Trends in Sexual Orientation Missing Data Over a Decade of the California Health Interview Survey

Abstract

Objectives. We explored changes in sexual orientation question item completion in a large statewide health survey.

Methods. We used 2003 to 2011 California Health Interview Survey data to investigate sexual orientation item nonresponse and sexual minority self-identification trends in a cross-sectional sample representing the noninstitutionalized California household population aged 18 to 70 years (n = 182 812 adults).

Results. Asians, Hispanics, limited-English-proficient respondents, and those interviewed in non-English languages showed the greatest declines in sexual orientation item nonresponse. Asian women, regardless of English-proficiency status, had the highest odds of item nonresponse. Spanish interviews produced more nonresponse than English interviews and Asian-language interviews produced less nonresponse when we controlled for demographic factors and survey cycle. Sexual minority self-identification increased in concert with the item nonresponse decline.

Conclusions. Sexual orientation nonresponse declines and the increase in sexual minority identification suggest greater acceptability of sexual orientation assessment in surveys. Item nonresponse rate convergence among races/ethnicities, language proficiency groups, and interview languages shows that sexual orientation can be measured in surveys of diverse populations.

Measuring sexual orientation in health surveys facilitates comprehensive public health surveillance. Accumulating evidence suggests that some lesbians, gay men, and bisexual individuals, compared with heterosexual persons, have higher smoking rates,^1,2 greater second-hand smoke exposure,³ more psychological distress^4–6 and depression,⁷ higher suicide attempt rates,⁸ worse general health status⁹ and higher disability rates,¹⁰ and lower preventive care use.¹¹ As a reflection on these emerging findings, calls for greater collection of sexual orientation data abound,^12–17 but the validity of sexual-minority research is threatened if survey respondents cannot or will not provide these data.

Several large health surveys now routinely measure sexual orientation. Since 2001, the California Health Interview Survey (CHIS) has included questions assessing self-identified sexual orientation.¹⁸ Twelve Behavioral Risk Factor Surveillance System state surveys also asked sexual orientation questions at least once between 2000 and 2014.¹⁹ Other large-scale surveys currently asking sexual orientation include the Los Angeles County Health Survey,²⁰ National Health Interview Survey (NHIS),²¹ National Health and Nutrition Examination Survey,²² and General Social Survey.²³

Results from these surveys indicate that most respondents provide a codeable sexual orientation response. One percent of NHIS respondents in 2013 did not respond when asked their sexual orientation. “Don’t know” responses comprised 0.4% and refusals made up 0.6%.²⁴ In 2003 to 2010 Washington State Behavioral Risk Factor Surveillance System data, 0.74% responded “don’t know” or “not sure,” and 1.12% refused.²⁵ Yet, African Americans, Asian Americans, and Hispanics in that study had higher odds of nonresponse than Whites. This raises questions about possible sociodemographic differences in sexual orientation measurement, but there have been few assessments of the combined roles of race/ethnicity and language in sexual orientation item nonresponse, and changes in those effects over time.^25,26 The independent effects of English proficiency and interview language remain largely unexplored as well. Linguistic and ethnic minorities who are also sexual minorities may be underrepresented in routine public health surveillance efforts if they are differentially likely to answer sexual orientation questions.²⁵ Understanding the relationship among sexual orientation item nonresponse, race/ethnicity, and language proficiency is important because these sociodemographic domains correlate with health disparities.^27-29

Sexual orientation nonresponse is likely attributable to social stigma of identification and a lack of understanding of the terminology used to discuss the topic.³⁰ Secular trends in lesbian, gay, and bisexual (LGB) social and legal recognition³¹ may increase LGB individuals’ willingness to disclose their sexual orientation in surveys. The non-LGB public may also become more comfortable with and knowledgeable about the topic as a result. Public opinion surveys now show majority support for gay marriage and LGB people in general.^32,33 Sexual orientation item nonresponse should decline, and the percentage of respondents identifying as LGB may increase as stigma recedes and familiarity grows. These potential effects may be more pronounced among racial, ethnic, and linguistic minorities.^25,26

Two primary research questions guided this study:

• (1) Does the sexual orientation item nonresponse rate change over time?

• a. If so, is this change constant across races/ethnicities, English proficiency levels, and interview languages?

• b. How strongly do race/ethnicity, English proficiency, and interview language predict sexual orientation nonresponse?

• (2) Does LGB identification vary over the same time period?

