We used data from the ACS 5-year summary file, 2009 to 2013, to obtain the total number of households, total number of SSM households (male householder and male partner), and total number of men aged 18 years and older for each county in the United States (Multimedia Appendix 1) [21]. The ACS is a supplement to the decennial census that provides annual updates to housing and demographic statistics for the United States [23]. Approximately 1 in 38 US households are randomly sampled each year, and the selected individuals respond using either Web-based or paper questionnaires. Staff from the US Census Bureau follow up with individuals who do not respond, in order to improve response rates.

ACS data are publicly available as 1-, 3-, or 5-year summary files or as a Public Use Microdata Sample (PUMS), which contains a de-identified and unaggregated sample of ACS data. The 1- and 3-year summary files are limited to areas with populations of 65,000 or 20,000 or more, respectively. However, the 5-year ACS summary files contain data at all available geographic areas. We did not include data from US territories.

To more accurately describe where MSM reside, we supplemented data from the ACS using the urbanicity categories produced by the National Center for Health Statistics (NCHS) [24]. According to the NCHS classification scheme, counties fall into six categories: central (ie, inner city) or fringe (ie, suburban) portions of large MSAs (population size ≥ 1,000,000 population), medium-sized MSAs (population size of 250,000-999,999), small MSAs (population of < 250,000), micropolitan area (counties that contain all or part of a city of 10,000 or more), and noncore (counties that do not contain any part of a city of 10,000 or more) [24]. In order to incorporate urbanicity-specific percentages of MSM among adult men, we then collapsed the categories according to the four-level urbanicity classification used by Oster et al. [22]: large central metropolitan county, large fringe metropolitan county, medium/small metropolitan county, and nonmetropolitan county.

We developed a method to estimate small-area MSM populations by combining two models reported by Lieb et al. [10]. The first, Model A, applied estimates of the percentage of MSM among adult men, stratified by urbanicity, to the adult male population. The second, Model B, weighted the national MSM percentage according to the relative representation of SSM households among all households in an area, referred to as the MSM Index. We combined these two models into a single model by stratifying the MSM Index formula to determine the urbanicity-specific relative representation of SSM households (Figure 1, Equation 1). We then multiplied this within-stratum MSM Index to the urbanicity-specific estimated percentage of MSM among adult men from NHANES, as reported by Oster et al. [22] (Figure 1, Equation 2), to arrive at the percentage of males who are MSM in each county. Next, for each county, the number of MSM was estimated by multiplying the MSM percentage by the total adult males (Figure 1, Equation 3).

By calculating the MSM Index within strata of urbanicity, we expected to reduce inter-urbanicity differences in same-sex cohabitation and reporting among MSM due to stigma. However, 35.4% (1112/3143) counties had no reported SSM households, and consequently had MSM Index values and estimated MSM population sizes of zero, which likely reflected these biases in detection of MSM. To impute MSM in these areas for our final estimates, while preserving the relative population sizes based on SSM households, we added households to both the numerator and denominator of the above equations. For each county, we increased the number of SSM households and the number of total households by adding the urbanicity-specific percentage of SSM households (Figure 2, Equation 4).

As an example of our imputation method, we will use two hypothetical nonmetropolitan counties. The total percentage of SSM households among all households in nonmetropolitan counties in our data was approximately 0.1%. For a county with 1000 households, of which zero were SSM households, we added one SSM household, or 0.1% of 1000. This meant that, for the part of our model that calculated urbanicity-specific indices, the new totals for that county were 1001 households, of which one was a SSM household. For another nonmetropolitan county with 20,000 households, of which 15 were SSM households, we added 20 SSM households, for a new total of 20,020 households and 35 SSM households. By adding a proportionate number of SSM households to all counties, we effectively maintained the relative representation of SSM households within urbanicity strata while estimating at least some MSM in counties with no SSM households. Because the index was used as a way of weighting the percentage of MSM among adult men in each county and not as a direct method of estimation, adding SSM households did not add MSM to our final population estimates.

For our analysis, we chose to estimate the number of men who had sex with men within the past 5 years, rather than the past 12 months or over the lifetime, as others have reported [2]. Using past-year estimates might underestimate the total sexually active population, particularly because we are using 5-year population estimates, while lifetime estimates would do the opposite: under that scenario, all men who had sex with another man would be counted, regardless of how recent or frequent the behavior. Because NHANES only has data regarding same-sex sex in the past 12 months and over a lifetime, we scaled our results to sum to 3.9% of the US adult male population, the estimated national percentage of adult men who had sex with a man within the past 5 years from the Purcell et al. meta-analysis [2].

