanging characteristics of a particular city or corridor. PTS may change significantly over time, particularly with introduction of a new service into an urban travel market, in a manner which planners may not anticipate1. PTS may also be influenced by policies encouraging long-term land-use changes to stimulate off-peak and reverse-peak transit patronage.
Peak service supply and peak vehicle occupancy serve as surrogates for offered capacity (Co) and utilized capacity (Cp), crucial parameters in transit capacity analysis (Vuchic 1981). The product of maximum practical service supply (vehicles per hour per direction, or vhd) and the maximum likely vehicle occupancy (p/v) equals achievable capacity. This is the maximum likely passenger volume (passengers per hour per direction, or phd), a useful parameter for analysis of modal choice. (Parkinson and Fisher (1996) define "achievable capacity" as the maximum number of passengers that can be carried "allowing for the dimension of demand.")
The postulated relationship among peak service supply, service consumption and travel demand is explicitly deterministic, and properly so given the purpose at hand -- establishment of likely maxima. There are well-defined limits to the number of vehicles which can be operated safely or efficiently along a single lane or track during a given interval. Although less well-defined, the number of people who will choose to occupy an enclosed space such as a transit vehicle also imposes a limit which may not correspond with the assumptions of service planners and vehicle designers. These are the principal factors determining the likely maximum service-consumption levels for BRT and RRT. There is, of course, some degree of uncertainty associated with the predicted consumption levels, and the authors have rounded many of the numbers to reflect this.
 
2)  Supply-Side Estimation of Modal Capacity
The maximum likely BRT and RRT passenger volumes in most U.S. and Canadian cities fall below -- sometimes well below -- the figures often assumed for modal "capacity." Supply-side analysis of fixed-guideway capacity estimates the level of peak-period service supply commensurate with various passenger volumes, per hour or per weekday. Selection of peak service supply and PVO levels for this analysis requires careful consideration of current practice.
In general, maximum BRT or RRT service-supply levels reflect the physical characteristics of the line, vehicle fleet size and vehicle dimensions. When fleet size is not the governing constraint, the major determinant of BRT service maxima is terminal capacity -- on-street or off-street. Maximum train length and service frequency are the major determinants of RRT service maxima. The former reflects station configuration (e.g. platform length). The latter may be based on track capacity and required dwell time at intermediate stations, or it may be established by safety considerations and vehicle performance, and may be enforced by the signal system. Rigid constraints on train length and service frequency are characteristic of light rail (LRT) lines using surface streets for central business district (CBD) access. Spacing between CBD intersections limits train lengths2, while headways are limited by traffic signals and operating rules (e.g. minimum separation between trains of one city block at all times). For BRT and RRT, there is also a trade-off between traffic volume and average speed: beyond a certain threshold, the latter falls as the former grows, owing to increased dwell time at stations when vehicles carry large numbers of standees. This is also true of headway and average speed; the minimum interval between vehicles is determined in part by safe stopping distance, which depends on operating speed.
It is important to understand the distinction between "headway" and "service frequency." "Headway" refers to the spacing between vehicles or trains in terms of time. "Service frequency" refers to the number of departures during a particular interval from a station or stop, or to the interval between departures ("Maximum service frequency" is synonymous with minimum interval between departures.) The two terms are often used interchangeably, but they are not equivalent, for headway does not account for the time required for the vehicle or train to move its own length.
2A)  Peak Service Supply: Bus Rapid Transit
Estimation of practical BRT modal capacity is made difficult by the scarcity of published data from actual operations. This, of course, reflects the rarity of such facilities in the U.S. and Canada, and the unfortunate practice of quoting hourly or daily ridership statistics with little or no accompanying information on service supply. The problem is critical with regard to peak vehicle occupancy (Section 2C). Various hypothetical and actual BRT peak service supply levels are illustrated in Figure 4.
The service-supply "feasibility threshold" for BRT modes is not clearly defined. Most busway and HOV facilities do not have signals, and road vehicles generally operate beyond the safety limit in vehicle spacing (the safe stopping distance at a given speed is greater than the average spacing between vehicles).
Roadway capacity (cw) does not impose significant constraints. Given the necessary conditions, more than 1,000 vhd may be operated over a single lane, as on the Lincoln Tunnel "XBL" (exclusive bus lane) between New Jersey and New York City (Werner 2001)3. But cw does not establish practical or "achievable" offered capacity (Co).
Station capacity (cs) is the usual limiting factor for BRT net capacity (Vuchic 1981). If intermediate stations are provided, maximizing of capacity requires some means, such as bypass lanes or "off-line" stops, for buses to operate through without stopping. This facilitates operation of different types of services whose schedules might otherwise conflict. Pittsburgh and Ottawa operate frequent, regular-interval local busway services and a network of peak-period express routes serving more distant destinations. Where buses cannot pass each other in service, operation of "platoons" or "convoys" -- closely-spaced groups of two or more buses, sequenced according to destination -- permits a service supply up to 270 vhd over a two-lane busway, nearly doubling its capacity (Kain 1992). Parkinson and Fisher (1996) conclude that the single most important factor in maximizing RRT capacity is controlling and reducing station dwell time through careful design of stations and vehicles. This is also true for BRT facilities where vehicles cannot pass each other.
If station capacity is not an issue, BRT net capacity reflects that of the associated CBD distribution or terminal facilities. In mixed traffic, the practical maximum service level on a single CBD street -- with good schedule reliability -- is about 100 vhd. If curb zones accommodate 2-3 buses each, and buses may pass one another (given streets wide enough, and congestion levels low enough, to permit this), the practical maximum may reach 150 vhd. Similar results may be obtained with CBD streets reserved for buses only.
The Portland Mall has two parallel one-way streets, each with two bus lanes and a discontinuous third lane for other traffic. Capacity is estimated at 195 vhd (Schumann 1996). The threshold of practicality is not well defined but once reached, speed and schedule reliability decrease with each increment of additional service, with no net increase in capacity.
Operation of BRT services in mixed CBD traffic is not an efficient strategy: service will be slow, erratic, labor-intensive in terms of staff-hours per passenger or per passenger-mile, and expensive to operate. Preferential lanes permit higher average speed, better schedule reliability and greater labor productivity. The practical capacity of about 150 vhd can be increased by adequate layouts of curb zones and refinement of operating practices. Multiple routes using preferential lanes may be developed if necessary. Buses using the Shirley Highway HOV facility south of Washington, DC, use two different routes through central Washington. Peak service supply was about 200 vhd during the 1970s (Charles River Associates and Levinson 1988); a more recent figure is 160 vhd (Institute of Transportation Engineers 1988).
Exclusive CBD alignments for BRT are costly, and scarce. Only one such facility has yet been built in the U.S. or Canada: a 1.3-mile tunnel beneath the Seattle CBD for dual-mode buses, which use electric power through the tunnel and diesel power elsewhere. The intended maximum service supply was stated at 145 vhd prior to construction (Municipality of Metropolitan Seattle 1985). Roughly half this figure has been achieved, and the operator now states the practical maximum at 125 vhd (Sound Transit et al. 2001). The principal constraint is vehicle fleet size, for the planned fleet expansion from 236 to 490 vehicles was not carried out4. Conversion for mixed operation of LRT trains and dual-mode buses is planned by 2009.
 
Several Brazilian cities offer very high service levels with segregated busways built in the medians of arterial streets. A two-lane busway, requiring a minimum street width of 75 feet, can accommodate up to 260 vhd. Three-lane busways require wider streets but permit buses to pass each other, facilitating limited-stop operation. Reported volumes range up to 20,000 phd on two-lane busways (Porto Alegre, São Paulo) and 30,000 phd on three-lane busways (São Paulo) (Gardner et al. 1992). However, Hensher (1999) states 1.) the maximum volume carried at an average speed greater than 12 miles per hour is 11,000 phd, in Curitiba, 2.) operating speeds drop toward that of the surrounding traffic flow when volumes exceed this level, and 3.) Curitiba and São Paulo have planned to serve 22,000 phd at speeds exceeding 16 mph, but have not yet achieved these volumes. The PVO levels implied by available data range from 77 p/v (Porto Alegre, São Paulo) to 91 p/v (Curitiba).
These impressive results deserve closer examination to determine whether they might be replicated under U.S. or Canadian conditions. This is by no means assured, given the land-use, economic5, social and cultural factors that distinguish Brazilian cities from their U.S. and Canadian counterparts. A recent case study of Curitiba (Parsons Brinckerhoff 1996) outlines a number of factors, including highly integrated land-use and transit planning, which have given rise to an urban form very different from U.S. and Canadian cities. Curitiba's "trinary" arterial corridors, which include busways6, were built on rights-of-way set aside under a 1943 plan to create Paris-style boulevards (Cervero 1998). CBD distribution, staffing levels, operating costs and environmental impacts are among the issues that should be explored in greater detail.
 
2B)  Peak Service Supply: Rail Rapid Transit
Much more data for RRT performance and operations are available than for BRT modes. This facilitates establishment of feasibility thresholds. There is no clear-cut division between "light rail" and "heavy rail." For the purposes of this paper, we define LRT as including some operation, or capability for operation, on surface streets. This, in turn, implies rigid constraints on train length and maximum frequency of service, and therefore on Co.
Experience worldwide suggests that an RRT service frequency of three minutes (180 seconds), or 20 trains per hour per direction (thd), is easily achieved. Higher service levels are achieved routinely by some operators (Figure 5), but require increased investment in signal and control systems. Operating costs also increase since more supervisory staff are required to ensure schedule reliability. With conventional signaling, the feasibility threshold is about 90 sec (40 thd), but marginal cost per unit of service grows rapidly as the 180-sec threshold is crossed7. The usual factors limiting RRT net capacity are station and terminal capacity (Vuchic 1981), as with BRT, unless vehicle fleet size imposes the governing constraint.
Co is also a function of train length. HRT lines built recently in the U.S. and Canada were designed typically for six-car trains. Longer trains require a trade-off: higher capacity vs. higher capital cost for larger stations. Hence, few HRT lines are built for trains longer than ten cars. The maximum train length for LRT lines using surface streets is generally considered to be four two-section articulated vehicles. Site-specific constraints, such as the spacing of CBD intersections, may limit this to as few as two vehicles, as in Portland and initially in Los Angeles8.
HRT lines with automatic train operation and "moving-block" signaling (MBS) may achieve remarkably high service levels. Co for existing systems of this type is not as great as headways suggest, since designers chose to reduce costs by using smaller vehicles and shorter trains. MBS permits more-frequent service for HRT systems using full-sized stock, and also for tunnel sections of LRT systems. The cost of conventional track-circuit signaling increases rapidly when the designed maximum service frequency is increased from 120 sec (due largely to the need for on-board cab signals to supplement conventional wayside signals). MBS provides greater economy, and permits operation of service supply based on the acceleration and braking capabilities of the rolling stock: 80 sec (45 thd) for 8-car trains, 77 sec (46 thd) for 6-car trains, and 73 sec (49 thd) for 4-car trains (Weir 1992). MBS has the potential for increasing service-supply (Co) levels by up to 30 percent (Parkinson and Fisher 1996).
Busy urban rail systems in large, densely-populated Asian cities such as Hong Kong, Ôsaka, Singapore and Tôkyô maximize Co through use of large vehicles and long trains, rather than maximize service frequency with automated train operation and high rates of acceleration and braking. Sone (1990) states 1.) the practical capacity of an "uncomplicated" HRT line, 40 thd (one every 90 sec) is set by a general rule of 40 sec between trains, plus the average dwell time at the busiest station, plus the time needed for a train to move its own length from a standing start, 2.) the practical limit on a "plain double-track" line is around 300 vhd, set either by dwell time at the busiest station or at the terminal station where train handling capacity is the limiting factor, and 3.) a four-track stub-end terminal has a practical limit of 30 thd.
Vuchic (1981) states that the maximum service frequency for a multiple-branch RRT system is lower than for a single line, or networks where each line operates as a self-contained unit (the typical HRT practice worldwide). In contrast, Parkinson and Fisher (1996) state that carefully designed junctions should not impose constraints on capacity. The BART HRT network has three branches, and was intended to provide a maximum service frequency of 90 sec over its trunk section (requiring trains to operate within five sec of schedule (Nock 1973)). However, BART did not operate trains more frequently than every 3.8 min. until recently (except during a 1974 test without passengers (Strapac 1974)). Following replacement of its train-control computer, BART's current maximum service frequency is 2.5 min. An advanced control system is now under development as a defense-conversion project, incorporating military position-location (GPS) technology to permit a maximum service frequency of 1.3 min. (Middleton 1996). Operation at this frequency should establish whether capacity is constrained by junctions that use flyovers to avoid crossing movements. The operator planned to place its "Advanced Automatic Train Control" system in service at mid-2002 (Baker 2002).
The technical maximum for surface LRT operation is about 120 sec (30 thd), set by operating standards that are characteristic of new installations (e.g. one train per traffic signal cycle; minimum spacing of one city block between trains at all times). Practical considerations such as traffic signal delays, and loading and unloading of wheelchair passengers without high platforms or low-floor vehicles, might place the feasibility threshold in the range of 180 sec (20 thd). More-frequent operation is possible, but this comes at the price of lower operating speed. Toronto operates up to 60 veh/hr over CBD surface tracks (Parkinson and Fisher 1996) over a long-established streetcar system, without traffic signal preemption or multiple-car trains. Two-car trains once operated over the busiest lines during peak periods.
Higher net capacities may be calculated for LRT (and BRT) based on characteristics of the "line-haul," exclusive right-of-way portion of the corridor, but the problem with this approach is that the CBD distribution or terminal element of a BRT or RRT corridor forms an integral part of the corridor "system." It influences the overall operating characteristics and may establish the governing constraint on system capacity. When this is true, achieving the greater capacity of the line-haul portion is possible only through increasing the capacity of CBD distribution or terminal facilities.
Service frequencies up to 20 sec (180 vhd) have been achieved with single railcars operating "on sight" (without a signal system) although Vuchic (1981) notes that these were recorded under exceptional conditions (e.g. very low levels of auto traffic) that generally do not exist today. This was typical of many traditional streetcar systems but is not an effective operating strategy. As with BRT in mixed traffic, service is likely to be slow, erratic and labor-intensive; Cp falls short of levels suggested by the headway, while labor productivity falls far short of theoretical HRT or LRT levels. Owing to inefficient utilization of staff, a rail service operated "streetcar" style achieves only 15-20 percent of the peak-period labor productivity possible with less-frequent, multiple-car trains (Pushkarev et al. 1982).
 
