The Effectiveness Of Government Masking Mandates On COVID-19 County-Level Case Incidence Across The United States, 2020

As COVID-19 emerged in the US in early 2020, federal, state, and local authorities responded in a variety of ways to try to curb community transmission of the SARS-CoV-2, the coronavirus that causes COVID-19. One of these responses was the wearing of face coverings, or masks. By April 2020 the wearing of masks in public became the recommendation of the Centers for Disease Control and Prevention (CDC), but the federal government did not issue a mandate. Subsequently, many states and counties established their own mandates of who should wear masks and when, and how public masking would be enforced. Between April and September 2020, the timing of these masking mandates was highly variable throughout the country. By the fall of 2020, guidance from local health authorities converged so that universal masking in public locations, alongside recommendations for social distancing, had become the norm in most regions.

The evidence for the effectiveness of masking at the individual level has been strong^1–4 for viruses that, similar to SARS-CoV-2, spread principally through close contact with an infected person.^5–9 However, the incremental benefit of community-level mandates beyond public health guidance has been less certain. Studies that have assessed the influence of community-directed masking mandates often compare states or countries that are quite heterogeneous in demography.^10–12 Smaller studies of county-level masking mandates have relied on pre-post comparisons of case incidence¹³ or have compared counties without adjusting for the propensity to mandate^14–17—approaches that may be fraught with bias from secular trends¹⁸ and unmeasured confounding.

Although randomized clinical trials, considered the “gold standard” for evaluating intervention effects, might not be practical for studying the effect of masking mandates, the variation in timing of the enactment of masking mandates at the county level during the early period of the pandemic offers an opportunity to assess the impact of county-level mandates on subsequent case incidence via a pragmatic quasi-experimental comparative effectiveness study. In this observational study we strove to emulate a trial that randomly assigned counties to a masking mandate versus no mandate by using a matched study design¹⁹ to compare the effectiveness of county-level masking mandates on subsequent county-level case incidence during the period of April–August 2020. The primary objective was to determine the impact of masking mandates on the trajectory of county-level COVID-19 case incidence and to further assess the impact of masking mandates across regions of different urbanicity and political leaning.

Study Data And Methods

County Eligibility And Case Incidence Data

Data on COVID-19 county-level case incidence for all US counties were obtained from USAFacts.org.²⁰ To create a representative sample of US areas affected by the pandemic, we designed the eligibility criteria for this analysis to include either densely populated areas or counties with moderate-to-substantial community transmission, including areas affected by the second wave of the pandemic in the summer of 2020. Starting in April 2020, thirty-eight US states and Washington, D.C., issued state-level masking mandates, and thirty-four (nearly 90 percent) of them started before September 2020.²¹ To be considered eligible, a county had to have at least three consecutive days of daily case count exceeding five as of August 31, 2020, and meet at least one of the following criteria: contain at least one city with a population exceeding 100,000 people, contain a state capital, be the most populated county in the state, or have an average daily case incidence that exceeded twenty during July 1–August 31, 2020. The incidence threshold of twenty was selected on the basis of empirical distribution of case counts in the observation period. A total of 569 counties, representing forty-six states and Washington, D.C., were eligible for this study.

Exposure

The exposure of interest was the date of initiation of a public masking mandate for a county, issued by either the state or county government (time zero). These data were captured from a review of local news reports, executive orders, the COVID Analysis and Mapping of Policies database at Georgetown University,²² and the New York Times state-level reopening tracker.²³ For a county to be considered as exposed for this analysis (that is, under a masking mandate), the mandate order needed to be posted between April 1 and August 31, 2020, and—at minimum—require that masks be worn by employees and customers inside public-facing buildings. If the state or county did not have a masking mandate but the largest city in the county had one, that county was deemed to have a masking mandate for this analysis. The first and last masking mandates among the 569 counties were observed April 4 and August 25, 2020. Case incidence data were collected from March 21 to October 20, 2020, to allow for a preintervention observation period of a minimum of two weeks and a postintervention period of at least eight weeks.

