Mask wearing in community settings reduces SARS-CoV-2 transmission

The effectiveness of mask wearing at controlling severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) transmission has been unclear. While masks are known to substantially reduce disease transmission in healthcare settings [D. K. Chu et al., Lancet 395, 1973–1987 (2020); J. Howard et al., Proc. Natl. Acad. Sci. U.S.A. 118, e2014564118 (2021); Y. Cheng et al., Science eabg6296 (2021)], studies in community settings report inconsistent results [H. M. Ollila et al., medRxiv (2020); J. Brainard et al., Eurosurveillance 25, 2000725 (2020); T. Jefferson et al., Cochrane Database Syst. Rev. 11, {"type":"entrez-nucleotide","attrs":{"text":"CD006207","term_id":"30322945","term_text":"CD006207"}}CD006207 (2020)]. Most such studies focus on how masks impact transmission, by analyzing how effective government mask mandates are. However, we find that widespread voluntary mask wearing, and other data limitations, make mandate effectiveness a poor proxy for mask-wearing effectiveness. We directly analyze the effect of mask wearing on SARS-CoV-2 transmission, drawing on several datasets covering 92 regions on six continents, including the largest survey of wearing behavior (n= 20 million) [F. Kreuter et al., https://gisumd.github.io/COVID-19-API-Documentation (2020)]. Using a Bayesian hierarchical model, we estimate the effect of mask wearing on transmission, by linking reported wearing levels to reported cases in each region, while adjusting for mobility and nonpharmaceutical interventions (NPIs), such as bans on large gatherings. Our estimates imply that the mean observed level of mask wearing corresponds to a 19% decrease in the reproduction number R. We also assess the robustness of our results in 60 tests spanning 20 sensitivity analyses. In light of these results, policy makers can effectively reduce transmission by intervening to increase mask wearing.

Face masks are one of the most prominent interventions against COVID-19, with very high uptake in most countries (1). However, global mask wearing fell substantially in 2021, even in countries with low vaccination rates (Fig. 1). Given ongoing epidemics, establishing the effectiveness of mask wearing in community settings is critical. The following sections review past work on the effectiveness of mask wearing in different settings and at different scales.Open in a separate windowFig. 1.Reported mask wearing in countries with <40% of population fully vaccinated, as of 1 October 2021 [wearing from the UMD/Facebook survey (1); vaccinations from ref. 2]. The y axis is the proportion who reported that, over the last week, they wore masks most or all of the time in public spaces.In the context of healthcare, N95 masks (as defined by ref. 3) work well when worn properly by trained users—reducing transmission of coronaviruses including severe acute respiratory coronavirus syndrome 2 (SARS-CoV-2) by at least half (4, 5). Cheng et al. (6) find that ideal surgical masking (7, 8) of a noninfected person corresponds to a 65 to 75% reduction in their risk of COVID-19.However, the effect of mask wearing in small-scale community settings is more difficult to detect.In particular, four meta-analyses have summarized studies on respiratory infections, conducted in community settings (4, 9–11). They estimate mean decreases in infection risk between 4% and 15% for surgical masks, but with large uncertainty: Individual results ranged from a 7% increase in infection risk to a 61% decrease in infection risk. In addition, few of these studies are randomized controlled trials (RCTs), and those that are RCTs have considerable issues: Bungaard et al. (12) found a small, nonsignificant reduction in infection risk. Abaluck et al. (13), found a significant, 8.6% decrease in symptomatic seropositivity linked to mask wearing. However, limitations of the study included a requirement for unblinded participants to self-report symptoms before testing, use of an antibody test with a very low 5 d sensitivity, and unclear generalization from the specific context (rural villages in Bangladesh).We focus on the effects of mask wearing or mandates (i.e., legal requirements to wear a mask) on transmission in large connected populations. To study mask impacts on transmission, many studies use the timing of mask mandates as a proxy for sharp changes in the level of mask wearing. Some such studies have inferred limited or inconclusive effects in cross-country analyses (14) and within-country studies (15), while others find cross-country evidence that mask mandates and recommendations lead to decreased transmission and mortality (16, 17).Other analyses provide evidence for reduced case growth following subnational mandates within countries such as the United States (18–20) and Germany (21). A potential explanation for the inconsistency and uncertainty of these results is that data on national mandate timing may be poorly suited for analyzing the effects of mask wearing on transmission.Epidemiological studies often use government mask mandates as a proxy for mask wearing. However, the existing literature on the relationship between mandates and actual levels of mask wearing has shown surprisingly weak effects. For example, studying US states, ref. 22 failed to find a statistically significant relationship between mandates and subsequent wearing, while other studies found postmandate increases in wearing of just 13% (23) and 23% (24). Betsch et al. (25) find a ∼40% increase in wearing after local mandates in Germany, but no other study finds a comparably large increase. Given that the link between mandates and wearing is surprisingly weak, it is likely that the link between mandates and transmission is difficult to detect. Three additional factors lead us to suspect that a link between mandates and transmission would be difficult to detect. First, introducing a mandate is a coarse, one-off event that necessarily loses signal by not tracking day-to-day changes in mask wearing. We also have fewer data on mandates: Less than half of the regions we study enforced any mandate during the study period. Second, past studies treat mandates as a binary on/off intervention that is fully implemented at a single point in time. However, modeling the effect of mandates as an instantaneous change in the reproduction number or mortality fails to capture changes in wearing behavior following the announcement of a mandate but before its enforcement (21). Nor does it account for gradual change in behavior after the implementation of a mandate. Finally, the circumstances of mandate policies are highly heterogeneous, both in terms of the preexisting level of voluntary wearing at the time of implementation and in terms of how exactly they are defined, enforced, and complied with. Consequently, averaging the international effect of mandates based on coarse data is unlikely to provide a useful summary of heterogeneous mandate effects. Importantly, these arguments point to the link between mandates and transmission being difficult to detect, not that it is absent.Because of these difficulties in studying the effect of mandates, we instead focus on estimating the effect of mask wearing on transmission, using a large (n = 19.97 million) global survey of self-reported mask wearing (1). Two other studies estimate mask effectiveness from self-reports: In their study of 24 countries, Aravindakshan et al. (26) use YouGov wearing data to infer an overall 3.9 to 10% relative decrease in case growth rate from whole population mask wearing. Rader et al. (22) study US states using a novel SurveyMonkey wearing dataset to infer a ∼10% decrease in transmission between the lowest and highest empirical quartiles of wearing (a 50 to 75% increase in wearing). Rader et al. use data limited to 12 US states during June–July 2020. Our data are richer: We study 56 countries on six continents, and our inferential analyses span May–September 2020.Our analysis goes further than past work in the quality of wearing data—100 times the sample size, with random sampling and poststratification—the geographical scope, the use of a semimechanistic infection model, the incorporation of uncertainty into epidemiological parameters, and the robustness of our results (59 sensitivity tests). See