Sexual Minority Status and Self-Rated Health: The Importance of Socioeconomic Status, Age, and Sex

Abstract

Objectives. I examined how sexual minority status, as indicated by sex of sexual partners, is associated with self-rated health and how socioeconomic status suppresses and age and sex moderate this association.

Methods. I used multinomial logistic regression to analyze aggregated data from the 1991 to 2010 General Social Survey, a population-based data set (n = 13 480).

Results. Respondents with only different-sex partners or with any same-sex partners reported similar levels of health. With socioeconomic status added to the model, respondents with any same-sex partners reported worse health than those with only different-sex partners, but only if sexual intercourse with same-sex partners occurred in the previous 5 years. Age and sex moderated this relationship: having any same-sex partners was associated with worse health for women but not men and among younger adults only.

Conclusions. The relationship between sexual minority status and self-rated health varies across sociodemographic groups. Future research should use population-level data to examine other health outcomes and continue to explore how the intersection of sexual minority status and other sociodemographic indicators shapes health.

Sexual minorities, defined as persons who are sexually attracted to people of their own sex; have sexual relations with people of their own sex; or identify as gay, lesbian, bisexual, or queer,¹ are disadvantaged in many physical and mental health outcomes, including cancer, depression, HIV/AIDS, obesity, anxiety disorders, cardiovascular disease, and disability, relative to heterosexuals.^1–5 Most previous studies did not examine variation in sexual minority health by age or sex nor how the sexual minority health disadvantage might be linked to socioeconomic status (SES), a composite measure of economic status (measured by income or wealth) and social status (measured by education).⁶

Because people occupy many social identities and may possess multiple marginalized statuses that both independently and relatedly influence health, I advocate for an intersectional approach to studying sexual minority health.^7,8 I examined how the relationship between sexual minority status and health is suppressed by SES and moderated by age and sex. I assessed suppression by SES because high SES improves health through multiple pathways (e.g., access to care and healthy foods, education, living conditions).^6,9 Sexual minorities, on average, report higher SES than do heterosexuals,^10,11 and this SES advantage may mask their health disadvantage. I also considered moderation by age and sex. Previous research indicated that both age and sex significantly shape racial, SES, and marital status health disparities.^12–14 Thus they likely influence sexual minority health disparities, although this has been largely unexamined.

FACTORS THAT INFLUENCE DISPARITIES

SES is inversely related to morbidity and mortality risk.^6,15,16 The relationship between sexual minority status and SES is complicated because sexual minority women have higher incomes than do heterosexual women, but this is not the case among men.^11,17,18 Sexual minorities also report higher levels of educational attainment than do heterosexuals.¹⁰ Sexual minorities with low SES report worse health than do those with higher SES.⁹ Previous studies, however, relied on samples composed entirely of sexual minorities (and thus no comparison group)⁹ or nonrepresentative samples.¹⁹ Thus we do not yet know the importance of SES in explaining or exacerbating sexual minority health disparities. However, I expect that SES suppresses sexual minority health disparities, particularly among women, such that these disparities are only apparent when SES is added to analytic models.

Age moderates the relationship between health and other sociodemographic indicators, namely, SES and race,^12,20 and thus likely moderates the relationship between sexual minority status and health. Past studies on sociodemographic health disparities and age focused on 2 contradictory hypotheses.¹² The age-as-leveler hypothesis states that health inequalities are smaller at older ages, mainly because of mortality selection. The hardiest individuals from all sociodemographic groups, even disadvantaged groups, survive to the oldest ages, reducing any health disparities.^21–23 The cumulative disadvantage hypothesis posits that health risks and conditions accumulate over the life course, leading to chronic strains and stress, which wear down the body, such that disadvantaged groups become even more disadvantaged in health at older ages.^12,24–26

Although worse health is reported by older than younger sexual minority individuals,⁹ it is untested whether the age-as-leveler hypothesis or the cumulative disadvantage hypothesis best describes patterns of health disparities by sexual minority status. I examined this at the aggregate level.

