Abstract
In many cities and towns throughout the United States, citizens with lower levels of education and skill confront challenges when seeking employment. The jobs best suited for their skills may be located in a different part of the metro area from their homes, and the existing public transportation system may not provide them with a practical way of interviewing for these jobs or commuting to them on a long-term basis. The rapid rise of ride sharing services, like Uber and Lyft, may provide a new opportunity to address these longstanding needs in a more cost-effective way. The public and private sectors may be able to share the costs of making the excess capacity in these ridesharing systems available to disadvantaged citizens whose needs are not being served by the existing public transportation system. Such a creative partnership could connect our least advantaged citizens to new opportunities while also connecting our employers and businesses to human resources in an increasingly tight labor market. To examine the effectiveness of this new technology in expanding the mobility of citizens with special transportation needs, we plan to combine a large-scale field experiment with sophisticated data analysis to evaluate the impact of ridesharing on individual mobility, employment outcomes, and access to training and social services.
Description
1. Introduction
The labor force participation rate in the United States among the poorly educated is very low. For adults aged 25 and over, it is less than 60% among those with a high school degree, and less than 50% among those without a high school education. One contributing factor to low participation in the labor market is relatively high transportation costs. At least since the classic work of Oi (1976), economists have understood that labor force participation can be sensitive to quasi-fixed costs, such as transportation costs, and there is a small but important empirical literature reinforcing this idea. However, this is a literature with many open questions. As we discuss in our literature review below, researchers seeking to estimate a causal impact of transportation costs on labor supply face a daunting challenge. The ideal empirical design would be a quasi-experiment in which transportation costs decline for some (treatment) individuals, while remaining fixed for other (control) individuals, and while holding all other factors fixed. As a counter-example, consider an investment in a city’s public transportation system. It is likely that the investment will reduce transportation costs more in some neighborhoods than in other neighborhoods, but because of equilibrium effects, many of which are anticipated, it is exceedingly difficult to tease out the causal impact of the change in transportation costs on labor supply. For instance, property values of homes near new convenient transportation hubs will likely increase, perhaps even before the completion of the new transportation project, as individuals with a high propensity to work relocate to those locations.
As is emphasized in the “spatial mismatch” literature, transportation costs are likely to be particularly burdensome for lower-paid poorly-educated workers, because these individuals often cannot afford housing near job opportunities. Prior research suggests this burden is likely to be especially intense for lower-income mothers with children at home, for whom the opportunity cost of time in transit may be especially high, given their parental responsibilities. For these individuals in particular, then, it seems likely that an exogenous decline in transportation costs might increase labor force participation.
Against this backdrop, we are proposing here a field experiment of moderate duration (6 months), in which we reduce the transportation cost for a treatment group, and compare labor force outcomes to a control group. In our proposed experiment, the reduction in transportation cost will be implemented via an innovative treatment that has high potential policy relevance—the provision of services from a ride-hailing service, Uber. We are excited by this particular intervention because of the possibility that future directions in public urban transportation may include technology-enabled individualized transportation services as an integral part of a broader transportation system—possibly a system that integrates smaller shared cars or vans (even small self-driving vehicles) into a public transportation system that will also continue to rely on traditional rail and bus lines.
Thus, we view our experiment as accomplishing two goals:
(1) Our research will be the first randomized controlled experiment designed to study the impact of transportation costs on the labor supply of a generally lower-paid population. As detailed below, we are focusing specifically on mothers with children at home, who do not have regular access to a car. We have reason to believe that this group may be particularly sensitive to a reduced cost of transportation.
(2) Our experiment can also be thought of as an innovative “pilot program” designed to examine how one new technology in transportation can be leveraged to improve the labor market outcomes in low-income families. In particular, if our work suggests that the flexibility afforded by ride-hailing services improves labor force participation, this result could be provide an impetus for future research on the role of transportation innovation in improving labor market outcomes. More broadly we hope that the resulting research can be a valuable input for the design of public transportation systems in the decades to come.
