Masking the Identification Problem: How Time Aggregation Manufactures Cross-Country Evidence

A Comment on Tausk and Spira (BMC Public Health, 2025)

Abstract

Ecological regressions that relate population averages to health outcomes are attractive be- cause they use readily available data, but they are vulnerable to sign reversals when aggregation obscures heterogeneity and timing. Tausk and Spira (2025) report a positive cross-country as- sociation between average mask usage and age-adjusted excess mortality across 24 European countries during 2020–2021. I reproduce this result and show that it does not survive a minimal robustness check. A cross-sectional regression on two-year averages yields the published positive slope, but when the same data are differenced within countries across the two years, the esti- mated association becomes negative. The reversal is consistent with time aggregation loading on persistent between-country differences and shared pandemic shocks, while the first-differences coefficient captures within-country changes after removing country heterogeneity and common year factors. Neither coefficient identifies the causal effect of masking. However, the sign rever- sal demonstrates that the reduced-form relationship is not robust to a standard re-specification, which undermines any policy interpretation of the original result.