METHODS

Lesbians were identified with the help of four organisations that serve the lesbian and gay population of Allegheny County, Pennsylvania and maintain large mailing lists. These organisations provided a paper copy of their mailing lists for this project. All of the four lists had been updated immediately before the study. To maintain the confidentiality of names appearing on the lists, representatives of the organisations were instructed on how to match the names, and they supervised use of the lists. After the study’s completion, the mailing lists that had been provided were destroyed. The University of Pittsburgh’s Institutional Review Board approved the study protocol in advance.

To limit the study’s results to Allegheny County, women with zip codes outside this area were excluded. The names on the mailing lists were manually cross referenced and linking the names and addresses identified those appearing more than once. Three of the mailing lists were obtained from a community centre, an event promoter, and a foundation, all of which primarily serve the local lesbian and gay population. There are no data available to estimate the age range or other characteristics of the individuals included on these three lists. The fourth list was provided by a lesbian health research project at the University of Pittsburgh.¹² Surveys for the project were widely distributed throughout the lesbian community by a number of methods including; social, political, and religious organisations, media outlets, community events, mailing lists, and social networks. Participants had the option of providing their name and address on a postcard, indicating their interest in future research projects, that was mailed independent of the anonymous survey. Thus, the list from the research project included the names of women who returned the postcard. The age range of women completing the anonymous survey was 18–74 years. However, as the survey was anonymous and we had no way of linking the return postcards with completed surveys, we cannot estimate the age range of women that returned a post card and were included in the study.

Capture-recapture method and log-linear modelling¹³ were used to estimate the total number of lesbians in Allegheny County based on these four lists. Source dependence was modelled by adding the corresponding interaction term to the model. The significance of the interaction was assessed using likelihood ratio statistics, and the goodness of fit of the model was measured by deviance. To model heterogeneity among the population, a method suggested by Agresti¹⁴ and Darroch et al¹⁵ was used. A confidence interval (CI) for the estimated number of cases was computed via the profile likelikhood method and Akaike Information Criterion (AIC) was used for the model selection. For the log-linear model the value for this criterion is AIC = Deviance B 2×df, where df is the number of degrees of freedom of the model, and the model with the smallest AIC is selected as the best model.¹⁶ The proportion of the county’s lesbian population was determined by dividing the number of lesbians estimated through the capture-recapture method by the number of adult women residing in the county according to the 1990 census. The statistical software SAS with GENMOD procedure was used for the analyses.

RESULTS

A total of 2185 unique women were identified from the four lists as follows: 947—community centre (list A), 873—event promoter (list B), 506—foundation (list C), and 542—research project (list D). However, we know that by their nature these lists are incomplete and that the total number of lesbians is larger than 2185 (0.6% of the adult female population of Allegheny County). But how much larger is it? Capture-recapture utilised the overlap between the lists (table 1) to determine the degree of underascertainment for the raw count of 2185 and thus provided an estimate of the total lesbian population. Most (74%) of the names appeared on only one list (n=1612). Nearly 20% of them appeared on two lists (n=436), 5% appeared on three (n=114), and only 1% appeared on all four (n=23). The log-linear model that best fit the data included both heterogeneity and dependence among the sources (table 2). On the basis of this model, the number of lesbians in Allegheny County was estimated at 7031 (95% CI 5850 to 8576). The number of adult women in Allegheny County in 1990 was 375 901. Therefore, the county’s adult lesbian population constituted 1.87% (95% CI 1.56% to 2.28%) of its entire adult female population.

DISCUSSION

This study shows the usefulness of the capture-recapture method in determining the population of lesbians in a defined geographical area. That usefulness must nevertheless be qualified. The study’s sources of data depended on a woman’s willingness to have her name included on a mailing list. Of course, the voluntary nature of being on a mailing list entails that some members of the lesbian population will have a low probability of capture. This may apply particularly to women who do not want to be identified as lesbian and whom our approach thus could not count. Yet a principle of the capture-recapture method is that its estimate is valid only for a population that can be captured without selectivity. Given this stricture, we must qualify our results because all four mailing lists used in the current study capture only those women who are visible or active in the lesbian community. In effect then, we have estimated the number of women in Allegheny County who identify with that community. It is also possible that a woman may be included on a list and not be a lesbian. We have no way of estimating this number but expect that it would be extremely small. In addition, while the matching of the names was done manually, there could have been problems with this process because of different spellings of names, incomplete names (providing only first or last initial), or the possible use of an alias on one of the lists). All of these factors could contribute to potential mismatches.

There are advantages to studying the population of lesbians that are out and identify with and participate in the lesbian community. Social learning theory postulates that a health behaviour is influenced by interaction among a given person’s characteristics, the person’s environment, and the behaviour itself.¹⁷ Therefore, possible differences between the health behaviours of lesbian and heterosexual women may be attributable in part to the shared environment of the lesbian community. Hertzman et al¹⁸ define special populations as those having shared characteristics. They further maintain that the social environment within such populations may condition certain health habits, producing a pattern of negative or positive behaviours that ultimately influence health. Thus, to understand the complexity of health behaviours within the lesbian population it seems necessary to enumerate and then study those women that identify with the community.

The capture-recapture method with log-linear modelling suggested that there was heterogeneity in capture probability among the lesbian population, that is, different individuals would have a different chance of being listed by the sources. In addition, the model also indicated the existence of source dependence in the lists. It is likely that some positive dependence exists among the four sources used for this study. In other words, appearing on one mailing list may increase a woman’s chance of being on the other three as women who are willing to have their names on a lesbian related mailing list are more likely to have their names on other such lists. In addition, the women participating in the research project (list D) were recruited from many sources in the lesbian community including publications and advertisements that would have likely been read by women on lists A, B, and C.

Using capture-recapture has the potential to provide accurate and reliable estimates of the lesbian population in any location. The method’s basic tenet is that one need not identify every member of the population to estimate its size accurately, provided that appropriate sources are used for case identification and an adequate model is selected for the estimation. Consequently, under proper conditions, capture-recapture is incredibly cost effective in counting any elusive population and could easily be used to enumerate the lesbian population locally, nationally, and globally.

Furthermore, studies such as this have important public health implications. After all, estimating the lesbian population is critical in planning for healthcare needs and developing relevant research programmes within the population. For example, if the proportion of lesbians thought to be overweight is 29%, then the number of lesbians in Allegheny County that could benefit from a weight reduction programme would be 2039. Statistics such as this would certainly be worth knowing. In summary, the results of this study in combination with epidemiological studies on health behaviours, will allow us to estimate not only the prevalence of risk factors for disease among lesbians residing in Allegheny County but how many women are in need of health promotion programmes and health care services.