2. Background
2.1 Spatial Mismatch in the U.S. Labor Market
The spatial mismatch hypothesis centers on the idea that many workers may have poor labor market outcomes because they reside far from the job opportunities appropriate to their skill level, and the monetary or time cost of transporting the workers from their residences to job sites is high. Formal study of this problem began in the 1960s, and was spurred in part by the investigations surrounding the 1965 Watts riots in Los Angeles (the McCone Commission, 1965). The investigating commission concluded that the low employment rates of Watts residents contributed to the riot, and these low employment rates were, in turn, driven by the geographic isolation of residents from skill-appropriate jobs elsewhere in the Los Angeles metro area. Lower rates of personal vehicle ownership among Watts residents not only cut off access to jobs but also access to many social services provided outside the neighborhood. The commission strongly recommended improvements in public transportation in order to boost employment outcomes and access to more services. However, city transportation budgets have limits, and the long-term shift of many low-skill jobs to the urban periphery, where population density is low and job sites are relatively far apart, has made it difficult to resolve this issue through traditional public transportation technologies. Decades after the release of the McCone Commission Report, researchers continue to find evidence consistent with a significant degree of spatial mismatch in American cities.
There is now little doubt that spatial mismatch is a serious social problem. One particularly persuasive study on spatial mismatch, by Andersson et al. (forthcoming), uses employer-employee administrative data, combined with a person-specific job accessibility measure, to show that after a mass lay-off, lower-skilled workers were disproportionately likely to face long unemployment spells due to poor job accessibility. The study finds that African Americans, females and older workers are more sensitive to travel time than other subpopulations. This research is important because it provides solid evidence for the important role of transportation time costs as a key factor shaping labor market success, particularly among lower-skilled workers.
A modest literature focuses more directly on transportation costs/time as a factor affecting labor market outcomes. Black, Kolesnikova and Taylor (2014) find that participation in the labor force of married women is negatively correlated with the city’s average commuting time, and provide some evidence of a causal link. Also, studies of local transportation systems provide useful evidence on the topic, e.g., Moeller and Zierer’s (2018) evaluation of highway expansion in Germany, and Thierry and Trevien’s study of the expansion of France’s Regional Express Rail (RER).
Of course, none of these studies mentioned in the previous paragraphs are a substitute for an experiment that exogenously varies transportation costs. The Black et al. (2014) paper, for example, shows a strong relationship between commuting time and labor force participation, but, as they acknowledge, part of that relationship could be driven by the sorting of households with high rates of labor force participation into cities where commuting times are shorter. Similarly, the research on the role of public transportation expansion demonstrates that reduced transportation costs shape labor markets, but as we mention in the introduction, these innovations have equilibrium effects—making it difficult to tease out causal effects on individual work behaviors.
Finally, it is important to note that limited access to public transportation can hinder access to job opportunities (Lichtenwalter, Koeske, and Sales, 2006) and can make job search difficult as well. Studies have shown that higher time and distance from jobs leads to lower search efforts. Not surprisingly, having access to a car makes job searching less costly, and those with cars tend to have higher search intensity (Patacchini and Zenou 2005). Dependence on personal vehicles for job searching stands as a major barrier for lower income households who cannot afford to own a car.
2.2 Spatial Mismatch in Allegheny County, Pennsylvania
The suburbanization of poverty has also changed the dynamics of spatial mismatch in American cities. Policies seeking to connect lower income and lower skilled workers living in the urban core to jobs may no longer work as these lower income populations move to the suburbs (Frey 2016). As the poor become more geographically dispersed along the urban periphery, the traditional hub and spoke model of most city mass transit systems becomes less useful. Disadvantaged residents need to move from one suburban area to another to pursue employment opportunities, but mass transit systems provide fewer links along the periphery. Some parts of the community are served by a single bus line that runs infrequently, making it challenging to secure employment, particularly third shift jobs or jobs in other suburban neighborhoods.
We see these trends and their consequences clearly in our own region of Southwestern Pennsylvania. The Allegheny County Department of Human Services Suburban Poverty report found that in many suburban census tracts, over 30 percent of residents do not own a vehicle. These residents face real transportation constraints, because 36 percent of suburban census tracts have limited access to public transportation and another 23 percent have only moderate access (Collins, Dalton, and Good, 2014). A study completed by the Shared Use Mobility Center argues that the suburbanization of poverty has led to longer commutes, poorer job access and increased reliance on personal vehicle ownership (APTA, 2016).
