Three Applied Economic Studies in Collaboration with Government Policymakers by Seth Aaron Levine Garz A dissertation submitted in partial satisfaction of the requirements for the degree of Doctor of Philosophy in Agricultural and Resource Economics in the Graduate Division of the University of California, Berkeley Committee in charge: Professor Alain de Janvry, Chair Professor Michael L. Anderson Professor Paul J. Gertler Summer 2016 Abstract Three Applied Economic Studies in Collaboration with Government Policymakers by Seth Aaron Levine Garz Doctor of Philosophy in Agricultural and Resource Economics University of California, Berkeley This manuscript features three independent applications of contemporary techniques in applied economics conducted in collaboration with three different governments. In the first essay, I present findings from an experimental intervention conducted in collaboration with the government of the Dominican Republic in which the government varies whether front-line volunteer community workers are recruited through a public advertisement or through local staff referrals. I find that governments likely face tradeoffs in selecting optimal recruiting mechanisms as publicly recruited candidates demonstrate superior observable characteristics to referred candidates across a wide variety of indicators except for the key indicator of cognitive skills. I also find public candidates are more likely to accept job offers and attend trainings conditional on being hired. To my knowledge, this is the first study to rigorously test the impacts of an open public recruitment process versus a private targeted recruitment process. In the second essay, my co-author and I evaluate impacts of a World Bank funded road improvement and employment generation intervention in Nicaragua. We employ a difference-in-difference research design, matching proprietary road building data from the government of Nicaragua’s Ministry of Transportation with three rounds of a publicly available household survey. We find strong evidence that the World Bank’s Fourth Roads Rehabilitation and Maintenance Project fulfilled its primary goal of improving road infrastructure and suggestive evidence of select economic and social impacts. Notably, we do not observe impacts on the likelihood of employment or incidence of poverty. In the third essay, I collaborate with the City of San Francisco’s Office of the Treasurer to investigate mechanisms that potentially mediate household decisions to save for distant future expenditures with particular relevance for education-oriented savings. Employing a randomized experiment in the natural setting of the Kindergarten-to-College school district-wide college savings program, I vary the messages of postcard savings reminders with information about either the availability of college financial aid or the increasing cost of tuition. My results suggest that savings reminders may overcome inattention to lumpy future expenditures, but only among those who intended to save. I find no evidence that manipulating the salience of college cost affects savings behavior.   1 Dedication To my children and grandchildren, that absolute poverty may not still be around to demand your attention when it is your turn.   i Contents Chapter One “Optimal Recruitment Design for Frontline Public Servants: Public Advertisements vs. Private Referrals.” August 2016. Chapter Two “Evidence on the Impacts of Short-term Employment Generation Projects to Improve Road Infrastructure in Rural Nicaragua.” April 2014. Chapter Three “Being Reminded to Save for College: The Impact of Inattention and Expected College Cost.” April 2013.   ii Acknowledgements First and foremost, I would like to acknowledge the members of my dissertation committee. Thanks to Alain de Janvry as my committee Chair and academic advisor for your encouragement, optimism, and wisdom, which was always available at a moment’s notice, and for introducing me to the world of development economics as my teacher. Thanks to Paul Gertler for your mentorship: for showing me how to do real research, engage in authentic open collaboration, and design academic work to answer real world problems. Thanks to Michael Anderson for your lucid instruction and ongoing counsel on causal inference and research design; when attending your Applied Econometrics class I began to see the world with new eyes and became determined to pursue the Ph.D. in ARE. Thanks to my classmates Sylvan Herskowitz and Kenneth Lee for the collective angsting, casual intellectualizing, and irreverent humor throughout this journey. Thanks to my officemates Yang Xie, Daniel Tregeagle, and Marieke Kleemans for your capacity to overlook molding coffee cups, to indulge frantic technical questions, and to take seriously my intellectual meandering. Thanks to my teachers and unofficial advisors Elisabeth Sadoulet, Jeremy Magruder, and Fred Finan. Special thanks to Peter Berck for reminding me that with an infant and a first year Ph.D. course load, I should have been satisfied just to still be alive, and for always being around for guidance when doubt crept in. Thanks to David Roland-Holst for employing me as a Reader in Chinese Economy and an instructor at the MDP boot camp for so many years. Thanks to Carmen Karahalios for always giving me the administrative inside scoop. Thanks to Señora Marjorie Spaeth for sending me to Mexico in 9th grade and catalyzing my interest in the foreign. Thanks to my collaborators on the project that is the focus of Chapter One, including Hector Medina, Ezequiel Volquez, José Antonio Pellerano, Yaneth Basora, Ricardo Cuesta, Elianny Medina, Sandro Parodi, Maricha Freidman, as well as the ProSoli, Gabinete, and CEGA staff. Special thanks to Justo Mirabal. Thanks to my Chapter Two co-author Elizaveta Perova and collaborators among the staff of MTI in Nicaragua. Thanks to my collaborators, advisors, and commenters on the