Prevalence of Face Mask Wearing in Northern Vermont in Response to the COVID-19 Pandemic

Abstract

Objectives

Information on the prevalence of face mask use to reduce the spread of SARS-CoV-2 is needed to model disease spread and to assess the effectiveness of policies that encourage face mask use. We sought to (1) estimate the prevalence of face mask use in northern Vermont and (2) assess the effect of age and sex on the likelihood of face mask use.

Methods

We monitored the entrances to public businesses and visually assessed age, sex, and face mask use. We collected 1004 observations during May 16-30, 2020. We calculated estimates of overall face mask use and odds ratios (ORs) for effects by age and sex.

Results

Of 1004 observations, 758 (75.5%) sampled people used a face mask. Our census-weighted estimate was 74.1%. A higher percentage of females than males wore face masks (83.8% vs 67.6%). The odds of face mask use were lower among males than among females (OR = 0.52; 95% CI, 0.37-0.73). Face mask use generally decreased with decreasing age: 91.4% among adults aged >60, 70.7% among adults aged 26-60, 74.8% among people aged 15-25, and 53.3% among people aged ≤14. The OR of an adult aged >60 wearing a face mask was 14.70 times higher, for young people aged 15-25 was 2.72 times higher, and for adults aged 26-60 was 2.99 times higher than for people aged ≤14. Females aged >60 had the highest percentage of face mask use (96.3%) and males aged ≤14 had the lowest (43.8%).

Conclusions

Educational efforts promoting the use of face masks should be targeted at males and younger age groups to limit the spread of SARS-CoV-2.

The use of face masks can be effective at reducing the transmission of SARS-CoV-2, the virus that causes COVID-19. ^1,2 Yet adoption of face mask wearing is highly variable in the United States, with some states adopting mandatory face mask mandates and other states leaving face mask wearing to individual choice. ³ Theoretical studies have shown that the extent of disease outbreak depends on both population compliance with face mask use and the effectiveness of face masks at reducing viral transmission. ⁴ Simulation studies further suggest a rapid phase transition from low to high prevalence of infection in populations as a function of the extent to which face mask use is adopted by the public. In these studies, face mask use below a threshold percentage of people results in widespread infection of the population in the absence of other mitigation strategies. ⁵ However, the likelihood of an individual engaging in protective behaviors such as wearing a face mask may vary, with demographic differences among individuals. For example, a meta-analysis of protective behaviors in response to respiratory epidemics found significantly lower compliance among males than among females, ⁶ as have studies of face mask use during the COVID-19 pandemic. ^7,8 Other studies showed that being older or having higher perceived susceptibility to disease is associated with higher rates of compliance with protective behaviors. ⁹ Information on overall population compliance with face mask use is important to understand and model the continuing dynamics of COVID-19, and information on demographic characteristics such as sex and age in adoption of face mask wearing is needed to increase overall compliance through targeted public education and outreach.

The objectives of our study were to (1) estimate the overall prevalence of face mask use in public places of business in northern Vermont and (2) assess the effect of sex and age on face mask use.

Methods

We visually assessed face mask use and estimated the age and sex of individuals. We collected 1004 observations from outside businesses and analyzed these data using Bayesian logistic regression.

Regional Context

Vermont reported its first case of COVID-19 on March 7, 2020. The governor declared a state of emergency on March 13, 2020, followed by a series of orders to mitigate the spread of SARS-CoV-2. ¹⁰ These orders included a “stay home, stay safe” order on March 24 that directed the closure of in-person operations for all nonessential businesses and mandated that the public remain at home unless leaving for reasons critical to health or safety. ¹⁰ On May 15, 2020, the order was updated to “be smart, stay safe,” which relaxed the previous restrictions, began the process of reopening businesses, and asked Vermonters to socially distance and use face masks in public. ¹⁰ Businesses were allowed to reopen and Vermont did not require face mask use. Thus, our study predates the face mask mandate issued on August 1, 2020, which made face mask use mandatory and provides baseline data on how various demographic groups responded to wearing face masks in the absence of government requirements.

