Sexual Fluidity: Implications for Population Research

Abstract

For the first time ever, national censuses have begun asking adults to report their sexual orientations. However, because such surveys provide only cross-sectional snapshots of populations, these data obscure one key complexity: that sexuality can be fluid, with sexual self-identification evolving over time. Drawing on unique, restricted-use data from the Population Assessment of Tobacco and Health, the current study documents the prevalence, correlates, and empirical consequences of sexual fluidity in the contemporary United States. Overall, about 1 in 11 American adults changed sexual identities over five annual surveys, including 6% of cisgender men, 11% of cisgender women, and 35% of gender minorities. Fluidity was particularly pronounced among young adults and among those who had ever identified as bisexual or “something else.” Despite the frequency and patterning of sexual fluidity, accounting for fluidity did little to change observed patterns of disadvantage on three measures of sexual minority well-being: mental health, financial insecurity, and substance use. Given these facts, I argue that demographic research should foreground the complexities inherent in quantifying sexuality, focusing less on how many people “are” a given sexual orientation and more on the social dynamics that continue to produce sexual minority disadvantage.

Introduction

After long rendering sexual minority populations invisible, major demographic surveys are increasingly inviting adults to report their sexual orientations. For the first time ever, the 2021 Census of England and Wales included a voluntary question asking all adults, “Which of the following best describes your sexual orientation? Straight or Heterosexual; Gay or Lesbian; Bisexual; Other sexual orientation” (Office for National Statistics n.d.). The 2023 New Zealand Census included a nearly identical question (Stats NZ 2021). In the United States, the Census Bureau included a similar measure in its Household Pulse Survey (U.S. Census Bureau 2021), a first step in the Biden administration's plan to expand the collection of sexual orientation data (Exec. Order No. 14075, 2022). As of 2019, a total of 15 OECD countries had assessed sexual orientation in at least one nationally representative survey (OECD 2019).

This rapidly expanding data landscape has enabled a new generation of research on the size and characteristics of sexual minority populations (National Academies of Sciences, Engineering, and Medicine 2020). However, because large demographic surveys are typically cross-sectional snapshots of populations, these data obscure a key complexity: sexuality can be fluid. For decades, research using nonprobability samples has documented the significance of sexual fluidity, particularly for cisgender women (e.g., Blumstein and Schwartz 1977; Budnick 2016; Diamond 2008; Rupp et al. 2014; Rust 1993) and gender minorities (e.g., Katz-Wise et al. 2023; Katz-Wise et al. 2016). Despite this research, less is known about the dynamics of sexual fluidity within the population at large.

The current study employs unique, restricted-use data to analyze the prevalence, correlates, and empirical consequences of sexual fluidity in the contemporary United States. Drawing on the Population Assessment of Tobacco and Health (PATH) Study—a large, nationally representative longitudinal study of American adults—I document within-person sexual identity changes across five survey waves, conducted roughly annually between 2013 and 2019. Because sexual identity—often conceptualized as “a person's core internal sense of their sexuality” (National Academies of Sciences, Engineering, and Medicine 2022:5)—constitutes just one dimension of sexuality, I also provide supplemental analyses of within-person changes in sexual attraction.¹ After establishing the prevalence and distribution of sexual fluidity, I explore whether accounting for sexual fluidity substantively changes any conclusions regarding the well-being of America's sexual minority populations. At a historic juncture in the measurement of sexual minority populations, the current study provides wide-ranging, new evidence on an aspect of sexuality typically unobservable in population research.

Background

Fixed and Fluid Conceptions of Sexuality: Consequences for Demographic Research

The most fundamental task faced by demographers is quantifying the size of populations. So a natural first question is: How many lesbian, gay, or bisexual adults are there in the United States? The Williams Institute (Conron and Goldberg 2020) provides a relatively precise answer to this question: 10,338,000. To reach this estimate, the authors drew on a large, nationally representative sample of adults selecting their sexual identity from a list of choices provided in the Gallup Daily Tracking Survey. As sexual identity items move from research surveys to official national censuses, estimates like these will only grow in precision and authority.

Given the historic erasure of sexual minority lives in demographic data, the inclusion of sexual identity items in population-representative surveys is a vitally important development. And yet, cross-sectional data like these are only able to capture individuals' sexual identification at a single point. As a result, these data freeze in time what many people experience as a much more dynamic and ongoing process. To produce stable estimates of how many people “are” a given sexual identity, demographers must sidestep this issue, implicitly assuming a fixed, static conception of sexuality.

Demographers' implicit conception of sexuality is consistent with a model of human sexuality that assumes “an individual's sexual predisposition . . . is an early-developing and stable trait that has a consistent effect on that person's attractions, fantasies and romantic feelings over the lifespan” (Diamond 2008:2). This model clearly captures the experiences of many people. Yet there are many others for whom this fixed model of sexuality fits more poorly, if at all. For example, in her foundational study of female sexual fluidity, Diamond (2008) described the experiences of cisgender women entering long-term romantic relationships with other women despite having no prior same-sex attractions. These women were not previously “closeted.” Instead, they experienced genuine changes in their sexual self-conception, occasioned by shifts in external circumstances. Diamond (2008) argued that cisgender women's capacity for this kind of ongoing self-discovery illustrates the limits of a fixed model of sexuality—a model, she noted, that is based largely on studies of cisgender men.