METHODS

The CHIS has asked sexual orientation via telephone interview with identical question wording and response options since 2003. The question reads: “Do you think of yourself as straight or heterosexual, as [gay/gay, lesbian] or homosexual, or bisexual?” Bracketed text fills with “gay” for men and “gay, lesbian” for women. The use of multiple terms for heterosexual and homosexual allows for various ways that sexual minority members may identify themselves, and variability in respondents’ terminology knowledge. Interviewers could define terms at a respondent’s request or uncertainty. Responses were recorded in 7 categories: straight or heterosexual; gay, lesbian, or homosexual; bisexual; not sexual, celibate, or none; other (with open-text specification option); refused; and don’t know. “Not ascertained” was recorded if the interviewer could not code the respondent’s answer or nonanswer into one of the 7 categories. The question was only asked of respondents aged 18 to 70 years after 2003.

Sample

This study focused on adults aged 70 years or younger who answered “I don’t know” or “refused,” or from whom an answer was “not ascertained.” As part of standard CHIS data production and preparation for general use, open-ended responses were coded into substantive categories (e.g., straight, lesbian, gay, bisexual, or a single nonsexual, celibate, or other category) when there was sufficient detail in the open-ended response to do so.

CHIS used a dual-frame landline and cell phone random-digit-dial sample of the noninstitutionalized California household population.³⁴ We used data from five 2-year CHIS cycles (2003–2011) in this study. Of the 226 661 interviews conducted, 182 812 respondents were used (Table 1 shows unweighted descriptive statistics and their percentage or mean among sexual orientation respondents and nonrespondents). The sample was mostly White (59.99%), mostly interviewed in English (87.52%), roughly balanced over cycles (18.17% to 22.32%), mostly living in the United States greater than 75% of their life (77.87%), relatively young (47.12 years old; because of the exclusion of respondents aged 71 years and older), relatively well-off (59.88% at or above 300% of the federal poverty level [updated with each cycle of data collection]³⁵), and relatively educated (51.06% with a college degree or higher). Ten percent of the sample was Asian or Native Hawaiian/Pacific Islander (55.06% limited-English-proficient [LEP]). Hispanics made up about 22% of the sample (54.92% LEP). We excluded American Indian and Alaska Native respondents because of small sample sizes that inhibited analysis. All the variables in Table 1 had statistically significant associations with sexual orientation nonresponse.

TABLE 1—

Unweighted Univariate Regressors Used in Logistic Regression Model and Their Bivariate Distributions by Sexual Orientation Nonresponse With χ² Test of Association: California Health Interview Survey, 2003–2011

		Sexual Orientation Response, % or Mean
Variable	Overall, % or Mean (n = 182 812)	Answered (n = 179 558)	Did Not Answer (n = 3 254)
Race/ethnicity and LEP**
White: LEP and non-LEP	59.99	98.86	1.14
Asian and NHPI: non-LEP	4.62	97.59	2.41
Asian and NHPI: LEP	5.66	96.05	3.95
Hispanic: non-LEP	9.99	99.10	0.90
Hispanic: LEP	12.17	95.28	4.72
Black: LEP and non-LEP	4.92	98.54	1.46
Multirace: LEP and non-LEP	2.63	99.00	1.00
Interview language**
English	87.52	98.67	1.33
Spanish	9.32	94.30	5.70
Chinese, Korean, Vietnamese	3.16	97.25	2.75
Survey cycle**
2003	19.66	97.89	2.11
2005	19.73	98.18	1.82
2007	22.32	98.33	1.67
2009	20.12	98.32	1.68
2011	18.17	98.38	1.62
Percentage life lived in United States**
> 75% of life	77.87	98.85	1.15
≤ 75% of life	22.13	96.01	3.99
Mean age,** y	47.12	47.11	47.76
Poverty**
≥ 300% FPL	59.88	98.73	1.27
200%–299% FPL	12.52	98.37	1.63
100%–199% FPL	15.55	97.77	2.23
0%–99% FPL	12.06	96.13	3.87
Education**
≥ college degree	51.06	98.51	1.49
Some college	16.47	98.95	1.05
High-school graduate	21.87	98.57	1.43
< high school	10.60	94.98	5.02
Gender**
Female	58.15	98.07	1.93
Male	41.85	98.43	1.57

Note. FPL = federal poverty level; LEP = limited English proficiency; NHPI = Native Hawaiian/Pacific Islander. Significance based on χ² test of cross-tabulation of each predictor and sexual orientation nonresponse indicator. For age, significance reflects a t test of difference in mean ages between respondents and nonrespondents. Unweighted tests are reported. Survey-weighted statistical conclusions where identical. Some variables do not sum to 100% because of rounding.