We assessed the prevalence of face mask use by observing the entrances to 16 businesses from May 16 through May 30, 2020. Businesses included grocery, hardware, and convenience stores, as well as a golf course. We collected data on multiple days of the week, including Sunday, Tuesday, Friday, and Saturday, at times ranging from mid-morning to late afternoon. The weather during most sampling times was sunny and warm, although rain occurred during several sampling periods. The time spent at any given location during a single period of observation ranged from 15 minutes to 60 minutes but was typically 30 minutes. The number of individuals sampled during any single period of observation ranged from 11 to 111. Each business was generally visited once, although some businesses were visited twice.

Fifteen of 16 business locations were in Chittenden County, Vermont, as were most observations (893 of 1004). Chittenden County is the most urbanized and densely populated county in Vermont (population: 163 774), containing 26.2% of the state population; population estimates and demographic characteristics for Chittenden County are available from the US Census Bureau. ¹¹ We also collected observations (111 of 1004) from adjacent Franklin County (population: 49 402).

The decision to wear a face mask may be influenced by regional disease prevalence. ⁸ We collected data on the number of new COVID-19 cases in Chittenden County, Franklin County, and Vermont during May 3-16, 2020, and May 17-30, 2020, ¹² coinciding with our sampling period. We calculated the prevalence of SARS-CoV-2 in our study region and in the United States as the number of new infections during the preceding 14-day period per 100 000 people. The prevalence of infections was substantially lower (ie, 0%-10%) in Vermont than in the United States (Table 1).

Table 1.

Number of cases of COVID-19 in Vermont and the United States, May 3-30, 2020 ^a

Location	May 3-16, 2020	May 17-30, 2020
Chittenden County
No. of new cases	12	16
No. of cases per 100 000 population	7.3	9.8
Franklin County
No. of new cases	1	0
No. of cases per 100 000 population	2.0	0
Vermont
No. of new cases	48	44
No. of cases per 100 000 population	7.7	7.1
United States
No. of new cases	335 552	303 916
No. of cases per 100 000 population	102.2	92.6

^aData source: COVID-19 Case Mapper. ¹³

We recorded individuals’ face mask use, sex, and estimated age. We recorded an individual as wearing a face mask if it covered the person’s nose and mouth. We assessed both age and sex visually. We assigned age into the following categories: ≤14, 15-25, 26-60, and >60 years. We intentionally chose broad age categories because of a high degree of uncertainty about the actual age of any individual and because these broad categories were relatively easy to assess visually. We determined face mask use as people were entering the businesses. The observers were unobtrusively stationed in a vehicle near the store entrance or outside the businesses and did not interact with the observed individuals. The University of Vermont Committee on Human Research in the Behavioral and Social Sciences granted our study an exemption.

Statistical Analysis

We calculated the probability and marginal distributions of face mask use for each age by sex category. For each demographic category, we computed the posterior distribution of the probability of face mask use as a beta-binomial conjugate pair. We used a non-informative beta (0,0) prior distribution. We reported the raw counts and the mean and 95% credible intervals (CIs, defined by the 2.5 and 97.5 percentiles) for the posterior probability of wearing a face mask for each category. We also computed an adjusted overall rate of face mask use by using demographic information from the US Census 2019 American Community Survey (ACS) data for Chittenden¹³ and Franklin counties. ¹⁴ The age categories in the ACS data closely aligned with the age categories we used; as such, we used the estimated percentage of the population in our age categories and the proportion male and female for Chittenden and Franklin counties to weight our estimates on the basis of the fraction of samples from each county. We sampled from the posterior distribution of each age by sex category and then weighted these posterior estimates by their demographic fraction from the census data to compute a weighted prevalence of face mask use (mean and 95% CI).