The fluidity that Diamond (2008) documented among cisgender women appears to be even more common among gender minorities. In a community sample of transgender and gender nonbinary adults in Massachusetts, Katz-Wise et al. (2016) found that 58% of respondents reported having previously experienced a change in sexual attractions. Similarly, in a diverse nonprobability sample of teens and young adults, Katz-Wise et al. (2023) found that 70% of gender minorities retrospectively reported having ever changed sexual identities. This high degree of change led Katz-Wise et al. (2016:74) to argue that sexual fluidity is “the norm rather than exception among gender minority people.”

For demographers, acknowledging this kind of fluidity raises a complex set of empirical issues. The fact of sexual fluidity implies that, in any given cross-sectional survey, two distinct groups are being collapsed together: those for whom a given sexual identity is fairly fixed and those for whom it is much more fluid. This has implications not only for our understanding of the size of sexual minority populations, but also for our ability to analyze the disparities that sexual minority populations face. If sexually fluid respondents are particularly vulnerable, then their inclusion alongside more stably identified sexual minorities could have the effect of making stably identified sexual minorities appear more disadvantaged than they actually are. By contrast, if sexually fluid respondents have been able to avoid some of the burden of discrimination and minority stress (Meyer 2003) experienced by other sexual minorities, then accounting for sexual fluidity could reveal that stably identified sexual minorities actually experience worse outcomes than previously known. In either case, researchers' inability to measure sexual fluidity in cross-sectional data could cause them to reach incomplete or inaccurate conclusions about the well-being of sexual minority populations.

Demographic research is not naive to these issues. Recent reports underscore the limitations of cross-sectional data, emphasizing the need for data that illustrate the fluidity of sexuality in individuals' lives. For example, in announcing the New Zealand Census's first ever sexual identity measure, Stats NZ (2019:7) emphasized that “sexual identity can be fluid, changing over time or across social contexts.” Given this fact, Stats NZ (2019:8) warned that the new data should be understood only as providing “a snapshot of a person's sexual identity—at that particular point in time.” Similarly, in its report on America's LGBTQ+ populations, the National Academies of Sciences, Engineering, and Medicine (2020:83) noted that one of the most “important considerations in research around sexual orientation [is] the potential fluidity of identity.” Despite this widespread recognition of sexual fluidity, few studies have actually documented dynamic patterns of sexual identification using population-based data.

Existing Research on Sexual Fluidity

Given long-standing data limitations, most research on sexual fluidity has drawn on nonprobability samples—samples often selected precisely because of their patterns of sexual behavior and identification (e.g., Blumstein and Schwartz 1977; Rust 1993; Savin-Williams 2017). This research documents how individuals' attractions, behaviors, and identities can change over time, shifting in response to what Rupp et al. (2014) called sexual “opportunity structures.” Although these studies illustrate the significance of sexual fluidity for individuals, they cannot establish the magnitude and consequence of sexual fluidity for populations.

Population-based studies confirm that a nontrivial proportion of respondents change sexual identifications across survey waves, but yield vastly different estimates of how common such changes are. Analyzing five waves of the Growing Up Today Study, Ott et al. (2013) found that 37% of females and 16% of males reported at least one change in their sexual identity between the ages of 12 and 27. By contrast, comparing two waves of the National Survey of Midlife Development in the United States, Mock and Eibach (2012) found that only about 3% of middle-aged females and 2% of middle-aged males changed sexual identifications between 1994/1995 and 2005/2006. Between these two extremes, Everett (2015) showed that about 18% of female young adults and 6% of male young adults changed sexual identifications across two waves of Add Health data.

Underlying these divergent estimates are differences in the age groups, study periods, and identity measures used across existing research. For example, the high level of fluidity found by Ott et al. (2013) and Everett (2015) may reflect not just their relatively young samples, but also the particular sexual identity measure used in their data. Both of these studies used a five-point sexuality scale that may have encouraged particularly fluid responses by including categories like “mostly gay/lesbian.” At the same time, the low rates of fluidity documented by Mock and Eibach (2012) may be a function not just of their middle-aged sample, but also of the time period when they assessed sexuality. Compared with more recent surveys, adults may have been far less willing to report a sexual minority identity in 1994/1995 and even in 2005/2006, artificially suppressing observed fluidity.

Summary and Current Research Questions

Taken together, existing research affirms the importance of sexual fluidity but provides only fragmented evidence regarding its prevalence and implications for ongoing population research. The current study uses unique, new data to advance current knowledge in several key ways. First, unlike prior studies that analyzed only particular age groups, I document patterns of sexual fluidity in a general sample of American adults. Second, unlike studies that assessed sexuality using novel categories like “mostly gay,” I analyze responses to the four-category sexual identity measure now being used in multiple national censuses. Third, unlike studies that analyzed only two waves of data—data that were often collected many years ago and several years apart—I document a full five waves of sexual identification, using surveys conducted roughly annually between 2013 and 2019. Finally, after documenting the prevalence and correlates of sexual fluidity, I test its empirical consequences for three distinct measures of well-being: mental health, financial insecurity, and substance use.

