Methods in Longitudinal Policy Evaluation

Abstract

This dissertation consists of three essays on methods and applications for longitudinal health policy evaluation. The first chapter studies the effects of random variation in the timing of two government benefits on hypoglycemic episodes. Households with limited savings that receive monthly pay and benefit checks often end the pay period without enough money for food. If income and benefits were distributed more frequently, households might be better equipped to smooth their consumption. I investigate the relationship between benefit distribution timing and health outcomes, focusing on older adults who are eligible for both Medicare and Medicaid. I leverage the quasi-random assignment to days on which Social Security and SNAP benefits are distributed. This enables causal conclusions about the impact of benefit timing on hypoglycemic events (emergency department visits and inpatient admissions). I find a statistically significant increase in the relative risk of hypoglycemic events during the week before benefit receipt compared to the week following receipt for beneficiaries who receive their two payments at the same time, but not for those who receive benefits two weeks apart. The effects are larger among black beneficiaries and beneficiaries living in lower-income zip codes. Tweaking the design of these programs to optimize benefit timing is an under-used policy lever to improve health outcomes.

Chapter two empirically investigates the common but often arbitrary choices researchers make when aggregating data over time in difference-in-differences (DID) studies. The choice of time aggregation, from fine (e.g., daily) to coarse (e.g., yearly), can affect the estimation of treatment effects and is currently not guided by statistical criteria. Through simulations and empirical analyses, we investigate the optimal level of time aggregation across four performance metrics and across a multitude of common scenarios research face in DID study design. We find that in the simpler cases, the choice of aggregation does not have much influence on the performance of the models; however, in more complex cases the choice of time aggregation can influence the precision and bias of estimates, especially in scenarios with staggered treatment timings and unbalanced panels. Our study highlights the importance of carefully considering time aggregation in DID studies to ensure robust and reliable results and offers guidance towards this aim.

The third chapter explores the landscape of pharmacist-prescribed contraceptive laws in the U.S. and explore the possible mechanisms of uptake of this provision among pharmacists. These laws aim to enhance the accessibility and affordability of contraceptive care, particularly for individuals facing barriers to traditional healthcare services. The COVID-19 pandemic brought about significant changes in the role of pharmacies, particularly as key points of service for vaccines, but also with broader implications for medication provision. Drawing from nationwide transactional prescription data and innovative linkage of pharmacist licensure and area-level census data, chapter three provides a comprehensive analysis of these policies across the country, with a specific focus on outcomes related to pharmacists and their prescribing trends. Results highlight that despite being a small component of hormonal contraception prescribing, it has potential to expand coverage and improve access across vast geographical areas for a broad population. However, its full utilization may be hindered by a lack of supportive policies.