Essays in Causal Inference

Sahil Patil

This dissertation studies how institutional investments, market-based funding mechanisms, and adaptive strategic behavior shape economic and social outcomes. Across three essays, I apply causal inference and computational methods to examine how public and private interventions influence behavior in settings ranging from local public goods provision to entrepreneurial finance and algorithmic competition. The first essay evaluates whether large, one-time capital investments in public libraries reduce juvenile delinquency. I construct a county--year panel for the United States spanning 1997--2019 by merging Juvenile Court Statistics with the Public Library Survey. Exploiting staggered adoption of high capital investment (HCI) projects---defined as library spending of at least $100 per juvenile following a year with less than $50 per juvenile. I implement a difference-in-differences design that accounts for heterogeneous treatment timing. The estimates indicate small effects: on average, HCI is associated with a decline in the juvenile delinquency rate of approximately 0.16 percent (less than 0.04 standard deviations), with event-study evidence suggesting that any reductions emerge only one to two years after investment. Results are robust to alternative control groups and inverse propensity score weighting. Overall, the findings suggest that large library capital infusions do not substantially reduce juvenile delinquency in the short run, and any longer-run effects are limited in magnitude. The second essay shifts from public investment to private capital access and examines the causal effect of acquiring crowdfunding finance on subsequent entrepreneurial outcomes, with a focus on gender heterogeneity. Using data from Kickstarter, I employ a regression discontinuity design around the campaign funding threshold to address non-random selection into successful fundraising. The results show that completing a first Kickstarter campaign increases post-campaign fundraising by approximately $3,185, or about 32 percent of mean future funding within a $500 bandwidth. Disaggregated estimates reveal positive effects for both male and female entrepreneurs---$3,066 for men and $2,757 for women in the baseline specification---while gender differences are economically modest and sensitive to bandwidth choice. The findings are robust across alternative kernels, functional forms, and bandwidths, with larger but directionally consistent estimates at higher bandwidths. The third essay studies strategic interaction in markets governed by adaptive algorithms. Using a Q-learning framework in a Cournot duopoly, I analyze how interaction frequency, asymmetries in firm flexibility, and learning speed shape market outcomes prior to equilibrium. The results show that convergence to supracompetitive profits through reduced quantities is most pronounced at intermediate interaction frequencies: infrequent interactions delay stabilization, while very frequent interactions yield slightly higher, more competitive quantities. Capacity asymmetries weaken coordination incentives and generate persistent payoff differences across firms. Faster exploration leads to early stabilization at higher quantities, whereas slower learning promotes more collusive outcomes under exploit-only terminal phases. Allowing periodic re-exploration temporarily disrupts coordination but does not eliminate it. These findings demonstrate that emergent collusion depends not only on algorithmic design but also on economic features such as market structure and interaction cadence that govern how adaptive agents internalize strategic feedback. Together, the essays highlight how the effectiveness of investments, incentives, and strategic learning depends critically on institutional context and dynamic adjustment processes. By integrating quasi-experimental and computational approaches, this dissertation contributes new evidence on the limits of public capital investments, the returns to early access to entrepreneurial finance, and the conditions under which adaptive agents coordinate in competitive markets.

Essays in Causal Inference

Files and links (1)

Abstract

Metrics

Details