Specifically, this study addresses three research questions:

1. 1.

   How common is sexual fluidity among American adults? How should sexual fluidity impact our understanding of the size of America's lesbian, gay, bisexual, and queer (LGBQ) populations?

2. 2.

   What are the demographic correlates of sexual fluidity? Are there groups for whom fluidity is especially common?

3. 3.

   What are the empirical consequences of sexual fluidity? Does accounting for sexually fluid individuals change any of the conclusions that would be reached based only on cross-sectional data?

Data, Measures, and Methods

Data

To answer these questions, I use data from the Population Assessment of Tobacco and Health Study. Cosponsored by the National Institutes of Health and the Food and Drug Administration, PATH is a longitudinal cohort study designed to produce population-representative evidence on Americans' patterns of tobacco use and health. PATH participants were selected using a four-stage stratified area probability sample design, yielding a baseline sample that is representative of the U.S. civilian, noninstitutionalized population (Westat 2022). PATH includes samples of both youth (12–17) and adults (18+). For the current study, I focus on adults because only a subset of older teens was asked about sexual identity and, within this group, attrition out of the sample was exceptionally high.

To date, PATH has completed five waves of data collection, surveying respondents roughly annually since 2013. Notably, every survey was completed under the same conditions: privately within respondents' homes using audio computer-assisted self-interviewing technology. This consistency across survey waves minimizes the possibility that reported changes in sexual identity are due to interviewer effects or other external circumstances.

To my knowledge, PATH is unique within the current landscape of American population-based data sources in terms of the frequency with which it has assessed sexual identity, its general population sampling frame, its concurrent assessment of gender identity, and its comparatively large sample size. For example, the General Social Survey has included repeated sexual identity measures in some of its rotating panels. However, these data provide at most three waves of sexual identification and yield a sample size that is about one-eighteenth the size available with PATH.

Measures

Outcome Variable: Sexual Fluidity

For my primary analyses, I measure sexual fluidity in terms of changes in sexual identity. In supplemental analyses, I also analyze changes in sexual attraction. Both of these items were asked of all adults in every survey wave. In the final section of the survey, following questions on their current employment status, educational attainment, and income, respondents were asked first about sexual attraction and then about sexual identity.

Specifically, the sexual attraction item read, “The next question asks about your level of sexual attraction to BOTH males and females. Please consider the response choices carefully, as it is important that you understand them and are as honest as you can be in your answer. To whom have you felt sexually attracted, even if you did not take any action based on feeling attracted?” Respondents were provided six response options: “Only to females, never to males,” “Mostly to females, and at least once to a male,” “About equally often to females and to males,” “Mostly to males, and at least once to a female,” “Only to males, never to females,” and “I have never felt sexually attracted to anyone at all.” Immediately following this item, the sexual identity item asked, “Do you consider yourself to be . . .” and provided four options: “Straight,” “Lesbian or gay,” “Bisexual,” and “Something else.” Notably, this item includes the same four identity categories currently used in census-sponsored surveys in the United Kingdom, New Zealand, and the United States. Both questions allowed respondents to select only one answer.

Key Independent Variable: Gender Identity

Existing research on sexual fluidity emphasizes its frequency among cisgender women and gender minorities. Therefore, when analyzing the demographic patterning of sexual fluidity, my key independent variable is a measure of respondents' gender identity. To construct this variable, I rely on two survey items: one question assessing sex and another assessing transgender identification. Taken together, these two items yield an imperfect measure of gender identity.

Specifically, in Wave 1—and only Wave 1—all respondents were asked, “What is your sex?” and provided with two options: “Male” or “Female.” Then, in Waves 2–5, all respondents were asked about transgender identity: “Some people describe themselves as transgender when they experience a different gender identity from their sex at birth. For example, a person born into a male body, but who feels female or lives as a woman would be transgender. Do you consider yourself to be transgender?” This “one-step” approach to assessing gender identity is problematic because it fails to capture gender-diverse individuals who do not identify with the specific term “transgender” (National Academies of Sciences, Engineering, and Medicine 2022) and because it has been shown to have very poor reliability in general population samples (Saperstein and Westbrook 2021). Despite these important limitations, the availability of a transgender identity measure still marks an advance over simply assessing respondents' assigned sex. Therefore, I use this question to create a three-category gender identity indicator, categorizing respondents as cisgender women (those who report their sex as female and never identify as transgender), cisgender men (those who report their sex as male and never identify as transgender), or gender minorities (those who ever identify as transgender, regardless of sex).² In the online supplement, I provide a more complete analysis and discussion of the PATH transgender identity item.