2C)  Peak Vehicle Occupancy
The importance of peak vehicle occupancy as an indicator of consumer choice has been outlined above. This consumer choice usually constrains achievable capacity during peak periods to levels below theoretical capacity. For BRT modes, a "benchmark" figure for average PVO with non-articulated buses is 45 p/v, which is well supported by empirical data (Charles River Associates et al. 1988, ITE 1988). This "benchmark" is significantly higher than the 37 p/v suggested by post-1990 data, but the post-1990 cases are dominated heavily by express bus services using freeways and HOV lanes. Peak vehicle occupancies are higher for BRT services that provide two-way, all-day service with intermediate stops.
Establishment of a similar "benchmark" for BRT services using articulated buses posed a difficult problem. Published data pertaining specifically to peak vehicle occupancies carried by articulated buses are not available. Data for busway and freeway/HOV services in cities using articulated buses provide no breakdown by vehicle type. A "benchmark" of 70 p/v reflects relative vehicle sizes and seating capacities, but the authors could find no supporting data for U.S. or Canadian busway, freeway or HOV services. Peak vehicle occupancies observed by the authors in Pittsburgh and Ottawa fell 30 to 50 percent below this figure9. Data reported by State Transport Authority, Ballment and Hammond (1993) imply a PVO of 72 p/v for the guided busway in Adelaide, Australia, and data reported by Turnbull and Hanks (1990) suggest that articulated buses may have carried such loadings on Pittsburgh's East Busway (in 1990)10. In addition, Seattle has found that planning for a PVO greater than 80 percent of seated capacity -- that is, 46 p/v -- aboard articulated buses using its downtown transit tunnel leads to periodic overloads and may discourage ridership (Sound Transit et al. 2001).
Peak vehicle occupancy in terms of passengers per unit of vehicle length may be lower for articulated buses than for standard buses, reflecting some aspect of consumer choice. Pittsburgh's East Busway has a clear "division of labor" between standard and articulated vehicles, with the latter assigned almost exclusively to "local" services. Observed PVO for "express" services with standard buses (3.0 p/m) exceeds "local" services operated by articulated vehicles (2.6 p/m).
In Ottawa, articulated buses share transitway "local" services during peak periods with standard buses, which dominate the peak-period "express" services. In 1992, observed vehicle occupancies were higher for "local" (1.8 p/m) than for "express" (1.3 p/m) services. But vehicle occupancies for both categories of service during the p.m. peak were lower for articulated (1.3 to 1.9 p/m) than for standard vehicles (1.5 to 2.3 p/m). Observed PVO was significantly higher during the a.m. peak in 2000, ranging from 2.1 to 3.6 p/m for local and 2.1 to 2.3 p/m for express services. The number of peak-hour services worked by articulated buses was reduced from about 25 percent in 1992 to 11-14 percent in 2000, and these vehicles are now mostly confined to local (base-service) routes during peak hours. Observed PVO for local services during the a.m. peak did not differ between standard and articulated buses, suggesting improved allocation of resources by the operator.
A consumer preference for standard over articulated buses, if it exists, would probably reflect perceived differences in ride quality. Vuchic (1981) notes that a disadvantage of articulated buses is "rather low riding comfort in the rear section." In Ottawa, passengers appear to cluster in the front section of articulated buses working peak-period transitway services, causing a significant difference in PVO between sections. The authors note that Ottawa's proof-of-payment fare system for articulated buses permits passengers to board using all three doors, and that standees aboard buses and railcars typically cluster near doors.
Data available are not adequate for supporting firm conclusions on these points. Further refinement would require additional performance and operational data for BRT facilities providing two-way, all-day service (Figure 6).
For RRT, empirical data (Charles River Associates et al. 1988, Werner 2001, Pushkarev et al. 1982) suggest that PVO levels significantly greater than 100 p/v are not likely to occur in most U.S. cities, particularly on LRT systems. In the U.S. and Canada, vehicle occupancy levels as high as 150 p/v have been recorded only in the most densely populated and congested metropolitan areas, and then only on the busiest lines. Hence the authors' use of 100 p/v as a "benchmark" for HRT and LRT (Figure 7).
Post-1990 data for 16 corridors in nine HRT systems away from the most crowded and congested metropolitan centers (Table 2c, below) suggest that PVO falls short of 100 p/v. Of these systems, only Washington, DC, carried peak loads greater than 100 p/v. The passengers per vehicle parameter-–although convenient and useful for this analysis-–masks differences in vehicle size. Parkinson and Fisher (1996) suggest passengers per meter of vehicle length (p/m) as the standard for vehicle occupancy, to place all systems and modes on an equal footing. Using this standard, the highest PVO among these 16 HRT corridors is 5.6 p/m in Chicago and Vancouver. The median is 4.3 p/m, corresponding to 99 passengers per 75-foot (23-meter) railcar.
 
Post-1990 data for 17 corridors in 12 LRT systems (Table 2b) suggest that the 100 p/v benchmark falls at the median peak vehicle occupancy carried by recently-constructed LRT facilities. In terms of passengers per meter of vehicle length, the median, 4.0 p/m, is just below that for HRT and corresponds to 104 passengers per 85-foot (26-meter) vehicle. The highest peak vehicle occupancies among the 17 corridors are 4.9 p/m in Portland, 4.7 p/m in Sacramento and 4.6 p/m in Calgary. The Portland figure approaches that for Boston's Green Line LRT system: 5.1 p/m (1994).
HRT and LRT medians for PVO (p/m) are 40-50 percent greater than the BRT median, 2.8 p/m. In fact, the post-1990 RRT medians exceed all of the BRT values shown in Table 2a.
 
2D)  Peak Service Supply and Peak Vehicle Occupancy: Theory vs. Practice
Previous modal comparisons and planning studies have often assumed significantly different, usually higher, service-supply and vehicle-occupancy levels (Figures 4 to 7) from those observed in actual operation. Results are therefore biased in a manner that does not reflect actual experience.
Boyd, Asher and Wetzler (1973) outlined a service supply of 576 vhd to provide "capacity" of 31,600 phd. This implies a PVO of 50 p/v, marginally higher than empirical data indicate for BRT with non-articulated vehicles. But the postulated Co level is nearly 300 percent greater than the maximum yet offered on any U.S. or Canadian BRT facility other than the Lincoln Tunnel with its unique CBD terminal11. Biehler (1988) assumed PVO levels of 200 p/v for LRT and 100 p/v for articulated buses, well above those suggested by empirical data. Meyer, Kain and Wohl (1965) postulated 79 seats per railcar, 50 seats per bus, and service levels sufficient to provide seats for all passengers12. This "no-standee" model underestimates peak vehicle occupancy, for RRT in particular, and therefore overestimates Co levels necessary to attract and move given traffic volumes. The projected modal Co levels were high: 480 vhd as the "capacity" of a two-lane busway (based on observed highway-lane capacity). This would require extraordinary preferential-lane measures involving multiple CBD streets or a large off-street terminal. RRT "capacity" figures, 320-720 vhd depending on line and terminal configuration, were based on theoretical maxima calculated by Lang and Soberman (1964), and were not compared with actual operating experience. Much has been published in the nearly four decades since this pioneering study, but more recent studies do not include peak service supply and PVO details.
It is now clear that U.S. and Canadian consumers will not accept the PVO standards used for many planning studies during the 1970s and 1980s. This is demonstrated by the stabilization of observed PVO at similar levels in corridors having a wide range of demand characteristics (Table 2a-2c). In Portland, operator staff members have concluded that Portland transit passengers will not accept LRT vehicle occupancies greater than 135 p/v (4.9 p/m), except for special events. Prior to construction, LRT planners estimated the capacity of each 90-foot vehicle at 166 passengers (6.1 p/m) based on 76 seated passengers and four standees per square meter. This loading standard was based on (West) German experience (Schumann 2000).
 
