“Scared Straight or Scared to Death? The Effect of Risk Beliefs on Risky Behaviors” [PDF] [Online Appendix] (Revise and Resubmit, American Economic Review)
This paper tests a model of risk compensation that allows for “fatalism”: higher risks lead to more risk-taking, rather than less. Fatalism can be rational if the risk of each act exceeds a threshold value. I test this prediction by randomizing the provision of information about HIV risks in Malawi, and break down the risk elasticity of sexual risk-taking by people’s initial risk beliefs. Matching the model’s predictions, this elasticity varies from -2.3 for the lowest to 2.9 for the highest beliefs. Fatalism is more pronounced among people who think they may be HIV-positive, consistent with the model’s mechanism.
“Making the Grade: The Sensitivity of Education Program Effectiveness to Input Choices and Outcome Measures” (with Rebecca Thornton) [PDF][Online Appendix] (Revise and Resubmit, Review of Economics and Statistics)
This paper demonstrates the acute sensitivity of education program effectiveness to input choices and outcome measures, using a randomized evaluation of a literacy program. The program raises reading scores by 0.64SD and writing scores by 0.45SD. A reduced-cost version instead yields statistically-insignificant reading gains and large negative effects (-0.30SD) on writing. Detailed classroom observations provide evidence on the mechanisms driving the results, indicating the importance of increased productivity of time, rather than increased time on task. Changes in the use of materials as well as machine learning and mediation analysis results also suggest important complementarities in the education production function.
“Peers and Motivation at Work: Evidence from a Firm Experiment in Malawi” (with Lasse Brune and Eric Chyn) [PDF]
This paper studies workplace peer effects by randomly varying work assignments at a tea estate in Malawi. We find that a 10 percent increase in mean peer ability raises productivity by 0.3 percent. In contrast to previous studies, neither production nor compensation externalities can drive the results because workers receive piece-rates and do not work in teams. Additional analyses provide no support for learning or socialization as mechanisms. Instead, peer effects appear to operate through “motivation”: when given a choice to be re-assigned, most workers prefer working near high-ability co-workers because these peers provide motivation to work harder.
“Pay Me Later: A Simple Employer-based Saving Scheme” (with Lasse Brune and Eric Chyn) [PDF]
Workers in developing countries often lack good savings options. We study a no-frills employer-based savings technology that piggybacks on existing payroll infrastructure to provide a safe and convenient savings method. Partnering with an agricultural employer in Malawi, we randomize offers to defer part of worker wages for three months, at zero interest, and receive a lump sum payout at the end of the main season. We find that this savings product has high initial take-up, high usage, and high repeat take-up. Take-up of the product changes behavior and outcomes: total savings increase over the deferral period, as does labor supply. Large purchases increase immediately after the payout, and asset holdings are higher three months later. We show that the seasonal timing is not crucial for take-up. Similarly, restrictions on access to savings has limited impact on take-up at the margin. In contrast, disbursing savings in a lump sum and automatic regular deductions are key ingredients for the product’s success.
“Teacher Effectiveness in Africa: Longitudinal and Causal Estimates” (with Julie Buhl-Wiggers, Jason T. Kerwin, Rebecca Thornton, and Jeffrey A. Smith) [PDF]
This paper presents the first estimates of teacher effectiveness from Africa using longitudinal data from a school-based RCT in northern Uganda. Exploiting the random assignment of students to classrooms within schools, we estimate a lower bound on the variation in teacher effectiveness. A 1-SD increase in teacher effectiveness leads to at least a 0.13 SD improvement in student reading at the end of one year. Using detailed survey and classroom observation data, we find no detectable correlation between teacher effectiveness and teacher characteristics, but do find patterns associated with teaching behavior in the classroom. Using the RCT we find that providing teacher training and support increases the variation in teacher effectiveness, by probably making the most-effective teachers relatively better than the least-effective teachers.
“Enumerator Knowledge Effects in Subjective Expectation Elicitation: Evidence from a Randomized Field Experiment” (with Natalia Ordaz Reynoso)
Subjective expectations about probabilities and the distributions of variables play a central role in modeling economic behavior. Direct elicitation of subjective beliefs via surveys is an increasingly popular tool, and recent research has shown that this is feasible even in low-income and low-numeracy settings. However, the enumerated surveys commonly used in developing countries have a potential weakness: respondents’ recorded beliefs may be affected by the enumerators’ knowledge about the belief questions. We show the importance of this issue using a randomized experiment that used enumerators to implement an information treatment. Reported beliefs are significantly shifted by the enumerators’ knowledge, decreasing by about 0.3 SD of the initial belief distribution. We discuss several mechanisms that may drive this phenomenon, including how the change in knowledge may have affected “probing” that enumerators did to encourage guesses when a respondent said they did not know the answer. These particular results are specific to my experiment; however, given the disparities in education levels between enumerators and respondents in typical developing-country surveys, the issue of enumerator knowledge contamination of subjective expectation data may be widespread. We suggest corrections for this issue from the perspectives of enumerator recruitment, survey design, experiment setup, and data analysis.