Other Demographic Correlates

In addition to my key independent variable, I also assess several other potential demographic correlates of sexual fluidity. In all multivariate models, I include measures of age, race and ethnicity, region, educational attainment, marital status, and survey language. To help assess whether observed changes in sexual identity simply reflect measurement error, I also include an indicator of whether individuals reported inconsistent and implausible responses on one other categorical item assessed in all survey waves: veteran's status. By flagging individuals who reported that they never served in the military after previously reporting that they were a veteran, this variable serves as a rough control for survey inattentiveness or a general propensity to mis-click responses.

Analytic Sample

My analyses require a number of sample restrictions. Of the initial sample of 32,320 adults who participated in the baseline interview, only 18,925 participated in every interview across the full five waves. From that sample, I retain the 18,035 respondents who provided answers to the sexual identity item in every wave.³ Finally, I drop the 1.8% of respondents with missing data on any of this study's other covariates, yielding a final analytic sample of 17,712.

Given the substantial attrition across survey waves and the nontrivial amount of missing data, this study's analytic sample is no longer strictly generalizable to the U.S. adult civilian, noninstitutionalized population. Even so, the patterns of sexual identification in my sample closely replicate patterns recently reported in nationally representative samples of American adults. For example, Gallup's Daily Tracking Survey indicates that 2.5% of American adults identify as lesbian or gay and 4.0% identify as bisexual (Jones 2022). Similarly, the latest wave of my data shows that 2.1% of respondents identified as lesbian or gay and 4.2% identified as bisexual. The racial and ethnic distribution of my sample also closely reflects current census estimates. Among adults aged 18 or older, the 2020 census showed that the U.S. population was about 64% non-Hispanic White, 12% Black, and 17% Hispanic or Latino (Jones et al. 2021). In my sample, about 61% identify as non-Hispanic White, 15% as Black, and 17% as Hispanic or Latino. Complete descriptive statistics for the analytic sample are presented in the online supplement.

Methods

To address the research questions outlined earlier, I draw on a variety of approaches. First, to document the patterns of sexual identification and fluidity in the sample, I use a combination of data visualization and descriptive statistics. Then, to more systematically analyze the correlates and empirical consequences of sexual fluidity, I estimate a series of multivariate logistic regressions. Together, these methods allow me to make the most of the PATH data and provide a rich quantitative portrait of sexual fluidity in the contemporary United States.

Results

Research Question 1: How Common Is Sexual Fluidity?

To begin to explore the sexual fluidity reported by PATH respondents, Figure 1 visualizes the distribution of sexual identities within each wave as well as the patterns of movement among identities between waves. In the top panel, I present results from the full sample; in the bottom panel, I more clearly illustrate the movement between categories by excluding the approximately 88% of respondents who identified as straight in every single wave.

Focusing only on the distributions within waves, Figure 1 depicts a rather remarkable degree of stability. The overall percentage of respondents who identified with each sexual identity category was almost exactly the same in every wave of the study. Online appendix Table A3 summarizes the distribution of identities in each wave, illustrating that the percentage of respondents within each category never changed by more than a few tenths of a percentage point.

Looking only at repeated cross sections, it would appear that sexual identity was almost entirely stable in this sample. However, by revealing the changes between waves, Figure 1 also illustrates that there is actually a fair amount of mobility behind this apparent stability. Between any two waves, an average of 3.6% of respondents reported a change in their sexual identity. Summarizing across all waves, 8.7% of respondents changed sexual identities at least once. This fluidity is not uniform across identities: there was a comparatively high degree of mobility between the categories of straight, bisexual, and “something else” and far less movement into or out of the gay/lesbian category.

Table 1 quantifies these patterns. Of the 2,551 total identity changes reported in the data, 46% were between the categories of bisexual and straight, 32% were between “something else” and straight, and 9% were between bisexual and “something else.” Overall, then, 86% of all identity changes occurred between the categories of straight, bisexual, and “something else.” Far fewer changes—only about 14%—involved movements into and out of the more stable category of gay/lesbian.

How should these differential mobility patterns impact our understanding of the “size” of the LGBQ population? To begin to address this question, Table 2 presents four related quantities: the percentage of respondents who reported each identity, on average, across all five survey waves; the percentage who ever reported each identity in any wave; the percentage who never reported each identity in any wave; and, finally, the percentage who always reported each identity in every wave. These results illustrate the complexity of incorporating sexual fluidity into estimates of LGBQ population size.

Comparing the first and second columns in Table 2, the data show that cross-sectional estimates substantially understate the percentage of respondents who ever reported an LGBQ identity. Within any given wave of the survey, about 7% of respondents identified as LGBQ. Summarizing across all waves, nearly 12% of respondents identified as LGBQ at least once. Hence, the population who ever identified as LGBQ was about 70% larger than the population who identified as LGBQ at any given time. Consistent with the differential patterns of mobility illustrated in Figure 1, these underestimates are particularly pronounced for the bisexual and “something else” categories. For example, the percentage of respondents who ever identified as bisexual was nearly double the percentage who identified in any given wave. From this perspective, the LGBQ population is much larger than cross-sectional estimates would imply.

