The prevalence of homosexual behavior and attraction in the United States, the United Kingdom and France: results of national population-based samples.
by Randall L. Sell , James A. Wells , David Wypij
INTRODUCTION
Accurate estimates of the prevalence of homosexuality are important for the development of public health and civil rights policy, but without reliable estimates available, the prevalence of homosexuality has been left to public debate. While some individuals postulate relatively small numbers of homosexuals implying that homosexuals are particularly HIV-infected (based upon the ratio of homosexuals with AIDS to their population estimates of homosexuals) or to diminish their ability to obtain civil rights, others may have inflated the percentages of homosexuals in order to obtain civil rights though the public recognition and acceptance that larger numbers would demand (Gonsiorek and Weinrich, 1991; Michaels, 1991). With little scientific data available to support the debates, discussion has often focused upon studies of homosexuality published by Kinsey et al. in 1948 and 1953, which have, until recently, provided the most scientifically obtained prevalence estimates.
Kinsey's studies found that 37% of postpubertal men and 20% of post-pubertal women had a sexual experience with someone of the same sex, and 13% of men and 7% of women had more sexual experience with people of the same sex than the opposite sex (Voeller, 1990). The most often cited figure from the Kinsey reports is that 10% of the population is homosexual. The exact derivation of the 10% figure from Kinsey's work is unknown, but several researchers have postulated sources including (i) an average of the 13 and 7% figures for men and women described above and (ii) a passage in Kinsey's volume on the human male that states: "10 per cent of the males are more or less exclusively homosexual for at least three years between the ages of 16 and 55" (Michaels, 1991; Kinsey et al., 1948, 1953; Voeller, 1990). However this number originated, the 10% population figure has become the most widely accepted and utilized estimate of the prevalence of homosexuality. Unfortunately, because Kinsey did not use probability samples to obtain his estimates and respondents were disproportionately drawn from the Midwest and from college campuses, the findings are not conclusive (Turner et al., 1989). However, unlike the major studies that followed his, Kinsey recognized the importance of examining a respondent's homosexual attraction (which Kinsey termed "psychic response") as well as homosexual behavior when investigating homosexuality (Kinsey et al., 1948, 1953).
Not until the onset of the AIDS epidemic, more than three decades after Kinsey, was additional research on the prevalence of homosexuality conducted and reported in the United States, the United Kingdom, or France that approached the scale of Kinsey's research. To better understand and control the AIDS epidemic accurate prevalence estimates of sexual behavior (and in particular male homosexual behavior) were needed, and studies to produce such estimates were performed. Table I summarizes the results of recent major national population-based studies conducted in the United States, the United Kingdom, and France. Fay et al. (1989) and Smith (1991) in the United States used self-completed questionnaires to measure sexual behaviors, while Billy et al. (1993) in the United States and Johnson et al. (1992) in Britain used face-to-face interviews to measure sexual behavior. Spira et al. (1992) measured sexual behaviors using a telephone survey in France, and Harry (1990), in the United States, used a telephone survey to ask respondents to self-identify their sexual orientation. None of these studies measured homosexuality as Kinsey measured it. Because these studies were primarily HIV studies, and were not intended to measure the prevalence of homosexuality per se, they measured the prevalence of same-sex sexual behavior and ignored respondent's homosexual attraction. These studies therefore only identified and examined one dimension of homosexuality. The Harry (1990) study similarly, by measuring sexual orientation identity, did not measure homosexuality as Kinsey measured it. Sexual orientation identity (how a subject identifies his/her sexual orientation to others) is problematic because it relies upon how the respondent personally defines sexual orientation. One person's homosexuality may be another person's heterosexuality, making the results a comparison of apples and oranges (Klein et al., 1985; Coleman, 1990).
To summarize, no nationally representative samples of either the United States, the United Kingdom, or France has measured anything beyond homosexual behavior or self-identification as homosexual in the examination of the prevalence of homosexuality. Our study is the first in national population-based samples to measure homosexual attraction as well as sexual behavior to assess the prevalence of homosexuality.
SURVEY METHODS
The Center for Health Affairs contracted with Louis Harris and Associates to conduct major national surveys of AIDS risk behaviors and knowledge between June and November of 1988. The surveys were designed to be representative of the populations in each country ages 16-50 with the exception of individuals living in Corsica, Northern Ireland, Alaska, Hawaii, or noncontiguous national territories; nor did we include those in prisons, hospitals, or religious and educational institutions. The methodology undertaken in the conduct of this survey was designed to produce comparable population based samples in the United States, the United Kingdom, and France. Our procedure stratified the sample on two [TABULAR DATA FOR TABLE I OMITTED] dimensions: geographic region and metropolitan versus nonmetropolitan residence, insuring that it would reflect within 1% of those living in different regions of the countries and those living in urban and nonurban areas. Within these strata we randomly sampled clusters of households and households within clusters. Interviewers screened respondents in each household on age, limited to the range 16-50, and sex to determine eligibility for the study. This procedure resulted in weights assigned to respondents reflecting the complex sampling design, and differential response rates.