3. Research Plan
3.1 A Field Experiment on the Impact of Reducing Transportation Costs to Work
The goal of this project is to test the impact of reduced transportation costs on labor force participation of one group of individuals who we believe are likely to be particularly sensitive to transportation costs—women with children. Black, et al. (2014) show that labor force participation decisions of women with children are highly sensitive to transportation costs. Also, the literature cited above points to the particular problems facing those who do not have regular access to a car. With this in mind, the target group of the study is women with a child or children under age 18, who have no regular access to a car. The goal of this research is to see if labor market participation increases among these women if they have access to convenient subsidized transportation.
Recruitment
Recruitment of a sample of low-income mothers in the Pittsburgh region will be facilitated by cooperation with Allegheny County Department of Human Services (DHS), which has built an impressive database on county residents that can help us identify eligible participants. DHS has access to Allegheny County birth records that will allow us to identify mothers whose children are in the appropriate age range (under 18) at the time of our study’s inception. DHS can match these data to current addresses, phone numbers, and email addresses. DHS already has detailed information on the degree to which neighborhoods in the county are served by public transportation, providing us with the ability to oversample poor neighborhoods that have limited access to bus and rail lines. Thanks to a data cooperation agreement with Commonwealth of Pennsylvania’s Department of Labor and Industry, DHS can also match in state administrative records on hours and wages from the unemployment insurance system, providing us with a direct measure of poverty. DHS can supplement these already-rich data with information on receipt of TANF. Using all these data, we will identify a stratified random sample of poor mothers in our region, and DHS can use the phone/email contact information to inform these women of our study. We will conduct phone interviews of potentially eligible participants, inquiring about access to cars and making sure that our interviewees meet the criteria for our study. Consent procedures and randomization into the treatment or control groups will be done at the time of the screening interview. Our goal is to recruit 650 women into the study—325 into the treatment group and 325 into the control group. To be included in the study, the subjects must have a smartphone.
The Treatment Group
“Treatment” is a six-month intervention, as follows:
• Individuals will be set up with an Uber account. They will be asked to complete an initial survey, for which they will receive $10 in compensation.
In Month 1 of the study individuals will be provided with $200 of ride-hailing credits, and will be told that these credits are intended “to help with job interviews, if you are looking for a new job or looking to change your job, and to help with transportation to work.”
In Months 2 through 6, individuals will continue to receive $200 per month in credits. Obviously, $200 per month would not be sufficient to offset the full cost of a long-distance daily commute that relied solely on Uber, but it would allow participants to supplement mass transit services with “first mile / last mile” transportation, and could also provide our participants with an alternative to mass transit or carpooling on days when time was short, the bus was running late, or typical carpooling arrangements were not feasible. In short it would provide substantial increased flexibility, which we hypothesize might be very valuable for these workers. A transit search app, described below, will facilitate the ability of our participants to combine ride-hailing and other transportation options.
• A job-search “tool” will be made available. This tool, accessible from a smartphone or a laptop, will link individuals to free services available to job seekers in the Pittsburgh area through Careerlink, a state-run program designed to connect job seekers to open jobs, thereby providing modest assistance to individuals seeking to find a job or move to a new job.
• An app will be installed on smartphones that will allow us to conduct a very short monthly survey in which we ask about labor market activity—hours worked, wages, commute times, and job search activities. Individuals will receive $10 per month (in the form of an electronic transfer to a pre-paid card) for survey completion.
• The same mobile app will also allow us to track locations via GPS. This is an important and innovative feature of our study. The tracker provides location data at 15 minute intervals, which gives us a means of verifying self-reports about employment and transportation times. For instance, if the individual reports having a job at Target on Centre Avenue in Pittsburgh, working 40 hours per week, we expect that the GPS location tracker will allow us to easily verify this. Also, we will be able to extract from the data very precise movement patterns, and it will allow us to estimate transportation times—as a supplement to self-reports on commute times.