2E)  Modal Capacity: Passengers per Hour and Passengers per Weekday
The analysis presented in Table 1a and Table 1b uses service supply levels of 100, 150 and 200 vhd for low, medium and high service-supply levels for a "typical" busway or HOV facility. The 200 vhd level slightly exceeds the maximum yet offered on a U.S. or Canadian busway. As noted above, higher service levels would require carefully-developed preferential measures for CBD distribution, and may be achievable only when there are no intermediate stops, as in the exceptional case of the Lincoln Tunnel XBL. For a standard busway, it may be overly optimistic to assume that service levels higher than the current "record" can be achieved and maintained with good schedule reliability. As noted in Section 2C above, a peak vehicle occupancy for articulated buses of 70 p/v is not supported by available data, and substantially exceeds the p/v levels observed in actual operation. Hence, for applications in the U.S. and Canada, the modal capacity of BRT with articulated buses may fall below the levels presented.
For LRT, the "low," "medium," "high" and "exceptional" service levels are based on eight, 20, 30 and 40 thd, respectively, implying that trains operate every 7.5, three, two and 1.5 minutes. The low and medium service levels can be attained with on-street operation without difficulty. However, it might not be possible to maintain good schedule reliability over street track at the high service level; this depends on factors such as number of traffic signal cycles per hour, and the number of seconds of green for LRT per cycle. The "exceptional" LRT supply level would require an exclusive CBD alignment such as provided for Boston's Green Line or San Francisco's Muni Metro LRT systems. Neither operates this level of service, owing to constraints imposed by terminal configuration and signal systems (San Francisco), or long station dwell times reflecting the lack of level boarding (Boston). San Francisco has installed a new MBS system, and once regarded a maximum service frequency of 70-90 sec (40-51 thd) as a practical short-term goal (Welty 1992). Boston anticipated that new low-floor vehicles would provide substantial reductions in station dwell time (Jones 1993). Both operators have experienced serious technical difficulties, and have not been able to implement the planned service improvements.
For HRT, the "low" peak service supply of 60 vhd implies six-car trains operating every 360 sec (10 thd). The "medium" 120 vhd level implies six-car trains every 180 sec (20 thd). The "high" 160 vhd level implies eight-car trains every 180 sec (20 thd). The "exceptional" 300 vhd level, based on Sone (1990), is not typical of U.S. or Canadian practice but is achieved in the Tôkyô area and is well within the prospective service range of HRT with MBS (e.g. eight-car trains every 96 sec, or 38/hr). Significantly higher capital costs are associated with high service levels.
In order to estimate maxima for service consumption (phd) commensurate with various service-supply levels, this analysis uses the "benchmark" modal vehicle occupancy figures outlined in Section 2C. To estimate commensurate levels of weekday ridership, the analysis uses three values of peak traffic share (PTS), an indicator of temporal distribution of weekday transit service consumption. The "medium" 13 percent PTS is a "benchmark" average for U.S. and Canadian cities13. The "low" 10 percent PTS value is a threshold; smaller values reflect characteristics of only a small number of corridors14. The "high" 18 percent value is characteristic of facilities dominated by peak-period, peak-direction traffic. BRT services tend to have higher PTS values than rail services, suggesting that BRT modes are more strongly dominated by CBD work trips.
Except in the most crowded and congested U.S. and Canadian cities, the maximum likely BRT consumption level along a two-lane facility is about 10,000 phd (rounding from Table 1a); significantly higher volumes would require very high service levels, a 100 percent articulated fleet and carefully-designed CBD distribution measures. LRT using CBD streets and two-car trains can attain about 6,000 phd. Calgary has attained more than 9,000 phd on CBD streets with 25 thd, three-car trains and high-platform loading. Given an operating environment permitting four-car trains and a frequency of service sufficient to provide 100 vhd, LRT can attain about 12,000 phd. The practical maximum for LRT, given a segregated CBD alignment, is about 15,000 phd15. Except under unusual demand conditions that lead to peak vehicle occupancies significantly greater than 100 p/v, 20,000 phd probably exceeds achievable capacity with LRT in most U.S. cities, and represents an upper limit for HRT unless very high service levels are provided. Even then, a volume significantly greater than 40,000 phd is probably not achievable unless multiple tracks are provided -- and, of course, sufficient demand exists in the corridor.
These maxima fall below levels often assumed for "modal capacity," and the associated service levels are higher than typically assumed. Such results come as no surprise. Service-supply levels on U.S. and Canadian fixed-guideway facilities reflect policy, technical and financial factors rather than systematic calibration of supply to demand. Initial vehicle fleet size for most U.S. LRT systems opened during the 1980s and 1990s was not adequate for the peak-hour traffic that developed following service start-up. This has occurred most recently in Salt Lake City, which opened a new LRT line at the end of 1999. Service-supply improvements had to await procurement of additional vehicles, a lengthy process even when not delayed by political wrangling.
These maxima -- predicted in the context of U.S. and Canadian cities, excluding the largest, most congested and crowded ones -- also fall below fixed-guideway passenger volumes reported for some cities in other countries (Vuchic 1981, Parkinson and Fisher 1996), and below the volumes some systems have carried in the past. Again, such results come as no surprise. Passenger volumes carried by a particular BRT or RRT facility reflect the characteristics of the city where it operates. There is a general similarity of characteristics among U.S. and Canadian cities where new fixed-guideway transit facilities are under serious consideration: city size, economic levels, automobile ownership and use. An identical facility in a much different city, however, might carry more traffic -- or less, depending on the dimensions of demand. The Tôkyô subway system's Chiyoda Line (14.9-mile, double-track HRT) provides a maximum service supply of 270 vhd. Consumption parameters include (at 1995): 1,200,000 pass/day, maximum peak volume of 85,618 phd, PVO of 317 p/v and 15.9 p/m of vehicle length, and 7 percent PTS (Takatori 1995). These results would not necessarily hold for a similar facility in another city with different travel-demand characteristics.
 
2F)  Regression Analysis of Modal Capacity
The regression analysis presented in Tables 2a-2c demonstrates a strong correlation between peak-period service consumption and service supply, and also provides least-squares regression models for supply-side analysis of modal capacity. This analysis used post-1990 data for busway, transitway and HOV services, and 1994-2000 data for RRT. Data were separated into BRT, LRT and HRT categories. The four most crowded and congested U.S. and Canadian metropolitan centers, Boston, Montréal, New York and Toronto, were excluded. Hence, Lincoln Tunnel and Boston Green Line data, although included in the tables for comparison purposes, were excluded from the analysis. Buffalo's hybrid RRT line, in many respects an HRT facility adapted for CBD surface operation, has an atypically low level of consumption per unit of service, suggesting a "background" of demand substantially different from other LRT corridors. In Dallas, the northward corridor appeared to have an atypically low level of service consumption per unit of peak-period service. This reflects a great increase in peak-period service to reduce overcrowding, permitted by delivery of additional vehicles, which took place a few months prior to data collection (the line has since been extended). Both were excluded from the analysis.
The Ottawa transitway network was excluded from the analysis and from the tabulation, owing to 1.) discrepancies among published data, and 2.) incompatibility with published data from other BRT and RRT facilities. The maximum passenger volume carried by the system is reported at 9,000 phd by Bonsall (1987) and FTA (1992), at 10,000 phd by Levine (2000), Martinelli (1996) and Sawka (1999), and at 11,000 phd by Turnbull and Hanks (1990). ITE (1988) reports maximum passenger volumes for the Southeast and Southwest transitways of 8,100 and 4,200 phd, respectively. Based on the configuration of the network and the anticipated location of the maximum-load point before construction (Wilkins 1985), the sum of the ITE numbers should approximate the reported system maximum volume. But this sum, 12,300 phd, is 12 to 37 percent higher than the published "maximum" figures.
This discrepancy, and reconciliation of published data with personal observations of peak-period volumes, led the authors to suspect that published data for Ottawa represent flows rather than volumes. Volume is defined as the number of passengers (or vehicles) passing a given point in one hour. Flow (or flow rate) is defined as the number of passengers (or vehicles) passing during some other interval, scaled up to an hourly rate (Meyer and Miller 1984). Flows of 9,000 - 11,000 phd imply, for example, 750-917 passengers during the busiest five-minute interval. Actual hourly volumes might be in the range of 3,500 - 5,000 phd, or 45-70 percent lower than published "maximum" figures16.
Service-supply data were converted from vehicles per hour (vhd) to meters of vehicle length per hour (mhd) to remove variations caused by differences in vehicle length.
The regression models explain 80 to 96 percent of the variation in peak-period service consumption, suggesting a strong relationship. Transit service supply, of course, does not "cause" service consumption. However, within a fairly broad range of demand characteristics, it is possible to make a reasonable prediction of peak-period service consumption for a given supply level.
The authors used t-tests to address the possibility of random chance; all regression coefficients were found to be significantly different from zero (5 percent level; he intercept terms were not significantly different from zero). Results for the regression coefficients were very highly significant (0.1 percent level). In addition, the three regression coefficients were found to be significantly different from each other (5 percent level).
Using the BRT and LRT regression models to predict the supply levels commensurate with a given consumption level, a peak-period volume of 3,000 phd is correlated with an LRT service supply equivalent to 34 vhd given 23-meter (75-foot) vehicles. The BRT service supply is equivalent to 86 standard buses or 58 articulated buses per hour. Implied peak vehicle occupancies are 88 p/v for LRT, 35 p/v for standard buses and 52 p/v for articulated buses. This suggests that the p/v figures in Table 1a may be higher than warranted by the broad parameters of transit service demand in U.S. and Canadian cities during the 1990s; more than 10 percent higher for LRT, and 30-35 percent higher for BRT. In addition, the BRT service-supply level is 36 percent higher than for LRT at the same consumption level. This comes as no surprise, given the differences in PVO characteristic of each mode. The HRT service level associated with a peak-period volume of 3,000 phd is six percent greater than for LRT, a difference which is probably not significant.
The authors sought to avoid problems related to uncertainty and bias by excluding certain data, presented in Tables 2a-2c, from the regression models. Inclusion of authors' personal observations and authors' estimates in Tables 2a (BRT) and 2c (HRT) would not result in significant changes in the regression coefficients. For the LRT regression model, inclusion of observations and estimates for Baltimore, Salt Lake City and St. Louis (Table 2b) would produce a significantly different, and smaller, regression coefficient (3.83) and a weaker relationship (R2 = .74). However, inclusion of these observations and estimates, together with those for Philadelphia and San Francisco (p.m.), would result in a significantly different, and greater, regression coefficient (4.44) and a stronger relationship (R2 = .89). The implication in the latter case is that a BRT service level would need to be up to 50 percent greater than LRT for a given consumption level. It is true that the Philadelphia and San Francisco networks do not resemble the "typical" suburb-to-downtown LRT facilities opened in the U.S. and Canada over the past two decades, however, these results indicate a need for additional research.
 
2G)  Regression Analysis: Development, Interpretation and Significance
The authors emphasize that the regression models presented in Tables 2a-2c are not the product of consumer surveys ("revealed preference" or "revealed choice"), but of direct observation of consumer behavior – "observed choice." Therefore, certain issues of statistical validity that are characteristic of preference and choice surveys do not apply. For example, a preference survey asking response to choose among one’s own auto, a red bus and a blue bus17 might lead to biased results, for the otherwise-identical transit options are not likely to attract identical shares of respondents. However, researchers able to demonstrate through direct observation that red buses do attract larger numbers of consumers than blue buses, all else equal, face the challenge of explaining why this behavior occurs. With reference to the consumer behavior documented herein, the models in Tables 2a-2c are a tentative first step towards this goal.
The models establish strong and statistically-valid correlation between peak-period service supply and consumption.  This correlation does not imply causation. On the other hand, it is the consumers who decide what levels of consumption will be observed. To deny the important role of service supply in shaping consumer choice is to overlook the obvious: transit service supply "permits" consumption, and given levels of consumption will therefore not occur without adequate supply.
The authors, based on the principle of parsimony, chose to include only service supply as the determinant of consumption in the regression models. Other factors certainly influence the outcome, but their influence is much less clear. For example, commercial (schedule) speed may influence the consumer choices that determine PVO. However, data are not available to permit quantification of the likely relationship between passenger speed, relative to that provided by autos in individual corridors, and PVO. It is also possible that factors apparently extraneous may influence the outcome. For example, overall vehicle length and height do not influence available floor space per meter of vehicle length, but might influence consumer choice and therefore PVO. Once again, data are not available; additional research is clearly indicated.
The authors did not include demand factors (e.g. corridor population, population density, CBD employment, CBD parking supply and cost) in the regression models, for this is not appropriate for supply-side analysis. Such analysis is conducted against a "background" of demand factors (those that "cause" consumption) that are postulated to remain constant as other factors are changed. Inclusion of demand factors as "controls" is therefore not appropriate. Statistical validity is, however, not compromised unless changes in supply or other "non-demand" factors also cause changes in demand factors, thereby introducing bias. In other words, such bias would occur as the result of "induced" (or "suppressed") demand – in the literal sense. Data are not available to establish whether this occurred in the various BRT and LRT corridors in Tables 2a-2c. (The authors advocate strict adherence to the definitions of "demand," "supply" and "consumption;" "induced consumption" is not at all synonymous with "induced demand.")
Strictly speaking, the models cannot "predict" the responses of consumers to changes in service supply. This fact comes as no surprise, because such prediction requires consideration of demand factors – which are not incorporated into supply-side models.  It should be obvious that, if service supply could be increased without limit, consumption would eventually reach the theoretical maximum established by demand factors.
However, from the supply-side perspective, observed consumer behavior makes it possible to determine peak-period consumption levels associated with a given supply level, and to do so within a rather broad range of demand factors. Supply-side analysis cannot replace demand-side analysis, of course, but provides useful verification of demand forecasts. The fact that peak service changes in Portland and Los Angeles (Section 4A, below) were soon followed by consumption changes consistent with the regression model (Table 2b) illustrates 1.) the strength of the correlation and 2.) the existence of demand not served previously.
 