Matching Algorithm

Each exposed county was matched one to one with a county from a risk set of unexposed counties that was assembled from within the same region—but not within the same metropolitan area—as the exposed county, based on information from the Bureau of Economic Analysis.²⁴ To be eligible to serve as an unexposed county match to an exposed county, a county had to have no masking mandates at least three weeks past the date of the mandate for the exposed county. When more than one unexposed county was available from a risk set, a nearest-neighbor matching was used to select the unexposed match on the basis of the following variables: population density, total population, presidential election voting patterns in 2016, and case incidence and instantaneous reproduction number (${\mathrm{R}}_{\mathrm{t}}$) in the two weeks before time zero. The matching procedure proceeded chronologically through each exposed county to identify a corresponding risk set of potential unexposed counties. Because risk sets were defined on the basis of the timing of making mandates, unexposed counties could be matched to multiple exposed counties, and an unexposed county could later become an exposed county (online appendix exhibit S1).²⁵

Population density and total population were log-transformed before analysis because of the significant skewness for the largest cities. County-level daily incidence (cases per 100,000) and ${\mathrm{R}}_{\mathrm{t}}$ were categorized as the twenty-fifth, fiftieth, and seventy-fifth percentiles for the two weeks before time zero.^26–28${\mathrm{R}}_{\mathrm{t}}$ was estimated using the method of Anne Cori and colleagues,²⁶ with a moving average window of three days, as commonly used in models of transmission of SARS-CoV-2 and other communicable diseases.^29–34 The generation time was assumed to follow a gamma distribution with a mean of 7.5 days and standard deviation of 3.4 days, according to a previous epidemiological survey of the first 425 COVID-19 cases in Wuhan, China.³⁴ Matching on these variables was based on the nearest Euclidean distance.^35,36 All matching variables were standardized by mean and standard deviation to give them equal weight.

Outcome

The primary outcome was the county-level daily COVID-19 case incidence over the course of six weeks after masking mandate initiation. The daily incidence (cases per 100,000) was captured from time zero to six weeks postmandate (primary and secondary analyses) and from time zero to eight weeks postmandate (sensitivity analysis in a subset of counties). The reported daily incident COVID-19 case counts were smoothed using a three-day rolling average to reduce extra noise that was a result of batch reporting.

Covariates

County-level covariates with the potential to confound the analysis results were considered for each model, including social distancing, population density, wet-bulb temperature, proportion of county residents with diabetes, and proportion of county residents earning less than 200 percent of the federal poverty level. Social distancing was quantified using daily cellphone movement provided by Unacast (which collects and aggregates human mobility data),³⁷ measuring the percentage change in visits to nonessential businesses within each county compared with visits in a four-week prepandemic baseline period between February 10 and March 8, 2020. Regarding wet-bulb temperature, it has been demonstrated that humidity and temperature both play a role in the seasonality of influenza, and earlier studies also reported significant effect of temperature and humidity on SARS-CoV-2 transmission.^34,38,39 For this reason, we used wet-bulb temperature, a metric that captures the complex thermodynamic relationship of temperature and humidity, has been shown to predict human health events with more precision than temperature and humidity separately, and avoids the associated problem of collinearity.^40,41 We used a rolling average of the social distancing metric and wet-bulb temperatures from four to fourteen days before case identification to account for the incubation period of COVID-19.³⁴ Nonlinear effects for wet-bulb temperatures, obtained from National Oceanic and Atmospheric Administration Local Climatological Data,⁴² were included, as described in previous work modeling the effect of climate on SARS-CoV-2 transmission.³⁴ Diabetes and poverty covariates were abstracted from the Census Bureau’s American Community Survey,⁴³ Behavioral Risk Factor Surveillance System,⁴⁴ ArcGIS Business Analyst,⁴⁵ and Multi-Resolution Land Characteristics Consortium.⁴⁶ County-level covariates were standardized by mean and standard deviation before analysis.