**P < .001.

Interview Languages and Measures

Interviews were conducted in English, Spanish, Cantonese, Mandarin, Korean, and Vietnamese. All new CHIS question content was forward-translated and adjudicated by multiple translators each cycle,^36,37 and was culturally adapted to avoid problems with multicultural issues beyond language.³⁸ Limited English proficiency was a 2-category variable (speaking English only or very well vs well, not well, or not at all).³⁹ We analyzed interview language as a 3-category variable (English, Spanish, or Asian language) because of the low frequency of individual Asian languages. Low numbers of White LEP, Black LEP, and multirace LEP respondents required creation of a 7-category composite race–LEP variable for regression analyses: White (all LEP levels), Black (all LEP levels), multirace (all LEP levels), Asian non-LEP, Asian LEP, Hispanic non-LEP, and Hispanic LEP.

Hispanic status was asked first, followed by a race question allowing multiple responses. Responses were recoded into US Office of Management and Budget categories⁴⁰ that separate Hispanic identification from race identification (e.g., Hispanic, White non-Hispanic, Black non-Hispanic). Spending 75% or less of one’s life in the United States was a proxy for exposure to US mainstream culture. This cut-off was a natural break point in the distribution and led to better-fitting models than other cut points. Age was a continuous variable. Education and poverty were 4-category ordinal variables.

Statistical Analysis

We first examined cross-sectional sexual orientation item nonresponse trends over 5 CHIS cycles. We then ran a binary logistic regression model predicting sexual orientation nonresponse from race–LEP, interview language,⁴¹ survey cycle, spending 75% or less of one’s life in the United States, age, poverty status, and education. We also ran gender-stratified versions of that model to assess differences between men and women because answering sexual orientation questions may be a different social experience for each gender.⁴² Models included postestimation contrasts to test main effects of race–LEP, interview language, and survey cycle, and the interactions of race–LEP and interview language with survey cycle individually.

We used an adaptation of the multicycle analysis approach by Lee et al.,⁴³ which included design-adjusted analyses to represent the California household population. We ran all analyses in Stata version 12 (StataCorp LP, College Station, TX) with CHIS survey weights and Taylor Series variance estimation.

RESULTS

Figure 1 shows design-adjusted cross-sectional trends of sexual orientation item nonresponse over 2-year CHIS cycles by race/ethnicity, English proficiency, and interview language. Item nonresponse rates among Hispanics and Asians were 2 to 4 times higher than among Whites, Blacks, and multirace respondents in 2003, but the difference decreased over cycles until most of these groups were not statistically different in 2011 (Figure 1a). Asians and Hispanics exhibited steeper declines than Whites, Blacks, and multirace respondents. Limited-English-proficient respondents’ item nonresponse rate declined from 5.6% in 2003 to 3.1% in 2011 (Figure 1b). Despite the large decline, LEP respondents still had a significantly higher rate of item nonresponse in 2011. Interviews conducted in Spanish had more sexual orientation item nonresponse than English interviews in all cycles (Figure 1c). Asian language (Chinese, Korean, or Vietnamese) interviews had item nonresponse rates statistically similar to English interviews in 2003, and 2011, and similar to Spanish interviews in 2005, 2007, and 2009. Figure 1a–1c suggests a convergence in sexual orientation nonresponse rates over time, with the largest declines in Asians, Hispanics, LEP respondents, and interviews conducted in Spanish and Asian languages.

FIGURE 1—

Sexual orientation item nonresponse over 2-year California Health Interview Survey cycles by (a) race and ethnicity, (b) English proficiency, and (c) interview language.

Note. CHIS = California Health Interview Survey; LEP = limited English proficiency; NHPI = Native Hawaiian/Pacific Islander. The sample size was n = 182 812. Whiskers indicate 95% confidence intervals.

Figure 2 shows that LGB self-identification increased between 2003 and 2011 while sexual orientation item nonresponse declined. The LGB self-identification rate is based on self-reported (i.e., not imputed) sexual orientation responses (n = 178 330).

FIGURE 2—

Percentage of respondents not reporting their sexual orientation (n = 182 812) and percentage identifying as lesbian, gay, or bisexual (n = 178 330) over California Health Interview Survey cycles.

