The Effect of Social Media on Voters: Experimental Evidence from an Indian Election
This paper uses a field experiment to study the effect of social media on voters in a large election in Tamil Nadu, India. I randomly invited study participants to join chat groups organized by political parties on WhatsApp, India’s most-used social media platform. Unlike Facebook and Twitter, WhatsApp groups are relatively small, and users’ feeds are neither curated by algorithms nor moderated by the platform. I find that joining a group increases political knowledge, improving participants’ ability to distinguish true from false news. Moreover, the groups have an effect on political preferences, pushing participants toward the party affiliated with the WhatsApp group they joined. This effect is driven by moderates who were indifferent between parties at the baseline. To disentangle the effect of direct party messaging from the effect of messaging between individual users, I designed a second treatment arm that exposed participants to the posts of party officials but not the posts and replies of other group members. This reveals that the horizontal exchange between individuals is key: the treatment effects of party messaging alone are smaller than those of the full groups. I provide evidence on possible mechanisms underlying this difference. Participants assigned to the full groups both receive more messages and pay more attention to the messages they receive.
The Endowment Effect and Collateralized Loans (with Michael Kremer, Xinyue Lin, and Gautam Rao)
Revise and Resubmit, Econometrica
Collateral requirements play an important role in credit markets. This paper shows that the endowment effect—the phenomenon where owing a good increases one's valuation of it—inhibits demand for loans which use a borrower's existing assets as collateral. Using a field experiment in Kenya, we show that borrowers instead strongly prefer loans collateralized using the new durable assets being financed by the loans themselves. They are willing to pay 9% per month higher interest for such Same-Asset Collateralized Loans (SACLs) despite the endowed and new assets being randomized, and thus similarly valued before ownership. Our findings imply that assets which are difficult to use as collateral—which cannot be financed by SACLs—will be invested in less, even if the borrower has other collateral. We argue that borrowers' preference for SACLs is driven by naivete: they initially perceive that they have little to lose when offered a SACL, but subsequently come to develop an attachment to the new asset, resulting in high repayment effort. Consistent with this, borrowers underestimate their future attachment to an asset before owning it, and SACLs do not have higher default rates despite having higher demand. We derive the conditions under which offering consumers SACLs increases or conversely decreases borrower welfare.
Influence of Authority Figures in COVID-19 Vaccine Acceptance in Kenya (with Michael Kremer, Elisa Maffioli, Leah Rosenzweig, and Wendy Wong)
Data collection complete
The Impact of Psychological First Aid on Mental and Economic Well-Being (with Lisa Ho, Gautam Rao, and Frank Schilbach)
Data collection complete
Better Borrowing to Promote Access to Water and Improve Dairy Farming in Kenya (with Suleiman Asman, Michael Kremer, and Gautam Rao)
Data collection in progress
Time Preferences Over News: Evidence from Diabetes Testing