project that is the focus of Chapter Three, including SF Treasurer José Ciscneros, staff of the SF Office of Financial Empowerment, Anee Brar, Minah Jung, Leif Nelson, Sofia Villas-Boas, Aluma Dumbo, and Lei Cheng. Thanks to the founders and maintainers of the East Bay Regional Parks, the fire trails of which often rejuvenated me during steep afternoon runs, and to Alexis Madrigal for joining me up those hills. Thanks to the architects of the Morill Act, the builders of UC Berkeley, and the California taxpayers that continue to support the University of California whose vision and work has afforded me the privilege of studying and teaching at a public university that is a model of diversity and intellectual effervescence to which the world can aspire. Thanks to Pat for being real. Thanks to Joshy for your intentionality. Thanks to my great-grandparents, my grandfathers, Uncle Doug, Grandma Rae, and Grandma Tillie for leading us on our passage from the shtetl to the ivory tower. Thanks to Mom and Dad for every opportunity in the world. Thanks to my sister Jess for sharing how I understand. Thanks to my Sherwood Family for the love and support. Thanks to Olea and Eshel for true joy. Thanks to Julie.   iii Optimal Recruitment Design for Frontline Public Servants: Public Advertisements vs. Private Referrals Seth Garz August 9, 2016 Abstract In designing an optimal recruitment process, managers must decide to what extent they should invest in open public searches or privately targeted referral-based searches. Numerousstudiesdocumentthewidespreaduseofreferralsinthelabormarket,butex- isting evidence comparing the performance of referred to non-referred workers suggests contradictory results. Furthermore, these studies generally rely on associations in data thatarelikelysubjecttoimportantselectionbiasesandoftenfailtodefinethecounter- factual to referral-based recruiting in a way that meaningfully reflects the recruitment decisions routinely faced by managers. In this study, I collaborate directly with the government of the Dominican Republic to experimentally vary whether front-line vol- unteer community workers are recruited through a public advertisement or through local staff referrals. I find that the government faces tradeoffs in selecting among re- cruitment methods with publicly recruited candidates revealing superior observable characteristics across a wide variety of important indicators, but referred candidates demonstrating unambiguously superior cognitive skills. I also find publicly recruited candidates are more likely to accept job offers and attend trainings conditional on be- ing hired. To my knowledge this is the first study to rigorously test the impacts of an open public search recruitment process versus a private targeted search recruitment process. Worker selection is particularly important in the public sector given govern- ment’s limited means of motivating workers with traditional incentives and the large scale of public front-line workers in security, education, health, and social protection. My findings suggest a number of future research topics that could collectively have important policy relevance for improving government recruitment efforts. 1 1 Introduction A large body of literature documents that workers both in developed and developing coun- tries frequently find jobs through referrals. A handful of studies also demonstrate that firms frequently encourage referrals as a recruitment method. Researchers have empirically val- idated specific mechanisms by which firms can leverage the network attachments between referring employees and referral candidates to improve worker screening, monitoring, match- ing, satisfaction, and effectiveness generally.1 However, to conclude that firms benefit more from using referrals than from using other recruitment mechanisms would misinterpret the existing evidence. Merely observing the frequency of referral-based job search and validating plausible benefits from a referral-focused recruitment process is insufficient to discount the potential benefits to firms of employing alternative recruitment mechanisms, such as open public searches, about which there is little evidence. This study seeks to contribute to the literature on the explanations for and consequences of different job recruitment designs by providing what is to my understanding the first rigorous experimental evidence comparing an open public search recruitment process to a privately targeted referral-based recruitment process. Heath (2013) argues that the prevalence of referrals among hired employees is prima facie evidence for the inferiority of open public recruitment processes. However, this need not be true of the public sector even if we take revealed preference seriously to conclude that firms disciplined by the profit motive in competitive sectors of the economy conscientiously and rationally rely heavily on referral-based recruitment. While publicly advertising competitive processes for senior bureaucrats may be a regulatory requirement among many governments, the hiring of low level government staff and community volunteers that make up the front- line workers at the base of the public sector pyramid in developing and developed countries alike are almost certainly more likely to depend on hiring through references. In our study context in the Dominican Republic (DR), for example, roughly 12,000 volunteer community workers (CWs) administer home visits to welfare beneficiaries with the recruitment of the CWs entirely left to the discretion of local supervisors. Front-line worker hiring choices in the public sector may be particularly distorted given the low cost of the workers and the difficulty of fully internalizing the productivity losses from these workers to an organization that is not subject to market discipline. Ineffective recruitment may be particularly damaging to the effectiveness of public agencies given the difficulties they face motivating workers to perform and their limited ability to use strong incentives.2 The role of these workers in national and civil defense, education, health, and administering social protection services to the most vulnerable citizens may render sub- optimal recruitment strategies and hiring decisions particularly costly in terms of social 1Afridi et al. (2015) provides a review of recent theoretical and empirical work on social networks and labor productivity. As an example of managers reporting the frequency of referrals, as opposed to workers themselves, Dhillon et al. (2013) uses responses of managers from the 2006 World Bank Enterprise Survey from India to show workplace referrals account for between 40-65% of hires among a diverse of array of industries. 