Note. CHIS = California Health Interview Survey; LGB = lesbian, gay, or bisexual. Whiskers indicate 95% confidence intervals.

Table 2 presents 3 binary logistic regression models predicting sexual orientation item nonresponse. Using all respondents over 5 survey cycles showed significantly higher nonresponse odds among Asian (LEP and non-LEP) respondents compared to Whites of both LEP levels. Hispanic non-LEP respondents’ nonresponse odds were statistically equal to those of Whites. As expected from Figure 1, Black and multirace respondents had odds of nonresponse similar to those of Whites. Race–LEP had a significant main effect (F_{6, 182 720} = 20.73; P < .001) on item nonresponse.

TABLE 2—

Adjusted Odds Ratios From Survey-Weighted Binary Logistic Regression Predicting Sexual Orientation Nonresponse in the California Health Interview Survey, 2003–2011

Variable	All Respondents, AOR (95% CI) (n = 182 812)	Women, AOR (95% CI) (n = 106 300)	Men, AOR (95% CI) (n = 76 512)
Race (non-Hispanic)/ethnicity and LEP
White: LEP and non-LEP (Ref)	1.00	1.00	1.00
Asian and NHPI: non-LEP	3.29** (2.22, 4.88)	4.59** (2.78, 7.60)	2.01* (1.08, 3.73)
Asian and NHPI: LEP	4.00** (2.61, 6.13)	5.04** (3.09, 8.24)	2.79* (1.25, 6.20)
Hispanic: non-LEP	0.74 (0.47, 1.17)	1.24 (0.73, 2.13)	0.29* (0.13, 0.67)
Hispanic: LEP	1.14 (0.63, 2.06)	1.61 (0.87, 2.99)	0.66 (0.19, 2.26)
Black: LEP and non-LEP	1.47 (0.92, 2.33)	2.20* (1.29, 3.74)	0.72 (0.28, 1.86)
Multirace: LEP and non-LEP	0.78 (0.31, 1.95)	0.94 (0.24, 3.74)	0.60 (0.24, 1.51)
Interview language
English (Ref)	1.00	1.00	1.00
Spanish	2.37* (1.34, 4.20)	2.01* (1.09, 3.70)	3.34* (1.01, 11.10)
Chinese, Korean, Vietnamese	0.29** (0.16, 0.54)	0.21** (0.10, 0.44)	0.50 (0.17, 1.44)
Survey cycle
2005 vs 2003	0.87 (0.69, 1.11)	0.88 (0.65, 1.19)	0.87 (0.60, 1.25)
2007 vs 2003	0.73* (0.58, 0.93)	0.85 (0.63, 1.14)	0.63* (0.42, 0.93)
2009 vs 2003	0.72* (0.53, 0.99)	0.71 (0.50, 1.00)	0.73 (0.44, 1.21)
2011 vs 2003	0.69* (0.52, 0.91)	0.87 (0.59, 1.26)	0.53* (0.36, 0.79)
Controls
≤ 75% of life in US = 1	1.40* (1.12, 1.75)	1.54* (1.21, 1.98)	1.30 (0.88, 1.92)
Age, y	1.01** (1.00, 1.02)	1.01** (1.01, 1.02)	1.01** (1.01, 1.02)
Poverty
≥ 300% FPL (Ref)	1.00	1.00	1.00
200%–299% FPL	1.14 (0.93, 1.40)	1.09 (0.86, 1.40)	1.17 (0.84, 1.63)
100%–199% FPL	1.18 (0.96, 1.45)	1.21 (0.97, 1.51)	1.10 (0.76, 1.59)
0%–99% FPL	1.85** (1.50, 2.27)	1.96** (1.57, 2.46)	1.56* (1.07, 2.27)
Education
≥ college degree (Ref)	1.00	1.00	1.00
Some college	0.76* (0.60, 0.97)	0.66* (0.51, 0.85)	0.92 (0.60, 1.40)
High-school graduate	0.94 (0.79, 1.11)	0.88 (0.71, 1.10)	1.03 (0.79, 1.35)
< high school	1.61** (1.31, 1.98)	1.40* (1.09, 1.80)	1.95** (1.39, 2.73)

Note. AOR = adjusted odds ratio; CI = confidence intervals; FPL = federal poverty level; LEP = limited English proficiency; NHPI = Native Hawaiian/Pacific Islander. The regression models a binary outcome where 1 = did not answer.