3. Rail vs. Bus Peak Vehicle Occupancy:A Matter of Consumer Preference?
The observed difference between BRT and RRT peak vehicle occupancy implies that RRT generates greater consumption per unit of peak-period service, all else equal. For a given consumption level, roughly 35 percent more service would have to be supplied by a BRT alternative. This suggests an observable consumer preference for RRT over BRT with reference to peak-period travel.
Some consumers may find that seated travel is equally acceptable aboard either mode, but that standing travel is significantly less acceptable aboard BRT. Such potential passengers would ride RRT even if not certain of obtaining a seat, but would be willing to ride BRT only if seats could be obtained.
Other consumers may prefer seated travel aboard a railcar to seated travel aboard a bus. Such individuals would therefore be more likely to ride RRT than BRT.
Still other consumers may prefer seated travel aboard a bus to seated travel aboard a railcar, and would therefore be more likely to ride BRT than RRT.
The difference between BRT and RRT peak vehicle occupancy must reflect factors in addition to differences in vehicle size or interior configuration. The average railcar is not 30 to 50 percent wider than the average bus, nor does it have 30 to 50 percent more net floor space per unit of vehicle length. This is true in particular of LRT vehicles.
The PVO difference may reflect differences in service characteristics, but on-site observations do not support this hypothesis. U.S. and Canadian BRT operations are dominated by freeway and HOV express-bus services, where vehicles operate without stopping between suburban neighborhoods or park/ride lots and CBD destinations. Lower PVO might be an inherent characteristic of such service, owing to the lack of intermediate stops where shorter-distance riders willing to travel as standees might board. However, this is not apparent in Pittsburgh and Ottawa, where "local" routes serving intermediate stops share facilities with "express" routes that "pass through" intermediate stations.
Lower levels of demand in existing BRT corridors than in existing RRT corridors might explain the difference in PVO, but available data do not support this hypothesis. Peak vehicle occupancies are lower through the Lincoln Tunnel, for example, than on several recent LRT systems, but Lincoln Tunnel peak period volumes are much higher.
Recent LRT systems are not able to provide the same service supply as existing BRT facilities (owing, in the majority of cases, to constraints imposed by vehicle fleet size), which might explain why most LRT lines carry consistently higher peak vehicle occupancies. Again, however, available data fail to support this hypothesis. In Los Angeles, the El Monte Transitway supplies 66 percent more service during the busiest hour (mhd) than the LRT Blue Line, but carries just eight percent more peak-hour traffic (p/h) (Tables 2a-2b). Los Angeles, it should be noted, charges premium fares for BRT services (based on distance traveled on freeway or transitway), but not for LRT or HRT18. In Pittsburgh, the East Busway supplies 2.3 times as much peak-hour service (mhd) as the light-rail system, and carries nearly twice as much peak-hour traffic. However, as in Los Angeles, LRT peak service levels (vhd) are at the practical maxima, while peak vehicle occupancies are high enough to suggest the existence of unserved demand (see Section 4A, below). The disparity between LRT and BRT peak vehicle occupancies persists in corridors with a wide range of demand characteristics, and over a very wide range of peak volumes: 1,000 - 10,000 phd for LRT, and 400 - 32,000 phd for BRT.
Most previous studies of consumer choice found no special attractiveness for RRT (Ben-Akiva and Morikawa 1991; Deen and Pratt 1992), but did not focus on peak-period markets. Consumers in a single corridor market are rarely able to choose between equivalent BRT and RRT services, so direct comparison to determine relative consumer attractiveness would be difficult. Few fixed-guideway corridors are closely similar, and BRT facilities are scarce in the U.S. and Canada.
Different user perceptions may lead to greater attractiveness of RRT per se over BRT or for standard buses per se over articulated buses. More plausibly, characteristics typical of each mode (Tennyson 1989, Vuchic 1992) might lead to different user perceptions of BRT and RRT service: the two may not be "perfect substitutes." For example, users may perceive RRT as offering a smoother ride, less discomfort owing to less-frequent starts and stops, and greater reliability of service. This in turn might lead to greater willingness among prospective customers to travel aboard railcars as standing passengers -- which may be inferred from available data, and has been recognized by Seattle planners.
Achievable-capacity estimates for mixed LRT/bus operation in the Seattle CBD transit tunnel are based on different loading standards for each mode. The LRT standard of 137 p/v is based on observed PVO in Portland, and implies 4.7 - 5.0 p/m given the planned vehicle length (90-95 feet). The bus standard of 46 p/v, or 2.5 p/m, reflects actual experience in Seattle. The operator has found that planning for an overall average PVO of 80 percent of seating capacity usually results in periodic overloads, requiring some passengers to stand. The operator has also found that regularly exceeding this PVO level results in a large number of complaints. It has concluded that a PVO greater than 80 percent of seating capacity may discourage ridership on bus services using the CBD tunnel (Sound Transit et al. 2001).
Vuchic (1992) states that RRT provides a higher quality of service than BRT and thus attracts greater ridership. Tennyson (1989) concluded that RRT would attract 34-43 percent more passengers given equivalent service conditions. Peak vehicle occupancies for RRT are 30-50 percent greater than for BRT. This, together with results of the regression analysis in Section 2F, appears to confirm Tennyson's prediction, at least for peak-period travel markets.
BRT service offering one-seat rides does not appear to attract greater patronage than a trunk/feeder network (with BRT or RRT as the trunk mode) requiring transfers. The "transfer penalty" suggested by consumer surveys should manifest itself in terms of peak-period vehicle occupancy: express bus services offering one-seat transportation between suburban residences and CBD destinations should have higher vehicle-occupancy levels than line-haul services requiring transfers, all else being equal. Comparison between BRT and RRT vehicle-occupancy levels reveals exactly the opposite pattern, and comparisons within the BRT mode are inconclusive19. These results are not surprising. Disutility created by the need to transfer would be offset by increased service frequency on busy portions of a trunk/feeder network, particularly if that network provides high connectivity between a large number of origins and destinations. High transfer rates are characteristic of such systems, which include the most successful transit operators in the U.S. and Canada (e.g. San Francisco, Toronto).
Martinelli (1996) states that "there are fewer passengers per vehicle for busways," but refers clearly to differences in vehicle size and seating capacity, not to observed PVO. He also states "The capacity of busways is far more flexible and often can be used more efficiently than that of light rail." Once again, this refers clearly to theoretical (roadway) capacity (cw), not to achievable utilized capacity (Cp).
While additional research is clearly needed to identify and quantify the underlying cause or causes, greater peak-period vehicle occupancy for RRT is well supported by available data. The authors emphasize that such research will need to give paramount importance to consumer behavior, perceptions and attitudes rather than the researchers' perspectives. Vuchic et al. (1994) state that the "image" of a transit system is derived from operating conditions rather than vehicle type: that is, bus services in mixed traffic present a much different image than high-speed BRT services on exclusive rights-of-way. But public input received by Louisville planners is rather different, summarized by a consultant as follows: "If this [project] is going to cost hundreds of millions of dollars, I'd rather be riding a train" (Levine 2000). Kain (1992) states that an "inflexible" system such as LRT, requiring "virtually all" users to transfer from other transit services or autos, cannot be strongly attractive. But observations from Los Angeles and Pittsburgh, where LRT peak vehicle occupancies are substantially greater than for BRT services (Tables 2a-2b), fail to support this hypothesis. (As noted above, Los Angeles charges higher fares for most BRT services, but Pittsburgh does not.)
It is striking that three recent study documents regarding BRT are silent on the issue of peak vehicle occupancy (Federal Transit Administration 1998, Smith and Hensher 1997, U.S. General Accounting Office 2001). The PVO issue will certainly influence the results of FTA's current "Bus Rapid Transit Demonstration Program."
The difference between observed BRT and RRT peak vehicle occupancies has important implications for cost analysis -- particularly when BRT is compared to LRT. If a BRT alternative had to provide a peak-period service supply 30 to 40 percent greater than LRT in a given corridor, operating and certain capital costs for BRT would have to be adjusted upward. Labor productivity, for example, is an important issue. Given equivalent operating speeds, LRT with self-service fare collection and one-person operation of trains can achieve greater peak-period labor productivity than BRT, by factors in the range of 6:1 to 11:1. This advantage would be narrowed if an equivalent BRT service achieved higher speed, and of course is only part of the comparison; vehicle and guideway maintenance labor and the overall capital costs must be included in the analysis.
 
4)  Further Comparisons and Comments
 
4A)  Does Unserved Demand Exist?
It is possible, in principle, that no U.S. or Canadian fixed-guideway corridor has unserved demand -- which would mean that there are no positive changes in service levels that would produce positive changes in consumption. If so, then peak vehicle occupancy levels would simply decrease with each increment of additional service, with no net increase in ridership, as illustrated in Figure 1. But such a negative conclusion is difficult to support. Data from several LRT systems in particular imply high PVO and service levels at or near their practical site-specific maxima. This suggests that, at least in some cities and corridors, unserved demand does still exist.
As a specific recent example, Portland implemented a service increase in September 1998 that was followed by a large increase in ridership. Weekday vehicle-trips were increased by 31 percent, from 388 to 510. Weekday services (train-trips) were increased by 19 percent, from 213 to 255. Eastside Line ridership at October 1998 was 37,200 passengers per weekday (p/w), up by 34 percent from the October 1996 level of 27,700 p/w. An expanded vehicle fleet permitted a peak-period service increase from 17 to 22 vhd, or 29 percent. This was accompanied by a 17.5 percent peak-hour ridership increase, from 2,100 to 2,500 phd. PTS declined from 8 to 7 percent, indicating that the majority of the weekday ridership growth occurred outside of the peak-period, peak-direction travel market. This may reflect "reverse direction" travel by passengers riding through from the new Westside Line, which opened in September 1998, to employment centers east of the CBD.
Another example of a supply increase meeting previously unserved demand is the apparent Los Angeles Blue Line data-discrepancy cited in Moore (1992). University of Southern California researchers counted 24,100 boardings in June 1991. Two months later, the operator counted 32,587 boardings. Moore (1992) labels the latter an "outlier," and implicitly questions its veracity. However, peak service frequency was increased from 10 minutes to 7.5 minutes prior to the second count. A 33 percent service-supply increase, during the hours when roughly 50 percent of weekday transit ridership occurs in the U.S., is a very significant change, and corresponds well to the 35 percent increase in reported boardings. The change to which Moore (1992) attributes Blue Line ridership, a reduction in parallel bus service, was much too small to account for this20.
 
4B)  Phased Introduction
BRT has the very important advantage that preferential measures, reserved lanes and segregated alignments may be implemented as short, unconnected initial segments, to be linked as traffic warrants and finances permit. Ottawa provides the textbook example of phased BRT implementation in North America (Kain 1992, Parsons Brinckerhoff 1996). Its current transitway network consists of three discontinuous segments of segregated busway linked by CBD preferential lanes and a section of mixed-traffic operation. Completion of a fully-segregated network is not likely for some years. Similar step-by-step upgrading of streetcar or tramway networks has been done in Pittsburgh, but is not an option for most North American cities, where surface rail transit no longer exists.
 
4C)  External Costs and Impacts
Relative capital and operating costs are important issues in the fixed-guideway modal choice, but not the only ones. External costs of traffic and environmental impacts of high-volume BRT operation over CBD surface streets, even when there is unused capacity (Kain 1992), may be considerable and would offset part of any capital-cost difference between BRT and RRT. These costs would, of course, be minimal if large numbers of BRT passengers switched from previous transit services which were then withdrawn (or from autos to BRT). Transporting 3,000 phd would require a service supply of 60 to 90 buses or 34 LRT vehicles (nine four-car trains) per hour. LRT has the advantage in terms of traffic impacts: nine LRT trains generate 85 percent fewer traffic movements than 60 buses21. Each train has the capacity of several buses, and loading is more efficient since passengers destined for many different destinations use the same trains, changing to connecting services at transfer points. Relative emissions, impacts on other street traffic and costs remain to be worked out.
 
