Advertisement

Connecting to Economic Opportunity: the Role of Public Transport in Promoting Women’s Employment in Lima

  • Daniel F. Martinez
  • Oscar A. MitnikEmail author
  • Edgar Salgado
  • Lynn Scholl
  • Patricia Yañez-Pagans
Original Article
  • 76 Downloads

Abstract

Limited access to safe transportation is one of the greatest challenges to labor force participation faced by women in developing countries. This paper quantifies the causal impacts of improved urban transport systems in women’s employment outcomes, looking at Bus Rapid Transit (BRT) and elevated light rail investments in the metropolitan region of Lima, Perú. We find large gains in employment and earnings per hour among women, and not for men, due to these investments. Most of the gains arise on the extensive margin, with more women being employed, but employment does not appear to be of higher quality than that for comparison groups. We find also evidence of an increase in the use of public transport. Results are robust to alternative specifications and we do not find evidence that they are driven by neighborhood composition changes or reorganization of economic activity. Overall, these findings suggest that infrastructure investments that make it faster and safer for women to use public transport can generate important labor market impacts for women who reside in the area of influence of the improved infrastructure.

Keywords

Urban transport Gender Employment Impact evaluation 

JEL Codes

J01 J16 O12 R40 

Introduction

Social and economic differences between women and men play a significant role in travel behavior (Curtis and Perkings 2006; Wachs 1996). For example, women tend to work close to their home to facilitate household-related travel (Sermons and Koppleman; 2001). In addition, as women usually oversee multiple household responsibilities, they make more stops and more chained trips than men (Taylor and Mauch 2000) and report making a considerable number of trips for family and personal business (Schintler et al. 2000). Women also make a higher proportion of their trips by transit and walking, even when a private vehicle is available in the household (Peters 1999, 2013). In addition to having different transport needs, women also frequently report feeling unsafe when using public transport systems, with sexual harassment and robbery being some of the key issues (Gardner et al. 2017; Gekoski et al. 2017).

There is limited research exploring women’s needs and issues concerning public transportation use in developing countries (Kash 2014). This poses a barrier for transportation planners seeking to effectively target policies to reduce the mobility and accessibility gap between men and women. Moreover, women tend to be underrepresented in the transportation-related jobs, from decision-making and planning roles, to operators of public transportation (Duchéne 2011; Kunieda and Gauthier 2007; Peters 2006) which many argue may contribute to and reinforce gender biases in transport systems (Peters 2006). Consequently, women in many developing countries continue to have reduced access to safe and adequate public transportation, which may potentially limit their mobility and accessibility to economic opportunities.

Unequal labor force participation between women and men is also prevalent and is more sharply observed in developing countries. While a myriad of socio-economic and overlapping factors affect the decision and ability of women to engage in the labor market, including the level of economic development of cities, individual educational attainment, social dimensions (such as social norms influencing marriage, fertility, and women’s role outside the household), and institutional settings (e.g., laws, protection, benefits) (Verick 2014), access to transport is increasingly emerging as a key issue affecting women’s labor force participation. A report by the International Labor Organization finds that limited access to safe transportation is the greatest challenge to labor force participation that women face in developing countries, reducing their participation probability by 15.5 percentage points (ILO 2017). Furthermore, due to wage inequality and a higher prevalence of part-time work, women tend to have lower earnings, and in turn, access to lower quality modes of transport (Astrop and Palmer 1996; Srinivasan and Rogers 2005).

The role of urban transport in facilitating access to employment opportunities becomes even more relevant in contexts of rapid urban growth, such as the case of Latin America and the Caribbean (LAC),1 where the increase in the value of centrally located land has pushed lower income and vulnerable populations to the outskirts of cities in search of affordable housing. As urban planning mechanisms are fragmented, urban peripheral growth tends to be sprawling, informal, and lacking in adequate transport infrastructure services. This, in turn, tends to increase both the monetary and time cost of transportation for the poor, and exacerbates the already low level of access to jobs and other economic opportunities (Carruthers et al. 2005). Data from CAF (2009, 2011) show that in the largest 15 metropolitan areas in Latin America while bus users spend on average 59 min per trip, car users spend on average 25 min per trip.

This paper studies the impacts of access to improved urban transport systems on women’s employment outcomes, contributing to a small literature on the effects of public transport on female labor participation.2 We exploit the opening of two modes of urban transportation in the metropolitan region of Lima, Perú, namely a Bus Rapid Transit System (BRT) and an elevated light rail, better known as metro Line 1. Both the BRT and Line 1 are major transit investments that have increased the formality of the public transit system in Lima. The systems have considerably reduced travel times and increased connectivity between peripheral areas to major employment centers. In addition, they are equipped with lighting, security personnel, and security cameras at stations and on-board trains, which represent substantial improvements relative to the safety of the rest of the public transit in the city. We hypothesize that women in areas close to the system react to these changes by increasing their usage, improving their accessibility to jobs.

To quantify the causal impacts of the introduction of the two transport systems (BRT and Line 1), we estimate difference-in-differences (DID) models, using annual data from the Peruvian National Household Survey (ENAHO, original Spanish acronym). Our identification strategy compares changes in employment indicators for men and women living in areas that are closer to these transport systems versus those living in comparable areas that are farther away and with limited access to these services. The comparison areas satisfy a common support condition defined by a propensity score model at the area level. We test for changes in neighborhood composition and reorganization in economic activity, and for potential spillover effects. In addition, we conduct tests to rule out the possibility that those living in areas closer and farther to the systems were experiencing different trends before the urban transport systems were implemented.

We do not find significant changes for men but find positive and significant results for women. Our results show, by the end of the analysis period, an increase of ten percentage points in the probability of being employed among women living closer to the systems when compared to women living farther away. We also find large increases in both total earnings and hours worked, resulting in at least 17% increase in earnings per hour. The analysis suggests that most of these results are being driven by women previously not in the labor force joining the labor market. To understand what drives the earning results, we measure job quality by looking at job benefits, the characteristics of the employing firm or self-employment activity, and the type of occupation. Overall, we do not find significant improvements in job conditions. Thus, while employment increases, and earnings per hour increase for women close to the improved public transport systems, these changes are not driven by women finding higher quality jobs.

We also find substantial increases in the use of public transport, again only for women. The share of women spending positive amounts on public transport increases eight percentage points, suggesting that the opening of the BRT and Line 1 strongly pulled women into using public transport. Given that pre-intervention data shows that a large majority of men and women rely on public transport for trips to and from work, this provides further evidence for the hypothesis that the increase in labor force participation is facilitated by the new transport systems. Given the available information, and that both BRT and Line 1 brought simultaneous improvements in speed and safety, we are unable to disentangle the impacts that may be attributed to each of these dimensions. Despite this, we present descriptive evidence from local perception surveys showing that speed is one of the main factors influencing BRT and Line 1 users to choose these systems. Also, we observe that women who use BRT and Line 1 are more satisfied with safety in public transport and experience less harassment. Several robustness checks confirm that our main results are stable and robust to specification changes.

The paper is structured as follows. The next section discusses in more detail the related literature highlighting the main contributions of this paper. The “Lima’s Urban Transport System” section explains how urban transport systems operate in Lima. The “Data and Outcome Variables” section describes the data used in the analysis. The “Gender Differences in Travel and Employment Patterns in Lima” section presents some descriptive statistics on the different patterns in travel behavior of men and women. The “Methodology” section describes the methodology, while the “Results and Analysis” section presents the main results of the paper and robustness tests. The “Conclusion” section summarizes the conclusions.

Related Literature

Much of the literature on gender and transport in developing countries has explored women’s perception of accessing and using transport systems, finding that sexual harassment3 is one of the main issues that affect women who use public transportation (Schulz and Gilbert 1996; Gwilliam 2003; Zermeno et al. 2009; Kash 2014; Neupane and Chesney-Lind 2014). Specifically, women report frequently feeling unsafe walking to a transit stop/station, waiting for the bus or train, and traveling in the system. Sexual harassment disproportionately affects lower income women, who rely on public transport, commute longer distances, and tend to travel through more dangerous neighborhoods (Zermeno et al. 2009).

A handful of studies examine these security issues in informal versus formal public transit. For example, in a study conducted in Mexico City, female respondents said that the informal transport service was the most unsafe mode and that higher quality public transportation (scheduled service, defined stops, cleaner buses) will lead to safer trips (Tudela Rivadeneyra et al. 2015). In Bogotá and Arequipa, riders of informal transportation services identified crime as one of the principal problems with the system, which was tied to the crowding during peak hours. In Bogotá, women were significantly more concerned about crime than men in transit (Kash 2014). Women in the slums in Delhi identified themselves as targets of sexual harassment while traveling to work, especially when walking to the stops of informal and public transportation, which in some cases affected their ability to retain jobs in distant areas from their homes (Anand and Tiwari 2006).

Formal surveillance, such as the presence of on-site security personnel, has been found to be an effective strategy to reduce sexual harassment at transit stations (Gekoski et al. 2017; Loukaitou-Sideris and Fink 2008). Other security measures that have been rated positively include good lighting at bus stops and adjacent streets, request-stop programs (which allow women to get out of the bus closer to their destination), public awareness campaigns denouncing sexual harassment, policing (in vehicles and stops), and public education (Zermeno et al. 2009; Loukaitou-Sideris and Fink 2008). Some authors have also found benefits in women-only vehicles4 (Zermeno et al. 2009Ishibashi, 2003); however, this solution does not necessarily change the behavior of the perpetrators and might be perceived as a segregation tool against women (Gardner et al. 2017).

Regarding the literature connecting transport infrastructure with employment, this relationship is theorized to occur due to several factors. The reservation wage hypothesis posits that the wage at which a person is willing to supply labor is likely to rise in response to increasing transport costs and therefore limit the geographic range of viable job opportunities, particularly for lower wage workers (Patacchini and Zenou 2005). In addition, early scholars have hypothesized that longer commute times can have an impact on hours worked (Oi 1976). Increased commute times may decrease hours worked particularly among women due to a higher opportunity cost for going to work relative to household production activities (White 1977). As women, particularly in developing countries, tend to assume a larger burden of household and care-related work, this is argued to leave less time to participate in paid work and contribute to gender gaps in labor force participation rates (Ferrant et al. 2014). Farré et al. 2018 explore the effects of commuting time on female labor force participation in the USA and find that increased commuting times decrease labor force participation rates by 0.05 percentage points, with larger effects for married women and women with children under the age of 5.

In developing countries, differences in gender also exist in the way transportation options are used and accessed (Babinard 2011) and, in turn, influence employment decisions. Women tend to have access to slower transport modes, as they walk and take transit more than men, while men tend to use private vehicles and taxi’s (Salon and Gulyani 2010; Anand and Tiwari 2006; Kunieda and Gauthier 2007). After walking, public transit is the most important transport mode for women who can afford it. Besides choosing a job closer to home to meet household duties, low-income women report taking this decision because they can walk from home to work and vice versa without spending money on any other mode (Astrop and Palmer 1996; Srinivasan and Rogers 2005).

