There is renewed interest in many developing and developed countries in finding ways of providing efficient and effective public transport that does not come with a high price tag. An increasing number of nations are asking the question—what type of public transport system can deliver value for money? Although light rail has often been promoted as a popular ‘solution’, there has been progressively emerging an attractive alternative in the form of bus rapid transit (BRT). BRT is a system operating on its own right-of-way either as a full BRT with high quality interchanges, integrated smart card fare payment and efficient throughput of passengers alighting and boarding at bus stations; or as a system with some amount of dedicated right-of-way (light BRT) and lesser integration of service and fares. The notion that buses essentially operate in a constrained service environment under a mixed traffic regime and that trains have privileged dedicated right-of-way, is no longer the only sustainable and valid proposition. This paper evaluates the status of 44 BRT systems in operation throughout the world as a way of identifying the capability of moving substantial numbers of passengers, using infrastructure whose costs overall and per kilometre are extremely attractive. When ongoing lifecycle costs (operations and maintenance) are taken into account, the costs of providing high capacity integrated BRT systems are an attractive option in many contexts.
KeywordsBus rapid transit Comparative analysis Infrastructure costs Service levels
Public transport investment is being touted as a key springboard for a sustainable future, especially in large metropolitan areas with growing populations. Whether such investment will turn the tide away from automobility is a big question; however regardless of the likely outcome, any commitment to improved public transport has a growing number of options to pursue. Although variations in rail systems typically loom dominant in many strategic statements on urban reform (Sislak 2000; Edwards and Mackett 1996), ranging from heavy rail through to metro rail and light rail, there is a growing interest worldwide in ways of making better use of the bus as a primary means of public transport, and not limited as a service that feeds a rail network (Hensher 1999, 2007; Canadian Urban Transit Association 2004; Federal Transit Administration 2004).1
There are many ways in which bus transport can be developed as part of an integrated network-based public transport system, typified by the best practice bus rapid transit (BRT) systems in South America such as Curitiba in Brazil and TransMilenio in Bogota, Colombia (Menckhoff 2005). Bus Rapid Transit is “…a high-quality bus based transit system that delivers fast, comfortable, and cost-effective urban mobility through the provision of segregated right-of-way infrastructure, rapid and frequent operations, and excellence in marketing and customer service. BRT essentially emulates the performance and amenity characteristics of a modern rail-based transit system but at a fraction of the cost. A BRT system will typically cost four to 20 times less than a light rail transit (LRT) system and 10 to100 times less than a metro system.” (Wright and Hook 2007, 11).
Wright and Hook (2007) have compiled details of many BRT systems to document the inherent advantages (and disadvantages) of many of the systems in terms of cost and performance. With a focus on delivering a cost efficient and service effective transport system, there are opportunities today to evaluate mixtures of bus and rail systems that can service the full spectrum of capacity requirements and patronage demands (Cornwell and Cracknell1990; Hidalgo 2005; Transit Cooperative Research Program 2007).
What is especially pertinent however is the recognition that the so-called natural evolution from a bus in mixed traffic through to heavy rail in terms of passenger capacity per hour (seating and standing) is no longer strictly valid. BRT systems such as the TransMilenio have shown that a BRT system can, if appropriately configured, carry more passengers per hour than many rail systems. The main trunk corridor in Bogota has peak maximum ridership2 of 35,000 trips per hour3 one way with 3 min maximum peak headways (5 min off-peak headways) with buses spaced much closer together much of the peak, average station dwell time of 25 s, with articulated buses having a carrying capacity of 160 passengers and off-vehicle smartcard fare payment. Curitiba, the forerunner to Bogota, has a peak maximum ridership of 20,000 trips per hour one way. This compares to the busiest rail line in Sydney, for example, of 14,000 trips per hour one way. In general Hidalgo (2005, 5) states “There is range, between 20,000 and 40,000 passengers/hour per direction, in which Metros and HBRT4 are able to provide similar capacity. Nevertheless, there are large differences in initial costs: US$5-20 million per kilometre for HBRT, US$30-160 million per kilometre for Metros”.
As interest in BRT systems grow, questions are being asked about the actual cost and carrying capacity of such systems. To investigate these and other matters, we have taken the tables prepared by Wright and Hook (2007 appendices) together with some enhancements, and developed a data base to assess the relationship between infrastructure cost ($US total and per kilometre), carrying capacity (passengers per hour per direction), and the specifications of each system in terms of service frequency, fares and fare payment system, trunk and feeder capacity and connectivity, extent of separation from other modes, speed, station spacing, dwelling times etc. In any comparison between countries, however, we recognise the difficulties where inputs have substantially different prices, time periods, and baseline conditions prior to construction5; nevertheless, there are very important insights that can be gained to provide broad guiding signals on the appeal of very specific investment strategies to grow public transport patronage and deliver value for money in terms of the cost of providing a given level of service relative to other forms of public transport.