All analyses were conducted using R Studio, version 0.98.953 [25]. Data were analyzed at the county level and aggregated to state and the Office of Management and Budget’s core based statistical areas (CBSAs). CBSAs refer to both metropolitan and micropolitan statistical areas. MSAs are CBSAs with at least 50,000 people. Micropolitan statistical areas have fewer than 50,000 people.

Using Purcell et al.’s [2] estimate of 3.9% MSM (past 5 years) among US adult men and summary data from the 2009 to 2013 ACS, we estimated that there are approximately 4,503,080 MSM in the United States. Table 1 presents the state-level population sizes of MSM, ranked from largest MSM population to least. California, which has 12.1% (13,997,953/115,463,694) of the US adult male population according to ACS estimates from 2009 to 2013, had the largest percentage of the US MSM population at 17.6% (792,750/4,503,080). Furthermore, over one-half of US MSM resided in that and the next six states: Texas at 8.3% of US MSM (371,781/4,503,080; New York at 8.2% of US MSM (371,087/4,503,080); Florida at 7.6% of US MSM (340,163/4,503,080); Illinois at 4.4% of US MSM (199,486/4,503,080); Pennsylvania at 3.6% of US MSM (162,745/4,503,080); and Ohio at 3.2% of US MSM (144,367/4,503,080). Despite 52.9% (2,382,379/4,503,080) of the US MSM population residing in those seven states, they represent only 44.6% (51,508,277/115,463,694) of the US adult male population.

Of the 3143 counties or county-equivalent areas in the United States, we estimated that over one-half of the total US MSM population resided in only 51 (Table 2). The largest number of MSM lived in Los Angeles County, California at 5.6% of US MSM (251,521/4,503,080), followed by cook County, Illinois at 2.8% of US MSM (125,923/4,503,080); Maricopa County, Arizona at 2.0% of US MSM (87,894/4,503,080); New York County, New York at 1.9% of US MSM (87,556/4,503,080); and Harris County, TX at 1.9% of US MSM (83,401/4,503,080). A total of 9.9% (310/3143) US counties had fewer than 20 MSM, and 39.5% (1242/3143) had fewer than 100 (Figure 3).

By aggregating our county-level findings to CBSAs, we found that 97.4% (4,384,172/4,503,080) of the MSM in our model resided in the 917 CBSAs in the United States. One-half (2,251,068/4,503,080) lived in one of 16 CBSAs (Table 3), all of which were MSAs. Of those residing in a CBSA, the largest population of MSM was in the New York-Newark-Jersey City CBSA at 8.8% of US MSM (397,399/4,503,080); followed by Los Angeles-Long Beach-Anaheim at 7.0% of US MSM (313,711/4,503,080); Chicago-Naperville-Elgin at 3.9% of US MSM (175,118/4,503,080); San Francisco-Oakland-Hayward at 3.2% of US MSM (145,972/4,503,080); and Miami-Fort Lauderdale-West Palm Beach at 3.1% of US MSM (141,088/4,503,080). Thus, 26.1% (1,173,288/4,503,080) of the MSM in the United States live in these five areas.

We used recent estimates of the population size of US MSM [2], data from ACS 2009 to 2015 [21], and recent estimated percentages of MSM among adult men by urbanicity [22] with an existing estimation method [10] to estimate state-, county-, and CBSA-level populations of MSM. Based on the relative representation of SSM households and prior estimates of MSM percentages in large central metropolitan areas, large fringe metropolitan areas, medium and small metropolitan areas, and nonmetropolitan areas [22], we found that a majority of MSM live within relatively few counties and CBSAs. Our method is a relatively simple, robust approach to estimating small-area population sizes for MSM that can easily be updated as new data become available.

Our findings are consistent with other studies, although ours is the first to use this method at this fine of a geographic level for the entire country. For example, Gallup’s March, 2015 [11] report on the LGBT population sizes found that the San Francisco-Oakland-Hayward, California, metropolitan area had the highest percentage of LGBT individuals among the general population. This same metropolitan area had the highest percentage of MSM among adult men, according to our method. Furthermore, our estimated number of MSM in San Francisco County, 66,586, was very close to a method that incorporated HIV prevalence estimates and HIV diagnoses: Raymond and colleagues [6] estimated 66,487 MSM in the same area in 2010, which is within the timeframe of our ACS data.

