Introduction

Our understanding of how lesbian, gay and bisexual (LGB) individuals experience the economy has grown rapidly the past twenty years. LGB individuals experience unequal labor market outcomes, have poorer health, and have less access to health care in general (Badgett et al., 2021). These differentials may be getting smaller. However, we do not have a full understanding of mechanisms generating disadvantage to be able to properly account for temporal changes. This is particularly important given the potential for underlying inequalities to lead to disproportionate burdens imposed by the pandemic on sexual minorities.

The need to track the status of this population is particularly important given the rapidly changing social and legal environment. However, such research is difficult to do because there is little high quality data on sexual orientation and economic outcomes. The first Census survey to collect self-reported sexual orientation data was the 2021 Household Pulse Survey, designed to track the economic effects of the pandemic. We utilize this data to estimate sexual orientation based differentials in employment, job loss, and economic vulnerability. We measure economic vulnerability as reporting expense difficulty, food insufficiency, and housing insecurity. We also investigate how sexual minorities cope with vulnerability by estimating differential likelihoods of borrowing money from friends and family, taking on debt from credit cards or loans, and relying on government assistance to meet expenses.

We find that despite having largely similar likelihoods of being employed, sexual minorities are more economically vulnerable than their heterosexual counterparts. For example, gay and bisexual men are approximately 25% (8 percentage points) more likely to take on debt to meet their daily expenses than heterosexual men. Bisexual men are 50% (1 percentage point) more likely to both expect losing their homes within two months and to experience food insufficiency (3 percentage points). Lesbian women are similarly more likely to report expense difficulty—though the differential is twice as large for bisexual women—and are approximately 30% (3 percentage points) more likely to report insufficient food. Our use of the Pulse data allows us to extrapolate our results to the entire LGB population in contrast to much of the existing literature that investigates outcomes for same-sex couples.

Literature Review

There are many channels through which LGB individuals may be at higher risk of job loss and economic insecurity. First, sexual orientation based discrimination may lead to unequal outcomes. Indeed, quantitative social scientists have been investigating the size and source of sexual orientation based earnings inequalities since Badgett’s seminal article published in 1995. Badgett (1995) found that behaviorally gay men earned significantly less, within the range of 11 to 27%, than behaviorally heterosexual men. She did not find a significant sexual orientation based earnings difference among women. Badgett’s study motivated a literature whose rate of growth has steadily increased (see Badgett et al. (2021) for a recent summary.) This literature has robustly documented earnings penalties for gay men as well as a premium for lesbian women (both typically in the range of 10–20%). Outcomes for bisexual individuals, when observable, are worse (Waite et al., 2020). Most of the research on sexual orientation utilizes data from the United States and Canada but differential outcomes have also been documented in Sweden, the UK, Greece, and elsewhere (Valfort, 2017).

Researchers have employed a variety of techniques shedding light on discrimination as a likely contributor to unequal outcomes. Some of this research does so by ruling out alternative explanations. For example, Berg and Lien (2002) explain how the asymmetric earnings effects of sexual orientation are consistent with lesbian women working more—and gay men working less—than their heterosexual counterparts. This outcome could arise due to preferences or household considerations; a woman with a woman partner may spend less time doing household labor (and more time in paid labor) as her partner may contribute more than a typical man. However, earnings effects of sexual orientation are robust to measuring differences in earnings per year, month, or in hourly wages (Allegretto & Arthur, 2001; Berg & Lien, 2002; Martell, 2013a). Further, women in same-sex couples (and men in same-sex couples) do not spend systematically less (more) time in household labor (Martell & Roncolato, 2016).

Recent research has shown how the earnings effects of sexual orientation are consistent with discrimination. Martell (2013a); Berg and Lien (2002) and Antecol et al. (2008) all show that the earnings differentials for behaviorally gay/lesbian individuals cannot be explained by differences in worker characteristics such as their educational attainment or workplace experience, a pattern typically associated with discrimination. This is in part not surprising because sexual orientation based discrimination was legal in most states until 2020. However, state laws making sexual orientation based discrimination illegal decrease the earnings penalty gay men experience and increase their labor supply (Burn, 2018; Martell, 2013b, 2014).

Of course, sexual orientation is largely a concealable characteristic and sexual minorities could try to avoid discrimination by managing the disclosure of their orientation. It is not surprising, then, that lesbian, gay, and bisexual individuals are more likely to work in occupations that are characterized by task independence (Tilcsik et al., 2015) and that earnings penalties for gay men are smaller in occupations with higher independence, which may allow gay men to avoid discrimination (Martell, 2018). Being able to avoid discrimination, proxied by living in states with lower levels of prejudice, also reduces the earnings penalty gay men experience (Burn, 2020).

