Michael J. Maloney, Nathaniel J. Rhodes & Paul R. Yarnold
Proof School, Midwestern University & Optimal Data Analysis LLC
SARS-CoV-2 is the beta-coronavirus responsible for COVID-19. Facemask use has been qualitatively associated with reduced COVID-19 cases, but no study has quantitatively assessed the impact of government mask mandates (MM) on new COVID-19 cases across multiple US States. We used a non-parametric machine-learning algorithm to test the a priori hypothesis that MM were associated with reductions in new COVID-19 cases. Publicly available data were used to analyze new COVID-19 cases from 37 States and the District of Columbia (i.e., “38 States”). We conducted confirmatory All-States and State-Wise analyses, validity analyses [e.g., leave-one-out (LOO) and bootstrap resampling], and covariate analyses. No statistically significant difference in the daily number of new COVID-19 infections was discernable in the All-States analysis. In State-Wise LOO validity analysis, 11 States exhibited reductions in new COVID-19, and reductions in four of these States (AK, MA, MN, VA) were statistically significant in bootstrap resampling. Only the Social Capital Index predicted MM success (training p<0.028 and LOO p<0.013). Results obtained when studying the impact of MM on COVID-19 cases varies as a function of the heterogeneity of the sample being considered, providing clear evidence of Simpson’s Paradox and thus of confounded findings. As such, studies of MM effectiveness should be conducted on disaggregated data. Since transmissions occur at the individual rather than at the collective level, additional work is needed to identify optimal social, psychological, environmental, and educational factors which will reduce the spread of SARS-CoV-2 and facilitate MM effectiveness across diverse settings.