We used logistic regression to analyze our data. We first selected the best-supported model of face mask use from a set of fixed-effect logistic regression models using the Akaike Information Criterion. ¹⁵ These models were fit using the generalized linear model function in R (R Foundation for Statistical Computing). We then added business type as a random effect to the selected model and fit this Bayesian random-effects logistic regression using the NIMBLE library in R. We did not treat business locations as the random effect but rather the business identity (ie, all grocery stores of a given chain were treated as a single random effect). This random effect implies that any effect of a given business is associated with the demographic population using that business identity (eg, grocery store chain 1 vs grocery store chain 2) as opposed to an effect of any particular store location. The random effect captures additional remaining variation in face mask use after accounting for the fixed effects of age and sex. We ran 4 Markov chain Monte Carlo simulations, each of length 2 million, used a burn-in length of 10 000, and thinned our chains by a factor of 100. We used the Gelman–Rubin convergence diagnostic to confirm convergence of our chains. ¹⁶

We calculated means and 95% CIs for the parameter estimates and odds ratios (ORs). The 95% CIs are defined by the 2.5 and 97.5 percentiles of the posterior distribution. The OR represents the odds that a person of a particular demographic class will wear a face mask compared with the reference group (eg, females for sex and age ≤14 for age). ^17,18

Results

Face mask use was 75.5% (95% CI, 72.8%-78.1%) overall but varied widely by sex and age (Table 2). The overall estimate weighted by US Census data was 74.1% (95% CI, 70.8%-77.3%). Across all age groups, females were more likely to wear face masks than males (83.8% vs 67.6%). Face mask use was highest among adults aged >60 (91.4%) and lowest among people aged ≤14 (53.3%). Males aged ≤14, 26-60, and 15-25 were the least likely to wear face masks (43.8%, 62.5%, and 67.8%, respectively). Adult females aged >60 (96.3%) were the most likely to wear a face mask, followed by adult males aged >60 (86.0%). Females aged 26-60 and 15-25 had similar face mask–wearing behavior (80.8% and 80.9%, respectively).

Table 2.

Face mask use during the COVID-19 pandemic, by age and sex, Vermont, May 16-30, 2020 (N = 1004 observations) ^a

Age, y	Male			Female			Total
	No. wearing a face mask	No. not wearing a face mask	% (95% CI) b	No. wearing a face mask	No. not wearing a face mask	% (95% CI)	No. wearing a face mask	No. not wearing a face mask	% (95% CI)
≤14	7	9	43.8 (21.3-67.7)	9	5	64.3 (38.6-86.1)	16	14	53.3 (35.6-70.5)
15-25	78	37	67.8 (59.0-76.0)	106	25	80.9 (73.8-87.1)	184	62	74.8 (69.2-80.0)
26-60	178	107	62.5 (56.8-68.0)	189	45	80.8 (75.5-85.6)	367	152	70.7 (66.7-74.5)
>60	86	14	86.0 (78.6-92.0)	105	4	96.3 (92.1-99.0)	191	18	91.4 (87.2-94.8)
Total	349	167	67.6 (63.5-71.6)	409	79	83.8 (80.4-86.9)	758	246	75.5 (72.8-78.1)

Abbreviation: CI, credible interval.

^aObservations were made by monitoring the entrances to 16 public businesses (ie, grocery, hardware, and convenience stores, as well as a golf course) in Chittenden and Franklin counties and visually assessing age, sex, and face mask use. An individual was recorded as wearing a face mask if it covered the person’s nose and mouth.

^bResults are reported as raw counts and 95% CIs of the percentage wearing face masks (ie, mean [2.5 percentile-97.5 percentile]).

The most parsimonious fixed-effects model of face mask use included age and sex but not their interaction, indicating that lower face mask use persisted in males across age groups (Table 3). Our final model, which included age and sex as fixed effects and business identity as a random effect, estimated that the odds of a male using a face mask were 52.4% the odds of a female using a face mask (Table 4). The odds of wearing a face mask were about 15 times higher among adults aged >60 than among people aged ≤14 and about 5 times higher than among people aged 26-60 and 15-25 (5.4 and 4.9, respectively), whereas the odds of face mask use among people aged 15-25 and 26-60 were approximately 3 times greater than among people aged ≤14 (2.72 and 2.99, respectively).

Table 3.

Comparison of fixed-effects models for predicting face mask use during the COVID-19 pandemic, by age and sex, Vermont, May 16-30, 2020 ^a

Model	AIC	∆ AIC
1. Intercept only	1120.05	74.63
2. Sex	1085.89	40.47
3. Age	1077.57	32.15
4. Sex + age	1045.42	0
5. Sex + age + sex:age	1049.98	4.56

Abbreviation: AIC, Akaike Information Criterion.