Overall, few studies rigorously estimate the causal relationships between urban transport investments and employment outcomes (Yañez-Pagans et al. 2018). This responds to the empirical complexities that arise when trying to distinguish between impacts that can be attributed to transport investments versus those that result from non-random factors driving their placement (i.e., demand considerations) that might benefit populations that were already better connected, were more employed, or had higher income. Another important aspect pertains to the attribution of measured benefits. What is measured might not necessarily reflect the benefits obtained by the original population living in project-served areas, but could reflect that new populations, with distinct characteristics, are moving in (i.e., compositional changes).

There are several non-causal studies that analyze the changes on access to employment opportunities resulting from urban transport systems (Bocarejo and Oviedo 2012; Delmelle and Casas 2012; Bocarejo et al. 2016; Guzman and Rivera, 2017; Venter et al. 2018). These papers examine the reduction in travel times generated by improved transport systems across different areas in a city and consider how well they serve to connect low-income or vulnerable populations to employment centers. Another group of studies examines the correlation between employment outcomes and distance to urban transport systems, showing that proximity to a system is correlated with lower levels of unemployment (Sanchez 1999) or with a lower probability of being informally employed (Oviedo-Dávila 2017).

Causal studies are more limited, and the majority have relied on a DID empirical strategy. For cities in the USA, studies have shown larger job growth in areas surrounding transport stations, particularly in downtown areas (Cervero and Landis 1992, for subways) and for white-collar and high-wage employment (Guthrie and Fan 2016, for BRT). Other studies find increases in the propensity of suburban firms, previously not near a metro line, to hire minority populations (Holzer et al. 2003). In a related study, Scholl et al. (2018) analyze the overall labor market impact of the BRT system (trunk and feeder lines) in Lima, finding positive impacts on labor outcomes concentrated on individuals living close to the trunk line, and no impacts on individuals living in low-income areas served by the feeders. Similarly, Tsivanidis (2019) finds that the BRT system in Bogota (Transmilenio) caused increases in welfare and output, especially for high-skilled workers.

The role of transportation in shaping economic opportunities for women has not been explored in the literature to date. To the best of our knowledge, the only other study we identified looking at a related topic is Abu-Qarn and Lichtman-Sadot (2019), who find evidence of a trade-off between investment in education and time allocated to work for women after the introduction of bus services to Arab towns in Israel. This study thus makes two important contributions to the literature. First, it contributes to the limited causal evidence on the impacts of transport systems on employment. Second, and more importantly, it presents novel empirical evidence on the impacts that improved urban transport systems can generate on women.

Lima’s Urban Transport System

Lima is the capital of Perú, and its rapidly growing metropolitan area (Lima-Callao), with a population of close to 10 million, represents about one-third of the population of the country. Its public transit system is highly chaotic, informal, and challenged by oversupply of vehicles and poor levels of service quality. In addition, the city’s transport network suffers from high levels of congestion, traffic accidents, and transport-related air pollution (Bielich 2009).

Levels of sexual harassment of women in Lima’s public transport system are among the highest in the Latin American region, with 78% of women reporting that they had been a victim in the past year while traveling in a transit vehicle or waiting at a bus stop or transit station (Galiani and Jaiman 2016). Sixty-four percent of women surveyed in the same study stated that they felt insecure or very insecure in Lima’s public transit system, and 77% reported feeling unsafe if traveling at night in the system (Galiani and Jaiman 2016).

Through a series of planning efforts over the past 20 years, the Metropolitan Area of Lima-Callao has begun slowly transforming its transport system. The Metropolitan Area Urban Transport Project, developed between 1996 and 2000, sought to increase mobility and reduce the social and environmental costs of transport by connecting the most populous areas of the city to important employment centers. The first part of this project implemented was the BRT line and was followed shortly after by the implementation of the metro Line 1. Although the two projects represent significant improvements to the city’s transport system, mobility remains mostly informal (Darido et al. 2015).5 While Lima’s public transit system is still one of the least secure in the region, women ranked Lima’s Metro Line 1 to be the safest, followed by taxis, the BRT, buses, and finally microbuses (Galiani and Jaiman 2016).

Bus Rapid Transit System

The BRT project in Lima, better known as the Metropolitano, connects two of the fastest growing areas of the city and lower income neighborhoods in the northern and southern cones of the city with the financial district, major universities, and the historic downtown. The Metropolitano is the first line of a larger system planned for the city and was one of the first mass public transit systems proposed for Lima. The corridor comprises 28.6 km of segregated busway, with 35 stations, two terminals, and a central transfer. It also includes feeder routes that extend from the two terminals up to 14 km into the surrounding and primarily low-income neighborhoods in the north and south cones. It serves one of the highest demand corridors and offers late night and weekend service. Passing lanes and multiple docking bays allow for express and super express services between high-demand stations (Scholl, et. al. 2015).

Beginning operation in mid-2010, the system opened with only 22% of the planned articulated buses and five feeder routes in operation, in part because of low-demand and unfinished infrastructure (Guerra García Picasso 2014). By 2014, the system was nearly fully operational, with the full fleet of 300 articulated buses operating and 222 feeder buses serving 20 feeder routes. In the same year, demand reached 360,000 card validations per day. By 2019, the system’s demand surpassed 400,000 daily validations (Protransporte 2019). Travel time savings of the system were considerable. Before the implementation of the system, the average trip time from one end of the trunk line to the other took on average 55 min, while the same trip would take 35 min on average in the BRT (Scholl et al. 2015).

Metro Line 1

Lima’s metro Line 1, the first metro line for the city, is a 34.6-km elevated light rail that runs north-south along the eastern portion of the city and in parallel to the BRT line. The line was built in two stages. The first segment of the line began operating in January 2012 and connects Villa El Salvador, a low-income area, to central Lima. The second 12.4-km stretch runs from downtown Lima to San Juan de Lurigancho and opened in July 2014. Trains reach a maximum velocity of 100 kph and carry up to 1000 to 2000 passengers (AATE 2013). As with the BRT system, several operational and infrastructure improvements have been implemented since it opened for service, including amplification of stations and the addition of trains to reduce overcrowding and headways. As of 2015, the system carried 320,000 passengers per day, surpassing demand forecasts. Currently, ridership is estimated to be 344,000 per day with headways of 4–6 min in the peak hours (Diario Correo 2018).

Data and Outcome Variables

We rely on data both before and after the implementation of the BRT and Line 1. Our main data source is the Peruvian National Household Survey (ENAHO), produced by Perú’s National Institute of Statistics and Informatics (INEI, original Spanish acronym). The ENAHO is a continuous quarterly survey that generates indicators for poverty levels, employment, income, and living conditions of households in both urban and rural areas. It surveys approximately 3000 households and 15,000 persons per year in the Lima metropolitan area. Our empirical analysis combines the annual cross-section ENAHO surveys for the period 2007 to 2017.

We also rely on three additional datasets that allow characterizing smaller geographical areas and measuring neighborhood characteristics prior to the entry into operation of the BRT and Line 1: the 2008 Economic Census, the 2007 Population census, and a 2004 Origin-Destination survey. The last one is a specialized transport survey capturing detailed individual travel behavior. Details of the variables used from each of these surveys to estimate the probability of treatment can be found in the online appendix.

We analyze employment and quality of employment outcomes for individuals ages 18 to 64. Employed individuals are defined as working-age individuals who respond affirmatively to the question of whether they worked in the week prior to their interview and report positive earnings. We characterize quality of employment in several ways: (i) as working in formal firms (dummies for registered with the tax authority, carrying accounting books, or with more than five employees); (ii) as contributing to social security or under a formal contract (dummies for each); (iii) or as being in occupations associated with the top or bottom 25% of the earning distribution in the ENAHO sample (dummies for each).6 We also create two summary measures of quality of employment based on these variables: an index that adds up the five dummy variables in (i) and (ii) plus the dummy variable for being in an occupation in the top 25% of the earning distribution (this index can assume values from 0 to 6); and a dummy variable equal to one when the index is positive.

Gender Differences in Travel and Employment Patterns in Lima

In this section, we use baseline data from the 2004 Origin-Destination Survey (OD) to characterize transport patterns for men and women living in the metropolitan region of Lima and prior to the introduction of the BRT and Line 1. To facilitate the presentation, we aggregate the 427 traffic analysis zones (TAZ) in the OD survey into 14 zones, following an aggregation proposed by JICA (2015), and calculate statistics and identify gender gaps within those aggregated zones.

Figure 1 compares average travel times in minutes for trips outside their zone, reported by men and women. Overall, regarding trips for all purposes (panel a), we see that women travel less time than men and this pattern is observed in almost all areas across the metropolitan region (statistically significant differences are marked with a black triangle). The average travel times by area for women indicates that travel times are longer for those living farther away from the city center. It is important to highlight that even though the BRT and Line 1 are depicted in the maps, these systems had not yet been built in 2004. Regarding average travel times for work-related trips (panel b), the gender differences tend to disappear, particularly for the more centrally located zones. This suggests that, conditional on working, men and women experience the same travel times when they are centrally located. For those who live farther away from the city center, men seem to have longer work-related trips. Figure A1 in the online appendix shows that while men use public transport more than women for trips for all purposes, those differences disappear when considering only trips related to work.
Fig. 1

Average travel times for all trips outside traffic zone in minutes (2004). Black triangles denote that difference between male and female are statistically significant at 5%

Thus, the data shows that men and women have different travel behaviors and suggests that some of these differences may be explained by heterogeneity in their employment status. The fact that a lower percentage of women work and that, in general, bear most of the household work, is reflected in the large gender differences in transport patterns for the overall trips. Conditional on labor force participation, the OD survey suggests that women demand public transportation in similar ways as men do, but that they travel shorter distances and stay more within their own traffic zone. This is consistent with the fact that they are traveling shorter distances and could be also a reflection of the security concerns associated with traveling by public transport.

Methodology

We estimate the impacts of the introduction of the BRT and Line 1 on employment outcomes using a difference-in-differences (DID) approach, after the selection of conglomerates7 that are similar, through a propensity score at the conglomerate level, as we explain below. Our DID compares individuals before and after the introduction of the BRT and Line 1 living in treatment and control areas and uses distance to these transport systems as an exogenous measure of exposure to the new infrastructure. We exploit the geographic coordinates (centroid of the conglomerate) assigned for each household surveyed in the ENAHO to calculate the Euclidian distances of each household to the (i) closest BRT station and (ii) closest Line 1 station.8 Treatment areas are defined as those conglomerates within 1 km of the BRT or Line 1. This cutoff is based on the standard convention of an average walk speed of 5 km per hour (Levine and Norenzayan 1999). It also considers that according to data from the 2011 OD survey, 90% of public transport users in Lima walk 12 min or less to reach public transportation (i.e., around 1 km).9 We set as control areas those conglomerates between 2 and 5 km from the BRT or Line 1. To prevent potential spatial spillovers on the control group, we drop from the sample households located within 1 km and 2 km. Figure 2 shows in blue the treatment areas for both the BRT and Line 1 and in red the control areas.10
Fig. 2

Treatment and control areas

Since the two systems run parallel, and at some points, very close to each other, it is not clear which of the two lines a person relatively close to both of them would ride, leading to potential overlap in the treatment samples. We define a single treatment group by pooling together households in the areas of influence to either system (BRT or Line 1). Even though we have limited power, in the online appendix, we also explore the differential effects of the two systems, and test for treatment effects heterogeneity across different distances to the systems, by comparing the impacts observed within the 0 to 0.5 km buffer versus those within 0.5 and 1 km.