2SeeChaudhuryetal.(2006)forareviewofabsenteeismamongpubliceducationandhealthworkersand Dixit (2002) for review of theory of incentives in the public sector. 2 welfare. Given that the public sector constitutes the largest source of employment in virtu- ally every national economy, it stands to reason that modest improvements in recruitment processes of public agencies could have important social and economic implications. Further- more, investigating ways to improve the selection of front-line social protection workers, in particular, is consistent with broad interest among governments and multilateral funders in developing cost-effective regimes for layering social intermediation services on top of existing cash transfers programs.3 Thus, rigorous evidence comparing the characteristics and perfor- mance of candidates recruited through referral versus alternative means may be particularly valuable for these front-line employment settings. In this study, I collaborate with the agency responsible for administering the DR condi- tional cash transfer (CCT) program, Progresando con Solidaridad (ProSoli), and for man- aging the associated CW volunteers. We experimentally vary for which neighborhood CWs are recruited through an open public advertisement intervention versus a private targeted referral process from local government staff, which was designed to closely resemble the sta- tus quo recruitment process. By design, the interventions resulted in areas for which there are overlapping samples of candidates from both public advertisement and private referral recruitment processes. Comparison of these overlapping samples will constitute my primary means of identifying differences in the candidates recruited through either intervention to answer the following key research questions: 1. Are candidates privately targeted by government staff for low skilled public service jobs systematically different from those that would be attracted through a public re- cruitment process? 2. Are publicly or privately recruited candidates superior according to observable criteria that would a priori be used to screen candidates by a hiring employer? 3. Are characteristics of publicly recruited candidates unambiguously superior across all criteriaorwouldahiringorganizationexpecttohavetoweightradeoffsinthecharacter of the candidates recruited through different mechanisms? 4. Does public recruitment lead to higher rates of job acceptance and persistence? 5. Among privately referred candidates, are those that are higher ranked by referrers better or worse candidates according to observable characteristics? Results from comparisons of candidates recruited through open public search and pri- vate targeted referrals suggest systematically superior characteristics for public candidates across a wide variety of indicators observable to the researcher and hiring agency, including characteristics pertaining to demographics, employment, economic status, civic and polit- ical involvement, and personality. Referred candidates do, however, out-perform public candidates on important cognitive characteristics, begging the question of whether referred candidates may have other unobservable characteristics that would lead to better future job performance. Tension between the superiority of referrals on the cognitive dimension and 3See Camacho et al. (2014) for a recent review of the role of social intermediation services in CCTs. 3 the superiority of public candidates on virtually all other observable characteristics suggests governments may face important tradeoffs in deciding among alternative recruitment mech- anisms. Nevertheless, results from a small scale sub-experiment within the broader study suggest public candidates are also more likely to accept jobs and more likely to show up at subsequent trainings, casting doubt on the idea that privately targeted referrals constitute a better candidate recruitment and selection process. Additional results from comparison of referred candidates ranked higher and lower by the referrers suggests that there is little variation among privately targeted referrals. In terms of operational efficiency and cost, results from the study suggest that the public search process was modestly successful, recruiting candidates to roughly half of the tar- geted areas intended at cost equivalent to only one to two months wages per position. By comparison, the private referral process only succeeded in covering 19% more of the same targeted areas. This modest operational success coupled with the apparent superiority of candidates recruited through the public search process across most observable characteristics begs the question of why these front-line public sector jobs rely on private referrals and offers a promising policy alternative to improving the quality and persistence of front-line com- munity workers. Interpreted more broadly, these findings question the conventional wisdom that the prevalence of referrals in job markets coupled with empirical evidence explaining various plausible benefits of referral processes demonstrate the superiority of private targeted searches