Women report worse health than men across the life course.^27,28 Sex also significantly shapes health disparities within other groups. Widowed men experience more adverse health outcomes than do widowed women.^13,29 In addition, health disparities by SES are greater for men than women.^14,30 Among sexual minorities, men and women experience different disadvantages across health outcomes. Sexual minority men have higher rates of eating disorders³¹ and lifetime mental health problems^2,5 than do heterosexual men, and sexual minority women experience higher rates of obesity^4,5,32 and alcohol abuse⁵ than do heterosexual women. Most previous studies relied on samples of 1 sex only or did not stratify by sex. Thus, although past research suggests that the intersection of sexual minority status and sex significantly shapes health disparities, this has not been formally tested.

My study advances our theoretical and scientific understanding of how sexual minority status affects health in 2 ways: by using nationally representative data and by treating sexual minorities as a heterogeneous group. Few reports in the literature included discussions of how sexual minority health disparities vary across sociodemographic groups, mainly because most past studies relied on small, nonrepresentative samples that lacked the statistical power for subsample analyses.^1,9 The few studies that looked at sexual minority disparities across groups, although they used small, nonrepresentative samples, provide support for the assertion that sexual minority health inequalities should be considered alongside other types of inequality.

I examined the relationship between self-rated health and sexual minority status in a large, nationally representative sample. I analyzed how SES suppresses the relationship between sexual minority status and self-rated health and how age and sex moderate this relationship. This approach acknowledges the inherent complexities in investigating sexual minority status and health by placing sexual minority health disparities in the broader context of socioeconomic, age, and sexual diversity.

METHODS

I used data from the General Social Survey (GSS) from 1991 through 2010.³³ The National Opinion Research Center, a social science research center at the University of Chicago, conducted the survey annually between 1972 and 1994 (except in 1981 and 1992) and biennially beginning in 1996. The GSS uses a multistage area probability sampling design and questions on a wide range of topics. Questions about sexual behavior were added in 1988. The GSS is one of the few national probability surveys that collects information on sex of sexual partners or any other proxy for sexual minority status (e.g., sexual identity) across all adult ages. Other surveys either do not include variables that can be used to identify sexual minorities or only ask these questions of respondents at certain ages.¹ Other researchers have used the GSS to identify population-level trends among sexual minorities,^10,11,34 and I built on their use of the GSS data for my analysis. I analyzed data from 13 480 respondents, after excluding those with missing data on variables of interest.

Measures

Self-rated health.

The primary outcome variable was self-rated health. Multiple studies demonstrate that single-item self-rated health measures serve as valid and reliable indicators of health status in the general population^35,36 and are highly predictive of subsequent morbidity and mortality.³⁷ GSS respondents were asked, “Would you say your own health, in general, is excellent, good, fair, or poor?” The Brant test and likelihood ratio test indicated that the proportional odds assumption was violated, so I fit a multinomial logistic model with excellent health as the comparison category rather than an ordered logistic model. Only a few respondents described their health as poor (n = 412), so I combined poor and fair into 1 category, as others have done.^36,37 As a sensitivity analysis, I fit the model with 4 separate categories, but this did not alter the results statistically or substantively.

Sexual minority status.

Sexual minority status is a complex construct incorporating dimensions of attraction, behavior, and identity¹; however, the GSS, until recently, only asked about sexual behavior. Since 1989, the GSS has asked, “Now thinking about the time since your 18th birthday (including the past 12 months), how many male partners have you had sex with?” A parallel question asked about number of female partners. In the total sample, 6.1% of respondents reported any same-sex partners since age 18 years. Since 1991, the GSS has asked respondents, “Have your sex partners in the last five years been exclusively male, both male and female, or exclusively female?” In the total sample, 3.6% of respondents reported any same-sex partners in the past 5 years.

From responses to these 2 questions, I constructed a measure with 3 categories: those who reported any same-sex partners in the past 5 years, those who reported any same-sex partners since age 18 years but not in the past 5 years, and those who reported only different-sex partners since age 18 years (reference). For ease of discussion, I refer to these categories as any recent same-sex partners, any lifetime but no recent same-sex partners, and only different-sex partners. I excluded respondents who reported no sexual partners since age 18 years (n = 1025) because, by definition, they provided no information on the sex of their sexual partners. Most respondents who reported any same-sex partners reported more than 1 same-sex partner, indicating that for most respondents, same-sex partners were not a temporary phase, but an enduring pattern.