We collected data in two parts: a face-to-face interview and a self-completed questionnaire. Face-to-face interviews have the advantage of encouraging rapport between respondent and interviewer and they insure completeness of data collection because the interviewer is in control. However, a self-completed questionnaire was used to assess potentially sensitive and explicit questions about sexual behavior. We felt that this would afford the respondent a greater degree of anonymity, would decrease potential embarrassment for both respondent and interviewer and would result in more completed instruments. In choosing this method we gave up some control over the completeness of the data collection as the interviewer could not check the form for unanswered questions. However, the interviewer collected the questionnaire before leaving the respondent's home.
The questionnaire contained over 100 questions in the face-to-face portion and 68 questions in the self-completed section. The face-to-face portion of the questionnaire included questions relating to demographics, general knowledge about AIDS, and exposure to AIDS educational messages. Questions on the self-completed portion of the questionnaire dealt with sexual and drug-using behaviors and changes in these behaviors over time.
The United States and Great Britain version which was written in English was created first and then translated into Spanish and French with the Spanish and French versions translated back into English to ensure questionnaire comparability across countries. The French version was used only in France and the Spanish version was used only in the United States with individuals more skilled in understanding and/or reading Spanish than English. The questionnaires were pretested in the respective countries and modifications made where necessary. The questions used in this analysis include the following (which were asked for all respondents except for Question 5 which was asked only of males):
1. Indicate how many sexual partners you have had in each time period:
a. Men in the last five years
b. Men in the last year
c. Women in the last five years
d. Women in the last year
2. In the last five years have you ever had vaginal intercourse (when a man puts his penis into a woman's vagina)?
3. In the last five years have you provided oral sexual gratification to a man (placed your mouth on a man's penis)?
4. In the last five years have you had anal sex with a man (when a man puts his penis into his partner's anus)?
5. In the last five years have you had anal sex with a woman (put your penis into her anus)?
6. Being completely honest, how would you describe your feelings toward your own sex since the age of 15? (Choose One)
a. I have absolutely never felt any sexual attraction towards someone of my own sex.
b. I have felt attracted towards someone of my own sex, but never had any sexual contact with anyone.
c. I have had sexual contact with someone of my own sex, but rarely.
d. I have had sexual contact with someone of my own sex fairly often.
e. I have only ever had sexual contact with people of my own sex.
From these questions, the prevalence of homosexual behavior (Questions 1-5), and the prevalence of homosexual attraction and/or homosexual behavior were examined (Questions 1-6). We examined homosexual behavior by determining the sex of sexual partners (same sex only, opposite sex only, both sexes, or no sexual contact) over the previous 5 years. Men were determined to (i) only have had sex with others of the same sex if they responded positively to 1a, 1b, 3, or 4 and they responded negatively to 1c, 1d, 2, and 5; (ii) only have had sex with the opposite sex if they responded positively to 1c, 1d, 2, or 5 and they responded negatively to 1a, 1b, 3, and 4; (iii) have had sex with both sexes if they responded positively to at least one question in each of the same-sex only and the opposite-sex only categories; and (iv) have had no sexual partners over the previous 5 years if they answered "0" or "no" to Questions 1-5. Women were determined to (i) only have had sex with others of the same sex if they responded positively to 1c or 1d and negatively to 1a, 1b, 2, 3, and 4; (ii) only have had sex with the opposite sex if they responded positively to 1a, 1b, 2, 3, or 4 and negatively to 1c and 1d; (iii) have had sex with both sexes if they responded positively to at least one question in each of the same sex only and the opposite sex only categories; and (iv) have had no sexual partners over the previous 5 years if they answered "0" or "no" to Questions 1-4.
We determined the prevalence of persons who report homosexual behavior and/or homosexual attraction since age 15 using Questions 1 through 6. A man was determined to have experienced homosexual behavior or homosexual attraction by responding positively to 1a, 1b, 3, 4, 6b, 6c, 6d, or 6e. A woman was determined to have experienced homosexual behavior or homosexual attraction by responding positively to 1c, 1d, 6b, 6c, 6d, or 6e.
SURVEY RESULTS
Response rates for the total pool of eligible households were 67% in the United States, 72% in the United Kingdom, and 80% in France yielding the number of interviews for each country and sex as presented in column 1 of Table II. Individuals who failed to participate in the entire survey or who could not even be contacted are assumed to have a distribution of homosexual behavior and attraction similar to that of respondents. We make this assumption because these individuals failed to cooperate before they were informed that the survey was measuring explicit sexual behaviors and sexual attractions.