Since the BRT and Line 1 operations had a slow ramp-up since their official opening in 2010 (BRT) and 2011 (Line 1), it is of interest to understand the timing of the effects of the two systems. The standard DID model, allowing for time heterogeneity in effects, would be
$$ {Y}_{it}=\alpha +{\sum}_k{\gamma}_k{P}_{kt}+\delta {T}_i+{\sum}_k{\beta}_k{P}_{kt}{T}_i+\theta {X}_{it}+{\upvarphi}_d+{\eta}_{dt}+{\varepsilon}_{it} $$
(1)
where Yit is the outcome of interest (e.g., employment status) computed for the working-age (ages 18–64) individual i in time t, Ti is a dummy variable equal to 1 if individual i lives in the area of influence of the BRT or Line 1 and zero otherwise, the k dummies Pkt are equal to one for different sub-periods after the introduction of the lines (i.e., 2010–2011, 2012–2014, 2015–2017) and zero otherwise, and βk are the coefficients of interest, measuring the effects of the improved systems in each sub-period k. Xit is a vector of individual- and household-level covariates for individual i in time t, 11 φd are district fixed effects, while ηdt represents district trends, to control for potential within-district (which is the level for many planning decisions, including transport and security) time-variant unobserved heterogeneity, and εit is an error term.12
As our interest is in the differential effects for men versus women, we could either estimate Eq. (1) separately for men and women, and compare the respective coefficients, or modify Eq. (1) to allow for an interaction with a female dummy variable. To improve efficiency in our estimates, we opt for the interacted model. This allows estimating the following model to capture the heterogeneous effects on women:
$$ {\displaystyle \begin{array}{l}{Y}_{it}=\alpha +{\sum}_k{\gamma}_k{P}_{kt}+\delta {T}_{si}+\pi {F}_i+{\sum}_k{\beta}_k{P}_{kt}{T}_i+{\sum}_k{\tau}_k{P}_{kt}{F}_i+\zeta {T}_i{F}_i+{\sum}_k{\lambda}_k{P}_{kt}{T}_i{F}_i+{\theta}_M{X}_{it}\\ {}\kern2em +{\theta}_F{X}_{it}{F}_i+{\kappa}_M{\varphi}_d+{\kappa}_F{\varphi}_{\mathrm{d}}{F}_i+{\psi}_M{\eta}_{dt}+{\psi}_F{\eta}_{dt}{F}_i+{\varepsilon}_{it}\end{array}} $$
(2)
where Fi is the female dummy, and everything else is defined in the same way as in Eq. (1). In Eq. (2), we are interested now in the coefficients βk and λk. The DID estimate for the treatment effect for men is βk and the DID estimate for the treatment effect for women is (βk + λk). The comparison of these two effects allows us to compute the differential treatment effects across gender. More specifically, the treatment effect for women and men differs by (βk + λk) − βk = λk, which is the coefficient of the triple interaction term in Eq. (2). Given that some district characteristics, for example, security-related such as the presence of street gangs, might differentially affect women’s and men’s mobility decisions and outcomes, we also allow district-time trends to vary by gender in Eq. (2).

The specification in Eq. (2) considers a pre-treatment period (2007–2009) and three post-treatment sub-periods (2010–2011, 2012–2014, 2015–2017). Although the BRT system started operations only in July 2010, we include the data for this entire year in the post-treatment period to avoid splitting the ENAHO sample within a year. In addition, Line 1 was also completed in stages, with the first stage being completed in 2011 and the second stage in 2014. Thus, the 2010–2011 period can be considered as the period when the BRT was transitioning into fully functional, while the 2012–2014 can be considered as the one where the BRT was already (mostly) working as planned, while Line 1 was transitioning into becoming fully operational. The period 2015–2017 is the one where both the BRT and Line 1 are fully operational.

There could be a concern that the covariates included in a linear way in Eq. (2) may not be enough to properly account for differences in observable characteristics at baseline between treatment and control areas. To address any potential bias, we select in a first stage the most comparable conglomerates, and then estimate Eq. (2) restricting the ENAHO sample to the selected conglomerates. To do this, we take data from the 2007 Population Census, the 2008 Economic Census, and the 2004 OD survey; aggregate them at the ENAHO conglomerate level; and assign conglomerates to treatment and control groups based on the ENAHO individuals living in those conglomerates, and their distance to the BRT or Line 1. We then estimate at the conglomerate-level a propensity score model for the probability of being a treated conglomerate.13 Based on the estimated propensity score, we impose overlap between the treated and control conglomerates (i.e., we identify the comparable conglomerates as those that have common support or where there is overlap in the propensity score distribution). For this, we follow the propensity score trimming strategy proposed by Crump et al. (2009).14 We then estimate (2) using only households in the conglomerates that satisfy the common support condition.

DID models rely on the identification assumption that treatment and control observations follow parallel trends prior to the start of the treatment or intervention. To test whether this assumption is reasonable in our case, we use information from the baseline years 2007–2009, prior to the opening of the BRT and Line 1. More specifically, we estimate model (2) classifying 2007–2008 as baseline years and assuming 2009 is the treatment year. Any significant differences between these two periods would be an indication that trends between treatment and control areas are not parallel from 2007-2008 to 2009, which could indicate a violation of the key assumption underlying the validity of a DID model.

Finally, it is important to note that outcome variables related to earnings and hours worked are zero for those not employed. The same is true for expenditures in public transport for those not using this type of transport or no transport at all. This implies that it is not possible to take logarithm of these variables to obtain percentage changes when running the regressions. We apply to those variables the inverse hyperbolic sine (IHS) transformation. This allows the same interpretation as a logarithm in a regression framework and it is defined at zero.15

Results and Analysis

In this section, we first discuss descriptive statistics of the outcomes and covariates used in our estimations and report the results of the parallel trend tests to justify the validity of our empirical approach. Then, we discuss the main results obtained following the methodology described above, analyze intermediate outcomes that can help understand the mechanisms behind our results, and test for the presence of spillover effects. We discuss the possible channels explaining our results and in the online appendix, we report the results from robustness and heterogeneity analyses.

Descriptive Statistics

Table 1 presents in panel A the balancing of covariates at baseline. This is done for the full sample and for the sample after imposing overlap (satisfying common support). Differences between treated and controls within gender groups that are statistically significant at the 5% significance level are highlighted in italic. Results in this panel suggest that there are no major differences between treated and control observations in the baseline period for both men and women. Within the sample for men, before overlap, there are less indigenous and married individuals in the treated areas in the baseline. After imposing overlap, only statistically significant differences in marital status remains. Within the sample for women, the sample before overlap indicates that women in treated areas live in smaller households, are more likely to be head of household, and are older. After overlap, the age difference disappears, but women in treated areas are still more likely to head a household or live in smaller families. It is important to note that it is not necessary for covariates to be balanced between treated and control individuals in a DID model, as one of the main advantages of this type of models is that it allows for systematic differences between the two groups, provided they are not changing over time. Nevertheless, if the treatment and comparison conglomerates are very different, it weakens the credibility of the ex ante assumption of parallel trends between the two groups.
Table 1

Descriptive statistics outcomes and covariates

In Table 1 panel B, we present, after imposing overlap, the summary statistics of the different outcomes of interest for control and treatment groups by gender in the period prior to the BRT and Line 1 opening (2007–2009) and in the post-intervention period (2010–2017). There are large gender differences in employment rates between men (around 85%) and women (around 65%). There is a predominance of self-employment over paid employment (employee); women are more likely to be homemakers and have lower quality jobs (according to our job quality index16) and be employed in occupations classified as in the bottom 25% of the ENAHO sample earning distribution. These gender differences are also evident for earnings, hours worked, and their ratio. A large proportion of the sample reports spending on public transport; however, men tend to spend more than women. In terms of education, the average number of years of education is 11.9 for men and 11.2 for women. The latter is also reflected in the lower proportion of women with an education level of high school or higher.

Parallel Trends Placebo Tests

If the treated and control groups do not follow parallel trends, then it would not be valid to use the observed post-treatment outcomes for the controls, as counterfactual for the post-treatment outcomes for the treated. Exploiting pre-treatment data, we run regressions estimating the treatment effects for the year 2009, with the 2007–2008 as the “placebo” pre-treatment period; we would expect the “placebo treatment effect” associated with the year 2009 to be zero. We test this in Table 2 for the main outcomes, that we will discuss below, and find that the null hypothesis of no treatment effects cannot be rejected for any of the outcomes, either for men and women. At the very least, the results suggest that we cannot precisely detect any systematic differences between the treatment and control areas in the pre-treatment period that may affect the credibility of the empirical strategy.17
Table 2

Tests of parallel trends assumption

Coefficient

Employment

Unconditional on employment

Job quality index

Public transport expenditure

Education

 

IHS (earnings)

IHS (earnings/h)

Index value

Index value > 0

Expenditure > 0

IHS (expenditure)

Years

High school or more

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

2009 × treated BRT/Line 1 × Female

− 0.042 (0.041)

− 0.262 (0.283)

− 0.072 (0.130)

0.210 (0.176)

0.007 (0.038)

− 0.008 (0.030)

− 0.076 (0.168)

0.138 (0.375)

− 0.015 (0.044)

2009 × treated BRT/Line 1 × Male

− 0.001 (0.032)

0.044 (0.262)

0.124 (0.106)

0.018 (0.210)

− 0.039 (0.045)

− 0.089* (0.051)

− 0.514* (0.272)

− 0.074 (0.415)

0.065 (0.067)

2009 × treated BRT/Line 1 × (Female − Male)

0.006 (0.041)

− 0.306 (0.384)

− 0.196 (0.167)

0.191 (0.272)

0.046 (0.058)

0.081 (0.059)

0.437 (0.318)

0.213 (0.556)

− 0.080 (0.080)

District-year fixed effects

Yes

Yes

Yes

yes

Yes

Yes

Yes

Yes

Yes

Controls (linear and interacted with female)

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Observations

4,374

4,374

4,374

4,374

4,374

4,374

4,374

4,374

4,374

Number of districts

25

25

25

25

25

25

25

25

25

Standard errors in parentheses, clustered at the (district × female) level. ***p < 0.01; **p < 0.05; *p < 0.10. The regressions use only data from 2007 to 2009 (see notes for Tables 4 and 5 for details on the definition of the outcomes in columns (2) to (5))