• The same mobile app also provides a transit-search tool that will enable our participants to make the most of their access to free ride-hailing by combining it with other transportation options. This novel component of our app will make it possible for our study participants to search for all possible transit options, including mass transit (bus/subway), ride-hailing (Uber), biking, walking, or any mixed combination of these transit options, between two geographic locations. The transit search tool will make it easier for participants venturing outside their neighborhoods to combine an Uber ride with mass transit or other options, getting much farther on their limited monthly allocation of ride-hailing credits.
For a “treatment” individual who maintains full employment over the 6 month duration of the experiment, monetary benefit costs to the study—in the form of Uber credits and survey-response compensation—will be $1270. However, we expect that not all respondents will work each month, and thus anticipate the cost may be closer to $1000-$1,100 per subject for members of the treatment group.
The Control Group
Individuals randomly assigned to the control group, will receive the same benefits as the treatment group, except they will not be provided job-contingent transportation assistance in Months 2 through 6. Thus in the control group:
• Individuals will be set up with an Uber account.
In Month 1 of the study individuals will be provided $200 of ride-hailing credits, and will be told that this credits are intended “to help with job interviews, if you are looking for a new job or looking to change your job, and to help with transportation to work.” But no additional transportation assistance will be provided. By providing individuals in the control group with this “free” $200 in credits, we suspect that we will have near 100% enrollment among individuals who we recruit. This is importance for the fidelity of the study.
• The job-search “tool” will be made available.
• An app will be installed on smartphones that will allow us to conduct a very short monthly survey in which we ask about labor market activity—hours worked, wages, commute times, and job search activities. As with the treatment group, respondents will receive $10 per month (in the form of an electronic transfer to a pre-paid card) for survey completion.
• The same mobile app will also allow us to track locations via GPS. Again, this is an important and innovative feature of our study, providing us with a means of verifying self-reports about employment and transportation times.
• Finally, the same app will provide transit-search capabilities to the control group, in the same way that it does for the treatment group, enabling members of the control group to combine Uber rides with other transportation options.
For a “control” individual, the expected cost will be approximately $270 (assuming some non-response to surveys) -- $200 for ride-hailing and $70 for compensation for submitted surveys.
Our experiment thus is designed to implement a test of the basic theory of labor supply with quasi-fixed costs (commuting costs in this case) using experimental methods. To our knowledge this will be the first study of its kind, and thus will constitutes a unique and potentially valuable contribution to labor economics.
In addition, as we have noted, our work can be thought of as a “program evaluation” of potential new IT-enabled modes of public transportation. In the future, new technologies might allow municipalities to include (to at least some degree) the kind of flexibility that we are providing subjects in our experiment through public support for access to ride-hailing and ride-sharing services, perhaps in the more distant future with driverless vehicles. Our study is a first step in what we hope will be an active literature on the public policy benefits of such transportation innovations. In particular, our work will constitute a potentially important assessment of the value of such transportation for increasing the labor force participation rates of a particularly vulnerable population.
3.2 Data Analysis
Impact of Treatment on Employment Outcomes. The major focus of this study is to examine the impact of our intervention on labor market outcomes, including:
effort seeking a new job (among those who don’t currently have a job)
labor force participation (LFP)
time spent traveling to and from work (for those who work)
hours worked per month
wages and earnings
Because we have a randomized control trial, basic analyses will be quite simple; we can use t-tests to compare mean differences. With our intended sample sizes (325 in each experimental condition) we anticipate having power to detect reasonable-sized treatment effects. In addition to our tests of mean differences in outcomes, we anticipate using regression-based methods, which can reduce the standard error of the estimated treatment effects, and potentially yield additional insights. Given randomization, we do not need control variables to form an unbiased estimate of the treatment effect, but inclusion of the variables can help with the precision of the estimate and would also be a useful check for randomization fidelity.
Importantly, our outcomes can include not only self-reports, but also outcomes determined from the GPS tracking data, and outcomes constructed from data provided by the DHS, which will eventually include data on hours and wages taken from official records of the state Department of Labor and Industry. This last feature of our research design is useful for two reasons. First, having administrative data will add to the credibility of our results (and will also allow for some methodologically interesting analysis of measurement error). Second, it will allow us to do a follow-up analysis in which we can examine longer-run impacts. For instance, we may find that our transportation treatment not only increases labor force participation during the 6 months of the study but also has a longer-term impact. For instance, we may find that at least some individuals in our treatment group continue to earn higher incomes – and therefore have less demand for social services -- years after the cessation of the treatment.