Primary Analysis

Data were assessed for outliers relative to the ${\mathrm{R}}_{\mathrm{t}}$ before analysis. Days with an ${\mathrm{R}}_{\mathrm{t}}$ value outside of the 2.5–97.5 percentiles (${\mathrm{R}}_{\mathrm{t}}$ of 0.35 at the 2.5 percentile and 3.50 at the 97.5 percentile) were excluded from the analysis. Generalized linear mixed effects models were fit in a matched pair analysis to evaluate the association between exposures and outcomes, using a log link function while adjusting for covariates. Hierarchical random intercept and slope were used to account for correlations within the matched group and each individual county separately. We modeled the nonlinear trajectories of disease transmission over time, using cubic B-spline functions with no interior knots, and we compared trajectories between exposed and unexposed counties, using interactions between time and exposure. The time variable was defined as the difference between the date of case identification and the date of onset for the masking mandate. The adjusted ratios of moment-in-time case counts were estimated at two, four, and six weeks and averaged over the course of six weeks after mandate initiation.

Secondary Analysis

The ratio of daily case counts was also assessed across two subgroups. The first subgroup analysis examined counties by presidential voting behavior in the 2016 election,⁴⁷ contrasting counties that voted majority Republican (more than 50 percent of the vote went to the Republican presidential candidate) versus counties voting majority Democrat (up to 50 percent of the vote went to the Republican candidate). The second examined counties with high population density, which included more urban settings with 200–2,000 people per square mile, versus those with low population density, which included the most rural or suburban settings in which population density was lower than 200 people per square mile.

Sensitivity Analysis

Our analysis of model robustness considered different criteria for selecting unexposed counties during matching. First, we removed pairs whose intercounty distance was greater than 1,000 miles to reduce heterogeneity between counties (reducing our sample by 9.1 percent of matched pairs), as well as counties that were within 100 miles of each other to reduce interdependence of effects within metropolitan areas (reducing our sample by 9.0 percent of matched pairs). Second, we examined the additional longitudinal effects of extending the postmandate period to eight weeks. For this analysis, the minimum time to be without a masking mandate to be considered an unexposed county was extended from three to five weeks after mandate inception in the exposed county. We also added constraints to matching with replacement to ensure that a selected unexposed county could not be chosen within eight weeks of its inclusion in the study. By doing this, we could still use a county multiple times as a control, but there were no overlapping periods included within counties. This analysis reduced the matched pairs to 45.7 percent of the original matched pairs.

Limitations

Our approach was not without limitations. First, unobserved heterogeneity across counties may have contributed to residual confounding. For example, exposed and unexposed counties could have had differential rates of local gathering events or different nonpharmaceutical interventions such as business closures. However, a tipping-point analysis,^48,49 described in the appendix,²⁵ revealed that the magnitude of effect would have needed to be quite large to overturn the results. It would also be difficult to invoke other community or state mitigation strategies as potential confounders, given that many businesses were reopening during this period but schools had not yet opened across the country. To the degree to which our measurement of social distancing did not fully capture the impact of business reopenings during this period, we would have biased results to the null. Second, exposed and unexposed counties could have sustained differential SARS-CoV-2 transmission rates related to higher or lower incidence rates in neighboring counties and regions. However, our findings were robust in sensitivity analyses that excluded matched pairs that were either close or distant in proximity. Third, it is possible that unexposed counties could have instituted a masking mandate in the latter days of the follow-up period, which could have biased estimates toward the null. In the sensitivity analysis that extended the follow-up period to eight weeks postmandate, we extended the criterion for no mask requirement in the unexposed county from three to five weeks; the analysis revealed similar findings. Finally, our results might not be transportable to counties with smaller population densities, given that most of the analyzed counties have higher population density, or generalizable to later periods, when higher population immunity was attained through either natural infection or acquired vaccination.