I included a measure of recent same-sex partners along with a measure of lifetime but not recent same-sex partners for 4 reasons. First, same-sex partners earlier in the life course may significantly affect self-rated health, because self-rated health at 1 point reflects the accumulation of health and experiences throughout the life course.³⁸ Second, many adults are celibate at older ages,³⁹ and they, along with other currently celibate adults, would be excluded from the analysis if I had used only the recent measure. Third, the recent measure more accurately reflected a participant’s current sexual minority status, which likely had a more important bearing on self-rated health than a lifetime measure, as has been demonstrated by studies of relationship status and self-rated health.⁴⁰ Fourth, the recent measure of sex of sexual partners might have been less affected by recall bias and thus more accurate than the lifetime measure.⁴¹

Socioeconomic status.

SES indexes a person’s position in the unequal distribution of socially valued resources, generally measured by references to education and economic standing.⁴² I assessed SES from questions about educational attainment and family income. Income and educational attainment represent separate components of SES, and studies indicate that they exhibit related but independent effects on health.⁴³

For educational attainment, respondents reported their highest degree earned. I created 3 dummy variables: less than high school, high school, and some college and higher (reference). For income, respondents reported their family’s income within 23 categories in $10 000 units. I recoded family income at the midpoint of these categories in 6-digit numbers and adjusted it for inflation so that income across all years reflected values from 2000. I used multiple imputation to replace the missing income values and to retain the maximum number of respondents. I used the log of income to reflect income’s curvilinear association with health.⁴⁴

Age.

Age was reported in years and treated as continuous. Respondents aged 89 years and older were coded as 89 in the original GSS data. Respondents ranged in age from 18 to 89 years.

Sex.

Sex was derived from reports of respondents, who identified as male or female. For my analysis, respondents who identified as male were the reference category, and respondents who identified as female composed the comparison group.

Covariates.

In each model, I controlled for age, sex, year of interview (data were pooled from 1991 to 2010), race (dummy variables Black and other, with White as the reference, self-reported by respondents), and marital status (dummy variables widowed, divorced or separated, and never married, with currently married as the reference). I included race and marital status as covariates because they significantly shape health.^45,46 It was particularly important to include marital status because of the hypothesis that the relationship between sexual minority status and health is explained by the health benefits of marriage legally denied to most sexual minorities.¹³

Missing data.

I excluded respondents with any missing data from the analysis, with the exception of those missing data on income, which I imputed. None of the GSS respondents were missing data on race or sex. Respondents with missing data on marital status (n = 3), age (n = 22), and education (n = 43) did not differ significantly from those not missing these data. Most missing information on the sex of sexual partners and self-rated health was missing because of the GSS design, in which these questions were part of a supplement rather than the main survey.⁴⁷ Respondents missing information on self-rated health and sex of sexual partners for reasons other than survey design were older and less educated than other respondents. These questions were asked at the end of the survey, when older and less educated respondents were more likely to have quit taking the survey for a variety of reasons, not just a reluctance to answer questions regarding sexual behavior.^47,48

As a sensitivity analysis, I fit the models with dummy variables for missing responses, and these variables did not predict self-rated health. This supports the decision to drop respondents with missing variables.

Analysis

I first calculated descriptive statistics for each variable, stratified by respondents who reported only different-sex partners, respondents with any recent same-sex partners, and respondents with lifetime but not recent same-sex partners (Table 1). I also examined whether the respondents who reported any same-sex partners were statistically different than those who reported only different-sex partners for each variable of interest. To test differences, I used the χ² test for categorical variables and ordinary least squares regression for continuous variables.

TABLE 1—

Descriptive Statistics of Participants by Sex of Sexual Partners: General Social Survey, 1991–2010