This paper is structured as follows. The next section provides a descriptive overview of the key dimensions of the 44 BRT systems, enabling an appreciation of capability. Such a commentary, while informative, is limited in that the role of each feature of the BRT system needs to be assessed in terms of its influence on cost and patronage, given the level of service. This is presented in Sects. 3 and 4 using a multivariate analysis to reveal candidate influences on variations in infrastructure costs and daily ridership. The paper concludes with suggestions for an ongoing monitoring program to keep the accumulating evidence current.
Descriptive contrasts of BRT systems
Data on 44 BRT systems around the world, compiled by Wright and Hook (2007), provides the only ‘comprehensive’ source of information that has reasonably comparable data. This data, focussed on the BRT component on integrated systems, including any connection to a feeder network, is not without a large amount of missing items across the 44 systems and indeed not all commentators agree with the actual information provided6; however, there is enough useful information to begin to appreciate the nature of each system and, in particular, to identify the key features that are systematically varying sources of influence on infrastructure costs. We have 70 data items per BRT system (see Appendix), some of which are more complete in details than others. We have selected the data indicators that are relatively complete and which are plausible candidates for testable hypotheses on what may be key sources of systematic variation in infrastructure costs overall and per kilometre.
Peak ridership for 26 systems for which we have data shows four South American systems (Transmilenio in Bogota, Sao Paulo, Porto Alegre, and Curitiba) with 20,000 passengers per hour per direction, which then declines to 12,000 (Seoul), with the majority of systems in the 2,000 to 8,000 passengers per hour per direction.
A careful assessment of these figures shows a significant amount of variation in the specifications of each system. Clearly a preferred scenario would support high commercial speeds, no operating subsidies (unless they are optimal in an economic welfare sense), low flow buses with at level boarding, totally dedicated corridors with no interference from other modes (which is an attractive feature of railways), smart card off-vehicle fare payment, seamless model interchange (where it occurs), and minimum access and egress time. There is no one system that comes closest to fulfilling all these conditions. The Australian and US systems deliver the highest commercial speeds, the Latin American systems are least dependent on operational subsidies, the Latin American and European systems dominate the provision of at-level boarding and alighting, the Latin American systems have been most effective in eliminating the need for signal priority or grade separation at intersections, and the Latin American, Asian, and French BRT systems have committed to pre-board fare collection and fare verification. Modal integration at stations is strongest in Australia, Europe, and USA. Finally, the majority of BRT systems have stations spaced apart on average 500 m, although this increases to over 1.5 km for Australian and USA systems including one in China and in Holland.
This brief commentary highlights a very important feature of this comparison. Given the interest in establishing the infrastructure cost and patronage profile of all BRT systems for which data is captured, and recognizing the large variation in design and service levels with no obvious mapping between features and specific systems, it is necessary to develop a more formal multivariate framework within which to identify the influence that specific design elements have in explaining differences in infrastructure costs in total and per kilometre, and patronage per day across the BRT systems.
Assessment of systematic sources of variation in infrastructure costs
A two-stage empirical process was used to establish sources of systematic variation in (i) infrastructure costs per kilometre and (ii) total infrastructure costs. It was, however, preceded by a review of the literature (Canadian Urban Transit Association 2004; Federal Transit Administration 2004; Menckhoff 2005; Cornwell and Cracknell 1990; and Hildago 2005)7 as the basis of establishing a series of research hypotheses. The main factors likely to have an influence on infrastructure costs are suggested to be input costs in construction, the year(s) of construction, the funding source, the number and size of stations and terminals, and the extent to which a full BRT treatment is implemented, or degrees of light treatment in terms of components such as intersections and signalization.
Within the limitations of the data, we began with an assessment of the nominal scaled variables listed in Appendix Table A1, using a technique known as nonlinear canonical correlation analysis (NLCCA) that searches for the optimal scale (or cut-off points) for each variable, including those variables that ‘represent’ the identified sources from the literature review. We report the evidence only for infrastructure costs per kilometre in the Appendix. What we found (see Appendix Table A2), with rare exception, was a high amount of association between infrastructure costs and explanatory variables that we would argue have little to do with infrastructure costs (e.g., ‘operating subsidy requirements’ and the ‘management of the fare system’); and only two statistically significant dummy variables that made causal sense, i.e., at-level boarding and signal priority at intersections; the latter being identified in the literature review. Given that the dependent variable is expressed as a cost per kilometre we also defined candidate explanatory variables in per kilometre units (e.g. number of intersections with priority signal control and grade-separated per trunk kilometre, and the ‘number of terminals per trunk kilometre); however this did not alter the essential statistical message reported in the Appendix.