Taken all together, the cumulative evidence strongly suggests discrimination plays a large role in determining labor market outcomes for gay men. It is somewhat less straightforward to reconcile lesbian earnings advantages with discrimination. Discrimination and the absence of an earnings penalty among lesbian women may arise if increased labor market attachment, work intensity, and specialized educational investments are some of the ways lesbian women cope with greater social and economic vulnerability, which is particularly pronounced among younger lesbians (Martell, 2019). This is a particularly important consideration as lesbian women do not appear to escape the marker of their gender. They earn less than heterosexual men and women in same-sex households are consequently at elevated risk of living in poverty (Allegretto & Arthur, 2001; Badgett, 2018).

There is some evidence that sexual orientation based earnings differentials are getting smaller (Jepsen & Jepsen, 2020). However, there is no consensus on the extent, and if so, the speed at which convergence is occurring (Badgett et al., 2021). Rapid convergence is unlikely because widespread stigma, labor market discrimination, and unequal treatment under the law may have motivated sexual minorities of many generations to make pre-labor market decisions that systematically alter their earnings trajectories. For example, stigma and discrimination appears to motivate sexual minorities to complete more college but with college majors comprised of more tolerant individuals even if they map into lower paying occupations (Burn & Martell, 2020). The lack of access to the rights and privileges that accompany marriage may lower the returns to investments in household assets thereby influencing labor market attachment and occupational attainment (Del Río & Alonso-Villar, 2019; Hansen et al., 2020). Indeed, such behavior has long been argued as an explanation of the marriage premium observed among heterosexual men (Loh, 1996) and more recently among men and women in same-sex marriages (Martell & Nash, 2020).

However, not all sexual minorities are married. And many currently married sexual minorities long lived with unequal access to marriage and its set of institutions that promote savings and asset accumulation, access to private health insurance, and government benefits that can act as a buffer in times of economic insecurity. Therefore, it is not surprising that, despite suggestive evidence of convergence, LGB individuals are also more likely to experience poverty (Badgett, 2018; Badgett et al., 2021; Elton & Gonzales, 2022; Whittington et al., 2020). Moreover, married and unmarried sexual minorities live in a society where, even though it has been declining, stigma against sexual minorities remains widespread (Gil et al., 2021; Hansen et al., 2021). Thus, sexual minorities, on the eve of the global pandemic, were exposed to a series of social and economic inequalities making them particularly vulnerable to economic downturns (Drydakis, 2022). Elevated vulnerability to economic downturns may, in part, be due to increased unemployment making it easier for employers to engage in prejudice motivated discrimination against sexual minorities. Increased unemployment may make it easier to discriminate because employers have a larger pool of potential workers and workers a smaller pool of potential employers to choose from. This discrimination could be on the compensation or employment (first fired, last hired) margins.

The same stigma and homophobia that motivates sexual orientation based discrimination likely leads to negative impacts in many other domains including health, mental well-being, and life-satisfaction (Mann et al., 2019). Stigma, or minority stress more generally (Meyer, 1995), has been linked to lower levels of mental health, higher rates of tobacco use and substance misuse, and diabetes. These outcomes may limit labor market attachment and lead to financial insecurity via increased medical expenses that are exacerbated by lower insurance rates and limited access to care (Gil et al., 2021; Konnoth, 2020). Limited access to care, and the resultant unmet medical needs (Everett & Mollborn, 2014), is often linked to occupational attainment and segregation, both determined in part by discrimination as mentioned above. Severe illness, can cause financial distress through medical costs, time away from paid work, and other related care costs. Furthermore, access to paid leave in the event of illness was reported to be less available to LGBTQ individuals compared to heterosexual counterparts at the start of the pandemic (Whittington et al., 2020).

Stigma and homophobia also meaningfully affect the familial and social networks on which many rely for social and economic support in times of need. Family rejection among LGB youth in particular has a long history. Older LGBTQ individuals, for example, are more likely to live alone and be isolated from family compared to heterosexuals in the same age group (Hsieh & Liu, 2021; Whittington et al., 2020)—an outcome that contributes to worse mental and physical health outcomes. Thus, in times of need, LGB individuals may have fewer friends and family members both willing and able to provide assistance to meet financial needs.