Impacts on Employment, Earnings, and Job Quality

We analyze now the results from estimating the DID model specified in Eq. (2). In all cases, the regressions include the full vector of individual and household level covariates discussed in the “Methodology” section, as well as gender-specific district fixed effects and trends. Standard errors are clustered at the district level by gender. All the tables follow the same structure, with each column representing the regressions using a different outcome. In the interest of space, the first three rows of Tables 3, 4, 5, and 6 show the estimated treatment impacts for women \( \Big({\hat{\beta}}_k+{\hat{\lambda}}_k \)). The next three rows show the estimated coefficients \( {\hat{\beta}}_k \)associated with the treatment effects for men. The last three rows provide the differential effect of the BRT and Line 1 for women compared to the effect for men (\( {\hat{\lambda}}_k \)), which is the triple interaction term in (2) . All the regressions discussed in Tables 2, 3, 4, 5, 6, and 7 use the estimation sample after imposing overlap.18,19
Table 3

Employment outcomes

Coefficient

Employment

Categories of employment

Homemaker

(1)

Employee(2)

Self-employed(3)

Domestic worker (4)

(5)

2010–2011 × treated BRT/Line 1 × Female

0.060*** (0.022)

0.018 (0.026)

− 0.002 (0.017)

0.045** (0.017)

− 0.058*** (0.021)

2012–2014 × treated BRT/Line 1 × Female

0.085*** (0.020)

0.048 (0.030)

− 0.000 (0.020)

0.036** (0.015)

− 0.053*** (0.016)

2015–2017 × treated BRT/Line 1 × Female

0.101*** (0.034)

0.062 (0.046)

− 0.011 (0.019)

0.050** (0.021)

− 0.073*** (0.024)

2010–2011 × treated BRT/Line 1 × Male

0.009 (0.019)

0.023 (0.039)

− 0.015 (0.025)

− 0.000 (0.004)

0.010 (0.011)

2012–2014 × treated BRT/Line 1 × Male

− 0.025 (0.021)

0.022 (0.029)

− 0.045* (0.026)

− 0.002 (0.004)

0.008 (0.012)

2015–2017 × treated BRT/Line 1 × Male

− 0.011 (0.012)

0.040 (0.032)

− 0.053 (0.034)

0.002 (0.005)

0.016 (0.010)

2010–2011 × treated BRT/Line 1 × (Female − Male)

0.052* (0.029)

− 0.006 (0.047)

0.013 (0.031)

0.045** (0.017)

− 0.068*** (0.023)

2012–2014 × treated BRT/Line 1 × (Female − Male)

0.109*** (0.029)

0.027 (0.042)

0.045 (0.033)

0.038** (0.016)

− 0.061*** (0.019)

2015–2017 × treated BRT/Line 1 × (Female − Male)

0.113*** (0.036)

0.022 (0.056)

0.043 (0.039)

0.048** (0.021)

− 0.089*** (0.026)

District-year fixed effects

Yes

Yes

Yes

Yes

Yes

Controls (linear and interacted with female)

Yes

Yes

Yes

Yes

Yes

Observations

21,338

21,338

21,338

21,338

21,338

Number of districts

31

31

31

31

31

Standard errors in parentheses, clustered at the (district × female) level. ***p < 0.01; **p < 0.05; *p < 0.10

Table 4

Earning and hour outcomes

 

Unconditional on employment

Conditional on employment

IHS (earnings)

IHS (hours)

IHS (earnings/h)

IHS (earnings)

IHS (hours)

IHS (earnings/h)

(1)

(2)

(3)

(4)

(5)

(6)

2010–2011 × treated BRT/Line 1 × Female

0.490*** (0.170)

0.389*** (0.116)

0.169** (0.066)

0.047 (0.074)

0.219*** (0.070)

0.026 (0.065)

2012–2014 × treated BRT/Line 1 × Female

0.657*** (0.162)

0.466*** (0.090)

0.173*** (0.051)

0.025 (0.073)

0.188** (0.072)

− 0.064 (0.049)

2015–2017 × treated BRT/Line 1 × Female

0.823*** (0.274)

0.546*** (0.147)

0.268*** (0.082)

0.105 (0.090)

0.192** (0.078)

0.040 (0.046)

2010–2011 × treated BRT/Line 1 × Male

− 0.078 (0.161)

0.063 (0.081)

− 0.088 (0.076)

− 0.152** (0.065)

0.034 (0.037)

− 0.117* (0.069)

2012–2014 × treated BRT/Line 1 × Male

− 0.355** (0.173)

− 0.112 (0.098)

− 0.196** (0.074)

− 0.180*** (0.056)

0.000 (0.050)

− 0.149** (0.061)

2015–2017 × treated BRT/Line 1 × Male

− 0.167* (0.088)

− 0.060 (0.059)

− 0.055 (0.068)

− 0.093 (0.086)

− 0.018 (0.062)

− 0.028 (0.086)

2010–2011 × treated BRT/Line 1 × (Female − Male)

0.568** (0.234)

0.327** (0.142)

0.257** (0.101)

0.199** (0.098)

0.185** (0.079)

0.143 (0.094)

2012–2014 × treated BRT/Line 1 × (Female − Male)

1.011*** (0.237)

0.578*** (0.133)

0.369*** (0.090)

0.205** (0.092)

0.188** (0.087)

0.084 (0.078)

2015–2017 × treated BRT/Line 1 × (Female − Male)

0.991*** (0.288)

0.606*** (0.158)

0.323*** (0.107)

0.198 (0.124)

0.210** (0.099)

0.068 (0.097)

District-year fixed effects

Yes

Yes

Yes

Yes

Yes

Yes

Controls (linear and interacted with female)

Yes

Yes

Yes

Yes

Yes

Yes

Observations

21,338

21,338

21,338

15,509

15,509

15,509

Number of districts

31

31

31

31

31

31

Standard errors in parentheses, clustered at the district level. ***p < 0.01; **p < 0.05; *p < 0.10. IHS() refers to the inverse hyperbolic sine transformation (see text for details)

Table 5

Job quality outcomes

Coefficient

Job quality index

Formality based on firm characteristics

Formality based on employee

Characteristics of occupation

Index value

Index value > 0

Keeps accounting books

Is registered

Has more than 5 employees

Contributes to social security

Has a formal contract

In top quartile of earnings

In bottom quartile of earnings

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

2010–2011 × treated BRT/Line 1 × Female

0.068 (0.098)

0.004 (0.029)

0.019 (0.018)

0.021 (0.017)

0.012 (0.024)

− 0.038 (0.030)

0.041* (0.022)

0.014 (0.019)

0.051** (0.022)

2012–2014 × treated BRT/Line 1 × Female

0.072 (0.141)

0.029 (0.027)

0.026 (0.023)

0.036 (0.025)

0.016 (0.037)

− 0.037 (0.027)

0.033 (0.034)

− 0.002 (0.017)

0.062*** (0.023)

2015–2017 × treated BRT/Line 1 × Female

0.134 (0.190)

0.042 (0.050)

0.016 (0.027)

0.041 (0.027)

0.019 (0.044)

− 0.013 (0.039)

0.045 (0.039)

0.026 (0.027)

0.045** (0.023)

2010–2011 × treated BRT/Line 1 × Male

− 0.092 (0.157)

− 0.042 (0.032)

0.018 (0.043)

− 0.023 (0.036)

− 0.056 (0.034)

− 0.043 (0.041)

− 0.001 (0.031)

0.013 (0.019)

0.076*** (0.023)

2012–2014 × treated BRT/Line 1 × Male

− 0.109 (0.119)

− 0.052* (0.028)

− 0.006 (0.023)

− 0.029 (0.021)

− 0.040 (0.030)

− 0.043 (0.038)

0.025 (0.028)

− 0.016 (0.018)

0.014 (0.025)

2015–2017 × treated BRT/Line 1 × Male

0.041 (0.121)

− 0.035 (0.021)

0.021 (0.033)

0.007 (0.031)

− 0.018 (0.033)

− 0.021 (0.031)

0.045 (0.031)

0.008 (0.016)

− 0.010 (0.032)

2010–2011 × treated BRT/Line 1 × (Female − Male)

0.160 (0.185)

0.046 (0.043)

0.001 (0.047)

0.044 (0.040)

0.068 (0.041)

0.004 (0.051)

0.042 (0.038)

0.001 (0.027)

− 0.025 (0.032)

2012–2014 × treated BRT/Line 1 × (Female − Male)

0.181 (0.184)

0.081** (0.039)

0.032 (0.033)

0.065* (0.032)

0.056 (0.047)

0.007 (0.046)

0.008 (0.044)

0.014 (0.025)

0.048 (0.034)

2015–2017 × treated BRT/Line 1 × (Female − Male)

0.093 (0.225)

0.077 (0.054)

− 0.005 (0.043)

0.034 (0.041)

0.038 (0.055)

0.008 (0.050)

0.000 (0.050)

0.018 (0.032)

0.055 (0.039)

District-year fixed effects

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Controls (linear and interacted with female)

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Observations

21,338

21,338

21,338

21,338

21,338

21,338

21,338

21,338

21,338

Number of districts

31

31

31

31

31

31

31

31

31

Standard errors in parentheses, clustered at the district level. ***p < 0.01; **p < 0.05; *p < 0.10. Job quality index adds up dummies for columns (3) to (8) (index can take values 0 to 6). The classification in columns (8) and (9) is based on occupation code, using the earning distribution for all occupation codes from 2005 to 2009

Table 6

Intermediate outcomes

Coefficient

Intermediate outcomes

Composition effects

Public transport expenditure > 0

IHS (monthly public transport expenditure)

Years of education

High school education level or more

(1)

(2)

(3)

(4)

2010–2011 × treated BRT/Line 1 × Female

0.019 (0.027)

0.195 (0.141)

− 0.000 (0.228)

− 0.007 (0.027)

2012–2014 × treated BRT/Line 1 × Female

0.045 (0.032)

0.349** (0.158)

− 0.076 (0.207)

− 0.015 (0.023)

2015–2017 × treated BRT/Line 1 × Female

0.082** (0.032)

0.505*** (0.173)

0.071 (0.296)

0.002 (0.037)

2010–2011 × treated BRT/Line 1 × Male

− 0.047 (0.035)

− 0.313* (0.161)

0.031 (0.259)

0.014 (0.027)

2012–2014 × treated BRT/Line 1 × Male

− 0.012 (0.032)

− 0.154 (0.152)

0.105 (0.280)

0.004 (0.025)

2015–2017 × treated BRT/Line 1 × Male

− 0.005 (0.033)

− 0.105 (0.165)

0.069 (0.278)

− 0.003 (0.027)

2010–2011 × treated BRT/Line 1 × (Female − Male)

0.066 (0.044)

0.508** (0.214)

− 0.032 (0.345)

− 0.021 (0.038)

2012–2014 × treated BRT/Line 1 × (Female − Male)