Study Results

The matching procedure produced 328 matched county pairs identified across 394 unique counties (55.2 percent of the US population). A list of the matched pairs and daily COVID-19 case incidence per 100,000 people during the two weeks before masking mandates in the mandate groups is in exhibits S2 and S3.²⁵Exhibit 1 demonstrates the distribution of baseline variables between matched counties. Imbalance was still observed after matching. Variables that had more than 0.1 standardized difference were included in the regression model as baseline covariates. Estimates of regression coefficients are in appendix section A4.²⁵

Primary Analysis

Across time, the estimated daily case incidence per 100,000 people in masked counties was 77 percent of case incidence in unmasked counties (95% confidence interval: 69, 86) at four weeks and 67 percent (95% CI: 58, 74) at six weeks postintervention (exhibit 2 and appendix exhibit S4).²⁵ Aggregated across all six weeks, the average case incidence in masked counties was 84 percent (95% CI: 77, 92) of the case incidence experienced in unmasked counties. Exhibit 3 shows predicted postestimation trajectories of daily case counts per 100,000 people over the period of two weeks premandate to six weeks postmandate, conditioned on all counties having or not having the mandate. Time to peak incidence was reduced (thirteen versus twenty-four days between counties with mandate and those without), alongside a widening difference in daily case incidence over time in the setting of masking mandates.

SOURCES USAFacts, US COVID-19 cases and deaths by state (see note 20 in text); COVID Analysis and Mapping of Policies, Data access (see note 22 in text); and the New York Times state-level reopening tracker (see note 23 in text). NOTES The analysis included all of the analysis windows from 394 unique US counties between April and August 2020. “Without mandate” assumes that a mandate would not have been implemented, and “With mandate” assumes that a mandate would have been implemented. Time is aligned at the intervention time (time zero). Solid lines indicate medians. Shaded areas indicate interquartile ranges. The plot of the mean is in appendix exhibit S5 (see note 25 in text).

Secondary Analyses

Across subgroups, the daily case incidence per 100,000 people among Republican-leaning counties with a masking mandate was reduced to 69 percent (95% CI: 58, 81) and 62 percent (95% CI: 51, 75) relative to Republican-leaning counties without a masking mandate at four and six weeks after onset of the intervention, respectively (exhibit 2 and appendix exhibit S4).²⁵ The daily case incidence per 100,000 people among Democrat-leaning counties with a masking mandate was reduced to 84 percent (95% CI: 71, 99) and 70 percent (95% CI: 57, 86) relative to Democrat-leaning counties without a masking mandate at four and six weeks after intervention, respectively.

Urban counties were also uniquely influenced by masking mandates compared with more suburban and rural counties. The daily case incidence per 100,000 people for urban counties with a masking mandate was reduced to 65 percent (95% CI: 58, 74) and 52 percent (95% CI: 45, 60) relative to urban counties without a masking mandate at four and six weeks after intervention, respectively. This compares with moment-in-time effects among rural and suburban counties of 94 percent (95% CI: 75, 119) and 122 percent (95% CI: 95, 159) at four and six weeks after intervention, respectively. Postestimation adjusted daily case incidence per 100,000 people across subgroups is graphically presented in appendix exhibit S5.²⁵

Sensitivity Analyses

Findings remained robust to more stringent inclusion and exclusion criteria and matching procedures (appendix exhibit S6).²⁵ Extending the follow-up window to eight weeks also revealed similar findings, albeit in a smaller group of 178 matched pairs from 334 unique counties that were somewhat more densely populated, were from colder regions, had more significant social distancing, and had lower baseline transmission rates than the original sample (appendix exhibit S7).²⁵ At eight weeks of follow-up, counties with a masking mandate had a daily case incidence per 100,000 individuals that was reduced to 51 percent (95% CI: 42, 63) of the daily case incidence per 100,000 people in counties without masking mandates (exhibit 4, appendix exhibits S8 and S9).²⁵ This was slightly less than the reduction noted at six weeks from this analysis (50 percent; 95% CI: 42, 63). The attenuation between six and eight weeks was most pronounced in Republican-leaning counties and in suburban and rural counties with population density below 200 people per square mile (exhibit 4, appendix exhibit S8).²⁵