*P < .05; **P < .001.

Spanish interviews had higher odds of item nonresponse than English interviews, but Asian-language interviews had lower odds of nonresponse when we controlled for race/ethnicity, English proficiency, and other model variables. There was a significant main effect of interview language for the entire sample (F_{2, 182 720} = 26.60; P < .001). Yet Spanish interview nonresponse odds increased while Asian language nonresponse odds decreased relative to English interviews, suggesting differential effects of each translation or the people who were interviewed in those languages.

Item nonresponse odds were significantly lower in 2007, 2009, and 2011 versus 2003, and the overall survey cycle main effect was significant (F_{4, 182 720} = 7.52; P < .001). Interactions of cycle with race–LEP (F_{24, 182 720} = 1.33; P = .127) and interview language (F_{8, 182 720} = 1.45; P = .169 overall) were not significant, suggesting that the rate of decline was not different across race–LEP and interview language groups when modeled covariates were taken into account. Spending 75% or less of one’s life in the United States, older age, living in poverty, and having less than a high-school diploma also increased item nonresponse odds.

Race–LEP had different effects for women and men. Asian women and men both had significantly greater item nonresponse odds than White women and men, but the effect was much stronger for women than for men. For Hispanics, only non-LEP men showed significantly lower item nonresponse odds than Whites. Black women had significantly greater nonresponse odds than White women, but the nonresponse odds for Black men were not significantly lower compared to White men. For Hispanics and Blacks, there was a pattern of increased, though not always significant, nonresponse odds ratios for women and decreased odds for men. Men and women both showed significantly higher item nonresponse odds in Spanish interviews compared to English interviews, but only women showed significantly lower nonresponse odds in Asian language interview. Nonresponse odds decreased over time for both genders individually, based on the main effects of cycle (F_{4, 198 940} = 6.09; P < .001 for women, and F_{4, 210 343} = 4.05; P = .027 for men), but each cycle was not necessarily different from 2003. Interactions of cycle with race–LEP and interview language were significant only for women (race–LEP × cycle: F_{24, 198 940} = 2.13; P = .001 for women, and F_{24, 210 343} = 1.29; P = .157 for men; interview language × cycle: F_{8, 198 940} = 2.78; P = .005 for women, and F_{8, 210 343} = 0.86; P = .549 for men). Finally, having lived 75% of one’s life or less in the United States only increased nonresponse for women, and the overall effects of other predictors were roughly the same between women and men despite minor differences in the significances.

DISCUSSION

Sexual orientation item nonresponse rates declined over the past decade, mostly because of declining nonresponse among Hispanics, Asians, and LEP respondents. Sexual orientation item nonresponse rates for racial/ethnic and linguistic minorities converged toward an overall rate of 1% to 2% with small or no differences between demographic groups. Sexual orientation questions in the NHIS have obtained similar nonresponse rates.²⁴ This lends confidence in CHIS data quality for sexual minority groups and measurement of sexual minority status in surveys broadly because lower item nonresponse rates reduce the risk of nonresponse bias and make data more complete.

Despite this optimism, higher nonresponse rates among Asian respondents (regardless of their English proficiency) and respondents interviewed in Spanish point to possible linguistic, attitudinal, and cultural mechanisms underlying sexual orientation nonresponse. Nonresponse declined most for LEP respondents (from 5.6% in 2003 to 3.1% in 2011). Yet regression models showed that Asians (LEP and non-LEP) were still significantly more likely than Whites not to answer. Interviewing in an Asian language reduced the odds of item nonresponse when we controlled for other variables in the model. This may seem counterintuitive because Asian and LEP respondents both showed higher nonresponse rates than Whites (Figure 1), but we interpret this finding as evidence that in-language interviews may increase LEP Asian respondents’ understanding of the question. This highlights the combined role of interview language, race, and English proficiency in producing item nonresponse.

Moreover, when we removed interview language from the model (results not shown), the effect of being Asian was significant but less strong, suggesting that accounting for interview language is important in isolating the race effect. Comparatively, Spanish interviews led to more item nonresponse whereas Hispanic ethnicity and LEP did not. Subpopulation analyses of LEP Hispanic and Asian respondents only (not shown) showed very similar effects of interview language. These results together may suggest that interview language could be more important than race/ethnicity and LEP for those interviewed in Spanish, whereas race/ethnicity and LEP are more important predictors of item nonresponse than interview language for Asians.

