key: cord-0960242-osvnfx6m
authors: Karimi, Seyed M.; Salunkhe, Sonali S.; White, Kelsey B.; Little, Bert B.; McKinney, W. Paul; Mitra, Riten; Chen, YuTing; Adkins, Emily R.; Barclay, Julia A.; Ezekekwu, Emmanuel; He, Caleb X.; Hurst, Dylan M.; Popescu, Martha M.; Swinney, Devin N; Johnson, David; Hollenbach, Rebecca; Moyer, Sarah; DuPré, Natalie C.
title: Stratified Random Sampling Methodology for Observing Community Mask Use within Indoor Settings: Results from Louisville, Kentucky during the COVID-19 Pandemic
date: 2021-02-25
journal: bioRxiv
DOI: 10.1101/2021.02.25.432837
sha: 22ed428eecaa65a12abba991241854197d6a0f1e
doc_id: 960242
cord_uid: osvnfx6m

Wearing a facial mask can limit COVID-19 transmission. Measurements of communities’ mask use behavior have mostly relied on self-report. This study’s objective was to devise a method for measuring the prevalence of mask-wearing and proper mask use in indoor public areas without relying on self-report. A stratified random sample of retail trade stores (public areas) in Louisville, Kentucky, USA, was selected and targeted for observation by trained surveyors during December 14−20, 2020. The stratification allowed for investigating mask use behavior by city district, retail trade group, and public area size. The average mask use prevalence among observed visitors of the 382 visited public areas was 96%, while the average prevalence of proper use was 86%. In 17% of the public areas, at least one unmasked visitor was among the observed visitors; in 48%, at least one improperly masked visitor was observed. The average mask use among staff was 92%, but unmasked staff were observed in fewer public areas, as an unmasked staff member was observed in 11% of the visited public areas. The average prevalence of proper make use among staff was 87%, similar to the average among visitors. However, the percentage of public areas where at least one improperly masked staff was observed was 33. Significant disparities in mask use and its proper use were observed among both visitors and staff by public area size, retail trade type, and geographical area. Observing unmasked and incorrectly masked visitors was more common in small (less than 1500 square feet) public areas than larger ones, also in food and grocery stores than other retail stores. Also, the majority of the observed unmasked persons were male and middle age adults.

240 In the second section, a surveyor was asked to observe the mask use of up to ten visitors in the PA.

241 That is, if there were more than ten visitors, they needed to observe only ten of them, but if there 242 were fewer than ten visitors, they needed to observe all of them. They could start observing as 243 soon as they entered the PA or start observing at any time or location after entering the PA. In any Table 1) .

Any type of facial covering-such as bandanna, cloth mask, neck gaiter, disposable surgical mask, 118 cone-style mask, N95, and other respirators-used to protect against aerosol transmission of 119 infectious particles was defined as "mask" or "facial mask

Indoor PAs were selected using a random sampling technique stratified by the city district, 126 industry, and PA size in order to observe mask-wearing in various parts of the community, in 127 different types of PAs and PA sizes. Seven city districts-namely

East & North 129 East-were constructed based on the geographical proximity of zip codes, demographic and 130 median income(Appendix Table 1, Appendix Fig 1). Notably, people who are black are more 131 highly concentrated in specific areas of the city, as the city suffers from a legacy of racial 132 segregation highlighted discriminative zoning practices. 18-20 According to data from the 2010 133 census, the non-Hispanic black population concentration was the highest in the North West district 134 of the city (75%: 54%−93% in different zip codes)

The city's wealthiest district was East & North East

Retail Trade (SIC division G, codes 52xxxx−59xxxx)

Among the subcategories 148 of Retail Trade, Automotive Dealers and Gasoline Service Stations (SIC codes 55xxxx) and Eating 149 and Drinking Places (SIC codes 58xxxx) were excluded to preserve the observational and non-150 interventional nature of the study, as most businesses in these two types of retail provide personal 151 assistance to customers. Therefore, the study focused on the remaining subcategories of Retail 152 Trade: Building Materials, Hardware

Apparel and Accessory Stores (SIC codes 56xxxx), Home Furniture, Furnishings and Equipment