The decreased safety net and higher risk of poverty and discrimination is of particular concern given the shocks brought about by the recent pandemic. The same patterns of occupational attainment mentioned above that allow sexual minorities to manage the disclosure of their identities also placed them at risk during the pandemic. These same occupations were also ones that experienced higher risk of exposure to Covid-19 (i.e. in medical care and service occupations).Footnote 1 The risk of a Covid-19 infection was exacerbated by sexual minorities also being more likely to work in industries that were affected by the economic contraction imposed by the pandemic (i.e. in restaurants and services).Footnote 2

Recent research has began to better understand economic vulnerability for sexual minorities but has been hampered by the availability of high quality data. While some have documented job loss, food insecurity, problems accessing health care, and a decline in mental health for LGBTQ individuals during the pandemic (Drabble & Eliason, 2021; Martino et al., 2021; McKay et al., 2020; Mitchell et al., 2022; Moore et al., 2021; Raifman et al., 2021), findings have so far been limited to either non-random samples or samples without a counterfactual of heterosexual individuals. One exception, to our knowledge, exists. Singh et al. (2022) using nationally representative data from the July 21 through October 11, 2021 PULSE survey, find that LGBT were more likely to be food insecure during pandemic. Not surprising, then, that Deal et al. (2023) find that receipt of public assistance throughout the pandemic varied by sexual orientation. Carpenter et al. (2022) also use the PULSE data from this time period and find significantly worse employment, poverty and food security outcomes for transgender women during the pandemic (Carpenter et al., 2022). We build on these studies by using the Pulse data to specifically examine the economic outcomes and subsequent vulnerabilities of gay, lesbian and bisexual individuals during the pandemic.

Our investigation of how bisexual, gay, and lesbian individuals experienced the pandemic contributes to a broader understanding of the experience of sexual minorities in general. For example, in addition to the benefits of a large and nationally representative sample, a clear advantage of the Pulse data is that individuals self-report their sexual orientation. This is in contrast to the bulk of social science research discussed above where researchers have had to infer sexual orientation via sexual behavior (e.g. Badgett, 1995) or cohabitation status (e.g. Jepsen & Jepsen, 2020) Inferring sexual orientation via these methods introduces measurement bias into empirical estimates. Estimates are biased because the three components of sexual orientation (sexual identity, sexual behavior, and sexual attraction) are distinct. They may overlap, but one does not imply the other (Laumann et al., 2000).

The bias introduced by measurement error depends on the type of researcher inference of sexual orientation used. All methods exclude some members of the sexual minority population of interest and contaminates the sexual minority sample with individuals who are not sexual minorities. This measurement error can be nontrivial when studying small populations such as sexual minorities (DiBennardo & Gates, 2014).

Inferring sexual orientation identity via sexual behavior, as did early work in this area using small samples of the General Social Survey (Badgett, 1995; Blandford, 2003), results in possibly misclassifying LGB individuals who are not sexually active as heterosexual and classifying heterosexual individuals who have sex with members of the same-sex as behaviorally homosexual.Footnote 3 Inferring sexual orientation via cohabitation status, as does the bulk of social science research investigating how sexual minorities experience the economy (Jepsen & Jepsen, 2020), similarly results in misclassification and sample contamination. Selecting a sample based on cohabitation excludes single individuals (whose labor market outcomes tend to be worse). It also excludes most bisexual individuals. Bisexuals are much less likely to cohabit and when they do it is often with a different sex partner (Martell, 2021). This is particularly important as outcomes for bisexual individuals, when observed, are generally meaningfully worse than those of gay, lesbian, and heterosexual individuals (Martell, 2021; Sabia, 2015; Waite et al., 2020). Selecting a sample based on cohabitation also incorrectly classifies some heterosexual individuals who cohabit with a member of the same sex. Self-reported orientation allows us to overcome these challenges and reduce measurement error (Badgett et al., 2021) even though we do not observe the gender or sex of the partner of cohabiting individuals. Moreover, the Pulse data allows us to contribute to a broader understanding of the economic status of bisexual individuals, which should be a priority given our limited understanding of the mechanisms generating their disadvantage.