4D)  Other Analytical Issues
A much stronger case for rail transit in the U.S. and Canada may be made than some previous studies have found. The first comprehensive analysis, Meyer et al. (1965), could draw on current experience with high traffic volumes in only one city, New York, and, for historically high traffic, on wartime conditions that restricted automobile use. LRT systems were few and BRT still an early experiment. This led to overestimates of achievable capacity for BRT, especially for high (hypothetical) traffic volumes, and therefore underestimates of BRT costs relative to RRT. Overestimation of practical maximum capacity leads to underestimation of capital and operating costs for a given traffic volume, and to overstatement of the cost difference between modes. Meyer et al. placed the economic "break-even" point, the traffic level at which RRT has a total cost advantage, at a very high traffic level -- 50,000 phd, well above the modal capacities implied by empirical data from the U.S. and Canada. Pickrell (1985) placed the BRT/HRT break-even point far above the levels that have ever been achieved, anywhere, over a single lane or track -- 200,000 - 340,000 phd for a ten-mile corridor, depending on HRT capital cost.
Accurate prediction of transit service consumption from demand parameters alone is not likely -- unless supply elements influencing consumption are correlated strongly with the demand parameters used for the analysis. These parameters would need to model the underlying relationship among demand, supply and consumption. Otherwise, results would be ambiguous at best. The model for RRT ridership prepared by Gordon and Willson (1984 and 1985) provides an example. This is based on multiple-regression analysis of a cross-section of data published by the International Union of Public Transport (UITP): daily passengers per kilometer of route, city population density, GNP per capita, city automobile registration per capita, and average distance between stops. Published results, as one might expect from the lack of service-supply parameters among the inputs, are mixed: good agreement with actual LRT ridership in Baltimore and Portland (prior to 1998), overestimation of HRT ridership in Miami (Gordon and Willson 1985), and underestimation of LRT ridership in Los Angeles (Kain 1988). Models including some measure of service supply may produce results of significantly greater utility (Setty 2002)
Various planning studies for specific corridors contain unrealistic assumptions regarding supply parameters and the demand-supply-consumption relationship. San Diego Transit Corporation (1975) proposed a busway alternative to LRT, and projected a corridor ridership increase from 15,000 to 64,000 p/w over 20 years (to 1995). But PTS was assumed to remain fixed at a remarkably low 5.3 percent as ridership quadrupled. Moore (1992) also assumes that consumption increases generated by introduction of fixed-guideway service will be distributed evenly throughout the service day. Such predictions conflict with results produced by modal-split models, actual experience elsewhere, and common logic. The increased ridership generated by BRT or RRT occurs as an increased share of travel in the dominant market -- CBD work trips -- is attracted by the new, faster service. This is reflected by an upward shift in PTS.
Such problems are characteristic of fixed-guideway planning in Los Angeles. "System capacity" for the HRT Red Line was initially estimated at 30,600 phd, based on a planned PVO of 170 p/v, or 7.4 p/m (U.S. Department of Transportation et al. 1983). In addition, ridership and maximum peak volume were forecast at 376,375 p/w and 30,000 phd, implying an 8 percent PTS. Such assumptions were carried forward throughout the planning process, and are typical of those used for planning of other rail corridors in Los Angeles. For the planned Eastside HRT corridor (replanned subsequently as LRT), Los Angeles County Metropolitan Transportation Authority (1994) used "normal" and "maximum" peak-period vehicle occupancies of 169 p/v and 301 p/v, corresponding to 7.4 p/m and 13.2 p/m, respectively. The forecast weekday ridership and maximum peak volume of 65,902 p/w and 4,000 phd, respectively, implied a PTS of 6.1 percent. For the LRT Blue Line, Parsons Brinckerhoff/Kaiser Engineers (1984) assumed a "design load" of 175 p/v, or 6.4 p/m, to estimate peak capacity. The forecast weekday ridership and maximum peak volume were 54,702 p/w and 2,668 phd, implying a PTS of 4.9 percent. These examples illustrate 1.) significant overestimation of likely maximum peak vehicle occupancies, and 2.) significant overestimation of off-peak and reverse-direction ridership relative to peak-period, peak-direction traffic.
Such problems are not confined to RRT planning. In Adelaide, the operator's estimate of the Northeast Busway maximum "capacity" (STA et al. 1993) implies that tripling of peak service supply (from 60 to 180 vhd) would produce a 39 percent increase in PVO (from 72 to 100 p/v). This is not likely without substantial increases in demand factors (e.g. population and employment levels, road congestion, fuel and CBD parking costs). With PVO at the "current" level, the corresponding maximum (about 13,000 phd) falls nearly 28 percent short of the specified "capacity."
A capacity analysis performed for Pittsburgh assumed 80 p/v for standard buses and a service supply ranging between 90-150 vhd, resulting in "capacity" figures of 7,200 - 12,000 phd (Baker). The assumed PVO (p/m) is nearly twice the value observed on Pittsburgh's busways, and observed peak volumes are 40-60 percent lower than projected.
An alternatives analysis performed for Ottawa assumed "capacity" levels of 54 p/v for standard buses, 102 p/v for articulated buses, and 117 p/v for LRT vehicles (Kain 1992). Published "maximum volume" figures imply PVO levels of 50-52 p/v (3.4-3.8 p/m). The analysis overstated the net capacity of BRT services by about 50 percent (or more, given the lower peak vehicle occupancies suggested by on-site observation: 30-38 p/v; 2.3-2.9 p/m). Hence, it underestimated the service supply corresponding to given consumption levels, along with associated costs, and overestimated BRT performance relative to RRT. These problems appear insignificant compared to non-technical and political factors which influenced the modal choice (Belobaba 1982). However, if the operator wished to realize the designed "maximum" of 15,000 phd, it would need to operate 300-500 vhd, rather than 200 vhd as stated prior to construction (absent measures to increase peak-period demand for transit services, such as higher parking or fuel costs, leading to higher peak-period Cp levels). This appears impractical with the current CBD distribution facilities.
In Los Angeles, peak capacity estimates for the Harbor Transitway assumed 66 passengers per standard bus and 96 passengers per articulated bus, 5.4 p/m and 5.2 p/m, respectively (Federal Highway Administration et al. 1985). Implied peak-hour service supply and passenger volume were 80-125 vhd and 5,000 - 12,000 phd. These figures far exceed the peak-period service levels, vehicle occupancies and passenger volumes carried by the facility to date (Table 2a). Anticipated vehicle occupancies were also 79-86 percent higher than those observed on the El Monte Transitway, and anticipated peak volumes were two to five times greater than carried by the El Monte facility (Table 2a).
Rubin and Moore (1997) postulate a "theoretical maximum" peak-hour ridership for the El Monte Transitway, with three-section double-articulated buses as used in Curitiba, 720 vhd and 270 p/v. The service supply is entirely unrealistic without a large off-street terminal, the PVO of 11 p/m is far higher than the levels observed in the U.S. and Canada, and the estimated 194,400 phd far exceeds peak volumes achieved in actual service.
Certain recent planning studies also contain unrealistic assumptions. New Jersey Transit (1996) forecast a maximum of 8,247 phd during the a.m. peak on the Hudson-Bergen LRT line, carried by a peak service supply of 30 vhd. This implied 275 p/v and 9.8 p/m, given the 92-foot (28-meter) vehicle length. These unrealistic PVO figures imply crowding levels higher than reported anywhere in North America, except on the busiest HRT lines in México City, Montréal and New York. Levinson and St. Jacques (1998) present "suggested bus passenger service volumes for planning purposes" ranging between 7,500-10,125 phd, requiring 81-135 vhd. These figures imply 75-96 p/v and 6.1-7.9 p/m. Again, these PVO figures are unrealistically high. The authors have found only one case since the crush-loading years of World War II where published data support an observed PVO exceeding 4.8 p/m for any U.S. or Canadian bus service. This is 5.6 p/m for bus services on Hillside Avenue, Queens, New York City, circa 1962 (Highway Capacity Manual 1965).
The maximum capacity of the Seattle CBD transit tunnel, as implied by the parameters used currently by the operator, is 5,750 phd, based on 125 vhd and PVO equal to 80 percent of vehicle seating capacity. The operator states that the 125 vhd figure is based on "more than ten years of experience with operating the only all bus tunnel with on-line passenger stations in the world (Sound  Transit et al. 2001). 125 vhd is nearly 80 percent greater than the maximum service level that has yet been operated.
Niles et al. (2001) and DMJM+Harris (2001) estimate the maximum capacity of the Seattle tunnel at 13,455 phd, assuming 65 seats per vehicle, full seated loads, and 200 vhd. The associated service supply is nearly three times greater than has yet been operated. These analyses also estimate a maximum capacity of 15,950 phd, under assumptions of 110 p/v (6.0 p/m) and  145 vhd. The vehicle-occupancy and service-supply figures are double the current levels. Rubin and Moore (1997) state a "theoretical" peak-hour capacity of 18,000 phd, based on 145 vhd. This implies 124 p/v, (6.8 p/m), roughly three times greater than currently carried by tunnel services.
Seattle's plan for mixed bus and LRT operation in its CBD transit tunnel include a maximum service level of 60 bus vhd, 10 LRT thd, and four-car trains. Six buses would be scheduled to operate in "platoons" during the six-minute interval between trains. The feasibility of mixed operation has been demonstrated in Essen, Germany, but with much less than 60 bus vhd. The planned bus and LRT service levels may prove impractical when operated over the same guideway. Another issue, which has not been addressed, is that safety standards for each mode are not identical. Road vehicles, including Seattle tunnel buses, typically operate beyond the safety limit, but RRT systems worldwide do not permit this.
Performance evaluations of existing U.S. fixed-guideway facilities seldom consider service-supply issues, focusing exclusively on demand parameters and demand analysis. This curious and consistent oversight is difficult to explain; only a few examples can be cited here. Hamer (1976) wrote before most projects he considered had been completed. But Webber (1976) considered only the BART system, and was certainly aware of the large peak-capacity shortfall (Section 2B). Hall (1980) could have analyzed the impact of service-supply levels on BART ridership over the initial six years of operation. But neither addressed service-supply issues.
The critical role of peak service supply in increasing transit use and attracting patronage from private autos has also been ignored. For example, Hensher (1998) states that the failure to attract significant new patronage to RRT over the past two decades is due largely to lack of disincentives to automobile use. But San Francisco planners recognized forty years ago that diversion of RRT patronage from automobiles would not occur uniformly throughout the day, but would instead be concentrated into peak commute hours (Proctor 1960). Large-scale diversion of auto trips to public transit is not likely unless existing and planned systems provide levels of peak-period comfort acceptable to consumers and provide this comfort on a scale large enough to accommodate a significant share of consumers who now use private autos.
 
5) Conclusions
The key finding of this analysis is that supply of transit service is a critical factor governing ridership and effective capacity in the U.S. and Canadian environment, where most consumers have alternatives and are not forced to travel under crush-loading conditions. Effective RRT capacity is a peak-hour vehicle occupancy of between four and five passengers per meter of vehicle length, corresponding to 100 - 120 passengers in a typical articulated LRT railcar. BRT peak-hour vehicle occupancy is consistently less, slightly below three passengers per meter of vehicle length. These are averages during the peak hour as vehicles traveling in the peak direction pass the busiest point on their route. If supplied capacity is not adequate to meet potential demand under these conditions, some potential customers will decide not to ride. In Los Angeles and Portland, LRT capacity increases attracted additional customers, indicating the previous existence of unserved demand. There are other, perhaps many, systems where similar results could be obtained.
This does not imply that transit service consumption should be predicted from supply parameters alone. Both demand and supply are necessary inputs to the planning and decisionmaking processes. The analysis presented here models the relationship among supply, demand and consumption where price is set by factors which may be little influenced by market conditions, is held rigid over the short term, and exerts very little influence on supply levels. Equilibrium between supply and demand, expressed as service consumption, must therefore be shaped by factors other than price (which may also be true in other public-service sectors). Observed consumption may fall short of the totality of demand unless supply levels are adjusted to the requisite levels. But service-supply levels may be frozen by constraints such as vehicle fleet size.
Despite the apparent difference in achievable capacity, there are several factors that make BRT an attractive alternative from the supply-side viewpoint. One is the option of staged implementation to give BRT an exclusive guideway where bottlenecks exist, while deferring other construction until ridership grows and finances permit. Another, important where guideway capacity is adequate but the vehicle fleet may prove too small, is that more buses can be bought, borrowed or transferred quickly, but railcar procurement is a time-consuming process.
Since service consumption arises from the interaction between travel demand and service supply, it is clear that consumption parameters such as PVO must be influenced by population density, CBD employment and other demand parameters. But quantification of such relationships would require, in the words of Cervero and Landis (1997),
'a far richer, more comprehensive and more dynamic data base . . . than we, or anyone else to date, had available.'
For this paper, the authors addressed demand in aggregate fashion: postulating the existence of various demand parameters, in magnitudes and combinations sufficient to generate the predicted consumption levels -- in concert with various service-supply levels. Additional research, better data and improved methodology should eventually permit disaggregate analysis of various demand factors with supply and consumption parameters, and extension of this analysis to off-peak and weekend service periods. Compilation and publication of nationwide data regarding peak service supply and consumption for U.S. fixed-guideway facilities, of the quality available for countries such as Japan, by an agency such as the Federal Transit Administration would expedite such research.
Over-optimistic demand forecasts would lead inexorably to ridership shortfalls, but veracity of demand analysis cannot be determined through observation of consumption levels alone. Supply parameters must also be addressed.
In conclusion, the authors believe there is a crucial relationship among travel demand and service supply parameters, which must be recognized as a key factor determining transit service consumption. It is hoped that additional research will provide a more comprehensive picture of these interactions and their impact.
 