Miscellaneous Retail (SIC codes 59xxxx). The last category included 156 drug stores and proprietary stores, liquor stores, used merchandise stores, and books stores, among 157 others. A surveyor could conveniently visit a typical business classified under any of the six retail

The third stratification criterion was PA size so that the team could investigate if mask-wearing 160 behavior differed in larger PAs where visitors may spend more time and cluster in greater numbers

23 A total of 4,648 PAs in Louisville were classified as Retail Trade (excluding SIC codes 166 55xxxx and 59xxxx) and were considered for observation. Information on the PAs was imported 167 into the statistical software STATA

Three variables representing stratification criteria were constructed: district, retail groups, and PA 172 size. The city zip codes were grouped as described above, and the seven districts were formed 173 (Appendix Table 1, Appendix Fig 1)

Furnishings 176 and Equipment Stores) were combined to form the furniture, equipment, and home improvement 177 stores group, and SIC codes 53xxxx (General Merchandise Stores) and 56xxxx (Apparel and 178 Accessory Stores) were combined to form general merchandise and apparel stores group. 827 PAs 179 were grouped into furniture, equipment, and home improvement stores, 996 in general 180 merchandise and apparel stores, 878 in food and grocery stores

) between 2500 and 4999, and (4) greater than 5000. The 184 specific recategorization was to increase the uniformity of PAs' distribution over size

In practice, PA size groups 1, 2, 3 and 4 included 1186

Each cluster included PAs specified by a 188 combination of either of the seven districts, four retail groups, and four PA sizes

Three PAs were randomly selected without 194 replacement from any cluster that fell in the first tertile, four PAs from the second tertile clusters, 195 and five PAs from the third tertile clusters (Appendix Table 4). As a result, 447 PAs we randomly 196 selected to be observed. Observing 447 PAs was perceived feasible based on the survey's cost 197 assessment and funding. Using the same method, an extra 447 PAs were selected as a back

To demonstrate the selected PAs' representativeness for observation, the proportion of each district 200 in all selected PAs with the district's human and PA populations proportions

The differences in proportions between the selected PAs and human population were small, 202 between -2.7% and 2.7% across the seven districts, and the differences between selected PAs and 203 PA population proportions were also small

As a more detailed representativeness test, the difference in each district's share in all selected PAs 206 of a specific size was calculated with the district's human and PA populations shares. Therefore, 207 two measures of error (equal to the absolute value of the two differences) were calculated for each 11

When the shares of selected PAs in the districts were compared 209 to the districts' human population shares, the mean error was 1.98%, and quartiles of the errors 210 were 1.43%, 1.95%, and 3.05%, respectively. Comparing the share of selected PAs in the districts 211 to the districts' PA population shares resulted in a mean error of 2.98%

Surveyors' assignment sheet

Each surveyor observed mask use in a subset of the 216 randomly selected 447 PAs and a subset of the 447 back-up PAs that were also randomly selected

A surveyor's assignment sheet was a Word document that included two tables: a table of selected 218 PAs and a table of reserve PAs. Each table had six columns: Unique ID, PA Zip Code, Retail 219 Group

A surveyor was asked to observe only the table of selected PAs. If a selected PA was not 221 operational, then the surveyor was asked to replace it with a PA in the same retail group and the

Unique IDs were generated 224 to connect the collected data from a PA to its other information reported in the Data Axle master 225 dataset. Surveyors would insert the PA's unique ID into the online questionnaire; hence the unique online questionnaire to the assignment sheet. The observation number was used as an extra 229 identification tool to connect a PA questionnaire (hence the survey data) to the PA's notes on the 230 assignment sheet

and its link was 254 emailed to surveyors to record observations from a PA. When a questionnaire was submitted, the 255 link could be used to start filling another questionnaire for another PA. Answers were stored and

The definitions 260 of a mask, an unmasked person, and an incorrectly masked person were reviewed. Moreover, 261 examples of a typical observation were presented to them, and their questions about the 262 questionnaire and the data collection process were answered before the survey began and during 263 the survey. The survey was conducted from

During the survey week, 382 PAs were observed. The largest number of PAs were observed in the