We anticipate that we may have enough power with our experiment to even include simple interaction terms of the treatment variable and one of our “control variables,” thereby investigating heterogeneity in treatment. For instance, we could form a dummy variable equal to 1 for individuals who live in “transportation deserts,” to determine if our intervention is larger for these individuals. Alternatively, we could see if treatment effects are larger for mothers with young children.
Finally, for some outcomes—such as employment—we may find it useful to use specifications that allow for time-varying dimensions (i.e., have a time subscript in the basic regression). For many of our analyses OLS will do, but for some outcomes (e.g., number of weeks employed over the 6 month period) other models will be employed (negative binomial regression methods, etc.)
Exploratory Analyses of Treatment on Other Outcomes. While the primary focus of our study is on the impact of the transportation experiment on employment, our design will allow us to do some “exploratory” analysis of the impact on other outcomes from the self-reported data, the GPS location tracking data, and, more importantly, from the DHS data. For example, DHS records individual-level use of some training programs and other social services. Also, the data include child-level outcomes, such as school truancy and disciplinary actions. It seems possible that improved access to transportation flexibility may improve family lives in ways that extend beyond employment.
Also, recall that all participants will have an app on their smartphones that regularly notes their GPS coordinates. These data will be recorded using an algorithm that assigns a unique identifier to every participant but protects his or her identity. In addition, for participants with access to ride-hailing services, the data generated by our cell phone “location tracker” app will be supplemented by the pickup and drop-off data collected by the rides-hailing company. It is quite possible that access to ride-hailing services may enhance the mobility of participants in ways that are not always directly connected to job search or efforts to access social services. For instance, ride-hailing services might enhance the ability of lower-income participants in poor, geographically isolated communities to access community amenities (parks, libraries, etc.). It could also expand their ability to consume a wider range (and higher quality) of commercial goods and services, and engage in more frequent social interaction through community events, religious services, musical performances, etc. The granularity of our user mobility data may allow us to detect or infer some of these changes in consumption and social interaction. As part of our initial efforts to explore these data, we will quantify differences in mobility between the control group and the treated groups along a number of dimensions. We plan to measure the number of unique neighborhoods (or zip codes) an individual visits per day/week. We can build upon prior analyses of individual mobility by measuring the spatial entropy of an individual’s movements over a period of time (e.g., a measure of the distribution of an individual’s movement through geographic space across distinct neighborhoods and locations). Finally, we can, in principle, compare the patterns of movement of participants with ride-hailing services who reside in certain neighborhoods to members of the control group who live in the same neighborhoods, but lack access ride-hailing services.
Timeline
In recent weeks, we have been completing our small sample pilot study. At the end of April, we plan to start the main experiment, recruiting low-income mothers in "batches" of roughly 50 participants, randomizing them into control and treatment groups, and tracking their activity over the 6 months of the experiment. Because recruitment will be phased in over the time, the entire study will extend well into 2020.
We anticipate that the funding in this grant will be mostly used to support our graduate research assistants.
Strategic Description / RD&T
Deployment Plan
As noted in our detailed project description, we plan to recruit a large number (approximately 650) low-income women with children who have limited access to a car into a randomized control trial assessing the impact of access to ride-hailing services on employment outcomes. We will randomly select recruited participants into a treatment group that receives access to a substantial amount of free ride-hailing for six months and a control group that does not. The GPS features built into modern cellular phones allow the mobility of ride-hailing fund recipients to be tracked with great accuracy across time and geographic space. These patterns could be analyzed and compared carefully to the mobility patterns of similar individuals who did not receive access to free ride-hailing services. Surveys and pickup and dropoff data from our partner ride-hailing company provide additional insight into the impact of this transportation resource on our participants, allowing us to measure the impact of access to ride-hailing on interviews, job offers, and wages. Our interdisciplinary research team provides a unique blend of faculty expertise that will allow us to run a state-of-the-art experiment and conduct follow-up mobility trajectory data analysis, building on the most recent advances in information technology, statistics and machine learning, transportation engineering, and behavioral economics. The Pittsburgh region offers an ideal context in which to undertake this experiment, because there are flows of workers from city neighborhoods to jobs on the periphery of the metro area, and there are also smaller communities located far from the urban core that send workers into the city. Like other American cities, Pittsburgh has seen some rejuvenation in its urban center, but a decline in the fortunes of many of the surrounding communities that are linked to it, and there are few existing public transportation links connecting some of these disadvantaged communities to the rest of the region. Thus, an experiment in the Pittsburgh region could have broad implications to the rest of the country and even for metro areas outside the United States. The project timeline, provided above, shows how we will allocate this activity over the next two years. Most of the funding in the current grant is expected to be spent in the next few months, as the timeline above indicates.