Discussion

In this matched comparison of US counties with and without masking mandates in the early months of the COVID-19 pandemic in the US, the mandates were associated with reduced case incidence six weeks after the onset of the mandates. Reduction in case incidence ranged from −5 percent to 27 percent during the six-week interval (exhibit 2), with peak impacts between three and four weeks after the initiation of mandates. The benefit of masking mandates was observed across all subgroups except for suburban and rural counties, potentially as a result of insufficient sample size. The effect was strongest among urban counties, where reductions, by inference, reached 27 percent (95% CI: 18, 34) by six weeks, and among Republican-leaning counties, where reductions, by inference, reached 24 percent (95% CI: 12, 23) by six weeks. Findings in the study were robust to a sensitivity analysis that extended the follow-up period to eight weeks, in which similar findings were observed but a possible attenuation of effect was detected between six and eight weeks in some counties. These latter effects were observed across all subgroups and were most profound in Republican-leaning and rural and suburban counties, whose moment-in-time risk reduction attenuated from 52 percent to 44 percent and 11 percent to 1 percent between six and eight weeks, respectively.

The design of this study, which matched on baseline community transmission and demographics, was a major strength. The study simulated a pragmatic trial by which to estimate the incremental effect of mandates on community COVID-19 transmission. A large observational study of counties across the United States in 2021 also suggested that masking mandates were beneficial in reducing SARS-CoV-2 transmission,¹⁰ but the approach in our study was arguably stronger, as it reduced the potential influence of secular trends. By matching unexposed and exposed counties, this study reduced potential observed confounding, and the county-level random effects used in the analysis also allowed estimation of between-county heterogeneity that was not explained by the observed variables. Another challenge in this study is that we were unable to examine the factors that mediate the benefit of public masking mandates. It is possible, for example, that mandates led to higher rates of sustained masking use in public locations; unfortunately, there were no longitudinal surveys with sufficient sampling across the counties to examine this influence, including whether fatigue over time may have contributed disproportionately to the attenuation in effects observed among some Republican-leaning and rural or suburban counties at eight weeks. This may make it more difficult as well to translate these findings to later periods of the pandemic, when fatigue with public health recommendations was likely heightened and when more transmissible variants (such as Delta and Omicron) were circulating across the population. It is also possible that mandates signal to the community that risk throughout the region is rising and thus may—at least temporarily—affect practices of social distancing or proclivity to stay home among worried residents. In that regard, the degree to which the social distancing measure in this analysis approximated individual behavior more broadly might also have led us to overfit models and somewhat bias results to the null. Finally, benefits of mandates may also vary by commitment of law enforcement, a variable we were unable to study.

Conclusion

Masking mandates likely conferred benefit in reducing community transmission rates and case incidence during the initial months of the pandemic.

This matched analysis, during a period of variable public masking orders across the US, suggests that on average, masking mandates likely conferred benefit in reducing community transmission rates and case incidence during the initial months of the COVID-19 pandemic. Although such benefits were not equally distributed in all regions, it appears that masking mandates may offer broad value in reducing community risk during periods of elevated SARS-CoV-2 transmission in the US.

ACKNOWLEDGMENTS

Support for this research was received from the National Institutes of Health (Grant Nos. R01CA-178744, R01CA-244845, and TR-001878). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. [Published online February 16, 2022.]