This district also included the largest share 271 of the city's population (20%) (Appendix Table 4). The smallest number of PAs were observed 272 in the North West district (n=32, 8%) (Appendix Fig 2); the district included 11% of the city's 273 population (Appendix Table 4). The observed PAs included 19% furniture, equipment, and home 274 improvement stores, 24% general merchandise and apparel stores, 35% food and grocery stores, 275 and 23% miscellaneous retail stores

In effect, the 280 difference between a district's share in total observed PAs and the district's share in total human 281 and targeted PA populations was greater than that for the district's share in total selected PAs

The larger error in observation than selection was the result of two implementation challenges

Firstly, several smaller-sized PAs either were not operational or could not be observed without an 287 appointment. Secondly, the number of PAs in furniture, equipment, and home improvement stores 288 and in miscellaneous retail stores that could not be observed without an appointment was greater

Both 294 mean and variation of incorrect mask usage were greater than those of no mask use: 14% wore a 295 mask incorrectly (SD = 22%). In 52% of the PAs, there were no observations of incorrectly masked 296 visitors (Fig 1a). Incorrectly masked and unmasked visitors were most frequently observed within 297 small public areas, those consisting of less than 1,500 square feet

Unmasked visitors were less frequently observed in the visited PAs from furniture, equipment, and 300 home improvement stores, where the unmasked prevalence among visitors was zero in 92% of 301 them, and the mean prevalence was 2% (SD=13%)

Unmasked visitors were observed most frequently in the South & South West district, where the 308 mean prevalence of unmasked visitors was 8% (SD = 21%) (Fig 3a). Nonetheless, the prevalence 309 of unmasked visitors was zero in 76% of the observed PAs in this district. Incorrectly masked 310 visitors were observed most commonly in the North West district, where the mean prevalence of 311 incorrectly masked visitors was 20% (SD = 29%), and at least one incorrectly masked visitor was 312

Among visitors wearing 314 masks incorrectly, 53% were perceived as male (Fig 4b). About 50% of the unmasked and 48%

Among the observed staff in the observed PAs, the mean prevalence of unmasked and incorrectly 319 masked staff was 8% (SD=25%) and 13% (SD=25%), respectively (Fig 1b). The proportion of the 320 unmasked and incorrectly masked were zero in 89% and 67% of the PAs

In PAs from the other retail groups, the 328 prevalence of no mask usage was higher, between 7% and 11% (Fig 2b). The incorrectly masked 329 staff were most frequently observed in food and grocery stores than the other three observed retail 330 groups

Unmasked staff were observed most commonly in the North West district, where the mean 333 prevalence of unmasked staff was 16% (SD=35%) (Fig 3b). Nonetheless, the prevalence of 334 unmasked staff was zero in 68% of the observed PAs in this district. Incorrectly masked staff were 335 observed most frequently in the South &

respectively (Fig 3b). Nonetheless, the proportion of incorrectly masked staff 337 of these districts was zero in 65% and 59% of the observed PA in the two districts

Kentucky (USA) in December 2020 and to explore potential 347 patterns in mask-wearing behavior. Results suggested there was a high prevalence of mask use: 348 96% of visitors and 92% of staff wore a mask in the 382 observed PAs. Improper masking appeared 349 more common than not wearing a mask at all in December 2020. Among the masked visitors and 350 staff, respectively, 86% and 87% used it properly

While the mean mask use among staff was smaller than that among visitors (92% versus 96%), 352 the proportion of PAs where at least one unmasked staff was observed was smaller than the share 353 of PAs where at least one unmasked visitor was observed

One explanation could be the way individuals follow perceived social 359 norms. 25 Perhaps individuals more intentionally follow a norm in areas with more people; there 360 is a greater expectation to follow a pre-set norm, such as wearing a mask correctly, in larger 361 rather than smaller public areas. Others suggest that health behaviors are more specifically 362 connected to one's appraised threat such that the lower the appraised threat, the less likely the 363 individual will adhere to protective behaviors. 26,27 In other words, if one assesses a lower risk of 364 contracting COVID-19 infection in a smaller establishment due to the location's smaller 365 capacity, then they are less motivated to completely adhere to those protective behaviors. 26 9% to 18%, respectively, across different retail stores (Fig 2). Also, unmasked and 371 incorrectly masked staff in the retail stores ranged from 5% to 11% and 8% to 16%, respectively 372 (Fig 2). A survey conducted by the Pew Research Center in August 2020 revealed that 373 approximately 85% of people reported wearing a mask over the past month, which has increased 374 from 65% seen in June. 8 The odds of observing a person wearing a mask in an urban or suburban 375 retail store were four times than those seen in the rural areas