By comparing the first and fourth columns in Table 2, one could also argue that sexual fluidity tends to inflate population estimates. Viewed in this way, cross-sectional data collapse together two distinct groups: those for whom a given sexual identity is effectively fixed and those for whom it is more fluid. If one brackets off this latter group, focusing only on those who reported the same identity in every wave, the resulting estimates are consistently smaller. For instance, although 93% of respondents identified as straight in any given wave, only 88% of respondents consistently identified as straight across all waves. Similarly, although about 7% of respondents identified as LGBQ in any given wave, only about 4% did so in all waves.

Quantifying population sizes is especially complex with respect to the two most fluid categories of sexual identification: bisexual and “something else.” The “something else” category, in particular, raises complex questions. Across all waves, ever identifying as “something else” was fairly common: about 4.2% of respondents identified as “something else” at least once. Yet always identifying as “something else” was extremely rare: this was reported by only 0.2% of all respondents.

Summary: Research Question 1

The results in this section established three key facts. First, in every wave, the overall distribution of sexual identities was remarkably stable. Second, behind this apparent stability, sexual fluidity was common: over a five-year period, about 1 in 11 respondents changed their reported sexual identity at least once. Third, empirically acknowledging sexual fluidity raises complex questions about how to characterize the size of LGBQ populations, generally, and the size of bisexual and “other” populations, in particular.

Research Question 2: What Are the Demographic Correlates of Sexual Fluidity?

The results for question 1 demonstrate that 8.7% of all respondents changed sexual identities at least once. By averaging across the entire sample, however, these overall results may be masking certain populations for which sexuality is especially fluid or in which it is comparatively fixed. To explore this possibility, Figure 2 presents the percentage of respondents who ever changed sexual identity—overall, by gender, and, for cisgender respondents, by gender and age.⁴ The observed distribution of sexual fluidity dramatically supports prior research.

Overall, cisgender women were almost twice as likely to report sexual fluidity as cisgender men: 11% of women but only 6% of men reported any change in sexual identification. Further stratifying by age reveals a striking gradient among cisgender women, with fluidity declining monotonically across every age category. Among cisgender women aged 18–24 at baseline, 18% reported at least one change in their sexual identification. For cisgender men, too, sexual fluidity was highest in the youngest age group: about 9% of young men reported at least one change. By age 45, cisgender women and men reported roughly equal levels of sexual fluidity, indicating that the large overall difference was driven by younger women. The amount of fluidity reported by both of these groups, however, was substantially surpassed by the fluidity reported by gender minorities. Across five annual surveys, 35% of respondents who ever identified as transgender reported a change in sexual identity.⁵

To more systematically analyze the demographic correlates of sexual fluidity, I estimate a logistic regression modeling the log-odds of sexual fluidity as a linear function of the full set of demographic correlates described earlier. All of these predictors are measured at baseline, except for marital status, which was first reported in Wave 2. Results are presented in Table 3.⁶ These findings reinforce the importance of gender and age, while also highlighting other demographic factors that structure respondents' likelihood of reporting sexual fluidity. I find no significant differences by region and little variation across racial and ethnic groups. One exception is that, compared with White-identified respondents, those who identified as multiracial had 53% higher odds of reporting sexual fluidity. This intriguing result suggests, perhaps, that those who embrace complexity in their racial or ethnic identities are also more open to fluidity in their sexual identity.

Even net of age, education and marital status both emerged as significant predictors of sexual fluidity. Perhaps unsurprisingly, currently married individuals had a lower likelihood of reporting sexual fluidity than those who were previously or never married. Less expected was the finding that sexual fluidity declined with educational attainment: compared with those with only a high school degree or less, college graduates had 37% lower odds of reporting sexual fluidity. This contrasts with qualitative research emphasizing how college campuses facilitate women's exploration of same-sex attractions (Rupp et al. 2014), but is consistent with demographic research showing that women with the lowest levels of education actually report the highest levels of same-sex sexual experience (Budnick 2016).

Finally, both inconsistently reported veteran's status and survey language were significant predictors of reported fluidity. In particular, whether a respondent had ever taken a survey in Spanish (the only non-English language available) emerged as a consequential predictor. Conditional on the other variables, Spanish survey takers had about 326% higher odds of changing sexual identifications. On closer examination, these high rates of fluidity were driven entirely by the fact that Spanish speakers were six times as likely to ever identify as “something else” (“otra cosa” in the Spanish survey). Among those who never identified as “something else,” there were no differences by survey language. This exact dynamic—Spanish speakers selecting “something else” at unexpectedly high rates—was previously documented by Michaels et al. (2017) in their qualitative study examining how sexual identity measures perform with older, straight-identified Spanish speakers. Cognitive interviewing indicated that these responses were driven by misunderstandings, not substantive choices. Therefore, together with the higher rates of fluidity observed among respondents who mis-clicked the veteran status question, these results raise the concern that some of what I have interpreted as “sexual fluidity” could simply reflect measurement error.