0.057 (0.045)

0.503** (0.219)

− 0.181 (0.348)

− 0.019 (0.034)

2015–2017 × treated BRT/Line 1 × (Female − Male)

0.087* (0.046)

0.610** (0.239)

0.002 (0.406)

0.005 (0.406)

District-year fixed effects

Yes

Yes

Yes

Yes

Controls (linear and interacted with female)

Yes

Yes

Yes

Yes

Observations

21,338

21,338

21,338

21,338

Number of districts

31

31

31

31

Standard errors in parentheses, clustered at the (district × female) level. ***p < 0.01; **p < 0.05; *p < 0.10

Table 7

Tests for spillover effects: main outcomes

Coefficient

Employment

Unconditional on employment

Job quality index

Public transport expenditure

Education

IHS (earnings)

IHS (earnings/h)

Index value

Index value > 0

Expenditure > 0

IHS (expenditure)

Years

High school or more

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

2010–2011 × (Original Control) × Female

− 0.020 (0.031)

− 0.117 (0.228)

− 0.000 (0.079)

− 0.004 (0.091)

− 0.011 (0.033)

0.058 (0.036)

0.152 (0.177)

− 0.032 (0.263)

0.026 (0.026)

2012–2014 × (Original Control) × Female

− 0.022 (0.031)

− 0.190 (0.251)

− 0.039 (0.089)

− 0.110 (0.109)

− 0.021 (0.030)

− 0.050 (0.034)

− 0.337* (0.175)

− 0.228 (0.191)

0.014 (0.018)

2015–2017 × (Original Control) × Female

0.024 (0.042)

0.113 (0.344)

0.019 (0.103)

− 0.013 (0.097)

− 0.008 (0.031)

− 0.006 (0.042)

− 0.131 (0.207)

− 0.471 (0.289)

− 0.032 (0.023)

2010–2011 × (Original Control) × Male

0.011 (0.019)

− 0.207 (0.192)

− 0.077 (0.125)

− 0.315* (0.160)

− 0.058 (0.043)

− 0.021 (0.026)

− 0.075 (0.118)

− 0.134 (0.417)

0.003 (0.046)

2012–2014 × (Original Control) × Male

− 0.006 (0.019)

− 0.036 (0.228)

− 0.003 (0.102)

− 0.222*** (0.077)

− 0.041 (0.039)

0.011 (0.025)

0.106 (0.142)

− 0.553** (0.236)

− 0.004 (0.021)

2015–2017 × (Original Control) × Male

0.008 (0.010)

− 0.450 (0.297)

− 0.192 (0.133)

− 0.402*** 0.116)

− 0.082 (0.051)

0.013 (0.055)

0.092 (0.253)

− 0.695** (0.287)

− 0.021 (0.032)

2010–2011 × (Original Control) × (Female − Male)

0.003 (0.039)

0.090 (0.298)

0.076 (0.147)

0.312* (0.184)

0.048 (0.054)

0.078* (0.044)

0.227 (0.212)

0.102 (0.492)

0.023 (0.053)

2012–2014 × (Original Control) × (Female − Male)

− 0.020 (0.041)

− 0.154 (0.339)

− 0.035 (0.135)

0.112 (0.133)

0.020 (0.049)

− 0.061 (0.042)

− 0.443* (0.225)

0.325 (0.303)

0.018 (0.028)

2015–2017 × (Original Control) × (Female − Male)

0.073 (0.054)

0.564 (0.454)

0.211 (0.168)

0.389** (0.151)

0.075 (0.059)

− 0.019 (0.069)

− 0.223 (0.326)

0.224 (0.406)

− 0.010 (0.039)

Conglomerate fixed effects

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Controls (linear and interacted with female)

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Observations

17,972

17,972

17,972

17,972

17,972

17,972

17,972

17,972

17,972

Number of conglomerates

30

30

30

30

30

30

30

30

30

Standard errors in parentheses, clustered at the (district × female) level. ***p < 0.01; **p < 0.05; *p < 0.10

Column (1) in Table 3 provides the results for an employment indicator equal to one for all the individuals who declare working and have positive earnings, and zero otherwise. Columns (2) to (4) split employment into three categories (employee, self-employed, domestic worker). Column (5) shows a specific group among the non-employed which is of particular interest in the case of women, homemakers. While for men, there are no statistically significant results, for women, there are large positive and statistically significant effects on employment and they are increasing over time in the order of 6 percentage points in the period 2010–2011, 8.5 percentage points in the periods 2012–2014, and 10 percentage points for 2015–2017. These effects imply increases of between 9.5 and 16% with respect to the pre-treatment employment rate among women living in the treatment area. The increase in employment appears to be driven by the increases in the domestic worker category (column 4) and by decreases in the homemaking category (column 5).

Table 4 explores the impacts of the BRT and Line 1 on earnings. Columns (1) to (3) show results unconditional on employment status while columns (4) to (6) show results conditional on employment. As the outcomes in columns (1) to (3) are zero for those not working, we use the IHS transformation discussed above. Its interpretation is equivalent to that of a logarithmic transformation, which means that the coefficients can be (roughly) interpreted as percentage changes.20 Columns (1) and (4) show total labor earnings, columns (2) and (5) show total hours worked, and columns (3) and (6) show earnings per hour, calculated as the ratio of the two preceding columns. The unconditional effects are clear, with large increases for women in the three post-intervention periods in total earnings, hours worked, and earnings per hour. As column (3) shows, there are increases in hourly earnings in the order of 17 to 27%. Conditional on employment, however, most effects go away or are marginally significant and, if any, there are some increases in hours worked but much smaller than those unconditional on employment (coefficients decrease to being as low as a quarter to as large as a half of the unconditional ones). In addition, in regressions not shown, we do not find any evidence that either the unconditional or conditional impacts on employment and hours are driven by significant changes in part-time employment. Overall, this suggests that most of the results are being driven by a reduction on zero earnings and hours. This is consistent with the increase in employment rates discussed above. For males, however, the coefficients associated to earnings and hourly earnings conditional on employment are negative in the order of 15 to 18%. This does not necessarily mean that earnings decrease for this group, but more precisely that they might be increasing at a slower rate than the earnings for males in the control group. A potential explanation for this finding will be discussed when presenting the results in the next table.

To understand what drives the earning results, we analyze in Table 5 different job quality measures. Columns (3) to (5) attempt characterizing job quality with indicator measures of job formality derived from the characteristics of the employing firm or self-employment activity. The outcome in column (3) classifies as a firm as formal if it keeps accounting books, in column (4) if the firm is registered,21 and in column (5) if the firm has more than five employees. In columns (6) and (7), formality is defined by whether the individual contributes to social security or if he or she has a formal contract. In columns (8) and (9), we characterize the type of occupation, based on its position in the earning distribution of the ENAHO sample. Finally, as explained above, to summarize all these measures, and to try to minimize the possibility of finding effects purely due to testing hypothesis on multiple outcome variables, we create an index of job quality, which is equal to the sum of columns (3) to (7), the alternative measures of formality, and column (8), the indicator for occupation in the top quartile of the hourly earning distribution. The index can assume values from 0 to 6. The results associated with the index are presented in column (1), while those associated with the indicator equal to one when the index is greater than zero (and zero otherwise) are presented in column (2). The results suggest that there are no significant changes in job formality for women closer to the BRT and Line 1, and, instead, that there are increases in their participation in occupations that are in the bottom of the income distribution (by between 7 and 10 percentage points). Thus, while employment and earnings per hour increase for women close to the BRT and Line 1, these changes are not driven by women finding high-quality jobs. For men, there are some negative effects on formality, which seem to be driven by a more rapid job inflow into smaller firms and this is also reflected in marginally higher participation in low-paid jobs and with no contributions to social security. For men, we also see an increase of 7 to 11 percentage points in the share working in occupations in the bottom 25% of the earnings, which may explain the negative impacts on earnings among men found in Table 3.

To better understand the effects estimated so far, Table 6 explores some intermediate outcomes of interest and tests for potential composition effects that could be driving our results. Column (1) presents the results associated with an indicator variable equal to one if spending in public transport by the individual is greater than zero, and zero otherwise. Column (2) looks at changes in the intensive margin by looking at the IHS of individual monthly transport expenditures. We observe important changes in transport expenditures for women that are closer to the BRT and Line 1, both in the intensive and extensive margins. Specifically, we see an increase between 35 and 51% in public transport expenditures among women in the treatment areas. The results suggest that the opening of the BRT and Line 1 strongly pulled women into using public transport (8 percentage points increase in positive expenditures, i.e., a 10% increase compared to the pre-treatment period, by 2015–2017). This aligns with the increases in employment found in Table 3. In contrast, effects on intermediate outcomes for men are mostly absent, except some negative changes in the first post-treatment period, 2010–2011.

Robustness Tests

Besides testing the validity of the parallel trend assumption, we conduct multiple robustness checks to support our empirical approach. The first one, to rule out the possibility that the effects could be driven by compositional changes in the characteristics of the individuals across areas, is shown in columns (3) and (4) in Table 5, to analyze changes in the years of education of the individuals and in the proportion of the individuals with a high school education level or higher. The regressions show no statistically significant results, suggesting that the observed impacts are not driven by compositional changes in the education characteristics of those men and women who live in the areas of influence of the BRT and Line 1.

Another important consideration is whether there are spillover effects. That is, treatment assignment should not affect outcomes of other non-treated individuals. In the context of spatial models, our results may be invalid if the effects we find are the consequence of changes in the control units due to the intervention. Redding and Turner (2015) recommend distinguishing between “growth” and “reorganization” effects to rule out finding treatment effects because there was a re-organization of economic activity from control to treated areas (or vice versa). Although the exclusion of the sample within the 1- to 2-km distance to the BRT/Metro system is an attempt to account for this, we further test whether control units reacted somehow differently than a sample located farther away. We compare the original control sample (i.e., 10,952 individuals within 2 to 5 km from the transport systems) with a sample of 7020 individuals living within 5- to 8-km distance to the BRT/Metro system and report the results for the main variables in Table 7. Results indicate that control areas did not react differently than areas located farther away.

Another concern would be that the supply of employment opportunities in the treatment area changed because of the intervention. In results not presented (available upon request), using data from the 2015 Economic Survey, we find that the distribution of firms’ age is not different between treated and control areas, suggesting that there are no changes in the distribution of economic activity, but that households are gaining improved access to existing employment centers.22

Finally, to test for the stability of our results, Table A1 in the online appendix presents results for our main outcomes without imposing overlap in conglomerate-level characteristics and confirms our findings.