Data: Household Pulse Survey

We utilize the Household Pulse Survey, phases 3.2–3.4, collected by the U.S. Census Bureau from July 21, 2021 to May 9, 2022.Footnote 4 Each survey-week of the Pulse data contains a random sample of 60,000–70,000 respondents. Throughout, we utilize person weights that adjust for household non-response and household size to create representative individual level estimates. The sample is designed to produce reliable state-level estimates as well as estimates for the 15 largest MSAs in the US. The survey sample is a random selection of housing units for whom the Census Bureau had an associated email or cell phone number. Each sampled housing unit received the survey via email or text, and the respondent was the household unit member for whom the contact information corresponded. The survey has a response rate of nearly seven percent and was designed to take approximately 20 min to complete over the internet.Footnote 5 Respondents provide information on their demographic characteristics as well their household. These were designed with the intention to understand the impact of the pandemic on labor market outcomes and markers of socio-economics vulnerability. However, the uses of the data extend beyond understanding the pandemic. Earlier waves of the Pulse data have been used to investigate related questions among heterosexual individuals including financial well-being (Garner et al., 2020), food and financial security (Bitler et al., 2020), and education (Bansak & Starr, 2021). It has also been used, as mentioned above, to better understand how transgender and other gender minority individuals experience the economy (Carpenter et al., 2022).

Phases 3.2 to 3.4 are the first surveys from the Census to collect self-reported sexual orientation (as well as gender identity), which explains our choice to utilize these phases and not earlier. Sexual orientation is not observable in prior phases. Specifically, the survey asks the individual who responds to the survey, “Which of the following best represents how you think of yourself?” Possible answers include “Gay or lesbian,” “Straight, that is not gay or lesbian,” “Bisexual,” “Something else,” and “I don’t know.” We do not observe the sexual orientation of other household members. Even though our primary estimation sample includes a small number (less than 2%) of individuals whose demographic characteristics were allocated by the Census Bureau, we exclude all individuals whose sexual orientation or sex assigned at birth was not reported, which we code as missing. Neither of these choices is consequential (see Section 7 “Robustness” below). Approximately six percent of the male sample is gay and three percent is bisexual. Among women, three percent of the sample is lesbian and six percent is bisexual. While larger than that observed in other nationally representative, yet smaller, surveys, these proportions are consistent with recent estimates based on younger samples of Americans (Burn & Martell, 2022).

We investigate if sexual minorities are more susceptible to negative labor market outcomes by estimating differential likelihoods of employment. We assess four outcomes. First, residing in a household where someone experienced a recent job loss.Footnote 6 Second, being employed in the past week.Footnote 7 Because individuals could not be working for reasons unrelated to the pandemic, we also investigate differential likelihoods of not working due to the pandemic. That is, if individuals did not work for pay or profit in the past week because their employer closed due to the pandemic or the individual was laid off due to the pandemic.Footnote 8

Negative employment outcomes, as well as other shocks brought about by the pandemic, may lead to unique vulnerabilities for lesbian, gay, and bisexual individuals. We measure vulnerability as having difficulty meeting household expenses,Footnote 9 having acquiring sufficient food,Footnote 10 or experiencing housing insecurity (reporting being somewhat or very likely to lose home within two months).Footnote 11 Because sexual minorities may have different resources to cope with financial shortfalls, we also investigate if lesbian, gay, and bisexual individuals have differential likelihoods of taking on debt,Footnote 12 borrowing from friends and/or family,Footnote 13 and being on governmental assistanceFootnote 14 when they are unable to meet their spending needs as they typically do.Footnote 15 We treat individuals who did not respond to these questions on employment or economic vulnerability as missing and do not include them in our estimation sample.

Descriptive Statistics

We include individuals aged 18 to 64 (inclusive) whose sex was not allocated by survey administrators. Our sample for analysis includes 189,314 heterosexual men, 12,825 gay men and 5480 bisexual men. For women we work with 299,247 heterosexual, 8311 lesbian and 19,574 bisexuals. Table 1 shows that, compared to heterosexual men, gay men are younger, more educated, less likely to be married, and less likely to have children in the home. Bisexual men are on average younger than both gay and heterosexual men. They are more likely to be white non-Hispanic, more likely to have completed some college, but less likely to have a bachelor’s degree or graduate degree compared to heterosexual men. Bisexual, like gay, men are also less likely than heterosexual men to be married (but there is no difference in likelihood of marriage between bisexual men and gay men). Bisexual men are more likely than gay, but less likely than heterosexual men, to have children in the home.Footnote 16 Table 2 shows that, compared to heterosexual women, lesbians are on average younger, more likely to be white non-Hispanic, less likely to be married, and have a lower average number of children compared to heterosexual women. Bisexual women are on average younger, less likely to be married and less educated than both heterosexual women and lesbians. They are more likely than lesbians, but less likely than heterosexual women, to have children in the household. These differences demonstrate that the characteristics—as well as outcomes which we discuss below—of bisexual individuals are meaningfully different than those of their lesbian/gay and heterosexual counterparts. This highlights the importance of data allowing individuals to self-report their sexual orientation so that outcomes can be measured for the full population of sexual minorities.