Acknowledgements
The authors wish to express sincere appreciation to the following: Doug Allen, Maria Batista, Bill Capps, Wilson Fernandez, Gerald Fox, James J. Hughes, Lawrence F. Hughes, Charles Jahren, Judy Leslie Clark, Dave Colquhoun, Jerry Eddy, Joel B. Freilich, Robert W. Gower, Linda Hancock-Ross, Larry A. Humiston, Nancy Jarigese, Ashok Kumar, William Lieberman, Fred L. Mannering, Clarence W. Marsella, Brian Matthews, William J. Mattock, Allen S. Morrison, Paul O'Brien, the late Stephen B. Renovich, Jake Satin-Jacobs, Michael D. Setty, Bruce Shelburne, John W. Schumann, E. L. Tennyson, the late Gordon J. Thompson, Scott Vetare and Van Wilkins.
Figures 1 through 5 were drawn in Illustrator 10 for Macintosh System X by Michael D. Setty, adapted from hand drawings by Leroy W. Demery, Jr.
 
References
Assessment of Rush Period Rail Service Levels. 2000. Washington, DC: Office of Operations Planning and Administrative Support, Department of Operations, Washington Metropolitan Area Transit Authority.
Baker, Michael, Jr., Inc. Undated. Capacity Analysis and Peak Hour Loading for PATWAYS. (Prepared for the Port Authority of Allegheny County.)
Baker, Jeffrey K. 2002. “Advanced Automatic Train Control pioneered in San Francisco.” Railway Gazette International 158,6: 311-312.
Belobaba, Peter P. 1982. “Rapid Transit Development in Medium-Sized Urban Areas: A Comparison of Planning and Decisionmaking in Two Canadian Cities.” In: Transportation Research Record 877 Washington, DC: Transportation Research Board, National Research Council.
Ben-Akiva, Moshe, and Takayuki Morikawa. 1991. Ridership Attraction of Rail Compared with Bus (final report, Jan 1987 - Jan 1990). (Prepared for Urban Mass Transportation Administration, U.S. Department of Transportation.) Cambridge, MA: Center for Transportation Studies, Massachusetts Institute of Technology.
Biehler, Allen D. 1988. “Exclusive Busway Versus Light Rail Transit -- A Comparison of New Fixed-Guideway Systems.” In: Light Rail Transit: New Systems at Affordable Prices (TRB Special Report 221). Washington, DC: Transportation Research Board, National Research Council.
Bonsall, John A. 1987. “Transitways, the Ottawa Experience.” In: Proceedings, Second National Conference on HOV Lanes and Transitways, Houston, TX, October 25-28, 1987, edited by Ann Lancaster and Timothy Lorax. Texas Transportation Institute, Texas A&M University.
Boyd, J. Hayden, Norman J. Asher, and Elliot S. Wetzler. 1973. Evaluation of Rail Rapid Transit and Express Bus Service in the Urban Commuter Market. (Prepared for Office of Transportation Planning Analysis, U.S. Department of Transportation.) Arlington, VA: Institute for Defense Analysis.
Cervero, Robert. 1998. The Transit Metropolis: A Global Inquiry. Washington, DC: Island Press.
Cervero, Robert, and John Landis. 1997. “Twenty Years of the Bay Area Rapid Transit System: Land Use and Environmental Impacts.” Transportation Research 31A, 4: 309-333.
Charles River Associates, Inc., with assistance from Herbert S. Levinson. 1988. Characteristics of Urban Transportation Demand - An Update. (Prepared for Urban Mass Transportation Administration, U.S. Department of Transportation.) Boston.
Deen, Thomas B. and Richard H. Pratt. 1992. “Evaluating Rapid Transit.” In: Public Transportation: Planning, Operations and Management, edited by George E. Gray and Lester A. Hoel. Englewood Cliffs, NJ: Prentice-Hall.
Demery, Leroy W., Jr. 1994. “Supply-Side Analysis and Verification of Ridership Forecasts for Mass Transit Capital Projects.” Journal of the American Planning Association 60, 3: 355-371.
DMJM+Harris. 2001. Review and Analysis of Sound Transit Evaluation of Joint Operation in the Downtown Seattle Transit Tunnel. Seattle. (Prepared for Integrated Transport Research, Inc.)
European Conference of Ministers of Transport. 1978. ECMT Roundtable 38: Scope for the Use of Certain Old-Established Urban Transport Technologies (Trams and Trolleybuses). Paris.
Federal Highway Administration and California Department of Transportation District Seven. 1985. Final Environmental Impact Statement, Interstate 110 Freeway Transitway.
Federal Transit Administration, U.S. Department of Transportation. 1998. Issues in Bus Rapid Transit. Washington, DC.
Gardner, G., P. R. Cornwell and J. A. Cracknell. 1991. The performance of busway transit in developing cities (Research Report 329). Crowthorne, England: Transport and Road Research Laboratory.
Gordon, Peter, and Richard Willson. 1984. “The Determinants of Light Rail Transit Demand -- An International Cross-Sectional Comparison.” Transportation Research 18A, 2: 135-140.
Gordon, Peter, and Richard Willson. 1985. “The Determinants of Fixed-Rail Transit Demand -- An International Cross-Sectional Comparison.” In: International Railway Economics, edited by Kenneth J. Button and David E. Pitfield. Aldershot, Hampshire, UK: Gower.
Hall, Peter. 1980. Great Planning Disasters. Berkeley, CA: University of California Press.
Hamer, Andrew M. 1976. The Selling of Rail Rapid Transit. Lexington, MA: D. C. Heath.
Hensher, David H. 1998. “The Balance between Car and Public Transport Use in Urban Areas: What Can We Do About It?” Transport Policy 5,4: 193-204.
Hensher, David H. 1999. “A bus-based transitway or light rail? Continuing the saga on choice versus blind commitment.” Road & Transport Research 8,3: 13-20.
Herbst, K. 1992. “Brazil's Model City: Is Curitiba too good to be true?” Planning 58,9: 24-27.
Highway Capacity Manual (Special Report 87). 1965. Washington, DC: Highway Research Board, National Research Council.
Institute of Transportation Engineers. 1988. The Effectiveness of High-Occupancy Vehicle Facilities. Washington, DC.
Jacobs, Michael, Robert E. Skinner, and Andrew C. Lemer. 1984. Transit Project Planning Guidance - Estimation of Transit Supply Parameters. Cambridge, MA: Transportation Systems Center, U.S. Department of Transportation.
Jones, Robert W. 1993. “Boston Faces a Challenge.” In: 1993 North American Light Rail Annual and User’s Guide. Glendale: Interurban Press.
Kain, John F. 1988. “Choosing the Wrong Technology: Or How to Spend Billions and Reduce Transit Use.” Journal of Advanced Transportation 21, 3: 197-213.
Kain, John F. 1992. Increasing the Productivity of the Nation’s Urban Transportation Infrastructure. Cambridge, MA. (Prepared by Department of Urban Planning and Design, Graduate School of Design, Harvard University, for University Research and Training Program, Federal Transit Administration, U.S. Department of Transportation.)
Lang, A. Scheffer, and Richard M. Soberman. 1964. Urban Rail Transit. Cambridge: MIT Press.
Levine, Lenny. 2000. “Bus Rapid Transit Duels With Rail.” Metro 96,1: 43-44.
Levinson, Herbert S., and Kevin R. St. Jacques. 1998. Bus Lane Capacity Revisited. Paper presented at Transportation Research Board annual meeting, January 1998.
Los Angeles County Metropolitan Transportation Authority. 1994. Final Environmental Impact Report, Los Angeles Eastside Corridor. Los Angeles.
Martinelli, David R. 1996. “A Systematic Review of Busways.” Journal of Transportation Engineering 122,3: 192-199.
Meyer, John R., John F. Kain, and Martin Wohl. 1965. The Urban Transportation Problem. Cambridge, MA: Harvard University Press.
Meyer, Michael D., and Eric J. Miller. 1984. Urban Transportation Planning -- A Decision-Oriented Approach. New York: McGraw-Hill.
Middleton, William D. 1996. “BART Leads San Francisco's Rail Transit Growth”. In: Developing Metros 1996. Sutton, UK: Reed Business Publications.
Moore, James E., II. 1993. “Ridership and Cost on the Long Beach - Los Angeles Blue Line Train.” Transportation Research 27A, 2: 139-152.
Municipality of Metropolitan Seattle. 1985. Downtown Seattle Transit Project, Final Environmental Impact Statement.
New Jersey Transit Corporation and Federal Transit Administration, U.S. Department of Transportation. 1996. Hudson-Bergen Light Rail Transit System Final Environmental Impact Statement.
Miles, John, Dick Nelson and Jim MacIsaac. 2001. Light rail or Buses in the Downtown Seattle Transit Tunnel: Assessment of Benefits to King County metro and Regional Public Transportation. Seattle: Integrated Transport Research, Inc.
Nock, O. S. 1973. Underground Railways of the World. London: A & C Black, Ltd.
Parkinson, Tom, and Ian Fisher. 1996. Rail Transit Capacity (Transit Cooperative Research Program Report No 13). Washington, DC: Transportation Research Board, National Research Council.
Parsons Brinckerhoff/Kaiser Engineers. 1984. Conceptual Design Operations and Maintenance Plan, Long Beach - Los Angeles Rail Transit Project. (Prepared for Los Angeles County Transportation Commission.)
Parsons Brinckerhoff Quade & Douglas, Inc. 1996. Transit and Urban Form, Volume 2 (Transit Cooperative Research Program Report No 16). Washington, DC: Transportation Research Board, National Research Council.
Peschkes, Robert. 1996. Conversation, 4 August 1996.
Pickrell, Donald H. 1985. Urban Rail in America: A Review of Procedures and Recommendations from the Regional Plan Association Study. Cambridge, MA: Transportation Systems Center, U.S. Department of Transportation.
Proctor, William A. 1960. Transcript of Joint Discussion of Alternative Rapid Transit Plans for San Francisco by staff members and engineering consultants of the Bay Area Rapid Transit District and the San Francisco Transportation Technical Committee, held at the Conference Room, San Francisco Municipal Railway, Geary Boulevard and Presidio Avenue, at 9:30 A.M., Wednesday, June 29, 1960.
Pushkarev, Boris S.; Jeffrey M. Zupan, and Robert S. Cumella. 1982. Urban Rail in America: An Exploration of Criteria for Fixed-Guideway Transit. Bloomington, IN: Indiana University Press.
Rail Fleet Management Plan. 1998. Miami: Miami-Dade Transit Agency.
Rubin, Thomas A., and James E. Moore II. 1997. Better Transportation Alternatives for Los Angeles (Policy Study No. 232). Los Angeles: Reason Public Policy Institute.
San Diego Association of Governments. 1984. San Diego Trolley: The First Three Years. San Diego.
San Diego Transit Corporation. 1975. A Comparison Between a Light Rail System and an Articulated Bus System. San Diego.
Sawka, Oxana. 1999. “Ottawa's Transitway - 750 million riders and still counting!!” Public Transport International 48,2: 26-27.
Schumann, John W. 1996. Letter dated 7 October 1996.
Schumann, John W. 2000. E-mail correspondence dated 12 December 2000.
Setty, Michael D. 2002. Urban Transit Supply vs. Observed Consumption: Is Consumption Dependent on urban Transportation Supply? (Review Draft #3). http://www.publictransit.us/cadocs/supplydemand/transportsupplyvconsump.htm
Skinner, Robert E. 1981. “Estimating Transit Supply Requirements for Alternatives Analysis.” In: Transportation Research Record 835. Washington, DC: Transportation Research Board, National Research Council.
Smith, Neil, and David H. Hensher. 1998. The Future of Exclusive Busways: The Brazilian Experience. Sydney: Institute of Transport Studies.
Sone, Satoru. 1990. “Squeezing Capacity Out of Commuter Lines.” In” Developing Metros 1990. Sutton, UK: Reed Business Publications.
Sound Transit and King County Metro. 2001. Evaluation of Joint Operations in the Downtown Seattle Transit Tunnel. Seattle.
State Transport Authority of South Australia, Hugh Ballment and Ian Hammond. 1993. “Busways 2: The Adelaide Experience.” Transit Australia 48, 12: 267-273.
Strapac, Joseph A. 1974. “BART: The Final Test!” Pacific News 14, 9: 18-19.
Takatori, Yoshinobu. 1995. Kaku rosembetsu ni mita yusô no genjo. Tetsudô-Pictorial 45, 7: 24-36.
Tennyson, E. L. 1989. “Impact on Transit Patronage of Cessation or Inauguration of Rail Service.” In: Transportation Research Record 1221. Washington, DC: Transportation Research Board, National Research Council.
Tennyson, E. L. 1997. Letter dated 7 July 1997.
Transit Development Program 1999. 1999. Miami: Miami-Dade Transit Agency.
Turnbull, Katherine F., and James W. Hanks, Jr. 1990. A Description of High-Occupancy Vehicle Facilities in North America. College Station, TX: Texas Transportation Institute, Texas A&M University System.
U.S. Department of Transportation, Urban Mass Transportation Administration and Southern California Rapid Transit District. 1983. Draft Environmental Impact Statement and Environmental Impact Report, Los Angeles Rail Rapid Transit Project.
U.S. General Accounting Office. 2001. Mass Transit: Bus Rapid Transit Shows Promise. (Report number GAO-01-984.)
Vuchic, Vukan R. 1981. Urban Public Transportation - Systems and Technology. Englewood Cliffs, NJ: Prentice-Hall.
Vuchic, Vukan R. 1992. “Comparative Analysis.” In: Public Transportation: Planning, Operations and Management, edited by George E. Gray and Lester A. Hoel. Englewood Cliffs, NJ: Prentice-Hall.
Vuchic, Vukan R., Eric C. Bruun, Nikola Krstanoski and Yong Eun Shin. 1994. The Bus Transit System: Its Underutilized Potential (Final Report). Washington, DC: Federal Transit Administration, U.S. Department of Transportation.
Webber, Melvin M. 1976. The BART Experience - What Have We Learned? (Monograph No. 26). Berkeley, CA: Institute of Urban and Regional Development and Institute of Transportation Studies.
Weir, Robert S. 1992. “Moving Block Signalling Offers Cost Savings.” In: Developing Metros 1992. Sutton, UK: Reed Business Publications.
Welty, Gus. 1992. “Muni Moves to Moving Block.” Railway Age 193, 10: 38-39.
Werner, Alexander Y. 2001. 1998 Hub-Bound Travel. New York: New York Metropolitan Transportation Council.
Wilkins, Van. 1985. “Rapid No Rail.” Bus World 8, 2: 8-13.
 