NOTES

• 1 Doung-Ngern P , Suphanchaimat R , Panjangampatthana A , Janekrongtham C , Ruampoom D , Daochaeng N et al. Case-control study of use of personal protective measures and risk for SARS-CoV 2 infection, Thailand. Emerg Infect Dis. 2020;26(11):2607–
  16. Crossref, Medline, Google Scholar
• 2 Brooks JT , Butler JC . Effectiveness of mask wearing to control community spread of SARS-CoV-2. JAMA. 2021;325(10):998–
  9. Crossref, Medline, Google Scholar
• 3 Howard J , Huang A , Li Z , Tufekci Z , Zdimal V , Van Der Westhuizen H-M et al. An evidence review of face masks against COVID-19. Proc Natl Acad Sci. 2021;118(4):e2014564118. Crossref, Medline, Google Scholar
• 4 Bundgaard H , Bundgaard JS , Raaschou-Pedersen DET , von Buchwald C , Todsen T , Norsk JB et al. Effectiveness of adding a mask recommendation to other public health measures to prevent SARS-CoV-2 infection in Danish mask wearers: a randomized controlled trial. Ann Intern Med. 2021;174(3):335–
  43. Crossref, Medline, Google Scholar
• 5 Ueki H , Furusawa Y , Iwatsuki-Horimoto K , Imai M , Kabata H , Nishimura H et al. Effectiveness of face masks in preventing airborne transmission of SARS-CoV-2. mSphere. 2020;5(5):e00637–
  20. Crossref, Medline, Google Scholar
• 6 Centers for Disease Control and Prevention, National Center for Immunization and Respiratory Diseases, Division of Viral Diseases. Science brief: community use of cloth masks to control the spread of SARS-CoV-2 [Internet]. Atlanta (GA): CDC; 2020 [updated 2021 Dec 6; cited 2022 Jan 26]. Available from: https://www.cdc.gov/coronavirus/2019-ncov/science/science-briefs/masking-science-sars-cov2.html Google Scholar
• 7 Nishimura H , Sakata S , Kaga A . A new methodology for studying dynamics of aerosol particles in sneeze and cough using a digital high-vision, high-speed video system and vector analyses. PLoS One. 2013;8(11):e80244. Crossref, Medline, Google Scholar
• 8 Smither SJ , Eastaugh LS , Findlay JS , Lever MS . Experimental aerosol survival of SARS-CoV-2 in artificial saliva and tissue culture media at medium and high humidity. Emerg Microbes Infect. 2020;9(1):1415–
  7. Crossref, Medline, Google Scholar
• 9 Liu Y , Ning Z , Chen Y , Guo M , Liu Y , Gali NK et al. Aerodynamic analysis of SARS-CoV-2 in two Wuhan hospitals. Nature. 2020;582(7813):557–
  60. Crossref, Medline, Google Scholar
• 10 Lyu W , Wehby GL . Community use of face masks and COVID-19: evidence from a natural experiment of state mandates in the US. Health Aff (Millwood). 2020;39(8):1419–
  25. Go to the article, Google Scholar
• 11 Gallaway MS , Rigler J , Robinson S , Herrick K , Livar E , Komatsu KK et al. Trends in COVID-19 incidence after implementation of mitigation measures—Arizona, January 22–August 7, 2020. MMWR Morb Mortal Wkly Rep. 2020;69(40):1460–
  3. Crossref, Medline, Google Scholar
• 12 Kaufman BG , Whitaker R , Mahendraratnam N , Smith VA , McClellan MB . Comparing associations of state reopening strategies with COVID-19 burden. J Gen Intern Med. 2020;35(12):3627–
  34. Crossref, Medline, Google Scholar
• 13 Adjodah D , Dinakar K , Chinazzi M , Fraiberger SP , Pentland A , Bates S et al. Association between COVID-19 outcomes and mask mandates, adherence, and attitudes. PLoS One. 2021;16(6):e0252315. Crossref, Medline, Google Scholar
• 14 Van Dyke ME , Rogers TM , Pevzner E , Satterwhite CL , Shah HB , Beckman WJ et al. Trends in county-level COVID-19 incidence in counties with and without a mask mandate—Kansas, June 1–August 23, 2020. MMWR Morb Mortal Wkly Rep. 2020;69(47):1777–
  81. Crossref, Medline, Google Scholar
• 15 Karaivanov A , Lu SE , Shigeoka H , Chen C , Pamplona S . Face masks, public policies, and slowing the spread of COVID-19: evidence from Canada. J Health Econ. 2021;78:102475. Crossref, Medline, Google Scholar