	Different-Sex Partners Only (n = 12 652)		Any Recent Same-Sex Partners (n = 435)		Any Lifetime But No Recent Same-Sex Partners (n = 393)
	No.	Mean (SD)	No.	Mean (SD)	No.	Mean (SD)
Self-rated health
Excellent	3929	0.31 (0.46)	124	0.29 (0.45)	119	0.30 (0.46)
Good	6033	0.48 (0.50)	219	0.50 (0.50)	171	0.44 (0.50)
Fair or poor	2690	0.21 (0.41)	92	0.21 (0.41)	103	0.26* (0.44)
Age, y		44.80 (16.46)		38.92*** (12.57)		45.39 (15.68)
Female	7114	0.56 (0.50)	236	0.54 (0.50)	207	0.53 (0.50)
Year of interview		2000.23 (5.36)		2000.93** (5.24)		2002.81*** (5.51)
Race
White	10 365	0.82 (0.38)	352	0.81 (0.39)	302	0.77* (0.42)
Black	1533	0.12 (0.33)	53	0.12 (0.33)	60	0.15 (0.36)
Other	754	0.06 (0.24)	30	0.07 (0.25)	31	0.08 (0.27)
Marital status
Married	6416	0.51 (0.50)	79	0.18*** (0.39)	159	0.40*** (0.49)
Widowed	1040	0.08 (0.27)	14	0.03*** (0.18)	14	0.05* (0.22)
Divorced/separated	2437	0.19 (0.39)	89	0.20 (0.40)	91	0.23 (0.42)
Never married	2759	0.22 (0.41)	253	0.58*** (0.49)	123	0.31*** (0.46)
Education
≥ some college	4303	0.34 (0.47)	198	0.46*** (0.50)	136	0.35 (0.48)
High school	6792	0.54 (0.50)	203	0.47** (0.50)	194	0.49 (0.50)
< high school	1557	0.12 (0.33)	34	0.08** (0.27)	63	0.16* (0.37)
Log (income), $		10.44 (0.99)		10.33* (1.02)		10.18*** (1.23)

Note. Lifetime was with partners reported from age 18 years; recent was defined as past 5 years.

*P < .05; **P < .01; ***P < .001.

I then fit a series of multinomial logistic regressions with my categorical measures of sex of sexual partner as primary predictor variable. Multinomial logistic regression uses maximum likelihood to estimate the log odds of being in a given self-rated health category (fair or poor; good) compared with the reference category (excellent), allowing for separate slope estimates.⁴⁹ In the first model, I controlled for age, sex, year of interview, race, and marital status. In the second model, I added SES controls: dummy variables for highest degree earned and log of family income. In the third model, I added interactions for same-sex partners and the age categories, and in the fourth model, I replaced these interactions with interactions between same-sex partners and sex. I also tested for moderating effects of SES (data not shown because not significant).

As a sensitivity analysis, I divided the full sample in half and fit each model to both halves. This verified that my results from the full sample were robust. I used Stata SE 11.0 for all analyses.⁵⁰ In addition, I used linear regression models to examine whether variables used in the model were too highly collinear. Tolerance levels were all greater than 0.64, indicating variance inflation factor values that were all below 1.56, within the acceptable range.⁵¹

RESULTS

Overall, many of the respondents (48%) reported good health. Significance tests revealed only 1 difference in self-rated health between the categories: those who reported any lifetime but no recent same-sex partners were slightly more likely to report poor or fair health than were those who reported only different-sex partners. This may be because those who reported any lifetime but no recent same-sex partners were significantly older than those who reported only different-sex partners. Compared with those who reported only different-sex partners, respondents who reported any recent same-sex partners were younger, were less likely to be married or widowed, were more likely to be never married, were more educated, and had higher incomes. Respondents who reported any lifetime but no recent same-sex partners were similar to those who reported only different-sex partners (Table 1).

For all models, I reported results in the form of relative risk ratios (RRRs; exponentiated coefficients), which indicated the probability of reporting either poor or fair health (first column for each model) or good health (second column for each model) divided by the probability of reporting excellent health. I also reported corresponding 95% confidence intervals. The baseline model, model 1 (Table 2), tested how sexual minority status, as indicated by any same-sex partners, influenced the risk of rating health as good or fair or poor relative to excellent. In this baseline model, the covariates were age, sex, year of interview, race, and marital status. Regardless of measure used, respondents reporting any same-sex partners had no higher RRR of reporting good or fair or poor health than did respondents with only different-sex partners. In all models, age, year of interview, race, and marital status significantly predicted the RRR of reporting good or fair or poor health.

