We used data from the California Health Interview Survey, the largest population-representative state health survey in the nation. The survey is a stratified random sample of households, for which one randomly selected adult from each sampled household is surveyed.20
Its biennial administration facilitates pooling of data to examine health access needs of smaller subpopulations. The survey's large sample and multiethnic/geographical representation are achieved by telephone administration, multiple language interviews, and oversamples of small counties and ethnic groups.
We combined three years of adult files from the California Health Interview Survey (2001, 2003, and 2005) to maximize the number of observations of sexual-orientation minorities. We excluded adults age sixty-five and older, who typically are covered by Medicare (n = 29,623), respondents who did not report their sexual orientation (n = 1,024), and those interviewed by proxy or who had other missing values (n = 433). The final sample, after exclusions, contained 63,719 females and 46,535 males.
Theoretical Basis and Empirical Approach
Coverage disparities in employer-sponsored health insurance may result from employment discrimination or compensation discrimination. Employment discrimination is manifest when employers who offer health insurance are less likely to hire employees with minority sexual orientation. If employers are discriminating against gay men, lesbians, and bisexuals, we would expect a lower likelihood of obtaining own or personal employer-sponsored insurance in this group compared to heterosexuals, after accounting for education, skill level, and other relevant individual characteristics and labor-market factors.
Compensation discrimination, as described by Badgett, is manifest in the customary practice of covering different-sex spouses and not domestic partners or same-sex spouses, thus penalizing employees with a same-sex spouse or partner. Evidence of compensation discrimination can be observed if the likelihood of acquiring dependent health insurance from the employer is lower for employees with a same-sex partner than for those with a different-sex spouse.19
Our study design addressed the empirical challenges in identifying the association of sexual orientation with the uptake of dependent and own health insurance. First, because the California survey includes information on a randomly sampled adult and not all household members, if we were to restrict our study only to partnered or married employees, then nonworking partners and spouses—who rely most on dependent coverage—would be excluded. Our analysis therefore examines both the overall nonelderly adult population and the employed nonelderly adult population. This approach detects potential disparities in dependent health insurance coverage both at the population level and among California's employees.
We coded three categories of sexual orientation: gay or lesbian, bisexual, and heterosexual. Partnered/married status was determined from a single interview item that assessed current marital status. Those responding as married were coded as married, and those responding as living with a partner were coded as partnered. In the employed sample, we excluded self-employed adults and those who typically work zero hours per week.
Our dependent variable was health insurance status, constructed as a categorical variable: uninsured, public insurance (Medicaid and other public programs), own employer-sponsored insurance, dependent coverage, and privately purchased health insurance from the nongroup market. We ascertained health insurance status through a series of questions that probed coverage status at the time of the survey interview.
In the multivariate models, we examined other factors relevant to health insurance coverage, including sociodemographic covariates such as race and ethnicity, age, income, education, citizenship status, partnership status, presence of minor children in the household, language of interview, and rural or urban status. Labor-market characteristics—including hours worked per week, firm size, and industry—were also factors. Possible health care need based on self-rated health status was also examined.
We estimated weighted multivariate multinomial logit models for the full sample, partnered or married adults, employed adults, and partnered or married employed adults. Each model was stratified by sex, yielding a total of eight models. Based on the regression models, we also estimated predicted probabilities of each health insurance status outcome and computed relative risks with bootstrapped 95 percent confidence intervals to evaluate whether there were significant differences by type of coverage by sexual orientation. In the partnered or married models, we compared gay and lesbian partnered adults with heterosexual married adults.
We excluded lesbians, gay men, and bisexuals who reported being married in our partnered or married analyses. Because lesbians and gay men rarely reported being married, as opposed to partnered, we anticipate small, but still potentially biasing, effects from this restriction. In contrast, a sizable minority of bisexual individuals reported current married status. Many of these, we suspect, were heterosexual marriages, but their precise classification as such was not possible. The California surveys did not assess the sex of spouses or partners.
Although we assumed that the great majority of partnered lesbians and gay men are in same-sex partnerships and married or partnered heterosexuals are in different-sex partnerships, this is indeterminable.We assumed that greater potential misclassification errors exist for bisexuals. For both brevity and clarity, we limit our detailed presentation of predicted probabilities and relative risks to comparisons between lesbians or gay men and heterosexuals. Results from the full regression models that include comparisons between bisexuals and heterosexuals are available in the Appendix.16