Our greatest concern is with those individuals that began the survey and then failed to respond to questions necessary for this analysis. Table II presents the percentage of respondents who completed Questions 1 through 5 (column 2), and completed Questions 1 through 5 and Question 6 (column 3). Overall, 4.3-8.2% of respondents did not complete Questions 1 through 5, and 12.3-14.9% of respondents did not complete Questions 1 through 5 and/or Question 6. For those individuals who failed to provide complete responses, responses were imputed using the method of multiple imputation which is described in the following section.
[TABULAR DATA FOR TABLE II OMITTED]
MULTIPLE IMPUTATION FOR NONRESPONSE
The method, validity, and advantages of multiple imputation for non-response in surveys is outlined in detail elsewhere (Rubin, 1987). Therefore, we present only an overview here illustrating the method as it was used to impute values for persons we could not classify into sexual behavior classifications due to their nonresponse. The first step in the multiple imputation process was to examine those individuals that were classifiable into sexual behavior categories (based upon their complete responses to Questions 1 through 5) and determine what responses to other survey questions were associated with their sexual behavior classifications. Using chi-square analysis we found that both (i) marital status and (ii) personally knowing someone diagnosed with AIDS or infected with the AIDS virus were strongly associated with the sexual behavior categories. These two variables were then combined into one variable with 4 possible responses: (i) living together as an unmarried couple or never married and know someone with AIDS, (ii) living together as an unmarried couple or never married and do not know someone with AIDS, (iii) married, separated, divorced, or widowed, and know someone with AIDS, and (iv) married, separated, divorced, or widowed, and do not know someone with AIDS; and cross-tabulated with the sexual behavior classifications of both male and female respondents in each country. In all three countries, for both male and female, a strong association ([[Chi].sup.2], p [less than] 0.001) was found between sexual behavior classifications and the marital status and knowing someone with AIDS combination variable. The respondents who previously could not be classified into sexual behavior categories were then classified into categories based upon their response to these questions. For example, unmarried men who knew someone with AIDS were probabilistically assigned sexual behavior classifications according to the distribution of unmarried men who knew someone with AIDS for whom sexual behavior classifications were known.
This process was performed five times for each nonrespondent to create five distinct data sets (hence the term "multiple imputation"). In each of the five new data sets, individuals who were previously classifiable into a sexual behavior category were assigned the same sexual behavior classification but for nonresponders sexual behavior categories could be different in each new data set as they were assigned probabilistically as described above. The five new data sets were each analyzed separately yielding five estimates for each sexual behavior classification. Imputed prevalence estimates (as reported in Table III) are the average of the five sets of values obtained.
The above multiple imputation method was repeated to impute missing values for the prevalence of persons who report homosexual behavior and/or homosexual attraction since age 15 (as reported in Table IV). As with the sexual behavior classifications, both (i) marital status and (ii) personally knowing someone diagnosed with AIDS or infected with the AIDS virus were highly associated with the variables in question and therefore the combination of these two variables was used to impute missing values.
RESULTS
The results of this analysis are presented in Tables III and IV. These tables each contain two sets of results: (i) based upon complete cases and (ii) using multiple imputation to fill in missing data. As the imputation process made only negligible changes in the results, and the imputed values correct for item nonresponse, only imputed values are presented in the text and discussed below. The results are outlined in the following two sections.
Homosexual Behavior
Table III presents the prevalence of male and female respondents who report at least one sexual contact with others of the same sex, both sexes, [TABULAR DATA FOR TABLE III OMITTED] the opposite sex, or with no one in the previous 5 years. The percentage of respondents who report sexual contact only with others of the same sex is not large and only men in the United Kingdom report a figure greater than 1%. However, we find a substantial proportion of the populations reporting sexual contact with someone of the same sex during the previous 5 years when the "same sex only" and "both sexes" categories are combined. For males this total is 6.2% in the United States, 4.5% in the United Kingdom, and 10.7% in France. For women the numbers are somewhat smaller, with 3.6% in the United States, 2.1% in the United Kingdom, and 3.3% in France reporting sexual contact with someone of the same sex in the previous 5 years.
The percentage of people ages 16-50 who report no sexual behavior during the previous 5 years ranges from 9.2 to 13.0% and the percentage of the population reporting sexual contact with only the opposite sex ranges from 80.1 to 88.1%.
[TABULAR DATA FOR TABLE IV OMITTED]
Homosexual Attraction/Behavior
Population prevalence estimates of homosexual attraction and/or homosexual behavior are presented in Table IV. We find that a total of 20.8, 16.3, and 18.5% of males, and 17.8, 18.6, and 18.5% of females in the United States, the United Kingdom, and France, respectively, report some homosexual attraction or homosexual behavior since age 15.