Significant geographical variation in mask use and proper use prevalence was observed in this 380 study. The mean unmasked prevalence varied from 0% to 8% in visitors and 3% to 16% in staff

Appendix healthy behaviors and often face complex social contexts 30,31 , and having the financial 391 means to purchase or make an appropriately fitted mask is likely to be an obstacle in lower income 392 communities. Researchers have reported a higher prevalence of high-risk health behaviors among 393 communities with low-income. 32 In other words, although smoking and physical inactivity differ 394 from mask-wearing, the communities facing the greatest health challenges for systemic reasons 395 are also struggling with masking. Further, researchers currently suggest that case and mortality 396 rates are elevated for populations with complex health challenges

Finding a high prevalence 400 of incorrect mask use among males coincides with previous research findings. A number of studies to one's caregiving responsibilities. 35 Some have argued that masking-wearing as a 405 public health practice may evolve like the utilization of seatbelts

For example, researchers have 408 reported that young men are the most unlikely to follow seatbelt regulations, 36-38 and others have 409 suggested that seatbelt use is connected with one's perceived risk. 39 Even other public health 410 prevention tools, such as condom use, has been connected with one's tendency to take risk or attention. One survey identified the political and polarizing nature of mask-wearing 415 among United States counties. 41 Louisville is an urban area within a predominately rural state with 416 midwestern and southern history. Mask-wearing behaviors may differ between rural and urban 417 settings as well. 29 Louisville voted for democratic representatives in the House of Representatives 418 eight out of the past ten election cycles

Even among the retail trade, two major 428 classifications -namely, (1) automotive dealers and gasoline service stations and (2) eating and 429 drinking places, constituting about 5% of all businesses in Louisville-were excluded. 23 Observing 430 eating and drinking places is especially important to understand the dynamics of the spread of 431 respiratory infectious disease, as they are environments where masks are taken off, at least 432 occasionally, and have been linked to an increase in COVID-19 cases. 44 planned, though the error was still small. The median representativeness error 435

Approximately a half of the error could be attributed to the sample selection mechanism 438 that resulted in a median representativeness error of 1.95%. The rest could be attributed to the 439 implementation challenges. For example, some of the selected PAs were either non-operational or 440 could not be visited by surveyors. This happened more often in small PAs (square feet less than 441 1499), especially in the West Center

For example, the majority of the unmasked and incorrectly masked visitors of the 444 observed PAs were male and middle age adults. However, one cannot be certain that males and 445 middle age adults exhibit the worst mask-wearing behavior without confirming the sex and age of 446 the masked and correctly mask visitors as well. The later pieces of information were not directly 447 collected from the visitors and staff in this study. Nonetheless, the cross-district variation in mask were more commonly observed in small public areas (Square footage < 1,500). In contrast, 453 the incorrectly masked and unmasked staff were more commonly observed in medium-small 454 public areas (Square footage = 1500 -2500). Both incorrectly masked visitors and staff were most 455 commonly observed in food and grocery stores than other retail stores. Among the observed 456 visitors and the staff, middle age adults made up the highest proportion of unmasked and 457 incorrectly masked. Despite mask mandates in place

There is a continued need to improve awareness of the effectiveness of appropriate 460 facial mask use, financial resources to provide masks

Conceptualization, Methodology, Fieldwork organization, Surveyor

Data curation, Formal analysis, Supervision, Writing the original draft, Reviewing

Salunkhe: Conceptualization, Methodology, Fieldwork organization, Surveyor

Data curation, Formal analysis, Supervision, Writing the original draft, Reviewing

Conceptualization, Methodology, Fieldwork organization, Surveyor, Data 471 curation, Formal analysis, Supervision, Writing the original draft

Little: Conceptualization, Methodology, Supervision, Reviewing and editing 473 5. W. Paul McKinney: Conceptualization, Methodology, Supervision, Reviewing and 474 editing 475 6. Riten Mitra: Conceptualization, Methodology, Supervision

