Primary Submission Category: Applications in Health and Biology
When Do Observational Causal Conclusions Survive? Evidence from Patient Access to Medical Records
Authors: Minh Nguyen,
Presenting Author: Minh Nguyen*
Causal claims in healthcare often rest on observational associations whose credibility depends more on identifying assumptions than on statistical precision. We study whether access to personal medical records increases the probability that individuals use information in healthcare decisions, treating the application as a laboratory for systematic stress-testing rather than single-model confirmation. Across outcome regression, weighting, and doubly robust estimators, access is associated with increases of roughly 15–20 percentage points. Alternative adjustment sets, learner choices, and subsampling perturb magnitudes but rarely direction. Strengthening overlap attenuates effects without eliminating them. Placebo and negative-control exercises yield estimates near zero, and perceived usefulness remains associated with access even among respondents who have not acted. No diagnostic is decisive alone, but together they substantially narrow simple null explanations. At the same time, an asymmetry emerges: while the average effect is stable, mechanisms and individual prioritization are not. Specifications that leave the mean largely intact often reorganize who appears to benefit. The results suggest that observational evidence can be informative about whether access matters, yet remain underdetermined about why or for whom. We offer both new evidence on behavioral responses to information access and a framework for evaluating credibility across multiple dimensions of inferential f