Attributing these effects to the translations alone, rather than the respondents who were interviewed in those languages, leads to the conclusion that the Asian-language translations were better at interpreting sexual orientation concepts than the Spanish translations. Yet respondents are routed to an interview language by choice or necessity, so inferences about the effect of interview language per se should be conservative. Asians with LEP were much less likely to be interviewed in an Asian language (55%) than LEP Hispanics were to be interviewed in Spanish (74%). Despite this correlation between interview language and races/ethnicities, we saw no evidence of problematic multicollinearity. Higher nonresponse odds among Spanish interviews were seen in women and men separately, but lower odds among Asian-language interviews were seen in women only. Thus, there seems to be a complex relationship between interview language, gender, and item nonresponse that should be explored further in future research.

Asian women and men (LEP and non-LEP) both had higher nonresponse odds than Whites, but the relationship between gender, race, English proficiency, and nonresponse was less straightforward for Hispanics. Only Hispanic non-LEP men showed significantly less item nonresponse than White men. The item nonresponse odds for Hispanic women were not significantly higher than for White women in this study, yet there was a general pattern of non-White women having higher odds of nonresponse than White women. Non-White men (other than Asians) had lower odds of nonresponse than White men, although not all of these differences were significant. Black and Hispanic LEP men had statistically similar odds to White men. Multirace respondents of both genders had nonsignificantly lower nonresponse odds than White respondents. Where the relationship of race–LEP and nonresponse differ by gender, men and women within the same race/ethnicity may have different attitudes about sexual topics.

Spending less of one’s life in the United States, older age, living in poverty, and having less education each positively predicted sexual orientation item nonresponse. Taken together, this paints a picture of older respondents who are poorer and less educated, who immigrated more recently (as a fraction of their life), and who may be less knowledgeable about or comfortable with sexual orientation questions. Future research should continue exploring the complex relationships between demographic factors, LEP, and race/ethnicity to refine the demographic and linguistic profile of sexual orientation nonrespondent types. This would be immensely helpful to survey designers when they are making decisions about sexual orientation question wording and translation.

Future research should also try to further isolate the sources of the independent effects of race/ethnicity, language ability, and specific question translations. Experimental tests of different translation methods could estimate true translation and question wording effects. Measuring underlying sexual attitude and opinion norms would also be helpful, as these are likely the psychological drivers of item nonresponse. Terminology awareness questions could help isolate knowledge effects from attitude or opinion effects. Qualitatively, it would be helpful to better understand how ethnic/racial minorities who are sexual minorities self-identify their sexual orientation, particularly when their English proficiency is limited. Limited English proficiency restricts a person’s ability to access the linguistic channels of social progression, such as popular television, radio, print, and online media, and thus may restrict their knowledge about, awareness of, and comfort with sexual orientation issues.

Item nonresponse is only 1 facet of item-level data quality. This study did not assess nonresponse bias directly because that requires knowing either (1) the LGB identification status of respondents and nonrespondents, or (2) the correlation between nonresponse propensity and respondents’ LGB status.⁴⁴ On the basis of the low item nonresponse rates in this study, the impact of respondent–nonrespondent differences in LGB self-identification estimates on bias should be minimal, particularly in recent CHIS cycles. Concern about hidden biases attributable to differential nonresponse rates alone does not seem warranted, particularly because of the convergence between language and sociodemographic groups over time. Bias potential may still be highest among those who do not speak English very well or exclusively, and among Spanish interviews given the higher rates of item nonresponse among those groups.

Finally, future research should explore why LGB identification is rising while sexual orientation nonresponse is declining. Are people who would have refused or said “I don’t know” in previous years now identifying as LGB, or has there been true population change in LGB identification?^45–47 Are non-LGB people becoming more familiar with sexual orientation terminology? Do they perceive sexual orientation as an innocuous demographic question like gender and race? Comparative or experimental research should address these questions.

In conclusion, there is a convergence of sexual orientation nonresponse rates among racial, ethnic, and linguistic minorities over a decade’s worth of data from a racially and linguistically diverse population-based survey. Although overall measurement of sexual orientation in CHIS is sound and improving, the remaining differences between LEP and non-LEP respondents, and between those interviewed in different languages, motivate future research into the linguistic and cultural causes of sexual orientation nonresponse. For now, our findings are encouraging about the state of sexual orientation measurement and the future of sexual minority health surveillance.