TABLE 2—

Multinomial Logistic Regression Predicting Self-Rated Health by Sex of Sexual Partners: General Social Survey, 1991–2010

	Model 1, Self-Rated Health		Model 2, Self-Rated Health
	Good, RRR (95% CI)	Fair–Poor, RRR (95% CI)	Good, RRR (95% CI)	Fair–Poor, RRR (95% CI)
Sexual partners
Different-sex onlya (Ref)	1.00	1.00	1.00	1.00
Any recent same-sexb	1.18 (0.94, 1.48)	1.19 (0.90, 1.57)	1.29 (1.03, 1.63)	1.51 (1.12, 2.02)
Any lifetime but no recent same-sexc	0.90 (0.71, 1.14)	1.14 (0.86, 1.50)	0.87 (0.68, 1.11)	0.98 (0.73, 1.32)
Age	1.01 (1.01, 1.01)	1.03 (1.03, 1.03)	1.01 (1.01, 1.01)	1.03 (1.03, 1.03)
Sex
Male (Ref)	1.00	1.00	1.00	1.00
Female	0.95 (0.88, 1.03)	0.99 (0.90, 1.10)	0.92 (0.85, 1.00)	0.90 (0.81, 1.00)
Race
White (Ref)	1.00	1.00	1.00	1.00
Black	1.34 (1.18, 1.53)	1.97 (1.69, 2.29)	1.19 (1.04, 1.36)	1.39 (1.18, 1.63)
Other	1.24 (1.05, 1.47)	1.76 (1.43, 2.15)	1.20 (1.01, 1.43)	1.51 (1.21, 1.87)
Marital status
Married (Ref)	1.00	1.00	1.00	1.00
Widowed	1.28 (1.06, 1.55)	1.86 (1.52, 2.28)	0.94 (0.78, 1.15)	0.89 (0.71, 1.11)
Divorced/separated	1.22 (1.09, 1.36)	1.87 (1.64, 2.12)	1.03 (0.92, 1.16)	1.21 (1.05, 1.39)
Never married	1.04 (0.93, 1.15)	1.37 (1.19, 1.58)	0.87 (0.78, 0.98)	0.79 (0.67, 0.92)
Education
≥ some college (Ref)	1.00	1.00	1.00	1.00
High school			1.51 (1.39, 1.65)	2.22 (1.96, 2.51)
< high school			2.04 (1.73, 2.42)	5.36 (4.42, 6.50)
Log (income), $			0.80 (0.76, 0.85)	0.57 (0.53, 0.61)

Note. CI = confidence interval; RRR = relative risk ratio. Reference category was excellent self-rated health. Sample size was n = 13 480. Models adjusted for year of interview. Model 2 also adjusted for educational attainment and income.

^aSince age 18 years.

^bIn past 5 years.

^cSince age 18 years, but not in past 5 years.

Socioeconomic Status as Suppressor

I included SES variables in the models to test their suppressing effects on the relationship between same-sex partners and self-rated health. SES did not alter the relationship between same-sex partners and self-rated health among respondents reporting lifetime but no recent same-sex partners. But respondents with any recent same-sex partners had a 51% higher likelihood of reporting fair or poor health (P < .01) and a 29% higher likelihood of reporting good health (P < .05) than did respondents with only different-sex partners (Table 2).

In these models, for both indicators, the relationships between self-rated health and race and self-rated health and marital status were partially attenuated, indicating that SES significantly drove these relationships. Also, for both indicators, SES was a significant predictor of self-rated health (P < .001), with adjustment for same-sex sexual partners and the covariates. As a test of the robustness of these analyses, I conducted the same analyses with ordered logistic regressions with standardized y* coefficients and ordinary least squares regressions treating self-rated health as continuous. In each additional model, the results remained consistent.

Age and Sex as Moderators

I fit models with interactions between age and the sexual minority status variables. The RRRs and 95% confidence intervals are reported in model 3 in Table 3. I controlled for the same covariates as in model 2 (not shown in table). The interaction between age and any recent same-sex partners was significant, indicating that age moderated the relationship between any recent same-sex partners and self-rated health, such that a sexual minority self-rated health disparity attenuated with increasing age. Figure 1 illustrates these relationships, showing RRRs of reporting fair or poor health or good health stratified by age category among only participants who reported recent same-sex experience. The relationship between sexual minority status and self-rated health was significant only for those in the youngest age groups, 18 to 29 years and 30 to 39 years, with adjustment for other factors (P < .05). The figure illustrates that respondents aged 18 to 39 years who reported any same-sex partners in the past 5 years had about 1.5 to 2.5 times the probability as respondents with only different-sex partners to report fair or poor health.