Despite similarities with other studies, our results were different from other recent publications, notably the ones from which we derived part of our method. We estimated fewer MSM at the state level than Lieb and colleagues [3] did in their 2011 article. We also estimated fewer MSM at the state and county level in Texas than Campagna et al. [17,18]. However, Lieb et al. [3] and Campagna et al. [17,18] used a higher estimated proportion of MSM in urban areas for their studies [13]. There are also several additional assumptions about geography in their models that likely contribute to differences: Lieb and colleagues [10] use “urbanized,” “within urban cluster,” and “rural” designations from the US Census Bureau to approximate urban, suburban, and rural, as reported by Laumann et al. [13]. However, “urbanized” and “within urban cluster” refer to metropolitan and micropolitan areas, respectively, or areas with more than 50,000 individuals and areas with greater than 2500 individuals but less than 50,000 individuals. Under these definitions, many suburban areas would be considered “urbanized,” while rural communities might be considered “within urban cluster.”

In addition to the different definitions of urbanicity, the percentages cited by Lieb et al. [10], 1% for rural areas, 4% for suburban areas, and 9% for urban areas, represent Laumann and colleagues’ [13] estimates for gay identity, rather than same-sex behavior. Within the identity category, those percentages were derived from individuals in the urban cores and suburbs of the “top 12 urban areas.” Thus, Model A appears to apply an identity-based measure from major metropolitan areas to different classifications from the US Census Bureau.

Our findings substituted the Laumann et al. [13] estimates with those from Oster and colleagues [22]. We also allowed MSM to exist in areas with no reported SSM households, and we scaled our findings to Purcell and colleagues’ [2] national estimate for the percentage of adult men who had sex with a man in the past 5 years. Consequently, our methods and its results represent a new, improved approach to the important work by Lieb and colleagues [3,7-10,19].

We made several assumptions and adjustments to prior methods that may limit the interpretation and use of our results. First, we decided that computing the MSM Index according to stratum would more accurately compare geographic areas, given possible within-urbanicity tendencies for MSM either not to cohabitate or to underreport SSM households. However, it may be that it is more accurate to compare all geographic areas, rather than to generate urbanicity-specific MSM Index values. Second, we used urbanicity-specific MSM percentages from Oster and colleagues [22], rather than the original estimates from Laumann et al. [13]. However, the urbanicity estimates from Laumann et al. [13] are identity-based, and the Oster et al. [22] estimates provided the most congruent urbanicity classifications for Model A. Finally, in order to avoid underestimating the number MSM outside of large urban areas, we imputed a proportional number of MSM to areas with no reported SSM households. It may be that some areas with no SSM households truly also have no MSM. However, the relative percentages of MSM (and resulting MSM population sizes) in all areas was mostly preserved because we altered the number of households and not the number of individuals, which was used only for weighting.

In addition to our method, our findings may be limited by our use of ACS data. The ACS is a sample of the population that is weighted, unlike the decennial census, which contains more data. As a result, inferences based on the ACS may be less accurate than data from the decennial census. ACS might also miss some of the same-sex households that are not in urban areas, particularly if they are less likely to respond to a survey other than the decennial census. It could also be due to more cohabitation, including marriage, among same-sex couples due to differences in legislation permitting marriage. However, because our data span several years, we cannot determine the extent to which policies and laws regarding marriage influence geographic differences.

Small-area estimates of MSM populations that incorporate the most recent data and estimates available may provide a useful tool to public health practitioners and policy makers for determining the burden of HIV and STIs among MSM in local contexts and planning prevention and treatment responses. Our method produced similar results to a recent effort to estimate MSM population sizes in San Francisco County but different from other studies that used a similar method, largely due to differences in the assumptions underlying the models. The method we presented can be updated annually as new ACS data are released, which would provide counties and larger geographic areas with up-to-date population sizes and, potentially, incidence and prevalence rates. These local statistics would allow for better resource allocation, intervention development, and service delivery. For data from the current analysis and for future updates, visit the study website [26].

The findings and conclusions in this manuscript are solely the responsibility of the authors and do not necessarily represent the official views of the Centers for Disease Control and Prevention or the Department of Health and Human Services.