Job Market Paper
When Caring Backfires: Experimental Evidence from Family Caregiving in India (with Camille Falézan, Shrddha Rajesh, and Girija Vaidyanathan). [Draft coming soon!]
Does concern for family caregivers discourage elders from obtaining healthcare? We develop a theoretical framework in which elders may withhold information about healthcare opportunities when they care about caregivers and believe their caregivers would be pressured to help. This inefficiently reduces take-up when caregivers would help willingly. We test three model predictions using two field experiments with elder-caregiver pairs in rural Tamil Nadu, India (N=1,175). First, directly informing caregivers about a check-up opportunity for the elder increases attendance by 30% (5.3 percentage points), relative to informing only the elder, suggesting elders withhold information relevant for take-up. The effect is strongest for elders who report being worried about burden at baseline, whose attendance more than doubles. Second, worried elders are less willing to share information about the check-up and more sensitive to randomized reductions in caregiver costs than not-worried elders, suggesting worried elders believe sharing information imposes costly pressure on caregivers. Third, caregivers’ decisions to help are not sensitive to randomized observability treatments that vary pressure, suggesting caregivers willingly help out of care for the elder. Our results imply that elders' caring preferences can interact with misperceptions of caregiving burden to lead to inefficiently low health investment among older adults.
Working Papers
Self-Detrimental Avoidance of Rest (with Alexandra V. Schubert). [Draft]
Across many cultures, resting instead of working is viewed as a barrier to higher earnings. This belief is also reflected in many canonical economic models. Recent empirical evidence highlighting the productivity benefits of rest challenges this belief. Yet, existing work tends to ignore individuals’ demand for restful activities and whether it aligns with their returns. In the context of an online labor market experiment in South Africa, we explore whether workers capitalize on the returns to short rest periods. After eliciting demand for rest, we estimate returns to rest for the same individuals and find that mandated rest boosts productivity by 0.3 standard deviations, thus making up for forgone earnings from resting. At the same time, only 19% of workers voluntarily choose to rest. Contrary to the notion of selection on returns, workers with high financial returns to rest do not select into rest. We provide suggestive evidence that misperceived financial returns are driving the disconnect between demand for and returns to rest. Our results provide proof-of-concept evidence that individuals may be misallocating effort between resting and working and could reach higher overall utility by working less. This highlights the importance of understanding misperceptions around rest, especially in light of the economic burden of long-term costs of overworking such as burnout.
AI-Enhanced Handheld ECGs to Screen for Prior Myocardial Infarction in Rural India
(with Alexander Schubert, Nikhil Kanakamedala, Madeline McKelway, Luke Messac, Cyrus Reginald, Frank Schilbach, T.S. Selvavinayagam, Girija Vaidyanathan, Esther Duflo, and Ziad Obermeyer) [NBER Working Paper]
Underdiagnosis is a major barrier to effective treatment, especially in low- and middle-income countries (LMICs), where health system constraints limit access to testing. Physiological data from low-cost sensors, interpreted by artificial intelligence (AI), could enable diagnosis outside of hospitals, but training data from LMICs are scarce. Here we build an AI system that screens for prior ('silent') myocardial infarction (MI), using a new dataset from rural India pairing consumer single-lead ECGs with same-day echocardiography. The resulting model estimates likelihood of prior MI, based on cardiologist interpretation of echocardiograms, directly from ECG waveforms. It achieves an AUC of 0.77 (95% CI 0.69–0.84; vs. 0.68 for conventional risk scores), and flags a high-risk group (2.5% of the sample) with 9.3% prevalence of prior MI. Community-based screening for prior MI using this tool would cost $1,984 per life-year saved, comparable to other screening policies in India. Notably, many of the model’s high-risk patients appear low-risk by traditional criteria -- e.g., blood pressure, blood sugar, cholesterol -- highlighting the potential for physiological data to improve on risk scores from Western populations.
Community-Based Mental Health Interventions for Elderly Women in Tamil Nadu, India (with Esther Duflo, Camille Falézan, Madeline McKelway, Miriam Sequeira, Frank Schilbach, and Girija Vaidyanathan). [Available upon request]
Widespread depression among the elderly poses a significant threat to healthy aging in low- and middle-income countries, with especially high rates among elderly women. In partnership with the Government of Tamil Nadu, India, we conducted a large-scale RCT to evaluate two cross-randomized interventions aimed at improving the mental health of elderly women: individual-level cognitive behavioral therapy (CBT), and village-level group activities (GA). The interventions were delivered by trained government workers as part of their regular activities and were designed to be scalable within existing administrative structures. We find that group activities – both on their own and in combination with CBT – lead to significant improvements in depressive symptoms immediately following the intervention, with effect sizes of approximately 0.07 standard deviations (SD). In contrast, CBT alone does not generate detectable improvements. Six months later, the effect of group activities on depression increases to 0.12 SD, and we also observe improvements in functional impairment (0.1 SD) and cognition (0.08 SD). This makes group activities run by government frontline workers a cost-effective intervention to improve the health, well-being, and cognition of elderly women.
“The Value of Waiting Time in the United States: Estimates from Nationwide Natural Field Experiments” Conditionally accepted at American Economic Journal: Economic Policy [NBER Working Paper Version, Dec 2020]
(with Ariel Goldszmidt, John List, Robert Metcalfe, Ian Muir, Kerry V. Smith)
Values of time are key inputs to a variety of economic and policy decisions. We estimate the value of waiting time (VOT) via two large-scale natural field experiments run by the ridesharing company Lyft. The experiments randomly varied wait times and prices in nearly 15 million passenger sessions across 13 large United States metropolitan areas. We find that the VOT is approximately $19 per hour -- about 67% (90%) of the pre-tax mean (median) wage rate in 2015 dollars -- varies predictably with choice circumstances, and is increasing in the amount of waiting time.