Heterogeneity Analyses

We conduct three heterogeneity analyses and report them in the online appendix. The first analysis, reported in Table A2, evaluates the differential impacts of proximity to the BRT versus proximity to Line 1; we classify households considering their closest distance to any of the lines.23 The results show that the impacts come from both lines. For the second analysis, we explore whether there are heterogeneous impacts across different distances to the new transport systems, comparing the impacts associated with individuals within a narrower 0–0.5-km buffer versus those associated with individuals within the 0.5 to 1-km buffer. Results are reported in Table A3 and show that the impacts are in general very similar in the two buffers. The third analysis considers proximity to local areas where most jobs are located. A priori, we hypothesized that those that are farther away, whom probably have less transportation options, might be the ones gaining more from improved transportation. To test this, we use the individual-level trips reported in the 2004 OD survey to identify a core area in Lima where most of the job trips take place24 and distinguish between areas that are within 30-min travel time to this core and areas that are above this travel time. Table A4 shows the results, which are very similar between those living under 30-min distance and those living in an area above 30-min distance. Other heterogeneity analyses (available upon request) yield similar results. We find no differential effects for “feeder” lines,25 or for households with a high dependency ratio.26

The Role of Speed and Safety in Transport Decisions

Although we cannot clearly disentangle the effects that are attributed to improvements in speed versus those coming from safety gains using the ENAHO data, we conduct additional descriptive analyses to better understand the role that these two dimensions have on transport users. Ideally, an OD survey before and after the intervention would help us understand this more accurately; however, the two OD surveys available for Lima were collected before the opening of the two lines. We produce descriptive statistics using yearly data collected by the Lima Como Vamos (LCM) surveys from 2011 to 2017, which focus on the population’s perception of the quality of living conditions in Metropolitan Lima.

We expect that the opening of improved transport systems, such as the BRT and Line 1, provided a substantial change in travel time, large enough to push employment supply. Figure 3 shows that both men and women users of BRT and Metro report speed as the main reason for using the lines. In addition, Fig. A2 in the online appendix shows that both men and women users of BRT and Metro are more likely to indicate that in their perception, they take less time to get to their main destination, although there are no statistically significant differences between men and women.27
Fig. 3

Main reason for using BRT/Metro, by gender and year Source: Lima como Vamos survey (http://www.limacomovamos.org)

To provide evidence on the importance of safety for women when deciding to use the BRT/Metro, we complement our analysis with data from Galiani and Jaiman (2016), a study of the perception of insecurity in Lima and Asuncion by women that are not users of private vehicles. Since this study was conducted in 2015, well into the post-intervention period, we can only provide suggestive evidence on the differences in perception among users of different public transport systems.

Galiani and Jaiman (2016)’s survey main advantage is that it offers detailed questions on perception on safety in public transport. However, its sample size is quite small: it surveyed only 619 women in Lima. These women were classified according to their main mean of transportation, which we re-categorize in three groups (excluding taxi users): pedestrians (141), users of BRT or metro (18), and user of other means of transportation (422). The resulting sample size is 578.

Figure 4 compares means (controlling for year of birth and education level through regression) across these three groups for four variables labeled as (i) considers public transport is safe, (ii) fears traveling within the city, (iii) feels safe traveling at night, and (iv) has been verbally harassed by a man in public transport. The design of the survey excluded pedestrians from the last question.
Fig. 4

Safety perception among women using public transport in Lima in 2015 (marginal effect). Data source: Galiani and Jaitman (2016). Panels a, b, and c show coefficients from the estimation of the linear regressions yi = a + β1(BRT/Metro)i + β2(Other public transport)i + β3Xi + εi, where β1 and β2 are the coefficients for dummy variables on usage of BRT/Metro and of other means of public transportation. The dependent variables are those specified in the title of each panel. Regressions include year of birth and education level as controls Xi. The excluded category is pedestrians. The regression in Panel d includes only users of public transport. The coefficient is the difference between BRT/Metro users and other (non BRT/Metro) users of public transport

Although this is not causal evidence, this simple exercise suggests that women who use the BRT and metro are more satisfied with safety in public transport and experience less harassment. Panel (a) indicates that compared to pedestrians, women users of non-BRT/Metro transport are not more likely to consider public transport safe, while the group of BRT/Metro users are, on average, 25.7 percentage points more likely to consider public transport to be safe. The difference with non-BRT/Metro users is statistically significant. Panels (b) and (c) suggest there are no differences across groups in the perception of fear while traveling in the city or while traveling at night, respectively. Finally, panel (d), within the sample of public transport users, shows that BRT/Metro users are less likely to report male harassment by 33.4 percentage points. Together, these results are an indication of perceived and actual safety benefits experienced by women users of the BRT/Metro system.

Conclusion

This paper investigates whether access to faster and safer modes of public transport impacts women’s labor market outcomes. We conduct the analysis in the metropolitan area of Lima, Perú, where access to this type of transportation remains a challenge. The identification strategy, based on a DID estimation, allows us to infer causal estimates of the implementation of two interventions, a BRT system and an elevated rail system, better known as Line 1 of the metro. Both systems have considerably reduced travel times and increased connectivity between peripheral areas of the city to major employment centers. In addition, the systems are equipped with infrastructure and technology that represent a substantial improvement relative to the safety of the rest of the public transit in the city. We hypothesize that these changes may have disproportionally encouraged women to use these systems and improved their accessibility to jobs when compared to men. We find large gains in employment for women. Proximity to the system results in 7.6 percentage points increase in the probability of women being employed right after the BRT system was implemented (2012–2014) and 10 percentage points increase by the time both the BRT and Metro Line 1 are fully functional (2015–2017). These effects imply increases of between 8.3% and 16.6% with respect to the employment rate of treated women in the pre-treatment period. There are no effects for men. Employment gains are also reflected in larger earnings for women, between 18% and 27% with respect to treated women in the pre-treatment period. However, most of these gains seem to arise from changes on the extensive margin, with more women working. We do not find any significant effects on earnings for already employed women.

The increased employment for women does not appear to be of higher quality than that for comparison groups. We find no significant effect on different measures of formal employment and it is more likely that women are employed in occupations at the bottom 25% of the hourly earning distribution. The magnitude of this effect ranges between 6 and 11 percentage points, which implies increases of between 10% and 18% with respect to the employment rate for treated women in the pre-treatment period. The lack of impacts on job quality is probably explained by the high degree of informality in the labor market in Lima, where over 50% of men and over 70% of women work in informal jobs. Thus, women who decide to look for jobs are more likely to find them in the informal sector. In addition, if we add to this that poor populations are also more dependent on slower modes of transport, such as public transit (CAF 2011), it is less surprising that job quality did not necessarily increase despite the increase observed in overall employment.

To understand these results, we explore several mechanisms and find evidence of an increase in the use of public transport. Treated women are 8 percentage points more likely to report expenditures on public transport, which is reflected in 51% higher expenditures with respect to untreated women. We find no evidence of demographic composition impacts in the treatment areas, particularly when looking at changes in the levels of education of those living closer to the BRT and Line 1 when compared to those living in control areas. We also test for spillover effects by looking at whether control units reacted differently than a sample located farther away. Results provided help to rule out that the effects we observe are the consequence of changes in the control units due to the intervention.

Heterogeneity analyses show that results are indistinguishable between the BRT and Line 1. Proximity to either one of them renders similar results. Exploring variations in impacts across the proximity to the trunk, we do not find much heterogeneity in results either, when we compare the areas that are in close proximity to the lines (i.e., 0–0.5 km) and those farther to the lines (0.5–1 km). Other heterogeneity analyses we perform show that results are broadly homogeneous.

Overall, we believe these are novel and robust results. Tests of the parallel trends’ assumption seem to justify the use of the DID method. However, one needs to be careful in interpreting the results, given that defining “treatment” and “control” groups is inherently difficult for these types of urban transport interventions. The definition of treatment and control areas is somewhat arbitrary, and individuals can move in and out of those areas over time (however, we do not find evidence of neighborhood compositional changes or economic activity reorganization). Even if the employment and (unconditional on employment) earnings and hourly earning impacts are as large as our regression results suggest, we only have suggestive public transport expenditure evidence on the role the new lines had in promoting better labor outcomes for women. We would need to have individual-level transport use information (i.e., an OD survey) to unequivocally show a direct link between the usage of the two new public transport lines and labor outcomes. Unfortunately, there is no such source of information after the implementation of the two lines for Lima.

Whether this is a story about the role of increased speed or the role of improved safety in encouraging more women to use these transport systems, data limitations do not allow us to fully disentangle these effects. Despite this, we present descriptive evidence from local perception surveys that shows that speed is one of the main factors influencing BRT and metro users to choose these systems. Also, we observe evidence that women who use BRT and metro are more satisfied with safety in public transport and experience less harassment.

Keeping in mind these caveats, our findings provide strong suggestive evidence that infrastructure investments that make it faster and safer for women to use public transport can generate large labor market impacts for those women who reside in the areas of influence of the improved infrastructure. The extent to which women enter the labor market and the quality of the jobs they hold once their accessibility opportunities increase is still an area that needs to be further examined. Increasing their labor market participation and job quality may require additional structural interventions that go well beyond the reach of the transport sector. Regardless of this, the power of transport investments in facilitating access to opportunities and encouraging changes in time allocation decisions for women appears to be quite remarkable.

Footnotes

  1. 1.

    According to the Atlantic Council (2014), Latin American countries have undergone unprecedented urbanization in the past 60 years. From 1950 to 2014, the share of the population in Latin America living in urban areas increased from 40% to around 80% and it is expected to increase to 90% by 2050.

  2. 2.

    To the best of our knowledge, the only other studies we identified looking at a related topic are Asahi (2016), who finds evidence of a positive effect of proximity to a new subway line in Santiago (Chile) on employment outcomes for women, and  Abu-Qarn and Lichtman-Sadot (2019), who find evidence of a trade-off between investment in education and time allocated to work by women after the introduction of bus services in Arab towns in Israel.

  3. 3.

    Sexual harassment issues experienced by women in transit include staring, unwanted comments on physical appearance, men touching or rubbing against women, and groping (Gomez, 2000). While in developed countries sexual harassment in public transport has been reported to be more verbal than physical, subtle groping and unwanted touching are common in rush hours (Hsu 2011; Gekoski et al. 2015). In developing countries, this pattern is more pronounced (Zermeno et al. 2009).

  4. 4.

    This strategy has been implemented in cities such as Mexico City, Rio de Janeiro, Tehran, and Tokyo.

  5. 5.

    There are no reliable figures on the share of trips of the system using the BRT or Line 1. An opinion survey on living conditions in the metropolitan area suggest that the two lines are used daily by around 10% of the population of Lima and 6.2% of the population of Callao (Lima Como Vamos 2017, p. 43).

  6. 6.

    To assign jobs to the top or bottom 25% of the earning distribution, we take the 341 occupations that appear in the ENAHO, rank them in the period 2005 to 2009 based on hourly earnings, and identify them as appearing in the bottom or top quartile of the earning distribution. We then classify all occupations in the period 2010–2017 based on the pre-intervention classification.

  7. 7.

    A conglomerate is a geographic area with approximately 140 private dwellings, defined by INEI to be the primary sampling unit in its surveys. It could comprise one or several city blocks, depending on the density of the area.