Summary: Research Question 2

These results show that sexual fluidity is not restricted to any one group, and yet it is also not randomly distributed. Instead, individuals' likelihood of changing sexual identifications was systematically structured by other demographic traits, such as gender, age, education, and marital status. Consistent with prior research, I find that young cisgender women and gender minorities reported especially high levels of sexual fluidity. At the same time, survey language and inconsistently reported veteran's status also emerge as predictors of sexual fluidity. These latter results highlight the possibility that identity changes may reflect a certain degree of measurement error. Therefore, before turning to question 3, I will more fully address this concern.

Robustness Checks: Could Sexual Identity Changes Simply Reflect Measurement Error?

One important concern about the results presented so far is that what I have characterized as “sexual fluidity” might actually just be measurement error. For any categorical variable that respondents had to click through five years in a row, one might expect some number of responses to have been accidently mis-selected. How big of a problem might this be?

Notably, the results already presented mitigate against the possibility that sexual identity changes were purely the products of measurement error. If fluidity were entirely a function of accidental mis-clicking, we would expect these patterns to be more or less random. Instead, I find that different sexual identities showed differential levels of fluidity and that fluidity was systematically structured by other demographic factors. Although these facts are reassuring, it would be valuable to identify further evidence that could help triangulate how much measurement error might be reflected in my estimates. I present two such pieces of evidence.

First, one straightforward approach is to remove the two categories of respondents that showed evidence of probable measurement error: those with inconsistent veteran's status (2% of respondents) and those who ever took the survey in Spanish (4% of respondents). Within this more constrained sample, I find that the reported level of sexual fluidity does fall, but only slightly. Compared with the 8.7% who changed sexual identities in the full sample, 7.9% changed sexual identities in this more constrained sample. Moreover, as shown in online appendix Figure A2, the demographic structuring of fluidity by gender and age remains exactly the same in this constrained sample.

Second, I consider whether the changes I observe in sexual identities are also reflected in changing sexual attractions. Prior research has shown that sexual identity and attractions do not always align in expected ways (e.g., Mishel 2019), a fact that may be especially true for sexually fluid individuals (Suen et al. 2020). Moreover, there is a temporal gap between the identity and attraction measures in the PATH data, with the identity question asking how respondents currently identify and the attraction question asking to whom they have ever felt attracted. For both of these reasons, we would not expect that responses to these two items would necessarily move in lockstep with one another. Even so, if a large number of changes in sexual identity occurred simultaneously with changes in sexual attraction, it would seem unlikely that both changes were due to accidental mis-clicking.

Summarizing across waves, I find that changes in sexual attraction were even more common than changes in identity: 24% of all respondents changed sexual attractions, including 14% of cisgender men, 31% of cisgender women, and 52% of gender minorities. As with identity, attraction fluidity was particularly common among the youngest cisgender women, with 46% changing sexual attractions. When examining identity and attraction changes together, I find that 52% of all identity changes were accompanied by a simultaneous change in attractions. Accordingly, identity changes were nine times as likely to occur between waves that also involved an attraction change. Given the conceptual space between identity and attraction, the co-occurring patterns of movement in both of these distinct aspects of sexuality provide further evidence against measurement error.

Neither of these tests is definitive and surely some of the changes in the data represent merely mis-clicks. However, a range of evidence suggests that the “sexual fluidity” reported by respondents provides more signal than noise. Given this fact, I now address this study's final research question and examine the empirical consequences of accounting for sexual fluidity.

Research Question 3: What Are the Empirical Consequences of Accounting for Sexual Fluidity?

Contrary to the fixed conception of sexuality implicitly assumed in most demographic research, this study demonstrates that changes in sexual identity are fairly common among American adults. Given this fact, how does accounting for sexual fluidity impact our understanding of LGBQ Americans' experiences and outcomes?

Prior research on sexual fluidity has worked to assess whether fluidity, in itself, is a cause of psychological distress (Campbell et al. 2022; Everett 2015). I address a different question. For three distinct measures of well-being reported in the final wave of PATH—mental health, financial insecurity, and substance use—I present two different sets of results. First, I present the pattern of results that one would find if they only had access to respondents' current sexual identity: the situation most commonly faced by researchers. Second, I present results that take advantage of respondents' full history of sexual identification, disentangling the two groups that cross-sectional data collapse together: those who always report a given identity and those who currently report that identity but who are actually sexually fluid.⁷ For each outcome, I estimate a logistic regression modeling the log-odds of the outcome as a linear function of the sexual identity indicator and the eight variables described earlier. I present all results in terms of average adjusted predictions, comparing straight-identified respondents with both LGBQ-identified respondents, overall, and with respondents in each specific sexual identity category. Complete regression results are included in the online supplement.

First, I consider sexual fluidity's empirical implications for mental health disparities. Respondents were asked, “In general, how would you rate your mental health, which includes stress, depression, and problems with emotions?” and given five options, ranging from “excellent” to “poor.” Consistent with prior research on self-rated health (Gorman et al. 2015; Lagos 2018), I construct a binary indicator of whether respondents reported “poor” or “fair” mental health. Figure 3 presents these results.