We undertook the same analysis using a dummy coded multivariate analysis (Table A3) with 11 variables that explain 74.2% of the variation in infrastructures cost per kilometre. We then investigated the ratio scale variables (listed in Appendix Table A1) while initially retaining the two significant casual effects above as dummy variables, handled through traditional ordinary least square regression.
Infrastructure cost regression models
Natural Log of age of BRT system
Number of intersections with priority signal control and grade-separated
Number of terminals
Boston BRT (1,0)
Pittsburgh West BRT (1,0)
In recognition of the significant variation in infrastructure costs, we introduced a series of BRT specific dummy variables to control for the possibility of bias at the upper end (see Fig. 1). We found initially that capturing this for the three most expensive systems (Boston, Nagoya, and Pittsburgh West) was sufficient, with non-statistically significant BRT system-specific dummy variables after the top three. However the final model eliminated the Nagoya system due to missing data, and hence only two dummy variables remained. Overall, 55.9% of the variation in unit infrastructure costs, and 59% of the variation on total infrastructure costs, across 28 systems can be explained by five variables. This is a satisfying result for disaggregated data on very different systems in terms of scale and configuration. In one sense the models in Table 1 are a reduced form in which variables such as the age of the BRT is capturing other effects as suggested below. We were unable to find any statistically significant influence of location based on developed vs. developing economies, and between developed economies (e.g., West Europe, USA/Canada, Asian economies (including Japan, Tawain and Korea)).
We can clearly see that there are two very influential design delivery features that are systematically linked to the costs of infrastructure and which were suggested by the literature review; namely the number of terminals and number of intersections with priority signal control and grade-separated.
Assessment of systematic sources of variation in ridership
Passenger trips per day regression model
Number of stations
Trunk vehicle capacity
Average fare per trip
As expected there is a negative relationship for fares, with an implied direct mean fare elasticity of −0.12. The available carrying capacity of buses has a positive influence on ridership.
This paper has taken a closer look at the cost of constructing the BRT infrastructure for 44 systems and the range of design and service specifications that are offered through BRT in serving the public transport market. Given the widely varying specifications, the question of what might be the possible reason(s) for such varying costs begs some response. In general, the great majority of systems with all manner of variation cost less than $US10m per kilometre, and what is most notable about this is that these systems are not all confined to economies with relatively low input costs (especially labour) but are spread throughout developed and developing nations (such as USA, UK, Australia, Canada, France, Mexico, Korea, Brazil, and China).
In seeking out any possible sources of explanation, implementing a two stage multivariate statistical analysis that can accommodate the differing scales of a range of descriptors of each system, we were only able to identify a few influences, other than some associative ones, that cannot be claimed to be causally defining sources of systematic variation in infrastructure costs. This is surprising as well as being an important finding. It signals, with the exception of the number of terminals and intersection treatment by signal priority or at grade separation, that there are no other features that we can identify that have a statistically significant impact on infrastructure costs per kilometre. One interpretation of this finding is that the data may be limiting. Alternatively, we are inclined to suggest that the differences are principally attributable to the context in which the costs were negotiated, including the number of bidders at the time where a franchised arrangement was in place, how the project was actually financed, the specific year in which the project as constructed (although all costs were converted to $US2006), and the extent to which there were major works such a bridges and tunnels.
In addition we have established further evidence to support the position that growing public transport patronage requires a system that recognises the important role that interpretations of connectivity and frequency play. BRT systems appear to be focussing in the right place in growing public transport patronage.
Finally, we make a plea for continuing efforts to improve the quality of data, especially given that BRT is growing in popularity, and hence the benefits that can be gained in guidelines linked to scientifically rigorous empirical assessment of the expanding number of systems throughout the world.
Discussions with Raymond Lam, Singapore’s Minister of Transport, have been useful in highlighting this position.
For 35,000 passengers with a load of 160, there would need to be 219 buses in the peak hour, or almost four buses each minute.
With recent claims of up to 45,000 trips per hour.
Hidalgo (2005) refers to high level BRT as HBRT.
One reason for differences in infrastructure costs relates to the physical conditions prior to start of construction, which are difficult to define. Adelaide, for example, started from scratch, although they had the advantage that most of the land was in Government ownership, but they had to build a lot of bridges. Bogota, in most cases, converted some existing road lanes into BRT lanes.
Like any highly aggregate analysis that summarises the dimensionality of each system by a single average indicator, the data will be subject to disagreement, and indeed would display varying deviations around specific averages depending on the source used to obtain the data. Despite this, there is some useful broad evidence that signals specific strengths of BRT systems in respect to costs and ridership.