TABLE 3—

Multinomial Logistic Regression Predicting Self-Rated Health by Sex of Sexual Partners, Interactions with Age and Sex: General Social Survey, 1991–2010

	Model 3, Self-Rated Health		Model 4, Self-Rated Health
	Good, RRR (95% CI)	Fair–Poor, RRR (95% CI)	Good, RRR (95% CI)	Fair–Poor, RRR (95% CI)
Sexual partners
Different-sex onlya (Ref)
Any recent same-sexb	2.59 (1.22, 5.53)	4.72 (1.86, 12.02)	0.99 (0.72, 1.38)	1.10 (0.71, 1.70)
Any lifetime but no recent same-sexc	2.26 (1.06, 4.85)	1.30 (0.52, 3.27)	0.78 (0.53, 1.14)	0.88 (0.55, 1.43)
Age	1.01 (1.01, 1.01)	1.03 (1.03, 1.03)	1.01 (1.01, 1.01)	1.03 (1.03, 1.03)
Interactions with age
Age * any recent same-sex partnersb	0.98 (0.96, 1.00)	0.97 (0.95, 0.99)
Age * any lifetime but no recent same-sex partnersc	0.98 (0.96, 0.99)	0.99 (0.98, 1.01)
Sex
Male (Ref)
Female	0.92 (0.85, 1.00)	0.90 (0.81, 1.00)	0.88 (0.80, 0.96)	0.87 (0.77, 0.97)
Interactions with sex (female)
Sex * any recent same-sex partnersb			1.70 (1.07, 2.68)	1.89 (1.05, 3.40)
Sex * any lifetime but no recent same-sex partnersc			1.69 (1.00, 2.87)	1.53 (0.80, 2.95)

Note. CI = confidence interval; RRR = relative risk ratio. Reference category was excellent self-rated health. Sample size was n = 13 480. Models adjusted for year of interview, race, marital status, educational attainment, and log of income. Model 3 included interactions between age and sex of sexual partner. Model 4 included interactions between sex and sex of sexual partner.

^aSince age 18 years.

^bIn past 5 years.

^cSince age 18 years, but not in past 5 years.

FIGURE 1—

Relative risk ratio of reporting fair or poor health or good health, stratified by age, for any recent same-sex partners compared with only different-sex partners: General Social Survey, 1991–2010.

Note. Reference is excellent self-rated health, adjusted for year of interview, sex, race, marital status, educational attainment, and log of income.

*P < .05; **P < .01.

Finally, I fit a model with interactions between sex and the sexual minority status variables. The RRRs and 95% confidence intervals are reported in model 4 in Table 3. I adjusted for the same covariates as in model 2 (not shown in table). The sex interaction was significant for recent same-sex partners (P < .05). The significance of this interaction variable indicated that sex moderated the relationship between the recent measure of same-sex partners and self-rated health, such that a sexual minority self-rated health disparity existed only for women. Figure 2 illustrates these RRRs of reporting fair or poor health stratified by sex among only participants who reported having recent same-sex partners. This figure indicates that the association between same-sex partners and self-rated health was only significant for women, not men, with adjustment for other factors. Women with any recent same-sex partners had about twice the likelihood of reporting fair or poor health as women with only different-sex partners.

FIGURE 2—

Relative risk ratio of reporting fair or poor health or good health, stratified by sex, for any recent same-sex partners compared with only different-sex partners: General Social Survey, 1991–2010.

Note. Reference is excellent self-rated health, adjusted for year of interview, age, race, marital status, educational attainment, and log of income.

**P < .01.

DISCUSSION

I used population-level data to examine the association between sexual minority status and self-rated health, as well as how SES suppresses and age and sex moderate this association. Previous studies of sexual minority status and health have primarily drawn on nonrepresentative samples and have not examined how sexual minority status intersects with other sociodemographic indicators to shape health outcomes. Examining sexual minority status in the context of other sociodemographic indicators with population-level data is critical for understanding and improving the health of sexual minorities in the United States. My analysis yielded several important new findings.