Table 1 Descriptive statistics for individuals assigned male at birth
Full size table
Table 2 Descriptive statistics for individuals assigned female at birth
Full size table

The characteristics of respondents to the Pulse survey are broadly similar to other nationally representative surveys utilized to study sexual minority populations. We compare our sample characteristics two surveys completed prior to the pandemic and one completed during: the National Health Interview Survey and American Community Survey 2013–2018 (as presented in Badgett et al. (2021)) and the NHIS 2020–2021 (authors’ calculations presented in Appendix Table 13). Overall the Pulse sample is, on average, approximately four years younger than that of the NHIS and ACS in 2013–2018. However, differences across sexual orientations are similar across samples.Footnote 17 Differences in educational attainment in the Pulse data and pre-pandemic (2013–2018) ACS and NHIS are minor and only exist for gay men.Footnote 18 However, the educational attainment we observe in the Pulse is similar to that in the NHIS during 2020–2021. For women, like the NHIS 2020–2021, there is no observed lesbian educational advantage in the Pulse (which is 8% in NHIS and 9% in ACS 2013–2018 samples). However, lesbian, gay, and bisexual individuals in the Pulse data report having completed less education than those in the NHIS. For example, lesbian women are less likely to have a Bacehlor’s degree in the Pulse data (19.7%) than the NHIS 2020–2021 (24.6%). We also note that marriage take up rates among heterosexual men and women do not differ from the Pulse and the NHIS 2013–2018, but lesbian, gay, and bisexual men and women in the Pulse are about 5 percentage points more likely to be married in Pulse data. This may relate to the Pulse being more recent data because the differences in marriage take up rates are less pronounced when comparing to the most recent NHIS 2020–2021. The Pulse sample has a higher share of households with children across all sexual orientations but comparisons between groups is consistent with NHIS and ACS, i.e. heterosexual women were most likely to have children in the home and gay men least likely. The similarities - and lack of differences arising due to reasons related to the pandemic - of the Pulse data we use to other nationally representative samples more commonly utilized in the literature improves our confidence that our results generalize. There are no systematic differences across the samples that could lead to the differentials we estimate reflecting sample composition instead of the experience of sexual minorities. Moreover, we interact each of these characteristics below and our main pattern of results persists. These two factors suggest our results are not driven by the nature of the pandemic or the Pulse sample itself. Therefore, they inform our knowledge of how sexual minorities experience the economy, in good times and bad.

Empirical Approach

To better understand how lesbian, gay, and bisexual individuals experience the economy we estimate sexual orientation based differences in employment outcomes and markers of socioeconomic vulnerability. While our estimation procedure follows standard practices, it is descriptive. We shed light on the association between economic vulnerability and sexual orientation but not the underlying causal mechanisms generating disadvantage. The employment outcomes we investigate include living in a household where an adult recently experienced a job loss, the respondent: being employed, being laid off due to the pandemic, and working for an employer that closed due to the pandemic. The markers of socioeconomic vulnerability we investigate include: having difficulty meeting expenses, having insufficient food, being housing insecure, taking on debt to meet expenses, borrowing from friends and family, and being on government assistance. We predict each of these outcomes, \(Y_{i}\), via the following linear probability models.

$$\begin{aligned} Y_{i}=\beta _0 + \beta _1 LG_{i} + \beta _2 Bi_{i} + \varvec{\beta } X_{i} + + \sigma _{s} + \tau _{t} + \epsilon _{i} \end{aligned}$$
(1)

Our coefficients of interest are \(\beta _1\), the coefficient on the indicator for being gay or lesbian (indicated by \(LG_{i}\)) and \(\beta _2\), the coefficient on being bisexual (indicated by \(Bi_{i}\).) \(X_{i}\) is a vector of individual controls including age and its square, race (Black, Asian, other; reference category white), Hispanic ethnicity, indicators for gender identity (not cisgender and other gender; reference category cisgender) and sexual orientation reports of “Something else” and “I don’t know”, educational attainment (indicators for less than high school, some college, bachelor’s degree and graduate school; reference category high school degree), an indicator for being married, number of adults in the household, number of children in the household, and an indicator for living in a metropolitan statistical area. We also include state (\(\sigma _{s}\)) and survey wave (\(\tau _{t}\)) fixed effects. We estimate robust standard errors (White) to correct for heteroscedasticity. We do not cluster the standard errors.

