Research
Working Papers
Optimal Monitoring and Bureaucrat Adjustments
Abstract
Monitoring policies designed to maximize deterrence must account for attempts by agents to evade detection. This paper examines the strategic responses of bureaucrats, who implement India's employment guarantee program, as their expectations of being audited change. Exploiting random assignment to audit timing over multiple waves (without replacement), I find the rate of deterrence for misappropriated expenditures is increasing in bureaucrats' expectations of being audited. In addition, bureaucrats evade detection by adjusting the timing and type of expenditure to misappropriate. Applying a model of Bayesian persuasion, I analyze how information communicated on the likelihood of being audited should be designed. I estimate a sufficient statistic from the model to solve for the optimal signal and analyze welfare under counterfactuals. Concentrated incentives, i.e. notifying of audit timing in advance, would have persuaded bureaucrats to misappropriate USD 35m less in expenditures (16% of average annual expenditures) when compared to dispersed incentives, i.e. messages are uninformative and audit timing is unpredictable.
Access and Invitations: Increasing COVID-19 Vaccination in Kenya with Kevin Carney, Michael Kremer, Elisa Maffioli, and Leah Rosenzweig
Abstract
We examine the impact of a vaccination campaign in Kenya that sent healthcare providers to homes inviting adults, with relatively proximate healthcare services, to get a COVID-19 vaccine nearby. The intervention increased the cumulative number of doses given by 8.7 per 100 people on the day of the intervention, equivalent to about a 10% increase over the baseline number of doses in the control group. The greater number of doses in the treatment group persisted in the 3 months following the intervention, indicating that the intervention induced people to get vaccinated who would not have done so otherwise. A machine learning analysis of heterogeneity reveals that treatment effects are largest among more disadvantaged groups - women, those with less income, and those with less education. To examine whether social image considerations influence vaccination behavior, we borrow a design from DellaVigna, List, and Malmendier 2012 and DellaVigna et al. 2016 used in the contexts of charitable giving and voting, and randomized an announcement of the home visit and vaccination offer ahead of time. This strategy allows those unwilling to be vaccinated to avoid the visit without facing the social repercussions of declining in-person. Contrary to expectations, there was no evidence that social pressure influenced vaccination. Instead, the announcement increased the probability of getting vaccinated by 3.8 percentage points, primarily driven by older participants who are at higher risk of severe disease. A cost-effectiveness exercise suggests that our intervention is comparable to other vaccination campaigns, at 34USD marginal cost per marginal dose.
Hard to Read: The Impact of Advanced Reading Assignments on Literacy with Guthrie Gray-Lobe, Mridul Joshi, Michael Kremer, and Joost de Laat
Abstract
We evaluate the impact of providing more advanced English reading assignments to children attending low cost private schools in Nigeria. Top-performing students do not benefit from advanced reading assignments, and short-run reading fluency may even be harmed: students assigned to advanced work read fewer words per minute and score lower on literacy assessments. Surprisingly this negative impact is driven by the highest-performing students. We argue that this result may indicate that only top-performing students benefit from the status quo level of reading assignments in this setting.
Can SMS Interventions Increase COVID-19 Vaccination Completion in Kenya? [submitted] with Florence Aketch, Kevin Carney, Elisa Maffioli, James Otieno, and Leah Rosenzweig
Abstract
We evaluate the impact of SMS interventions on COVID-19 vaccination completion in Kisumu County, Kenya and address barriers to completion among people who receive a first but not a second dose of a two-dose COVID-19 vaccine sequence. We conducted four experiments between July 2022 and January 2023 testing the effect of SMS messages that provided people with information about the vaccines and reasons to get fully vaccinated, including incentives. We find no significant treatment effects of any of the SMS messages on vaccination completion rates (point estimate: 0.0031; 95% confidence interval: -0.0016-0.0078). Vaccination completion rates increase over the study period but do not increase significantly more in any treatment condition than in the control group. Phone surveys reveal that 85% of people recalled receiving the message, but that concern about COVID-19 could be a substantial barrier, with approximately one-third of the sample saying they are "not at all worried" about COVID-19.
Selected Works in Progress
Women Driving Women (field experiment completed; data collection and analysis in progress) with Pablo Peña
The Value of Concision (field experiment and data collection completed; writing in progress) with Guthrie Gray-Lobe, Michael Kremer, and Joost de Laat
Abstract
Two experiments by a firm delivering highly-detailed lesson plans to teachers found that providing teachers with more concise and clearer lesson plans raised student test scores by 0.17 standard deviations. Effects are concentrated among the lowest-scoring students and in larger classes, consistent with the hypothesis that more concise and clearer lesson plans freed up teacher time for unscripted tasks, such as providing students with individual feedback.
Improving Local Tax Compliance in India (field experiment and data collection completed; writing in progress) with Yusuf Neggers and Shreya Singh
Scaling Up Personalized Adaptive Learning in India (a portfolio of work with multiple field experiments in progress) with Emily Cupito, Guthrie Gray-Lobe, and Michael Kremer
Abstract
Personalized Adaptive Learning with edtech (PAL) has the potential to provide individually-customized education at scale within existing education systems. PAL tailors content to create a learning journey designed to address individual learning needs, and has been shown to foster remedial learning and mitigate learning gaps in settings where children lag behind grade-level skills (de Barros and Ganimian 2023; Muralidharan, Singh, and Ganimian 2019; Escueta et al. 2017). However, while PAL has been shown to be consistently effective in researcher-controlled settings after school, it has yet to be shown to be effective when run by a government partner during the school day (Banerjee et al. 2023). Additionally, the effectiveness of education interventions, including learning through the use of edtech, when conducted at scale can be sensitive to the model and quality of implementation (Banerjee et al. 2017; Kulik and Fletcher 2016). An edtech program that leads to poor adoption by teachers and students would not only lead to no gains in learning, but would also include costly investments in hardware and infrastructure which could have been better spent elsewhere. In 2019, the Government of Andhra Pradesh (GovAP) in India launched one of the first government-run PAL programs in over 500 schools with ConveGenius, the main provider, and other software companies. In collaboration with the GovAP, this project is using RCTs to 1) evaluate the impact of the GovAP's PAL program, and 2) evaluate two additional PAL program variants to inform the program's design at scale.
The Impact of Conversational AI on Learning with Education Technology in India (preparing for field experiment) with Emily Cupito, Guthrie Gray-Lobe, and Michael Kremer
Informal Relationships between Bureaucrats and Beneficiaries, and Access to Public Resources (data collection completed; analysis on hold)
Publications
Assessing the Costs and Effects of Antiretroviral Therapy Task Shifting From Physicians to Other Health Professionals in Ethiopia, with Johns, B., Asfaw, E., Bekele, A., Minior, T., Kebede, A., & Palen, J., Journal of Acquired Immune Deficiency Syndromes (2014)
Lessons Learned from Stakeholder-Driven Sustainability Analysis of Six National HIV Programmes, with Katz, I., Glandon, D., Kargbo, B., Ombam, R., Singh, S., Osika, J. S., Health Policy and Planning (2014)
Patient Satisfaction with Task Shifting of Antiretroviral Services in Ethiopia: Implications for Universal Health Coverage, with Asfaw, E., Dominis, S., Palen, J. G. H., Bekele, A., Kebede, A., & Johns, B., Health Policy and Planning (2014)