TABLE 1a. Supply-side prediction of modal capacity
Mode
Peak Service Supply, vhd
Peak       Vehicle   Occupancy, p/v
Predicted Peak        Ridership1, phd
Predicted Weekday Ridership2 with
LOW (10%)
MED (13%)
HIGH (18%)
peak traffic share (PTS)
BRT
LOW  (100)
45
4,500
45,000
35,000
25,000
non-articulated
MED  (150)
"  "
6,800
68,000
52,000
38,000
 
HIGH (200)
"  "
9,000
90,000
69,000
50,000
             
BRT
LOW  (100)
70
7,000
70,000
54,000
39,000
articulated
MED  (150)
"  "
10,500
105,000
81,000
58,000
 
HIGH (200)
"  "
14,000
140,000
108,000
78,000
             
LRT
LOW  (16)
100
1,600
16,000
12,000
9,000
2-car trains
MED  (40)
"  "
4,000
40,000
31,000
22,000
 
HIGH (60)
"  "
6,000
60,000
46,000
33,000
 
EXC   (80)
"  "
8,000
80,000
62,000
44,000
             
LRT
LOW  (32)
100
3,200
32,000
25,000
18,000
4-car trains
MED  (80)
"  "
8,000
80,000
62,000
44,000
 
HIGH (120)
"  "
12,000
120,000
92,000
67,000
 
EXC   (160)
"  "
16,000
160,000
123,000
89,000
             
HRT
LOW  (60)
100
6,000
60,000
46,000
33,000
 
MED  (120)
"  "
12,000
120,000
92,000
66,000
 
HIGH (160)
"  "
16,000
160,000
123,000
89,000
 
EXC   (300)
"  "
30,000
300,000
231,000
167,000
1  rounded to nearest hundred.                                                                             2  rounded to nearest thousand.
Under travel-demand conditions typical of corridors in the U.S. and Canada where BRT or RRT is under consideration, where operation on CBD streets is planned, and under service-supply conditions typical of existing installations outside the four largest, most congested urban centers, weekday ridership levels shown in Table 1b may be predicted.
TABLE 1b. Capacity Prediction with Operation on CBD Streets
MODE
PEAK SERVICE SUPPLY
(Table 1a  MED  levels)
WEEKDAY RIDERSHIP
(18% to 10% PTS)
BRT - non-articulated
150 vhd
38,000  to    68,000 p/w
BRT - articulated
150 vhd
58,000  to    105,000 p/w
LRT - 2-car trains
20 thd (40 vhd)
22,000  to    40,000 p/w
LRT - 4-car trains
20 thd (80 vhd)
44,000  to    80,000 p/w
The range in total weekday ridership levels reflects the uncertainties of predicting ridership during non-peak hours, when guideway capacity is available and vehicle capacity can easily be provided, and consumer demand is the limiting factor. Ridership projections for articulated buses assume that the rear section of an articulated bus has the same ease of boarding and alighting and the same riding comfort as the front section.
 
TABLE 2a. Regression analysis of modal capacity - BRT
City and Corridor
Year
Peak Service Consumption phd
Peak Service Supply
Peak Vehicle Occupancy
vhd
mhd
p/v
p/m
Houston
I-10 (Katy) Transitway
1990
1,820
45
548
40
3.3
"          "
I-45N (North) Transitway
1990
2,810
76
926
37
3.0
"          "
I-45S (Gulf) Transitway
1990
840
26
320
32
2.6
"          "
US 290 (N.W.) Transitway
1990
600
17
209
35
2.9
Los Angeles
El Monte Transitway
post-1995
2,750
70
857
39
3.2
Pittsburgh
East Busway, Negley, p.m.
2000
4,002
102
*1,399
39
* 2.8
"          "
South Busway, South Hills Junction, p.m.
2000
1,858
57
695
33
2.7
"          "
Interstate 279 HOV
1997
783
20
245
39
3.2
San Diego
Interstate 15
1990
350
14
171
25
2.0
Seattle
Interstate 5, a.m.
post-1995
2,750
70
**1,067
39
**2.6
"          "
SR-520, a.m.
1990
3,140
56
**855
56
**3.7
"          "
Interstate 90
1990
1,250
34
**515
37
**2.4
Vancouver
Lions Gate Bridge
1990
1,080
27
329
40
3.3
Washington
Shirley Highway Transitway
1996
4,500
140
1,714
32
2.6
"          "          
Interstate 66
post-1995
2,920
85
1,027
34
2.8
               
Los Angeles
Harbor Transitway
a. 2000
300
22
268
14
a. 1.1
Miami
South Dade Busway, a.m.
a. 2000
(bus) 460
13
158
35
a. 2.9
     
(van) 140
6
36
23
3.8
     
(total) 600
     
3.1
New York
Lincoln Tunnel XBL
b. 1998
34,796
1,003
12,227
35
b. 2.8
Seattle
Tunnel, International Dist., s.b.
a. 2000
2,500
54
988
47
a. 2.6
"          "
Tunnel, Convention Pl., n.b.
a. 2000
2,600
48
879
54
a. 3.0
Notes for Table 2a: Peak service supply in vehicles per hour per direction (vhd) converted to meters of vehicle length per hour per direction (mhd) assuming vehicle lengths of 40 feet (12.19m) and 60 feet (18.29m) for standard and articulated buses, respectively.
* based on composite vehicle length of 13.715 meters, assuming 25 percent articulated vehicles.
** based on composite vehicle length of 15.24 meters, assuming 50 percent articulated vehicles.
a. Los Angeles (Harbor Transitway), Miami and Seattle (downtown transit tunnel) data are based on authors' personal observations, were included for information only and were not used in the OLS regression model.
b. New York data included for information only and were not used in the OLS regression model.
Ottawa data excluded as incompatible with data from other cities (see text Section 2F).
Ordinary Least Squares Regression Model (t-statistics in parentheses):
PSCBRT  =  103.91 + 2.75PSS          R2 = .96
(0.758)     (17.022)
PSC: Peak Service Consumption, passengers per hour per direction (phd).
PSS: Peak Service Supply, meters of vehicle length per hour per direction (mhd).
Note: The model above is based on the 15 BRT corridors in the first group above. Corridors are unweighted in the model calculations.
TABLE 2b. Regression analysis of modal capacity - LRT
City and Corridor
Year
Peak Service Consumption phd
Peak Service Supply
Peak Vehicle Occupancy
vhd
mhd
p/v
p/m
Calgary
South Line
2000
4,660
39
1,064
119
4.4
"          "
Northwest Line
2000
3,650
36
982
101
3.7
"          "
Northeast Line
2000
4,550
36
982
126
4.6
Cleveland
Shaker Heights (Blue/Green)
2000
1,200
20
470
60
2.6
Dallas
LRT s.b., Union Sta., a.m.
2000
1,900
28
800
68
2.4
Denver
LRT, n.b., 10th/Osage, a.m.
2002
2,400
31
756
77
3.1
Edmonton
Northeast LRT
1994
3,219
36
866
89
3.7
Los Angeles
Blue Line
2001
2,618
22
599
119
4.4
"          "
Green Line
2001
1,112
8
218
139
5.1
Newark
City Subway
1994
1,769
30
421
59
4.2
Pittsburgh
LRT, Station Square, p.m.
2000
2,448
24
614
102
4.0
Portland
Eastside MAX, a.m.
1999
2,375
22
584
108
4.0
"          "
Westside MAX, a.m.
1999
2,380
18
478
132
4.9
Sacramento
LRT east
1997
1,727
16
368
108
4.7
"          "
LRT north
1997
1,600
16
368
100
4.3
San Diego
South Line LRT, a.m.
1999-2000
2,015
24
552
84
3.7
"          "
East Line LRT, a.m.
1999-2000
1,174
14
322
84
3.7
San Jose
LRT south
1997
1,327
14
380
95
3.5
               