^aObservations were made by monitoring the entrances to 16 public businesses (ie, grocery, hardware, and convenience stores, as well as a golf course) in Chittenden and Franklin counties and visually assessing age, sex, and face mask use. An individual was recorded as wearing a face mask if it covered the person’s nose and mouth.

Table 4.

Parameter estimates and odds ratios for face mask use during the COVID-19 pandemic corresponding to age and sex for the final model that includes random effects of businesses on face mask use, Vermont, May 16-30, 2020 ^a

Model effect	Parameter estimates b (95% CI)	Odds ratio b (95% CI )
Intercept	–0.24 (–2.73 to 2.11)	—
Sex
Female	1.00 [Reference]
Male	–0.65 (–0.99 to –0.31)	0.52 (0.37 to 0.73)
Age, y
≤14	1.00 [Reference]
15-25	1.00 (0.15 to 1.85)	2.72 (1.16 to 6.36)
26-60	1.10 (0.28 to 1.91)	2.99 (1.32 to 6.76)
>60	2.69 (1.72 to 3.68)	14.70 (5.61 to 39.6)

Abbreviation: CI, credible interval.

^aObservations were made by monitoring the entrances to 16 public businesses (ie, grocery, hardware, and convenience stores, as well as a golf course) in Chittenden and Franklin counties and visually assessing age, sex, and face mask use. An individual was recorded as wearing a face mask if it covered the person’s nose and mouth.

^bThe mean and 95% credible intervals (2.5 and 97.5 percentiles).

We found large variation in the random effect of business identity on face mask use after accounting for fixed effects (Table 5). The size of the random effects was of the same magnitude as the fixed effects of age and sex. The lowest random effect (indicating reduced face mask use) was at a golf course (OR = 0.01; 95% CI, 0.000048 to 0.21), followed by a gas station convenience store (OR = 0.30; 95% CI, 0.03 to 3.20). The highest face mask use (OR = 4.04; 95% CI, 0.42 to 44.79) was associated with a grocery store.

Table 5.

Random effects and odds ratios of business brand or identity on face mask use during the COVID-19 pandemic at 16 locations in Vermont, May 16-30, 2020 ^a

Business brands or identities	Parameter estimates b (95% CI)	Odds ratiob (95% CI)
Convenience store	–1.19 (–3.46 to 1.16)	0.30 (0.03 to 3.20)
Department store	0.60 (–1.66 to 2.99)	1.83 (0.19 to 19.96)
Golf course	–4.44 (–9.95 to –1.56)	0.01 (0.000048 to 0.21)
Grocery store chain 1	0.73 (–1.51 to 3.10)	2.08 (0.22 to 22.24)
Grocery store chain 2	1.06 (–1.20 to 3.45)	2.90 (0.30 to 31.52)
Grocery store chain 3	0.98 (–3.06 to 7.22)	2.67 (0.05 to 1359.76)
Grocery store chain 4	1.40 (–0.86 to 3.80)	4.04 (0.42 to 44.79)
Hardware store	0.89 (–1.44 to 3.39)	2.43 (0.24 to 29.66)

Abbreviation: CI, credible interval.

^aObservations were made by monitoring the entrances to 16 public businesses (ie, grocery, hardware, and convenience stores, as well as a golf course) in Chittenden and Franklin counties and visually assessing age, sex, and face mask use. An individual was recorded as wearing a face mask if it covered the person’s nose and mouth.

^bThe mean and 95% CIs are reported as 2.5 and 97.5 percentiles.

Discussion

The overall face mask use of 75.5% (census corrected, 74.1%) found in our study was high compared with the 41.5% face mask use observed in retail stores in Wisconsin during the same period. ⁷ The Wisconsin study found similar trends in sex and age groups: face mask use was higher among older adults aged >65 (57%) than among younger people aged 2-30 (37%) or adults aged 31-65 (41%) and among females than among males (45% vs 38%). The corresponding per-capita incidence of new SARS-CoV-2 infections was 15 times higher in Wisconsin than in Vermont during the respective study periods. ¹⁹ The lower prevalence of SARS-CoV-2 infections in Vermont compared with Wisconsin corresponded to higher face mask use in Vermont and was consistent with observations of reduced prevalence of SARS-CoV-2 infections in states after implementation of mandates requiring face mask use. ²⁰