Sexual minority status, even indicated by a crude measure of sex of sexual partners, was associated with self-rated health status. I found a sexual minority health disadvantage, such that respondents with any recent same-sex partners reported worse health than did respondents with only different-sex partners, but only when SES was taken into account. As found in other studies, sexual minorities reported higher levels of education than did heterosexual participants, and sexual minority women reported higher levels of income (not shown).^11,17 When I controlled for these SES differences, a sexual minority health disadvantage emerged. This finding highlights the importance of examining sexual minority health in the context of SES to more accurately gauge the extent of this inequality. Previous studies that did not consider SES likely underestimated the extent of the sexual minority health disadvantage.

The self-rated health disadvantage experienced by sexual minorities was not apparent among men and middle-aged and older adults. The absence of disadvantage among older adults at the aggregate level supports the age-as-leveler hypothesis, which states that health disadvantages within sociodemographic groups decrease with age. This finding may be attributable to the cross-sectional design of the GSS, which does not take mortality selection into account. In addition, this suggests that sexual minority status may affect the health of younger adults but not middle-aged and older adults through unique processes. For example, discrimination, sexual minority stress and health behavior disparities may be most salient at the youngest ages.^3,52,53 Future longitudinal work as well as more attention to cohort-specific processes is required to elucidate this finding.

I observed a health disadvantage for sexual minorities among women but not men. Most past research on sexual minorities and health focused on gay and bisexual men and the AIDS epidemic.³ My analysis indicates that important health inequalities exist for women, suggesting that the intersection of sex and sexual minority status remains an important topic for future research on sexual minority health.

The sexual minority disadvantage regarding self-rated health was more pronounced when I used the more recent measure of sexual minority status (same-sex partners in the past 5 years) rather than the lifetime measure (same-sex partners since age 18 years but none recently). This indicated that the design of surveys and questions importantly shaped results. A more recent measure of sexual partners appeared to be more important for self-rated health than a lifetime measure. The use of other health outcomes (e.g., chronic conditions) and other sexual minority status measures (e.g., attraction or identity measures) could lead to different conclusions, as demonstrated by other studies of life course processes, sexual minority status, and health.^2,40

Limitations

Because of small sample sizes, I pooled 19 years of data collections. Important social, political, and cultural changes occurred for sexual minorities during this period, but I was unable to take this into account in my analysis. Although I adjusted for year of interview in every model, this did not sufficiently account for the important period and cohort effects that likely influenced the results, especially the age interactions.

My measure of sexual minority status was a crude measure of sexual behavior (whether a respondent ever, even once, had sexual intercourse with someone of the same sex). This might also have been subject to recall bias.⁴¹ More comprehensive and multidimensional measures would likely produce different results. Bostwick et al. found that mental health outcomes vary by the dimension of sexual minority status examined²; this is also likely the case for self-rated health.

Respondents who reported both same-sex and different-sex partners were grouped with respondents who only had same-sex partners, rather than considered separately. Many respondents who reported any same-sex partners also reported a different-sex partner, but because of sample size, I did not stratify these groups. Past studies emphasized the importance of considering bisexuals separately, because they often experience an even greater health disadvantage than do other sexual minority persons.³ Preliminary analysis of our data (not shown) also revealed such a disparity.

Because I used multinomial logistic regression, which meant that the variance of the outcome variable shifted as other variables were added to the model, I was unable to formally test for mediation or suppression as I could in a linear regression model.⁵⁴ I performed supplementary analyses (fitting both an ordinary least squares regression and an ordered logistic regression) to confirm that education suppressed the relationship between reporting any same-sex partners and self-rated health, but I was unable to estimate the extent of this suppression effect, nor was I able to formally test for mediated moderation or moderated mediation.

Conclusions

My analysis demonstrates the value of using population-level data and considering other sociodemographic indicators alongside and in confluence with sexual minority status. My findings also reveal the important ways that sexual minority status shapes health, showing that sexual minority status is an important sociodemographic indicator that should not be ignored by public health, medical demography, sociology, or epidemiology.

Future research on sexual minority health should use population-level data to examine other health outcomes beyond self-rated health and continue to explore how sexual minority status intersects with multiple sociodemographic indicators. Using population-level data to examine how sexual minority status intersects with other sociodemographic indicators to shape self-rated health allows for better targeted care and policies aimed at improving the health of sexual minorities.