By 2050, 2.5 billion people will move into cities with the vast majority doing so in the developing world (United Nations 2014). This has the potential to lift millions out of poverty by increasing the productivity of firms and workers who benefit from agglomeration. However, rapid and unplanned growth can lead to sprawling, inefficient cities with hours wasted stuck in traffic. Governments will spend vast sums on mass transit systems to reduce commute times (McKinsey 2016), but measuring their benefits is challenging. While individuals save time on any particular commute, their decisions of where to live and work will change as new alternatives become attractive and land and labor markets adjust. The lack of detailed intra-city data in less developed countries coinciding with the construction of large transit systems makes evaluating their causal impact even more daunting.
In my job market paper, I ask the question: how large are the economic gains to improving public transit within cities and how are they distributed between low- and high-skilled workers? I construct detailed data across 2,800 census tracts from before and after the opening of the world’s largest Bus Rapid Transit (BRT) system–TransMilenio–in Bogotá, Colombia. I develop a new reduced form methodology derived from general equilibrium theory to empirically assess TransMilenio’s impact on city structure and use this framework to quantify its aggregate and distributional effects.
TransMilenio: The World’s Most-Used Bus Rapid Transit System
BRT provides an attractive alternative to subways in developing country cities: it delivers similar reductions in commute times at a fraction of the cost. Opened in 2000, TransMilenio operates more like a subway than the informal bus system that preceded it. Buses run in single-use lanes with express and local services, passengers pay at station entrances using smart cards, and buses are boarded at stations rather than at roadside. Today TransMilenio is recognized as the “gold standard” of BRT and is the world’s most patronized system (Cervero 2013).
To examine the impact of the changing transit network, I collect a host of administrative datasets covering residence, employment, commuting patterns, and land markets available before and after the system’s construction. Most are available at the census tract or below. My analysis proceeds in the following three steps.
Step 1: A Quantitative Framework for Evaluating the Aggregate and Distributional Effects of Public Transit
Prior to TransMilenio, poor, low-skilled workers relied on informal buses to commute to work. Due to their informal organization, these were 30% slower than cars preferred by the high-skilled. To understand the implications of improving public transit on worker welfare, I develop a general equilibrium model of a city where workers choose where to live, where to work, and how to commute. My approach extends a recent urban literature (e.g. Ahlfeldt et. al. 2015, Allen et. al. 2016) by incorporating multiple skill groups of workers and multiple transit modes. The model connects directly to discrete, tract-level data, making it particularly amenable to quantitative analysis.
Step 2: Using the Theory to Inform My Empirical Approach
A large literature estimates average treatment effects of transit by comparing locations near and far from stations (e.g. Gibbons and Machin 2005, Glaeser et. al. 2008). However, causal inference from relative comparisons can be confounded in the presence of spillovers between treatment and control units (Donaldson 2015). This is especially likely for small units within cities which are linked through commuting.
Instead, I show that for a wide class of models that feature a gravity equation for commute flows – which fits data on commute behavior very well – the full direct and indirect effects of the entire transit network on firms and workers can be summarized by a single variable: commuter market access (CMA).
Resident CMA reflects access to well-paid jobs: it is greater when a location is close (in terms of commute costs) to other locations with high employment, particularly so when these locations lack access to workers (increasing the wage firms are willing to pay). Firm CMA reflects access to workers: it is greater when a location is close to other locations with high residential population, particularly so when these locations lack access to jobs (lowering the wage individuals are willing to work for). For any gravity commuting model, these can be recovered using data on the location of residence and employment (as well as a measure of commute costs). These data are often available from censuses, or alternative sources such as cellphone metadata records.
Figure 1: Change in Commuter Market Access due to TransMilenio
Figure 1 plots the change in resident and firm CMA as a result of TransMilenio. In contrast to the distance-based approach, these capture a rich heterogeneity in treatment intensity. Tracts at the outskirts far from the high density of jobs in the center experience the largest improvement in resident CMA. In contrast, firm CMA increases the most in central locations that benefit from increased access to workers along all spokes of the network.
In a special case of my model, outcomes such as population, employment and house prices can be written as log-linear functions of CMA. I use this regression framework to empirically evaluate the impact of TransMilenio. One worry is that the government may have chosen routes to serve growing neighborhoods or stimulate lagging ones. I therefore create two instruments to predict their placement. The first is based on the location of a historical tram system that stopped operating in 1951. The second is based on the least cost routes I calculate the government would have built if their sole aim was to connect “portals” at the edge of the system with the central business. I then construct instruments for the change in CMA had TransMilenio been built along these predicted routes.
I find changes in CMA capture the heterogeneous response of population, employment and land markets (see Figure 2). Improvements in residential CMA led to growth in commute distances and wages, supporting the intuition that it measures access to good jobs. The system also caused a re-sorting of workers that increased residential segregation between skill groups.
Figure 2: Residential Floorspace Prices and Residential CMA
Step 3: Quantifying the Effect of BRT and Counterfactual Policies
Using the estimated model from Step 1, I find the system increased average welfare by 3.5% and output by 2.73% (net of construction and operating costs). This supports the notion that BRT can be a profitable investment for cities. TransMilenio enabled productive locations to “import” more workers through the commuting network, improving the spatial allocation of labor in the city.
However, high-skill workers benefit slightly more than the low-skilled. I find that the incidence of public transit across skill-groups depends on three channels:
- Mode Choice The group that uses public transit benefits along this channel, favoring the low-skilled.
- Commuting Elasticities The incidence of commute costs also depends on individuals’ willingness to bear high costs to work in a particular destination. In the model, this is determined by the heterogeneity of workers’ match-productivities with firms in different locations. For example, a high-skilled IT worker may be more willing to incur a long commute to an especially well-paid position. A low-skilled cleaner who receives similar wages wherever they work may instead substitute towards less costly alternatives. In the data, I find high-skilled workers are less sensitive to relative differences in commute costs across locations (suggesting a higher dispersion of match-productivity consistent with existing evidence) and bear a greater incidence along this channel.
- Geographic Characteristics The geography of the city and transit network matter: where house prices appreciate, whether the system connects where people live with well-paid jobs, and how these differ where each group lives and works. In Bogotá, these favor the high-skilled.
Accounting for the additional channels suggested by the theory and supported by the data, my methodology sheds new light on the distributional effects of commuting infrastructure. These differ from what I find using a traditional approach based on the “value of travel time savings” that accounts for mode choice alone (e.g. Small and Verhoef 2007).
Policy Implications Using the model to evaluate the impact of counterfactual policies, I find:
- Distributional Effects Depend on Geography Lines that connect poor neighborhoods with jobs in low-skill intensive industries benefit unskilled workers the most. If governments seek to target a particular group, routes should connect high densities of residence with high densities of well-suited jobs.
- Large Returns to Running Cheap, Complementary Bus Services TransMilenio’s “feeder” system – buses that transport individuals in the outskirts of the city to terminals at the end of lines using existing road infrastructure at no additional fare – increases welfare more than any other single line of the network. This underscores the benefits from reducing the last-mile problem of traveling between stations and final destinations.
- Increased Gains from an Integrated Transit and Land Use Policy In Bogotá, restrictive zoning laws meant that housing supply failed to respond to the BRT system. I simulate the effect of a “land value capture” scheme under which the government increases permitted densities where CMA rises and sells these permits to developers. Welfare gains are around 25% higher under one candidate policy, as increased housing supplies allow more individuals to take advantage of improved commuting opportunities, and revenues cover 18% of construction costs.