A referee suggested we undertake a comparison with other public transport systems (i.e., heavy and light rail). We have resisted this since the focus herein is on BRT per se, and understanding what factors may contribute to variations in unit and total infrastructure costs. The debate on the comparative cost of BRT, light rail and heavy rail is reported in other papers in the literature (see for example, Edwards and Mackett 1996; Hensher 1999, 2007a, Chap. 17; Vuchic 2007).
We adjusted the ridership figures by dividing by kilometres of corridors to correct for any comparisons that would tend to deliver higher patronage simply because of the amount of coverage regardless of the role of other factors.
Tom Wilson has pointed out that both enthusiasts and professionals in public transport often use “maximum” and “minimum” jointly with “frequency” and “headway” without thinking about what they really mean. In the case of very high frequency services, “frequency (or services) per hour” is a better concept.
The Thredbo 10 conference in August 2007 concluded that frequency and reliability are increasingly becoming the major contributors to evidential growth in public transport bus patronage in many parts of the world.
Optimal scales have a specific advantage over the dummy variable specification. If one uses multiple dummies from the same nominal variable, they will naturally be highly (negatively) correlated. In forecasting the effects of a change in categories, the analyst may have to decrease one category while simultaneously increasing the other.
When we initially adopted a dummy variable specification without the insights from NLCAA, we obtained, after extensive estimation, very few statistically significant effects. When we used NLCCA as the guiding framework, the selection of final statistically significant dummy variables was immediate as well as producing a much better model in terms of explanatory power.
The contribution of Zheng Li in assisting in preparing the data is acknowledged as is the ongoing discussions with Lee Schipper. Detailed comments from Tom Wilson (Department of Transport, Energy and Infrastructure Adelaide) and Alejandro Tirachini (ITLS and University of Chile) and four referees are appreciated.
- Canadian Urban Transit Association.: Bus Rapid Transit: A Canadian perspective, Issues Paper #10. CUTA, Toronto (2004)Google Scholar
- Cornwell, P., Cracknell, J.: The Case for Busway Transit, PTRC 18th Summer Annual Meeting (1990) (This paper is a summary of TRL Research Report 329 and Overseas Road Note 12 of the Transport Research Laboratory, Berkshire, UK)Google Scholar
- De Leeuw, J.:The Gifi system of nonlinear multivariate analysis. In: Diday, E. et al. (eds.) Data Analysis and Informatics, IV: Proceedings of the Fourth International Symposium. North Holland, Amsterdam (1985)Google Scholar
- Federal Transit Administration: Characteristics of Bus Rapid Transit for Decision Making, FTA-VA-26-7222-2004.1, US Department of Transportation, Washington DC (2004)Google Scholar
- Gifi, A.: Nonlinear Multivariate Analysis, Department of Data Theory. Leiden University, The Netherlands (1981)Google Scholar
- Gifi, A.: Nonlinear Multivariate Analysis. Wiley, Chichester (1990)Google Scholar
- Hensher, D.A.: Bus-based transitway or light rail? continuing the saga on choice versus blind commitment. Roads Transp. Res. 8(3), 3–21 (1999)Google Scholar
- Hensher, D.A.: Bus Transport: Economics, Policy and Planning, Research in Transportation Economics, Volume 18. Elsevier, Oxford (2007a)Google Scholar
- Hildago, D.: Comparing transit alternatives after recent developments in BRT in Latin America, paper presented at the January 2005 Transportation Research Board Annual Meeting, Washington DC (2005)Google Scholar
- Menckhoff, G.: Latin American experience with bus rapid transit, paper presented at the Annual Meeting, Institution of Transportation Engineers, Melbourne, August (2005)Google Scholar
- Sislak, K.G.: Bus rapid transport as a substitute for light rail: A tail of two cities, paper presented at 8th Joint Conference on Light Rail Transit, Transportation Research Board and American Public Transportation Association, http://www.apta.com/research/info/briefings/documents/sislak.pdf
- Transit Cooperative Research Program (TCRP).: Bus Rapid Transit Practitioner’s Guide, TCRP Report 118. Transportation Research Board, Washington DC (2007)Google Scholar
- Van der Burg, E., De Leeuw, J.: Nonlinear canonical correlation. Br. J. Math. Statist. Psychol. 36, 54–80 (1983)Google Scholar
- Vuchic, V.: Urban Transit Operations, Planning, and Economics. Wiley, New York (2007)Google Scholar
- Wright, L., Hook, W.: Bus Rapid Transit Planning Guide, 3rd edn. Institute for Transportation and Development Policy, New York (2007)Google Scholar