YuTing Chen: Methodology, Data Curation 477

Adkins: Surveyor

Surveyor 479 10. Emmanuel Ezekekwu: Surveyor 480 11. Caleb X. He: Surveyor 481 12. Dylan M

David Johnson: Organization of the survey's logistics at the University of Louisville 485 16. Rebecca Hollenbach: Organization of the survey's logistics at the LMPHW 486 17

DuPré: Conceptualization, Methodology, Supervision, Reviewing and editing

the corresponding author of this manuscript, is responsible for (is the 491 guarantor of) the content of the manuscript, including the data and analysis

Metro Department of Public Health and Wellness (LMPHW) funded this study through the

Coronavirus Aid, Relief, and Economic Security Act (the CARES Act)

Speech can produce jet-like transport 499 relevant to asymptomatic spreading of virus

Face 502 coverings and mask to minimise droplet dispersion and aerosolisation: a video case study

Physical distancing, 505 face masks, and eye protection to prevent person-to-person transmission of SARS-CoV-2 506 and COVID-19: a systematic review and meta-analysis

Low-cost 509 measurement of face mask efficacy for filtering expelled droplets during speech

Masks Do More Than Protect Others During COVID-512 19: Reducing the Inoculum of SARS-CoV-2 to Protect the Wearer

Visualizing the effectiveness of face masks in 515 obstructing respiratory jets

A Detailed Map of Who Is Wearing Masks in the U.S. 518 The New York Times

More Americans say they are regularly wearing masks in stores and other 522 businesses

Understanding America Study. The University 529 of Southern California Center for Economic and Social Research

cdc-finalize-mask-observation-study-results-with-plans-to-continue-in-the-spring. 540 14. Newman MG. An Observational Study of Mask Guideline Compliance In An Outpatient 541 OB/GYN Clinic Population

Use of face 543 coverings by the public during the COVID-19 pandemic: an observational study. 544 medRxiv : the preprint server for health sciences

Louisville Confronts Its Redlining Past and Present. Bloomberg CityLab

19. LouisvilleKy.gov. City begins community conversation to combat redlining 2017 553

Racial Capitalism and Real Estate

Survey Redesigns Make Comparisons to Years Before

North American Industry Classification System Association. SIC Codes by Industry 561

University of Louisville Journal 567 of Respiratory Infections. Forthcoming 2021. 568 25. Cameron KA, Campo S. Stepping back from social norms campaigns: comparing 569 normative influences to other predictors of health behaviors

The importance of coping appraisal in 572 behavioural responses to pandemic flu

Understanding face mask use to prevent coronavirus and other illnesses: 574 Development of a multidimensional face mask perceptions scale

Misinformation with Face Mask Wearing and Social Distancing in a Nationally 578 Representative US Sample

Who is 580 wearing a mask? Gender-, age-, and location-related differences during the COVID-19 581 pandemic

The Behavioral Model for Vulnerable Populations: 583 application to medical care use and outcomes for homeless people

Do Patient-Centered Medical Homes 586 Improve Health Behaviors, Outcomes, and Experiences of Low-Income Patients? A 587 Systematic Review and Meta-Analysis

Socioeconomic 589 factors, health behaviors, and mortality: results from a nationally representative 590 prospective study of US adults

Subgroup comparison of 592 COVID-19 case and mortality with associated factors in Mississippi: findings from 593 analysis of the first four months of public data

The effect of messaging and gender on intentions to wear a face 595 covering to slow down COVID-19 transmission

Presidential Address: Geographies of Care and Responsibility. Annals of the 597 Association of American Geographers

Driver and passenger 599 seatbelt use among U.S. high school students

The influence 601 of demographic factors on seatbelt use by adults injured in motor vehicle crashes

Buckle up: non-seat belt use and antisocial 604 behavior in the United States

The role of perceived risk, reasons for non-seat belt 606 use and demographic characteristics for seat belt use on urban and rural roads

Young men's condom use resistance tactics: a latent profile analysis

Politicizing the Mask: Political, Economic and Demographic Factors 612 Affecting Mask Wearing Behavior in the USA

Mobility network 617 models of COVID-19 explain inequities and inform reopening