Expected Outcomes/Impacts
For more than 50 years, social scientists have shown that spatial mismatch prevents many lower-income Americans from attaining higher incomes and higher levels of participation in the labor force. This scholarship has pointed out the limitations of conventional mass transit, and called for more flexible forms of support for the transportation needs of lower-income residents. The rise of ridesharing offers the possibility of a new transportation technology that may offer a new solution to this longstanding problem.
To the best of our knowledge, our project is the first effort to use a randomized control trial (RCT) methodology to evaluate the impact of access to ridesharing on the job search activities, employment outcomes, and labor force participation rates of the poor. If we can demonstrate, through a successful field experiment in the Pittsburgh region, that deployment of this technology measurably raises the incomes of the poor, then it could lead to more effective transportation and poverty reduction policies across the nation and around the world. It could lead to better lives for working families. It could help establish Carnegie Mellon and Pittsburgh as centers of policy innovation, complementing our well-deserved reputation as a center of technological innovation. Last, but certainly not least, our work is likely to lead to several peer-reviewed publications in the scientific literature.
We have described in the rest of this proposal a number of novel ways in which we be able to track the job search activities of our participants, as well as their employment outcomes. Funding from the Mobility 21 program will allow us to generate sample sizes sufficiently large such that we can quantify the impact of this program on participants, and formally test the hypothesis that it changes search behavior and yields better employment outcomes.
Expected Outputs
TRID
Individuals Involved
| Email |
Name |
Affiliation |
Role |
Position |
| branstet@cmu.edu |
Branstetter, Lee |
Carnegie Mellon University |
PI |
Faculty - Tenured |
| beibeili@andrew.cmu.edu |
Li, Beibei |
Carnegie Mellon Heinz College |
Co-PI |
Faculty - Untenured, Tenure Track |
| lt20@andrew.cmu.edu |
Taylor, Lowell |
Carnegie Mellon Heinz College |
Co-PI |
Faculty - Tenured |
Budget
Amount of UTC Funds Awarded
$90000.00
Total Project Budget (from all funding sources)
$200000.00
Documents
| Type |
Name |
Uploaded |
| Data Management Plan |
Data_Management_Plan_Mobility_21.docx |
Jan. 12, 2018, 7:06 a.m. |
| Presentation |
Can Ridesharing Help the Disadvantaged Get Moving? |
Sept. 28, 2018, 8:59 a.m. |
| Presentation |
Can Ridesharing Help the Disadvantaged Get Moving |
Sept. 28, 2018, 8:59 a.m. |
| Progress Report |
170_Progress_Report_2018-09-30 |
Sept. 28, 2018, 8:59 a.m. |
| Progress Report |
170_Progress_Report_2019-03-30 |
March 28, 2019, 2:41 p.m. |
| Presentation |
Panel on the Future of Work |
Sept. 29, 2019, 5:08 p.m. |
| Presentation |
Panel on the Future of Work |
Sept. 29, 2019, 5:08 p.m. |
| Progress Report |
170_Progress_Report_2019-09-30 |
Sept. 29, 2019, 5:11 p.m. |
| Final Report |
170_-_Final_Report.pdf |
Jan. 27, 2020, 4:57 a.m. |
| Progress Report |
170_Progress_Report_2020-03-30 |
March 28, 2020, 2:29 p.m. |
Match Sources
No match sources!
Partners
| Name |
Type |
| Uber Technologies |
Deployment Partner Deployment Partner |
| Allegheny County DHS |
Deployment Partner Deployment Partner |
