EuHEA seminar: Leo Zabrocki (Paris School of Economics)

Leo Zabrocki

Abstract

Accurate and precise measurements of the short-term effects of air pollution on health play a key role in setting air quality standards. Yet, statistical power calculations are rarely—if ever—carried out. We first collect estimates and standard errors of all available articles found in the epidemiology and economics literatures. We find that nearly half of them may suffer from a low statistical power and could thereby produce statistically significant estimates that are actually inflated. We then run simulations based on real data to identify which parameters of research designs affect statistical power. Despite their large sample sizes, we show that studies exploiting rare exogenous shocks such as transport strikes or thermal inversions could have a very low statistical power, even if effect sizes are large. Our simulation results indicate that the observed discrepancy in the literature between instrumental variable estimates and non-causal ones could be partly explained by the inherent imprecision of the two-stage least-squares estimator. We also provide evidence that subgroup analysis on the elderly or children should be implemented with caution since the average number of events for an health outcome is a major driver of power. Based on these findings, we build a series of recommendations for researchers to evaluate the design of their study with respect to statistical power issues.

Discussant: Ludovica Gazze, University of Warwick

Chair: Sverre Kittelsen, Frisch Centre