in comparison to well-defined and sufficiently resourced open public searches. Nev- ertheless, the results demonstrate that employers face tradeoffs and further investigation to better establish results consistent with those presented here is needed before applying these findings to more general policy recommendations. In future research, I intend to track the performance of the different types of candidates and their impacts on CCT beneficiaries to complement my current findings. This study seeks to overcome important limitations in existing studies that attempt to compare the characteristics and performance of referred versus non-referred workers. First and foremost, numerous studies define “being referred” in binary terms through self-reports of job candidates, identifying who is referred without rigorously defining the counterfactual to being referred.4 However, it is the counterfactual to referrals that is precisely what is of interest to public and private organizations when making the key management decision of how to optimally design an employee recruitment process. “Referral” in the sense that the term is currently used in many studies does not meaningfully correspond to a specific recruitment option that an organization may choose to pursue. Whetherornottouseorignorepeerreferralsdoesnoteffectivelycapturethemanagement decision an organization faces. More relevant to most organizations is the choice between whether to pursue some form of open public search process or to rely on the private targeted search processes of its employees. More specific recruitment options include: i) investing in a robust public search process, ii) relying on employees natural proclivities to pursue their own private search techniques, which often leverage their own social networks, iii) investing in incentivizing employees to pursue a private search process that is more compatible with the firm’s recruitment priorities, or iv) letting nature run its course and seeing which candidates 4See Burks et al. (2015) and Munshi and Rosenzweig (2006) for examples. 4 walk in the front door, so to speak. The precise counterfactual recruitment process to pure referrals can certainly vary and may drastically alter the comparison of referred versus non- referred candidates. In the extreme, just accepting applications from people that happen to walk in the front door would be a form of untargeted public search, but is unlikely to yield many qualified candidates. In Pallais and Sands (2015), the existing study that is closest in design to my study, the counterfactuals to referred workers are workers randomly invited to apply for piecemeal work from the pool of qualified temporary workers that had previous work experience with the hiring firm. This random recruiting approach is not completely unreasonable in the study’sstylizedsettingofaprivatesectoroffshorebusinessprocessingcenterthatmaintainsa networkofpre-screenedtemporaryworkers, butlikelylacksexternalvalidityformoregeneral recruitmentsettingsandpublicsectorsettings, inparticular, wherethereisrarelyanexisting pool of pre-screened workers from which to randomly select. Furthermore, the superior performance of referred workers in Pallais and Sands (2015) is not entirely unexpected given that there is no real open public pool from which to refer or randomly select new recruits, implying that non-referred workers may be systematically avoided by referrers by construct. Like in many other studies, referrals in Pallais and Sands (2015) come from peers and not managers who would likely apply a different private search screening process that may be more consistent with local hiring dynamics in low skilled job settings where mass candidate screening can be costly. Notably, the stylized design features of the Pallais and Sands (2015) study were intentional and enabled the authors’ to focus on a variety of other compelling questions related to monitoring and collaboration. In contrast to a highly stylized alternative to referral-based recruiting, the public search method in my study setting is the intervention the government deemed the most practical and obvious “counterfactual” innovation to the status quo approach of private search by local government staff. Specifically, the research team posted posters and handed out fliers in the exact neighborhoods from which the privately referred candidates were also recruited. This was possible because our setting requires candidates to live in close proximity to the welfare beneficiaries they are being selected to serve. A handful of existing studies suggest that referred and non-referred candidates with simi- lar observable characteristics at the time of recruitment have quite different levels of turnover and productivity, though the direction of the results in favor or against referrals is contra- dictory. Fafchamps and Moradi (2015) use associations in historical data in a low income country context on recruitment and desertion in Ghana’s Colonial Army to suggest refer- rals reduce the quality of recruits with higher desertion and dismissal rates among referred soldiers. Importantly, the authors find that, compared to non-referrals, referred candidates actually had superior observed characteristics, specifically height and chest circumference, at the time of recruitment. In contrast, using data from nine large firms across three industries mostly in the United States, Burks et al. (2015) find that referred workers are associated with higher rates of job offer acceptance, similar if not higher productivity, and lower rates of turnover despite having similar characteristics at the time of hire. These studies suggest caution in using results from my study demonstrating systematically different observable characteristics among public versus referred candidates to predict future job persistence and performance. Nevertheless, these existing studies with their contradictory conclusions suffer 5