Wisconsin and Vermont differ in income, education, and political affiliation, population characteristics that may be associated with differences in face mask use. Median annual household income and education levels are higher in our study region than in Wisconsin ²¹ : the median annual household income in Chittenden County, where 89% of observations were made, is approximately $69 896, and 50.7% of the population has ≥bachelor’s degree. The remaining observations came from Franklin County, which has lower levels of education (23.7% of the population has ≥bachelor’s degree) and income (median annual household income, $64 528). In Wisconsin, 29.5% of the population has ≥bachelor’s degree and the median annual household income is $59 209. Wisconsin and Vermont also differ in political affiliations: Wisconsin voters are evenly split between Republicans and Democrats (42% each), whereas in Vermont, 29% of voters identify as Republican and 57% as Democrat. ²²

The prevalence of face mask use in Vermont was lower than the rate reported in the Hong Kong Special Administrative Region (HKSAR), a special administrative region of the People’s Republic of China. ²³ The prevalence of face mask use in the HKSAR was 96.1%, corresponding to a low incidence (12.9 cases per 100 000 population) of SARS-CoV-2 during the initial 100 days of the COVID-19 pandemic in that region (December 31, 2019, through April 8, 2020). The level of compliance with face mask use was approximately 20% higher in the HKSAR than in Vermont, although the incidence of disease was similar (eg, 12.9 cases per 100 000 population in the HKSAR compared with <10 cases per 100 000 population in Vermont). The incidence of COVID-19 in the United States was nearly an order of magnitude higher than in the HKSAR during our study period (102.2 and 92.6 cases per 100 000 population in the United States for May 2-16, 2020, and May 17-31, 2020, respectively). Although the epidemiologic contexts are different, these disease incidences in Vermont, the United States as a whole, and the HKSAR are consistent with the existence of a threshold level of compliance of face mask use (and other personal protective behaviors) above which disease spread is greatly reduced and below which it is not. ⁵ One study has placed this threshold near 73% face mask use, ⁴ which would be consistent with a low incidence of COVID-19 in the HKSAR and Vermont.

We observed higher rates of face mask use by females and adults aged >60 compared with other demographic groups, and this observation is broadly consistent with patterns found in other studies. A meta-analysis of 85 publications found that women were 50% more likely than men to engage in nonpharmaceutical protective behaviors such as face mask wearing in the context of respiratory infectious disease epidemics. ⁶ Another review of 26 studies found that being older and female was associated with greater adoption of protective behaviors during respiratory infectious disease epidemics. ⁹ These demographic patterns were also consistent with patterns observed in the Wisconsin face mask study: 7.6% more females than males used face masks, and the use of face masks was 16.1% and 19.8% higher among adults aged >65 than among adults aged 30-65 and young people aged 2-30, respectively. ⁷

Limitations

Our study had several limitations. First, the use of convenience sampling could have led to biased sampling, either oversampling or undersampling of demographic categories. For example, people most at risk of SARS-CoV-2 and, thus, most likely to wear face masks, might avoid public venues such as businesses, biasing our results downward. However, potentially biased sampling of demographic categories (eg, age, sex) was partially offset by the use of census data to correct the overall estimate of face mask use. Second, age and sex could have been misclassified, thereby leading to biases in parameter estimates. ²⁴ However, our demographic characterization of face mask use is consistent with that found in other studies of use of personal protective equipment and behaviors in response to respiratory infectious disease epidemics.

Conclusion

Social mitigation strategies such as wearing face masks in public are important to reduce the spread of SARS-CoV-2. Information on compliance with the use of face masks is required both for constructing models of disease spread and for targeted messaging to increase face mask use, particularly among demographic groups with low rates of face mask use. Studies such as ours provide information for assessing both demographic effects on face mask use and baseline information for regional comparisons or to assess regional trends in face mask use across time. Achieving high levels of face mask use (eg, >70% of the population) may be effective in reducing the spread of SARS-CoV-2, and a focus on messaging to males and younger people, who have lower rates of face mask use than females and older adults, is needed to increase the overall prevalence of face mask use to help meet this goal.