Focusing only on respondents' current sexual identity, the results in Figure 3 replicate well-established patterns of LGBQ mental health disadvantages (e.g., Liu and Reczek 2021). Disaggregating by sexual fluidity, however, reveals a more complex picture. Considering LGBQ identification, overall, Figure 3 shows that sexually fluid adults reported much worse mental health than consistently straight-identified adults but somewhat better mental health than stably identified LGBQ adults. Even so, removing fluid respondents from the larger LGBQ sample has little effect on the main LGBQ mental health estimate. Separating out each specific sexual identity, I find that removing sexually fluid respondents has little effect on the results for lesbian/gay and bisexual adults. Accounting for fluidity does yield worse mental health among those always identifying as “something else.” However, because only about 0.2% of respondents identified in this way, this estimate is highly unstable.

Moving to financial insecurity, I model whether participants answered yes to the question, “In the past 30 days, because of a shortage of money, were you unable to pay any important bills on time, such as rent, electricity, or telephone bills?” Consistent with prior research (e.g., Badgett et al. 2019), Figure 4 shows that currently identified LGBQ adults—both overall and in each specific LGBQ identity—reported higher levels of financial insecurity than straight adults. Disaggregating by sexual fluidity, sexually fluid adults reported significantly worse outcomes than consistently straight-identified respondents and slightly (though not significantly) better outcomes than consistently LGBQ-identified respondents. When considering each specific sexual identity, Figure 4 shows little difference among consistently LGBQ and fluid adults. In all cases, separating out sexually fluid respondents has little discernable impact on the estimates for stably identified respondents.

Finally, for substance use, I model whether respondents reported having ever used any one of five separate categories of illegal drugs. Consistent with prior research (Rosner et al. 2021), Figure 5 shows that adults who currently identified as LGBQ—both overall and in each specific LGBQ identity—reported significantly higher levels of lifetime substance use than adults who currently identified as straight. Disaggregating by sexual fluidity shows that fluid adults are almost exactly at parity with stably identified LGBQ respondents. Again, accounting for sexually fluid respondents yields estimates that are substantively identical to those that would have been produced by considering only respondents' current sexual identities.

Summary: Research Question 3

Across three distinct measures of well-being, I find that sexually fluid adults reported significantly worse outcomes than adults who consistently identified as straight. Compared with consistently identified LGBQ adults, sexually fluid adults reported somewhat better outcomes on two measures—mental health and financial instability—and effectively identical outcomes on the third: substance use. In all cases, sexually fluid adults reported experiences that were much more similar to those of stably identified LGBQ adults than they were to those of stably identified straight adults. Because of these similarities, accounting for sexual fluidity ultimately has little impact on estimated patterns of LGBQ well-being. Taken together, these results suggest that, although sexual fluidity complicates our estimates for the size of the LGBQ population, accounting for sexual fluidity appears unlikely to substantively change our conclusions regarding the disparities that the LGBQ population faces.

Discussion

Across the English-speaking world and beyond, efforts are underway to add measures of sexual orientation into national censuses and other major demographic surveys. The data produced by these efforts will present historic opportunities to generate knowledge and shape policy. Yet these data will also be limited in one important regard: by only capturing individuals at a single point in time, these data will obscure the impact of sexual fluidity in many people's lives.

To understand the implications of sexual fluidity for ongoing population research, the current study drew on the Population Assessment of Tobacco and Health—a large, nationally representative cohort study of American adults. Analyzing five waves of survey data, I find that 8.7% of adults—about 1 in 11—changed sexual identities at least once between 2013 and 2019. Although all groups reported some degree of sexual fluidity, fluidity was especially common among gender minorities and among young cisgender women aged 18–24 at baseline. Roughly 1 in 3 gender minorities and 1 in 5 young cisgender women changed sexual identities at least once. An even greater proportion of respondents reported changes in sexual attractions, with attraction changes reported by 24% of all respondents, 46% of young cisgender women, and 52% of gender minorities. Because the PATH data cover only five years and begin at age 18, all of these estimates are surely lower bounds on the full extent of sexual fluidity that individuals experience over the course of their lives.

This degree of fluidity sits uneasily with the dominant—though often unstated—conception of sexuality employed in demographic research. In order to use cross-sectional data to quantify how many people “are” LGBQ in a given population, demographers must implicitly assume a static model of human sexuality—one that imagines sexual orientation as a fixed individual attribute just waiting to be counted. Although this is a useful heuristic, it is a vision of sexuality that many LGBTQ+ people explicitly reject (Suen et al. 2020). It is also radically different, for example, from the vision long put forward by humanistic queer theory, wherein sexual “subjectivities are fluid, unstable and perpetually becoming” (Browne and Nash 2010:1). To make sexuality tractable as something that can be collected on a census form, much of this complexity needs to be ironed out and bracketed off. The critical question is: How much information is lost when sexual fluidity is hidden from view? Does making sexual fluidity visible yield a fundamentally different picture of LGBQ populations? The results of this study suggest that the answer to that question is both yes and no.