  8. 8.

    We exclude the feeders from our analysis as they run in non-segregated roadways, do not have dedicated stations, and do not provide information on headways, while both BRT and Line 1 stations do. This can have important implications on the safety of women traveling in these systems. Moreover, Line 1 does not have an established system of feeders, only the BRT does. As we are pooling both systems together, we focus only on the areas of influence around the BRT trunk line and around Line 1.

  9. 9.

    The survey indicates that 90% of passengers walk no more than 12 min (91% walk no more than 15 min) and 99% walk 20 min or less to reach public transportation.

  10. 10.

    Based on our empirical strategy, 32,229 observations in the ENAHO fall within the treatment and control areas between 2007 and 2017. We eliminate 615 observations due to missing information in key variables of interest, and 4947 observations due to improper geocoding. Our final estimation sample is 26,668 observations.

  11. 11.

    The covariates included in Xit are as follows: age (and its square), an indicator variable for married or cohabitating status, a dummy variable for indigenous language as mother tongue by the individual, a dummy for whether the individual is currently enrolled in school, a dummy for single-parent household, an indicator for female-headed household, number of household members, number of children under the age of 6 in the household, and the household dependency rate, defined as 1 minus the ratio of income earners over total members of the household. A possible concern of including time-varying covariates is that they could be endogenous to the treatment. Thus, we run regressions without covariates, with results that are in line with the ones we obtain. The results from these regressions are available upon request.

  12. 12.

    In all regressions, we cluster the standard errors at the district level × female level to allow for arbitrary correlation within districts by gender.

  13. 13.

    The propensity score is estimated by a logit regression with the following conglomerate-level covariates: percentage of households who use gas as cooking fuel; are connected to a public source of electricity; have a toilet inside the premises; have a water connection; have mud, wood, or other low-quality material walls; have dirt of bare concrete floor; live in an apartment; live in rental housing; the average number of rooms in the premises, members of the household, years of education of the working age population (18–64), years of education of household head, age of household head; percentage of indigenous population, female-headed households; first principal component of household assets and services; establishments per inhabitants (in logs); value added per employed individuals (in logs); conglomerate strata according to the poverty map; average income per capita according to the poverty map; road density; and share of high-skilled activities and non-tradable activities.

  14. 14.

    We drop those conglomerates for which the propensity score is lower than an optimal cutoff value q or higher than (1-q). We obtain the value of q following Crump et al. (2009), 0.104—close to the rule of thumb suggested by Crump et al. (2009), 0.10. This implies that our estimation sample decreases by 5330 observations to 21,338 observations.

  15. 15.

    The IHS transformation of yi is equal to \( \log \left({y}_i+{\left({y}_i^2+1\right)}^{1/2}\right) \) (see Burbidge, Magee, and Robb (1988) for details).

  16. 16.

    This index groups six dummy indicators associated with formal employment and type of occupation: firm keeps accounting books, firm is registered, firm has more than five employees, employee contributes to social security, employee has a contract, and occupation is in the top 25% occupation rank according to average earnings per hour.

  17. 17.

    Although infrastructure investments may bring some anticipated effects, our data do not lend well to study anticipated shocks. The ENAHO started providing geo-localization of households only in 2007, and this is the key variable that allows us to differentiate between households in treated and control areas. Moreover, the announcement of both infrastructure projects is much earlier. Line 1 was originally designed in the early 1980s. Its construction started in 1986 but was stalled due to the start of an economic and political crisis that lasted until the early 1990s. Construction was only resumed in 2010. The BRT was planned since 1996 and its construction started in 2006. Thus, our data unfortunately cannot capture anticipation effects. If these effects existed, we would probably be potentially underestimating the impacts of these projects if anticipation impacts are positive.

  18. 18.

    The full results for all regressions are available upon request.

  19. 19.

    Results before overlap are consistent with those after imposing overlap. For the sake of space, the full set of results before imposing overlap is available upon request.

  20. 20.

    When regression models have log transformed outcomes, the impact of a one-unit change in a covariate (X) is calculated by exponentiating the coefficient. In this case, the interpretation of impacts should be done as exp. (\( \hat{\beta} \)) − 1. For example, for a coefficient of 0.23, the effect is calculated as exp. (0.23) − 1 = 0.26. When the estimated coefficient is less than 0.10, the simple interpretation that a unit increase in X is associated with an average of 100*\( \hat{\beta} \) percent increase in Y works well. When the coefficient is above 0.10, the simple interpretation will underestimate effects. For simplicity, we report percentage changes using the simple interpretation throughout the text.

  21. 21.

    This includes small firms registered in special categories with small tax burden.

  22. 22.

    We are only able to conduct this analysis with the 2015 data only, as the National Economic Surveys do not repot the geolocation for firms for prior years. The hypothesis of equality of distributions cannot be rejected by the Wilcoxon rank-sum test (z = 1.36).

  23. 23.

    These results should be taken with caution, given that the two systems run parallel and very close to each other and there could be some ambiguity in treatment assignment (i.e., it is not clear which of the two systems would a person relatively close to both take for their work trips).

  24. 24.

    The 2004 Lima and Callao Master Transport Plan provides an individual-level trip diary survey for a sample of 35,000 households. This means that for each person in the sample, we know the purpose of each trip (e.g., work, leisure, shopping) travel time and mode of transport taken in the 24 h prior to the survey. The Master Transport Plan defined 427 traffic analysis zones (TAZ) around the city to be used as origin or destination for all trips in the survey. Thus, we identify the cluster of TAZ that receives the largest number of work-related trips using Anselin’s local indicators of spatial association and designated that as the “job city center.” We then calculate the average travel time to this cluster from every other TAZ using the reported travel times from the survey.

  25. 25.

    It is possible that feeders to the BRT act as a faster (when combined with the trunk) but not safer mode of transport (when compared to the trunk). We test this hypothesis by selecting individuals close to the BRT feeders and add them to our estimations as an additional treatment group. None of the “feeder” interactions are statistically significant. Although this might suggest that safety is playing a role in the results obtained, we cannot unequivocally conclude that gains in speed are not as important. The reason is that gains in speed for populations in the feeder might come only from those combining a feeder and a trunk trip, and we are not able to distinguish this in the ENAHO data.

  26. 26.

    We explore whether women living in a household with a high dependency ratio (above the 75th percentile of an index defined as the proportion of non-income earners over total members of the household) react differently after the intervention and find no difference in results.

  27. 27.

    Among the dimensions monitored by the LCM survey, there is a categorical question on the time required to travel to a main destination compared to the previous year. In the answer, individuals choose between three categories: (i) it takes longer, (ii) it is the same, and (iii) it takes less time. We re-classify this variable to take the value of one for those who responded it takes less time (iii) and zero otherwise. We compare mean differences for this new variable between users and non-users of BRT/Metro for men and women, over time, in Figure A2 in the online appendix.

Notes

Acknowledgments

The authors would like to thank four anonymous reviewers for the useful comments on earlier drafts of the paper and Rafael Capristan for his support with the meetings and interviews with relevant government authorities and in procuring access to key data for this research.

Funding Information

The Inter-American Development Bank (IDB) provided funding for this project through the ESW RG-E1502.

Compliance with Ethical Standards

Conflict of Interest

The authors declare that there is no conflict of interest.

Disclaimer

The opinions expressed in this paper are those of the authors and do not necessarily reflect the views of the IDB, IDB Invest, their respective Boards of Directors, or the countries they represent.

Supplementary material

41996_2019_39_MOESM1_ESM.docx (368 kb)
ESM 1 (DOCX 367 kb)