Baltimore
Central LRT - north, p.m
a. 2000
650
10
286
65
a. 2.3
"          "
Central LRT - south, p.m.
a. 2000
920
14
400
66
a. 2.3
Boston
Green Line
b. 1994
10,000
90
1,958
111
b. 5.1
Buffalo
LRRT
c. 1997
1,240
25
510
50
c. 2.4
Dallas
LRT n.b., Mockingbird, a.m.
c. 2000
1,385
27
771
51
c. 1.8
Philadelphia
Subway-Surface LRT
d. 1994
4,100
60
914
68
d. 4.5
St. Louis
MetroLink
e. 1996
2,000
18
491
110
e. 4.0
Salt Lake City
TRAX, Ballpark, a.m.
a. 2000
1,400
17
391
82
a. 3.6
San Francisco
Muni Metro, Van Ness, a.m.
a. 1999
3,870
43
953
90
a. 4.1
"          "
Muni Metro, Van Ness, p.m.
a. 1999
6,100
60
1,330
102
a. 4.6
Notes for Table 2b: Peak service supply in vehicles per hour per direction (vhd) converted to meters of vehicle length per hour per direction (mhd) using vehicle lengths reported by Parkinson and Fisher (1996).
a.          Based on authors' personal observations, included for information only; and were not used in the OLS regression model.
b.          Boston data included for information only and were not used in the OLS regression model.
c.          Excluded from OLS regression model as an outlier (see text Section 2F).
d.          Authors' estimate, assuming service frequencies as shown in public timetables. Included for information only; excluded from the OLS regression model.
e.          Authors' estimate, assuming 8 percent PTS for the busier (eastward) segment of the line, 2-car trains and service frequency as shown in public timetable. Included for information only; excluded from the OLS regression model.
Streetcar operations using historic vehicles (e.g. Memphis, New Orleans, San Francisco F Line, Seattle Waterfront Streetcar) were excluded from this tabulation.
Ordinary Least Squares Regression Model (t-statistics in parentheses):
PSCLRT = 36.62   + 3.83PSS          R2 = .81
(0.124)    (8.395)
PSC: Peak Service Consumption, passengers per hour per direction (phd).
PSS: Peak Service Supply, meters of vehicle length per hour per direction (mhd).
Note: The model above is based on the 18 LRT corridors in the first group above. Corridors are unweighted in the model calculations.
TABLE 2c. Regression analysis of modal capacity - HRT
City and Corridor
Year
Peak Service Consumption phd
Peak Service Supply
Peak Vehicle Occupancy
vhd
mhd
p/v
p/m
Atlanta
East-West
1994
2,986
60
1,380
50
2.2
"          "
North-South
1994
5,093
58
1,334
88
3.8
Chicago
Dearborn Street Subway
1994
9,376
112
1,651
83
5.6
Cleveland
Rapid (Red Line)
2000
1,200
20
436
60
2.8
Los Angeles
Red Line
2001
3,400
60
1,380
57
2.5
Miami
Metrorail, Vizcaya, a.m.
1998
3,854
52
1,184
74
3.3
Philadelphia - PATCO
Lindenwold Line
1995
5,650
90
1,829
63
3.1
S.F. - BART
Transbay Tube, a.m.
1999
16,700
180
3,863
93
4.3
"          "
Mission St. Subway, a.m.
1995
8,069
130
2,789
62
2.9
Vancouver
Skytrain
1994
6,932
100
1,232
69
5.6
Washington
Blue / Orange, e.b., Rosslyn, a.m.
2000
15,800
140
3,286
113
4.8
"          "
Bl / Or, e.b., L'Enfant Plaza, p.m.
2000
10,400
106
2,488
100
4.3
"          "
Red, s.b., Dupont Circle, a.m.
2000
12,300
120
2,816
103
4.4
"          "
Red, s.b., Union Station, a.m.
2000
11,800
120
2,816
98
4.2
"          "
Green, s.b., Mt. Vernon Sq., a.m.
2000
4,200
40
939
105
4.3
"          "
Gr / Ye, n.b., L'Enfant Plaza., a.m.
2000
8,100
80
1,878
101
4.3
               
Baltimore
Metro, s.b., State Center, a.m.
a. 2000
4,200
108
2,213
39
a. 1.9
Chicago
Brown (Ravenswood)
b. 1994
7,051
97
1,507
73
b. 4.7
"          "
Green (Lake / South)
b. 1994
2,952
42
653
70
b. 4.5
"          "
Orange (Midway)
b. 1994
4,287
66
1,026
65
b. 4.2
"          "
Purple (Evanston)
b. 1994
3,479
42
659
82
b. 5.3
"          "
Red (Howard / Dan Ryan)
b. 1994
11,533
120
1,865
96
b. 6.2
Philadelphia - SEPTA
Broad St., n.b., Girard, p.m.
a. 2000
4,200
108
2,213
39
a. 1.9
"          "
Market St., e.b., 15th St., a.m.
a. 2000
5,500
96
1,600
57
a. 3.4
S.F. - BART
s.b. from Ashby, a.m.
c. 1995
4,400
70
1,502
63
c. 2.9
"          "
s.b. from Rockridge, a.m.
c. 1995
7.627
81
1,738
94
c. 4.4
"          "
n.b. to Lake Merritt, a.m.
c. 1995
6,413
86
1,846
75
c. 3.5
Notes for Table 2c: Peak service supply in vehicles per hour per direction (vhd) converted to meters of vehicle length per hour per direction (mhd) using vehicle lengths reported by Parkinson and Fisher (1996).
a.          Based on authors' personal observations, included for information only; and were not used in the OLS regression model.
b.          Authors' estimate, assuming service frequencies as shown in public timetables. Included for information only; excluded from the OLS regression model.
c.          BART volumes into downtown Oakland are shown for information only, and were not used in the OLS regression model since the lines to Concord, Fremont and Richmond form branches of the transbay "corridor."
Boston, Montréal, New York and Toronto were excluded from this tabulation.
Ordinary Least Squares Regression Model (t-statistics in parentheses):
PSCHRT = -928.80 + 4.50PSS          R2 = .86
(-0.896)     (9.376)
PSC: Peak Service Consumption, passengers per hour per direction (phd).
PSS: Peak Service Supply, meters of vehicle length per hour per direction (mhd).
Note: The model above is based on the 16 HRT corridors in the first group above. Corridors are unweighted in the model calculations.
Data sources for Tables 2a-2c: Levinson and St. Jacques (1998), Parkinson and Fisher (1996), Tennyson (1997), Turnbull and Hanks (1990), and Buffalo, Calgary, Cleveland, Denver, Los Angeles, Miami, Pittsburgh, Portland, Sacramento, St. Louis and San Jose operator staff members who kindly responded to the authors' requests for information.
Glossary
(Capacity notation after Vuchic, 1981.)
 
BRT: Bus Rapid Transit.
CBD: Central Business District.
Co: Offered Capacity (vehicles, seats or places per hour).
Cp: Utilized Capacity (passengers per hour).
cs: Station Capacity (vehicles per hour).
cv: Vehicle Capacity (passengers per vehicle, or passengers per unit of floor space).
cw: Way (roadway or track) Capacity (vehicles per hour).
HOV: High-Occupancy Vehicle.
HRT: Heavy Rail Transit.
LRT: Light Rail Transit.
MBS: Moving-Block Signaling (also known as "Transmission-Based Signaling").
mhd: Meters of Vehicle Length per Hour per Direction.
phd: Passengers per Hour per Direction.
p/m: Passengers per Meter of Vehicle Length.
p/v: Passengers per Vehicle.
p/w: Passengers per Weekday.
PTS: Peak Traffic Share, the ratio of traffic during busiest hour, in busier direction, to     two-way, all-day traffic.
PVO: Peak Vehicle Occupancy, the average passenger load carried aboard each vehicle, past the maximum-load point, during the busiest hour, in the busier direction.
RRT: Rail Rapid Transit.
thd: Trains per Hour per Direction.
vhd: Vehicles per Hour per Direction.
 
1 In Portland, off-peak ridership almost immediately required two-car trains throughout most of the day, which was not anticipated prior to opening. More recently, after several years of operation, off-peak ridership in Sacramento grew to require operation of two-car trains. Both examples suggest that off-peak ridership was relatively higher than anticipated, with PTS therefore lower than anticipated.
2 In the Sacramento CBD, intersections are 320 feet apart, permitting operation of four-car trains of 80-foot articulated vehicles. By contrast, the Portland CBD has intersections every 200 feet, limiting LRT trains to two cars.
3 Lincoln Tunnel bus services use Manhattan's Port Authority Bus Terminal, which has two structures, one built during the early 1950s and the other 30 years later. These together have more than 150 gates, many but not all having direct access to the tunnel. More than 100 gates are used by cross-river suburban and transit services. The terminal has two separate unloading ramps for eastbound buses, and one gate is designated specifically for wheelchair passengers. The terminal has 800 berths for individual buses. Subway trains, surface bus routes and taxis distribute passengers to various destinations.
4 Useful peak-period capacities of bus lines using the downtown Seattle transit tunnel are constrained by operating practices. Northward and southward lines are not through-routed as planned prior to construction. Instead, they are operated separately to insure schedule reliability. Between 20 and 30 percent of peak-hour, peak-direction vehicles using the tunnel are terminating services that do not continue beyond Convention Place or International District, the two portal stations.
5U.S. per-capita Gross Domestic Product (GDP) is six times greater than that of Brazil. U.S. central government expenditures per capita, and passenger car registrations per capita, are six times greater. Brazil has more than five times as many people as Canada, but only slightly more automobiles -- 14 million.
6Curitiba's population grew from 140,000 in 1940 to 500,000 in 1965, when a new master plan concentrating growth in five radial corridors was adopted. High-rise buildings were limited to a four-block strip on either side of the busway axes. Fifteen years later, the city had 1.29 million people and the metropolitan area had nearly two million (Herbst 1992). Employment was concentrated in two centers: retailing in the CBD, and industrial development in the "Curitiba Industrial City" (Smith and Hensher 1998). It is clear that Curitiba's transit services were developed concurrently with a unique urban form.
7 Peschkes (1996) observed 42-second headways on the Moskva (Moscow) Metro, and the authors’ subsequent observation in Sankt-Peterburg (St. Petersburg) made clear that this report was correct. Recalling that "headway" does not account for the time required for a train to move its own length, the 42-second headway corresponds to a maximum service frequency (i.e. minimum interval between departures) of 80-90 seconds.
8In Los Angeles, Blue Line station platforms have been lengthened to permit operation of three-car trains. The first began operation in September 2001, and most daytime services used three-car trains by spring 2002.
9Recent observations in Ottawa suggest that the operator has curtailed the use of articulated buses on peak-period transitway services, and has deployed those which remain on the busiest individual services, leading to significantly greater PVO than observed previously.
10Peak-period vehicle occupancy may have been significantly higher, or lower, than 70 p/v, but this cannot be determined from the aggregate data presented: 5,892 phd and 103 vhd.
11Boyd et al. (1973) estimated a peak capacity of 144 vhd per CBD street, and stipulated four parallel CBD streets for their hypothetical BRT service.
12Enforcing a "seat for everyone" rule on most existing U.S. fixed-guideway facilities would require controls on boarding during peak periods. Consumers who derive greater utility from traveling as standing passengers aboard the first available vehicle rather than from waiting for a seat -- and those who prefer to stand even when seats are available -- would view such controls as unjustified marketplace interference.
13This was suggested by Professor Jerry B. Schneider, Civil Engineering Department, University of Washington, and is supported by empirical data from several sources (Demery 1994).
14 In Portland, at 1996, when the LRT fleet was not adequate for peak-period traffic, maximum peak volume was about 2,100 phd and average ridership was about 27,000 p/w, implying a PTS of 8 percent. This may have reflected 1.) displacement of passengers to "shoulder" periods due to crowding, and 2.) stronger coordination between land use and public transit than is typical for U.S. cities. In Los Angeles (1994), LRT Blue Line ridership counts implied a PTS of 6.5 percent and PVO of 120 p/v. This suggests 1.) significant reverse-peak and off-peak traffic, and 2.) high levels of peak-period demand.
15Seattle planners (1995) outlined the following in order to achieve 15,000 phd: 1.) Four-axle cars, 66 feet (20 meters) long, coupled in six-car trains to fully utilize the 396-foot platforms in the CBD transit tunnel, 2.) three-minute maximum service frequency to provide 120 vhd during peak periods, and 3.) PVO of 125 p/v. The implied 6.2 p/m of vehicle length is unrealistically high.
16The maximum hourly volumes observed by the authors range from 3,400 to 4,700 phd.
17This example is adapted from lecture material presented by Fred L. Mannering, Professor and Head, School of Civil Engineering, Purdue University.
18The operator, seeking to reduce overcrowding on peak-period LRT Blue Line trains, eliminated the express surcharge on Harbor Transitway services as a promotion from November 2000 to January 2001. Ridership increased by about 10 percent (to 2,700 - 2,800 p/w). Most new passengers continued to use Harbor Transitway services following the end of the promotion.
19The Ottawa transit network operates in trunk-feeder configuration during most of the day, shifting to the one-seat pattern during weekday peak periods. The peak-period transfer rate should therefore be less than during other times. However, supporting data are not available.
20Two freeway express bus routes, 456 and 457, carried about 2,700 p/w, less than three percent of pre-rail corridor ridership of 101,000 p/w (FY 1990). Line 456 ridership fell by more than 45 percent (to 1,300 p/w) after rail service began, and this line was discontinued in June 1991. This followed the USC count of June 1991 but preceded the operator's August count. Line 457, when discontinued early in 1995, averaged 90 p/w, down from 300 p/w in FY 1990. Large-scale forced abstraction of rail ridership from parallel surface bus routes is not supported by operator statistics and public timetables, which show little evidence of peak service-supply reductions like the 40-60 vhd that would be commensurate with rail peak consumption level of 2,400 phd (1994).
21This example is based on Sacramento, where eight LRT trains replaced 60 buses entering the CBD during the busiest hour, generating 87 percent fewer traffic movements.