This study clearly illustrates that accounting for fluidity substantially changes our understanding of the size of LGBQ populations. Across identities, I find that cross-sectional data consistently understate how many people ever identify as LGBQ but overstate how many always identify as such. These disjunctures are most pronounced for the two categories that showed the greatest fluidity: bisexual and “something else.” I find, for example, that the number of respondents who ever identified as bisexual is almost three times higher than the number who always identified as bisexual.

Viewed alongside the frequency of sexual identity changes among young cisgender women, the fluidity of bisexual identification recasts one established demographic trend in an important new light. Previous research has documented the rapid growth of bisexual identification among younger cohorts of cisgender women (e.g., England et al. 2016), a fact that drives much of the overall increase in LGBTQ+ identification among “Gen Z” (Jones 2021). The current study suggests that these trends might be understood more accurately as an expansion in sexual fluidity rather than as a fixed shift in the “number” of bisexuals.

To illustrate this point, online appendix Figure A6 presents the gender-specific cohort trends across three measures of bisexual identification: those who currently, ever, and always identified as bisexual. Although cisgender women's bisexual identification increased by all measures, accounting for fluidity yields vastly different estimates of the magnitude of these trends. Comparing women born between 1971 and 1980, 1981 and 1990, and after 1990, ever identifying as bisexual increased from 8% to 14% to 19%, but always identifying as bisexual only changed from 2% to 3% to 4%. I present these divergent trends not to imply that either of them reflects the “true” number of bisexuals. Instead, the validity of both sets of estimates underscores just how complex it can be to quantify how many people “are” a given sexual orientation, particularly among young people today. These critical complexities are swept aside in the kind of year-by-year tallying of LGBTQ population size—from 4.5% (Newport 2018) to 5.6% (Jones 2021) to 7.1% (Jones 2022)—widely reported in the press.

Even as this study destabilizes current estimates of LGBQ population size, it reinforces current understandings of LGBQ well-being. Across three distinct outcomes, I find that accounting for sexual fluidity does not substantively change any of the conclusions that researchers would have reached on the basis of cross-sectional data alone. In terms of their experiences of mental health, financial insecurity, and substance use, sexually fluid adults reported much more like stably identified LGBQ adults than like stably identified straight adults. As a result, disaggregating sexually fluid respondents does little to either magnify or offset the patterns of disadvantage reported by LGBQ respondents more broadly.

Given these facts, the results of this study suggest that researchers' efforts are better spent analyzing the social dynamics that continue to disadvantage sexual minorities than attempting to precisely quantify how many people “are” any given sexual orientation. More specifically, I propose a few key implications for ongoing population research. First, whenever possible, researchers should analyze sexual orientation measures across multiple survey waves, noting within-person patterns of fluidity and assessing whether study results vary between fluid and stably identified subgroups. Second and relatedly, study designers who intend to assess sexual orientation should include measures across multiple survey waves as a matter of course. When it is not possible to include sexual orientation measures in multiple waves, study designers might consider including a follow-up question asking respondents whether they have always identified with their selected sexual identity. Third, when only cross-sectional data are available, researchers should acknowledge directly that they only have access to respondents' sexual identification at a given moment in time. When discussing limitations, researchers should reflect on how the presence of sexual fluidity may impact their results. This acknowledgment of fluidity is particularly important for research focused on young people, gender-diverse populations, and those identifying as bisexual or “another” sexual identity. Finally, researchers should treat estimates of LGBQ population size with considerable caution, explicitly foregrounding the complexities of quantifying sexuality and emphasizing the fluidity of sexuality in people's lives.

These conclusions should be interpreted in light of the important limitations that the current study faced. First, as noted above, although my data are unique in documenting a full five waves of sexual identification, five years is still a relatively short window within the sweep of someone's life. A longer panel, beginning at an earlier age, would surely reveal that this study's results are just a lower bound on the amount of sexual fluidity that individuals experience. Second, although this study documents the correlates of sexual fluidity, it does not attempt to establish the causes of fluidity. Future work could attempt to test how changes in life circumstances—such as pregnancy, marriage, or gender transitions—shape changes in sexual identification. Third, although I demonstrate the negligible empirical consequences of fluidity for three measures of sexual minority well-being, it is unclear how these results would generalize to other outcomes or other datasets. In any given cross-sectional analysis, the consequences of sexual fluidity rest ultimately on untestable empirical assumptions. Researchers should acknowledge this.

Conclusion

In January 2023, the British government released a report summarizing the first national census to ever ask all adults about their sexual orientation (Office of National Statistics 2023). At this historic juncture, the current study reminds demographers of a fact that has been apparent since the establishment of modern sex research: human sexuality is more complex than the categories we create to contain it. As Kinsey et al. (1948:639) famously concluded: “the world is not to be divided into sheep and goats . . . only the human mind invents categories and tries to force facts into separated pigeon-holes.” Although dividing individuals into quantifiable categories lies at the heart of demographic research, the current study reminds researchers that the boundaries between these categories are much more fluid than our data might suggest.

Acknowledgments

The author gratefully acknowledges the Demography editor and anonymous reviewers for their valuable feedback on early drafts. This research was made possible by the National Addiction & HIV Data Archive Program at the Inter-university Consortium for Political and Social Research.

Notes

References