• 16 Guy GP , Lee FC , Sunshine G , McCord R , Howard-Williams M , Kompaniyets L et al. Association of state-issued mask mandates and allowing on-premises restaurant dining with county-level COVID-19 case and death growth rates—United States, March 1–December 31, 2020. MMWR Morb Mortal Wkly Rep. 2021;70(10):350–
  4. Crossref, Medline, Google Scholar
• 17 Dasgupta S , Kassem AM , Sunshine G , Liu T , Rose C , Kang GJ et al. Differences in rapid increases in county-level COVID-19 incidence by implementation of statewide closures and mask mandates—United States, June 1–September 30, 2020. Ann Epidemiol. 2021;57:46–
  53. Crossref, Medline, Google Scholar
• 18 Brauner JM , Mindermann S , Sharma M , Johnston D , Salvatier J , Gavenčiak T et al. Inferring the effectiveness of government interventions against COVID-19. Science. 2021;371(6531):eabd9338. Crossref, Medline, Google Scholar
• 19 Dagan N , Barda N , Kepten E , Miron O , Perchik S , Katz MA et al. BNT162b2 mRNA Covid-19 vaccine in a nationwide mass vaccination setting. N Engl J Med. 2021;384(15):1412–
  23. Crossref, Medline, Google Scholar
• 20 USAFacts. US COVID-19 cases and deaths by state [Internet]. Seattle (WA): USAFacts; [last updated 2022 Jan 18; cited 2022 Jan 5]. Available from: https://usafacts.org/visualizations/coronavirus-covid-19-spread-map/ Google Scholar
• 21 Guy GP , Lee FC , Sunshine G , McCord R , Howard-Williams M , Kompaniyets L et al. Association of state-issued mask mandates and allowing on-premises restaurant dining with county-level COVID-19 case and death growth rates—United States, March 1–December 31, 2020. MMWR Morb Mortal Wkly Rep. 2021;70(10):350. Crossref, Medline, Google Scholar
• 22 COVID Analysis and Mapping of Policies. Data access [Internet]. Washington (DC): COVID AMP; 2021 [cited 2022 Jan 5]. Available from: https://covidamp.org/data Google Scholar
• 23 Lee JC , Mervosh S , Avila Y , Harvey B , Matthews AL , Gamio L et al. See reopening plans and mask mandates for all 50 states. New York Times [serial on the Internet]. 2020 [last updated 2021 Jul 1; cited 2022 Jan 5]. Available from: https://www.nytimes.com/interactive/2020/us/states-reopen-map-coronavirus.html Google Scholar
• 24 Bureau of Economic Analysis. How is a metropolitan area defined? [Internet]. Washington (DC): BEA; 2008 Jan 10 [cited 2022 Jan 5]. Available from: https://www.bea.gov/help/faq/459 Google Scholar
• 25 To access the appendix, click on the Details tab of the article online.
• 26 Cori A , Ferguson NM , Fraser C , Cauchemez S . A new framework and software to estimate time-varying reproduction numbers during epidemics. Am J Epidemiol. 2013;178(9):1505–
  12. Crossref, Medline, Google Scholar
• 27 Fraser C . Estimating individual and household reproduction numbers in an emerging epidemic. PLoS One. 2007;2(8):e758. Crossref, Medline, Google Scholar
• 28 Grassly NC , Fraser C . Mathematical models of infectious disease transmission. Nat Rev Microbiol. 2008;6(6):477–
  87. Crossref, Medline, Google Scholar
• 29 Pan A , Liu L , Wang C , Guo H , Hao X , Wang Q et al. Association of public health interventions with the epidemiology of the COVID-19 outbreak in Wuhan, China. JAMA. 2020;323(19):1915–
  23. Crossref, Medline, Google Scholar
• 30 Aylward B , Barboza P , Bawo L , Bertherat E , Bilivogui P , Blake I et al. Ebola virus disease in West Africa—the first 9 months of the epidemic and forward projections. N Engl J Med. 2014;371(16):1481–
  95. Crossref, Medline, Google Scholar
• 31 Faria NR , Quick J , Claro IM , Thézé J , de Jesus JG , Giovanetti M et al. Establishment and cryptic transmission of Zika virus in Brazil and the Americas. Nature. 2017;546(7658):406–
  10. Crossref, Medline, Google Scholar
