Study population and setting
This study related US state-level self-reported mask use to transmission control, as measured by an estimate of the reproduction number (Rt) of SARS-CoV-2. Mask use was assessed with a survey delivered by the SurveyMonkey.com online platform and answered by 378,207 individuals 13 years or older between June 3 and July 27, 2020. Results were analyzed as unweighted data and by weighting for age, race, sex, education, geography, and political affiliation to reflect the composition of the U.S. population. Respondents were asked how likely they were to wear a mask “while grocery shopping” or “while visiting with family or friends in their homes,” on a four-point scale ranging from “very likely to “not at all.” A binary classification of mask-wearing was created, defined as responding “very likely” to both questions. Individual mask use data were then aggregated at the state-level each week: this was the primary exposure measurement of mask community use used in subsequent models. Logistic regression models were fit to aggregated state-level weekly estimates of Rt that were dichotomized as either below or above 1 (epidemic slowing vs. epidemic growing). Models were adjusted for several possible confounding variables, including physical distancing (defined by state-level weekly time spent at home relative to a baseline period, measured with aggregated mobility data from Google), state population density, proportion of non-white respondents, and a linear time trend.
Summary of Main Findings
A high proportion (84.7%) of respondents reported that they would be very likely to wear masks at the grocery store, while only 40.2% reported they would be very likely to wear masks while visiting family and friends; 39.8% answered “very likely” to both questions. Self-reported mask use increased linearly with age and was higher among women, nonwhite people, and people with lower income. Mask use varied considerably by geographic region, and was highest on both coasts, along the southern border, and in urban areas. In multivariable logistic regression, self-reported mask use was associated with transmission control (defined as Rt <1): a 10% increase in mask use had an estimated odds ratio for epidemic control of 3.53 (95% CI: 2.03 to 6.43). Results were broadly similar, though attenuated in some instances, under other assumptions including alternative Rt estimates, an alternative definition of mask wearing, dichotomizing Rt at different thresholds, and using self-reported community contacts rather than mobility data. A separate analysis found no association between state-level mask mandates and changes in self-reported mask use.
This study employed self-reported mask-wearing behavior, rather than mask policies, as the exposure of interest. The sample size was large and was weighted to match the distribution of some US demographic variables. The authors performed a range of sensitivity analyses.
Self-reported mask-wearing behavior is subject to bias; for example, respondents may have provided answers in line with perceived social desirability, and this may have occurred differently across geographic regions. The survey was administered via a web platform, and thus respondents are more likely to have internet access than the broader U.S. population, and may have been non-representative in other ways. Moreover, those who responded to the survey may have systematically differed from those who did not respond, which is particularly concerning given that the response rate was only 11%. The measure of dichotomous mask-wearing was crude and may have ignored meaningful gradations in behavior. The outcome measure was similarly crude, and more problematic, since the time-varying reproduction number at the state level is determined by a heterogeneous array of factors, many of which relate to geographically specific transmission dynamics. It is unlikely that the potential confounding variables included in the model (including a linear time trend) adequately accounted for determinants of Rt that may also be related to self-reported mask use. Lastly, self-reported mask use may have been affected by characteristics of local epidemics that were not entirely accounted for by adjusting for prior peak Rt (e.g., test positivity rate, local hospital capacity, local social distancing policies).
Many prior ecological studies of mask-wearing effectiveness relied on mask mandates and policies; this is one of the few to measure self-reported mask use at a large scale.
This review was posted on: 19 February 2021