 Hee Kwon Seo ("Samuel")
Ph.D. Economics, Chicago Booth ('20, expected)
A.B. Applied Mathematics, Harvard ('13)
Abstract: Providing students with either "demand-side" incentives (such as performance-based rewards) or "supply-side" materials (such as books) often produces null effects; might there be complementary returns to providing both, and if so, why? In a three-year field experiment with 170 high schools in Tanzania, where mathematics pass rates remain below 20 percent nationally, students were provided with (1) money pegged to math test scores; (2) technologies to ease effort costs of learning; or (3) both of the above. Money or technologies alone make limited impact on test scores, while both together make a large, complementary effect, especially on the scores of students just below the top 20 percent (0.3σ), revealing an inverse-U-shaped relationship between entering grade-level performance and treatment complementarity. I first present a pre-specified benchmark model in which heterogeneous students balance achievement returns against strictly convex effort costs. I then generalize the model to allow minimum interest and knowledge thresholds: students not interested enough or prepared enough to learn have unproductive effort, possibly discouraged by large "entry" costs of learning new material on the curriculum. The thresholds model, structurally estimated based on detailed surveys of study habits matched to outcomes, generates the treatment-effect patterns while the benchmark model does not. Counterfactual simulations suggest that both providing the experimental technologies and motivating the students by doubling their chance of promotion would induce a modest but meaningful endogenous response of student knowledge, by reducing the share of students who are giving up on learning new material from 79 percent to 52. By explaining treatment complementarities, the proposed analytical framework extends models of classroom learning and welfare implications of policies outlined in the previous literature to reflect a higher degree of realism about developing community contexts. In particular, although the cost of providing technologies through the program is lower than the estimated cost of asking students to attain the equivalent preparation on their own, taking into account the effort costs associated with the "entry" portion of individual learning curves reduces benchmark estimates of the interventions' revealed-preference-based welfare impacts by more than two thirds.
© 2020 Hee Kwon Seo
|