• 32 Cowling BJ , Ali ST , Ng TWY , Tsang TK , Li JCM , Fong MW et al. Impact assessment of non-pharmaceutical interventions against coronavirus disease 2019 and influenza in Hong Kong: an observational study. Lancet Public Health. 2020;5(5):e279–
  88. Crossref, Medline, Google Scholar
• 33 Flaxman S , Mishra S , Gandy A , Unwin HJT , Mellan TA , Coupland H et al. Estimating the effects of non-pharmaceutical interventions on COVID-19 in Europe. Nature. 2020;584(7820):257–
  61. Crossref, Medline, Google Scholar
• 34 Rubin D , Huang J , Fisher BT , Gasparrini A , Tam V , Song L et al. Association of social distancing, population density, and temperature with the instantaneous reproduction number of SARS-CoV-2 in counties across the United States. JAMA Netw Open. 2020;3(7):e2016099. Crossref, Medline, Google Scholar
• 35 Rubin DB . Matching to remove bias in observational studies. Biometrics. 1973;29(1):159–
  83. Crossref, Google Scholar
• 36 Rosenbaum PR . Modern algorithms for matching in observational studies. Annu Rev Stat Appl. 2020;7:143–
  76. Crossref, Google Scholar
• 37 Unacast. Social distancing scoreboard [Internet]. New York (NY): Unacast; [cited 2022 Jan 5]. Available from: https://www.unacast.com/covid19/social-distancing-scoreboard Google Scholar
• 38 Shaman J , Kohn M . Absolute humidity modulates influenza survival, transmission, and seasonality. Proc Natl Acad Sci U S A. 2009;106(9):3243–
  8. Crossref, Medline, Google Scholar
• 39 Lowen AC , Mubareka S , Steel J , Palese P . Influenza virus transmission is dependent on relative humidity and temperature. PLoS Pathog. 2007;3(10):1470–
  6. Crossref, Medline, Google Scholar
• 40 Cheng Y-T , Lung S-CC , Hwang J-S . New approach to identifying proper thresholds for a heat warning system using health risk increments. Environ Res. 2019;170:282–
  92. Crossref, Medline, Google Scholar
• 41 Ross ME , Vicedo-Cabrera AM , Kopp RE , Song L , Goldfarb DS , Pulido J et al. Assessment of the combination of temperature and relative humidity on kidney stone presentations. Environ Res. 2018;162:97–
  105. Crossref, Medline, Google Scholar
• 42 National Oceanic and Atmospheric Administration, National Centers for Environmental Information. Data tools: local climatological data (LCD) [Internet]. Washington (DC): NOAA; [cited 2022 Jan 5]. Available from: https://www.ncdc.noaa.gov/cdo-web/datatools/lcd Google Scholar
• 43 Census Bureau. American Community Survey (ACS) [Internet]. Washington (DC): Census Bureau; [cited 2022 Jan 5]. Available from: https://www.census.gov/programs-surveys/acs Google Scholar
• 44 Centers for Disease Control and Prevention. Behavioral Risk Factor Surveillance System [Internet]. Atlanta (GA): CDC; [cited 2022 Jan 5]. Available from: https://www.cdc.gov/brfss/index.html Google Scholar
• 45 Esri. ArcGIS Business Analyst [Internet]. Redlands (CA): Esri; [cited 2022 Jan 5]. Available from: https://www.esri.com/en-us/arcgis/products/arcgis-business-analyst/overview Google Scholar
• 46 Multi-Resolution Land Characteristics Consortium [home page on the Internet]. Washington (DC): MRLC; [cited 2022 Jan 5]. Available from: https://www.mrlc.gov/ Google Scholar
• 47 MIT Election Data and Science Lab. U.S. president precinct-level returns 2016 [Internet]. Cambridge (MA): Massachusetts Institute of Technology; 2018 [cited 2022 Jan 5]. [Harvard Dataverse Version 11.0]. Available from: https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/LYWX3D Google Scholar
• 48 Rosenbaum PR , Rubin DB . Assessing sensitivity to an unobserved binary covariate in an observational study with binary outcome. J R Stat Soc Series B Stat Methodol. 1983;45(2):212–
  8. Google Scholar
• 49 Greenland S . Basic methods for sensitivity analysis of biases. Int J Epidemiol. 1996;25(6):1107–
  16. Crossref, Medline, Google Scholar