key: cord-0909206-m0l4zo0h
authors: Wallace, Jacob; Lollo, Anthony; Ndumele, Chima D.
title: Evaluation of the Association Between Medicare Eligibility and Excess Deaths During the COVID-19 Pandemic in the US
date: 2021-09-24
journal: JAMA Health Forum
DOI: 10.1001/jamahealthforum.2021.2861
sha: a0fd43ce672d304fe3e8feba37539e71ba73ba6a
doc_id: 909206
cord_uid: m0l4zo0h

This cross-sectional regression discontinuity analysis compares deaths slightly younger and older than 65 years to examine the relationship between access to health insurance coverage and COVID-19 mortality.

Our preferred specification is a regression discontinuity (RD) design with a quadratic age trend and a 4 year bandwidth around age 65. However, as is common in RD, we assess the sensitivity of our primary results to alternative bandwidths and statistical models. The table below reports primary results alongside alternative specifications. Each column indicates a different approach to specifying the bandwidth around the discontinuity at age 65. The rows are split by time periods, with each of the models (i.e., "Linear", "Quadratic", and "Local Linear") indicating a different approach to modeling the age trend of death counts in our data. The table above reports the sensitivity of regression discontinuity results to alterations of the statistical model, including changes in our bandwidth and differences in functional form. Robust (i.e., "Local Linear") estimates rely on the Calonico, Cattaneo, and Titiunik (2014) package, 1 which uses a data-driven process to select the optimal bandwidth and construct bias-corrected confidence intervals.

Below we report a second set of robustness tests, where we alter the time period or the dependent variable. The estimates in the table below are based on our preferred specification, a quadratic age trend (allowed to vary on both sides of the discontinuity) with a 4-year bandwidth. 

Robust nonparametric confidence intervals for regressiondiscontinuity designs