References

  1. Abu-Qarn A, Lichtman-Sadot S. The trade-off between work and investment in higher education: evidence from the introduction of public transportation to Arab communities in Israel, Mimeo; 2019.Google Scholar
  2. Anand A, Tiwari G. A Gendered perspective of the shelter–transport–livelihood link: the case of poor women in Delhi. Transp Rev. 2006;26(1):63–80.CrossRefGoogle Scholar
  3. Asahi, Kenzo, Better Urban Transport Improves Labour Market Outcomes: Evidence from a Subway Expansion in Chile (May 20, 2016). Available at: https://www.ssrn.com/abstract=2781807.
  4. Astrop A, Palmer C. The urban travel behaviour and constraints of low-income households and females in Pune, India. National Conference on Women’s Travel Issues; 1996. pp. 23–26.Google Scholar
  5. Atlantic Council. Urbanization in Latin America. Washington, DC: Adrienne Arsht Latin America Center, 2014.Google Scholar
  6. Autoridad Autónoma del Sistema Eléctrico de Transporte Masivo de Lima y Callao. Metro de Lima (AATE), Cada vez mas cerca, Revista Institucional de la AATE; 2013. https://www.aate.gob.pe/wp-content/uploads/2015/04/Revista-Metro-de-Lima.pdf. Accessed 10 Oct 2018.
  7. Babinard, Julie. World Bank gender transport surveys: an overview: an overview, transport notes, TRN-43, September 2011.Google Scholar
  8. Bielich SC. The war of the penny: Aa current look at public transport in Metropolitan Lima, Peru, Institute for Peruvian Studies, Working paper no. 155, Economics series 49; 2009.Google Scholar
  9. Bocarejo JP, Oviedo DR. Transport accessibility and social inequities: a tool for identification of mobility needs and evaluation of transport investments. J Transp Geogr. 2012;24:142–54.CrossRefGoogle Scholar
  10. Bocarejo JP, Portilla I, Meléndez D. Social fragmentation as a consequence of implementing a Bus Rapid Transit system in the city of Bogotá. Urban Stud. 2016;53(8):1617–34.CrossRefGoogle Scholar
  11. CAF. Desarrollo urbano y movilidad en América Latina. Panama. 2009.Google Scholar
  12. CAF. Desarrollo urbano y movilidad en América Latina. Panama; 2011. Retrieved from https://www.caf.com/media/4203/desarrollourbano_y_movilidad_americalatina.pdf.
  13. Carruthers R, Dick M, Saurkar A. Affordability of public transport in developing countries. Transport Papers series; no. TP-3. Washington, DC: World Bank; 2005.Google Scholar
  14. Cervero R, Landis J. Suburbanization of jobs and the journey to work: a submarket analysis of commuting in the San Francisco Bay Area. Journal of Advanced Transportation. 1992. 26 (3): 275–297.CrossRefGoogle Scholar
  15. Crump RK, Hotz VJ, Imbens GW, Oscar A. Mitnik; Dealing with limited overlap in estimation of average treatment effects. Biometrika. 2009;96(1):187–99.CrossRefGoogle Scholar
  16. Curtis C, Perkings J. Travel behavior: aA review of recent literature, impacts of transit led development in a new rail corridor. Perth: Department of Urban and Regional Planning Curtin University; 2006.Google Scholar
  17. Darido G, Pulido D, Targa F, Alvim B, Peralta-Quirós T. Lima urban transport: on the way to transformation, Connections, Transport and ICT, Note 22; 2015. http://pubdocs.worldbank.org/en/425451444329686088/1604860-TransportICT-Newsletter-Note-22.pdf. Accessed 11 Oct 2018.
  18. Delmelle E, Casas I. Evaluating the spatial equity of bus rapid transit-based accessibility patterns in a developing country: tThe case of Cali, Colombia. Transp Policy. 2012;20:36–46.CrossRefGoogle Scholar
  19. Diario Correo, Metro de Lima transports 355 thousand users per day; 2018. https://diariocorreo.pe/edicion/lima/metro-de-lima-transporta-355-mil-usuarios-por-dia-820935/. Accessed 11 Oct 2018.
  20. Duchéne C. Gender and Transport. International Transport Forum on Transport Society. Germany: Leipzig; 2011. p. 7–20.Google Scholar
  21. Farré, Lídia, Jordi Joffrey- Monsey, Juan Torrecillas. Commuting Time and Female Labor Force Participation, unpublished manuscript, University of Barcelona, June 2018.Google Scholar
  22. Ferrant G, Pesando LM, Nowacka K. Unpaid care work: The missing link in the analysis of gender gaps in labour outcomes. Paris: OECD Development Centre; 2014.Google Scholar
  23. Galiani S, Jaitman L. Public Transport from a gender perspective: perceptions of insecurity and victimization in Asuncion and Lima, Inter-American Development Bank Technical Note No. IDB-TN-1124; 2016.Google Scholar
  24. García G, Picasso G. Informe de Rendición de Cuentas de la Presidencia del Instituto Metropolitano ProTransporte de Lima 2011-2014, unpublished document, Protransporte; 2014.Google Scholar
  25. Gardner N, Cui J, Coiacetto E. Harassment on public transport and its impacts on women'’s travel behavior. Australian Planner. 2017;54(1):1–8.CrossRefGoogle Scholar
  26. Gekoski A, Gray JM, Adler JR, Hovarth MAH. The prevalence and nature of sexual harassment and assault against women and girls on public transport: an international review. J Criminol Res Policy Pract. 2017;3(1):3–16.CrossRefGoogle Scholar
  27. Gomez, L. Gender Analysis of Two Components of the Workd Bank Transport Projects in Lima, Peru: Bikepaths and Busways. World Bank, 2000.Google Scholar
  28. Guthrie A, Fan Y. Developers'’ perspectives on transit-oriented development. Transp Policy. 2016;51:103–14.CrossRefGoogle Scholar
  29. Guzman L, Oviedo D, Rivera C. Assessing equity in transport accessibility to work and study: tThe Bogotá region. J Transp Geogr. 2017;58:236–46.CrossRefGoogle Scholar
  30. Gwilliam K. Urban transport in developing countries. Transp Rev. 2003;23(2):197–216.CrossRefGoogle Scholar
  31. Hansen BE. Threshold effects in non-dynamic panels: eEstimation, testing, and inference. J Econ, Elsevier. 1999;93(2):345–368CrossRefGoogle Scholar
  32. Holzer H, Quigley JM, Raphael S. Public transit and the spatial distribution of minority employment: eEvidence from a natural experiment. J Policy Anal Manag. 2003;22(3):415–41.CrossRefGoogle Scholar
  33. Hsu, H.-P. How Does Fear of Sexual Harassment on Transit Affect Womens Use of Transit. Womens Issues in Transportation: Summary of the 4th International Conference. Transportation Research Board Conference Proceedings, 2 (46), pp. 85-94. Washingto D.C. 2011.Google Scholar
  34. Japan International Cooperation Agency, 2015, The Master Plan for Lima and Callao Metropolitan Area Urban Transportation in the Republic of Peru.Google Scholar
  35. ILO. World Employment and Social Outlook: Trends for women 2017 International Labour Office – Geneva: ILO, 2017.Google Scholar
  36. Ishibashi E. Constructing sexual risk: ‘Chikan’, collapsing male authority and the emergence of women-only train carriages in Japan, BKC.ILO. In: World Employment and Social Outlook: Trends for women 2017. Geneva: International Labor; 2003.Google Scholar
  37. Kash, G. Gendered perspectives on transit crime in Arequipa, Perú and Bogotá, Colombia. Transp Res Rec 2014:1–16.Google Scholar
  38. Kunieda M, Gauthier A. Gender and urban transport: smart and affordable, Module 7a, Sustainable transport: a sourcebook for policy-makers in developing cities, GTZ, Sector Project Transport Advisory Service; 2007.Google Scholar
  39. Levine RV, Norenzayan A. The pace of life in 31 countries. J Cross-Cult Psychol. 1999;30(2):178–205.CrossRefGoogle Scholar
  40. Lima Como Vamos. VII Informe de Percepción Sobre Calida de Vida en Lima y Callao; 2017. https://bit.ly/2GPj6UM. Accessed 12 Oct 2018.
  41. Loukaitou-Sideris A, Fink C. Addressing women’s fear of victimization in transportation settings: a survey of U.S. transit agencies. Urban Aff Rev. 2008;44(4):554–87 2017.CrossRefGoogle Scholar
  42. Magee, L. & Robb, A. L. & Burbidge, J. B., On the use of sampling weights when estimating regression models with survey data, Journal of Econometrics, Elsevier, vol. 84(2), pages 251–271, June 1998.Google Scholar
  43. Neupane G, Chesney-Lind M. Violence against women on public transport in Nepal: sexual harassment and the spatial expression of male privilege. Int J Comp Appl Crim Just. 2014;38(1):23–38.CrossRefGoogle Scholar
  44. Oi W. Residential location and labor supply. J Polit Econ. 1976;84(4, Part 2):S221–38.CrossRefGoogle Scholar
  45. Oviedo-Dávila NM. Does proximity to massive transport systems reduce the probability of being informally employed? Evidence from Bogota. The London School of Economics and Political Science (LSE); 2017.Google Scholar
  46. Patacchini E, Zenou Y. Spatial mismatch, transport mode and search decisions in England. J Urban Econ. 2005;58(1):62–90.CrossRefGoogle Scholar
  47. Peters D. Gender issues in transportation: a short introduction. In: UNEP Regional Workshop “Deals on wheels: sustainable transportation initiatives in developing countries”. San Salvador: The Institute for Transportation and Development Policy; 1999. p. 1–4.Google Scholar
  48. Peters D. Gender issues in transport-applying an integrative perspective. Berlin: Center for Metropolitan Studies, Technical University Berlin; 2006.Google Scholar
  49. Peters, D. Gender Sustainable Urban Mobility, Thematic study prepared for Global Report on Human Settlements, Available from http://www.unhabitat.org/grhs/2013, United Nations, 2013.Google Scholar
  50. Protransporte. Datos Abiertos – Estadísticas; 2019. http://www.protransporte.gob.pe/datos-abiertos/. Accessed 8 Jul 2019.Google Scholar
  51. Redding SJ, Turner MA. Transportation costs and the spatial organization of economic activity, Handbook of regional and urban economics. Amsterdam: Elsevier; 2015.Google Scholar
  52. Salon D, Gulyani S. Mobility. poverty. and gender: travel ‘choices’ of slum residents in Nairobi. Kenya. Transp Rev. 2010;30(5):641–57.CrossRefGoogle Scholar
  53. Sanchez T. The connection between public transit and employment, the cases of Portland and Atlanta. J Am Plan Assoc. 1999;65(3):284–96.CrossRefGoogle Scholar
  54. Schintler L, Root A, Button K. Women’s travel patterns and the environment. Transp Res Board. 2000;1726:33–41.CrossRefGoogle Scholar
  55. Scholl L, Guerrero A, Quintanilla O, Cels L’Heloste M. Comparative case studies of three IDB-supported urban transport projects. Washington, D.C.: Inter-American Development Bank; 2015. https://publications.iadb.org/handle/11319/6967#sthash.6naWTDtl.dpuf.CrossRefGoogle Scholar
  56. Scholl L, Martinez D, Mitnik OA, Oviedo D, Yañez-Pagans P. A rapid road to employment? The impacts of a bus rapid transit system in Lima. Inter-American Development Bank Working Paper Series No. 00980; 2018.Google Scholar
  57. Schulz D, Gilbert S. Women and transit security: a new look at an old issue. Women’s Travel Issues: Proceedings from the Second National Conferences. Washington D.C.: FHWA U.S. Department of Transportation; 1996. p. 550–62.Google Scholar
  58. Sermons M, Koppleman F. Representing the differences between female and male commute behavior in residential location choice models. J Transp Geogr. 2001;9:101–0.CrossRefGoogle Scholar
  59. Srinivasan S, Rogers P. Travel behavior of low-income residents: sStudying two contrasting locations in the city of Chennai, India. J Transp Geogr. 2005;13:265–74.CrossRefGoogle Scholar
  60. Taylor B, Mauch M. Gender, race, and travel behavior: aAn analysis of household-serving travel and commuting in the San Francisco Bay Area. Women’s Travel Issues Second National Conference, Baltimore; 2000. pp. 373–405.Google Scholar
  61. Tsivanidis N. The aggregate and distributional effects of urban transit infrastructure: evidence from Bogota’s TransMilenio; 2019. Mimeo.Google Scholar
  62. Tudela Rivadeneyra A, Lopez Dodero A, Raj Mehndiratta S, Bianchi Alves B, Deakin E. Reducing gender-based violence in public transportation. Transp Res Rec. 2015;2531:187–94.CrossRefGoogle Scholar
  63. Venter C, Jennings G, Hidalgo D, Pineda AFV. The equity impacts of bus rapid transit: aA review of the evidence and implications for sustainable transport. Int J Sustain Transp. 2018;12(2):140–52.CrossRefGoogle Scholar
  64. Verick S. Female labor force participation in developing countries. IZA World of Labor. 2014:87.Google Scholar
  65. Wachs M. Chapter 6: The automobile and gender: an historical perspective. Second National Conference. U.S. Department of Transportation, Federal Highway Administration Office; 1996.Google Scholar
  66. White MJ. A model of residential location choice and commuting by men and women workers. J Reg Sci. 1977;17(1):41–52.CrossRefGoogle Scholar
  67. Yañez-Pagans P, Martinez D, Mitnik O, Scholl L, Vázquez A. Urban transport systems in Latin America and the Caribbean: challenges and lessons learned. Inter-American Development Bank Technical Note No. IDB-TN-01518; 2018.Google Scholar
  68. Zermeno M, Pacido E, Soto E, Yadin M. La violencia sexual hacia las mujeres en el Sistema de Transporte público de la Ciudad de Mexico. Ciudad de Mexico: Ciudad de Mexico, Instituto de las Mujeres del Distrito Federal, CIADEM, GEO PROSPECTIVA, Secretaria de Desarrollo Social; 2009.Google Scholar

Copyright information

© Springer Nature Switzerland AG 2019

Authors and Affiliations

  1. 1.Inter-American Development BankWashingtonUSA
  2. 2.IDB InvestWashingtonUSA

Personalised recommendations