We note that our pattern of results is not sensitive to our use of OLS or the construction of our dependent variables. Our results are robust to the use of ordered probit models that replace the OLS assumption of cardinality with ordinality as well as take into account the discrete nature of the dependent variable. The general non-importance of the estimation approach is consistent with the existing literature on estimating qualitative outcomes (Ferrer-i Carbonell & Frijters, 2004). We present least squares results for our primary results for computational and interpretive ease.

Results

Men

We find few differences in employment status by sexual orientation for men (Panel A of Table 3).Footnote 19 Gay men have small and statistically insignificant differences in the likelihood of residing in a household with job loss, being employed, or losing employment due to the pandemic. Bisexual men are four percentage points (column 1, 22 percentFootnote 20) more likely to reside in a household with a job loss, but do not exhibit individual differences in the likelihood of currently being employed (column 2) or having lost their job due to the pandemic (columns 3 and 4).

Table 3 Employment outcomes and sexual orientation
Full size table
Table 4 Economic vulnerability and sexual orientation
Full size table

Even though gay men do not experience differential likelihoods of employment, this employment does not appear to completely shield them from economic vulnerability. Gay men do not report higher incidences of expense difficulty (Panel A of Table 4, column 1), food insufficiency (column 2), or housing insecurity (column 3). However, they are 8 percentage points (just over 25 percent) more likely to take on debt, 2 percentage points (20 percent) more likely to borrow from friends and family, and 2 percentage points (11 percent) more likely to be on government assistance than heterosexual men. Outcomes for bisexual individuals are worse. Bisexual individuals are 9 percentage points (16 percent) more likely to report difficulty with their expenses, 3 percentage points (nearly 50 percent) more likely to report food insufficiency, and 1 percentage point (nearly 50 percent) more likely to be housing insecure. Bisexual men were also more likely to pursue strategies to cope with this precarity by being 8 percentage points (34 percent) more likely to take on debt, 5 percentage points (50 percent) more likely to borrow from friends or family, and 4 percentage points (25 percent) more likely to be on governmental assistance than heterosexual men. While these differentials are estimated relative to heterosexual men, outcomes for bisexual men are also worse than those of gay men. The coefficient estimates for gay and bisexual men are statistically significantly different at five percent levels.

Women

Sexual orientation based differentials are larger among women (Panel B of Table 3). Lesbian women are more than 2 percentage points (11 percent) more likely to reside in a household that experiences a household job loss. However, they themselves are not statistically significantly less likely to be currently employed, even though lesbian women are one percentage point (50 percent) more likely than heterosexual women to have been laid off due to the pandemic at some time. Bisexual women are more likely (3 percentage points; about 50%) to reside in a household that experienced job loss, but two percentage points (20 percent) more likely to be employed themselves.

In Panel B of Table 4, we show that lesbian women are 4 percentage points (just over 6 percent) more likely than heterosexual women to report expense difficulty and 3 percentage points more likely to report food insufficiency (26 percent). However, lesbian women are not more likely to experience housing insecurity. The difference in the likelihood of experiencing expense difficulty (8 percentage points or 12 percent) relative to heterosexual women is estimated to be twice as large for bisexual women, but the point estimate is not significantly different from that of lesbian women. Bisexual women also have similar differentials in reporting food insufficiency or housing insecurity. These differentials persist even though bisexual women are also 5 percentage points (16 percent) more likely than heterosexual women to take on debt, 7 percentage points (54 percent) more likely to borrow from friends and family, and 5 percentage points (19 percent) more likely to be on government assistance than heterosexual women. These differentials are similar to those experienced by lesbian women with one exception; the increased likelihood of borrowing from friends and family, relative to heterosexual women, is half as large for lesbian women compared to bisexual women. For all of these outcomes, the confidence intervals for the coefficient estimates for bisexual and lesbian women overlap. Similarly the confidence intervals for bisexual men and bisexual women also overlap. It seems the effects of sexual orientation on these outcomes do not statistically significantly vary by gender with only one exception. The increased likelihood of experiencing expense difficulty and food insufficiency is statistically significantly larger for lesbian women than gay men.

Robustness

Our pattern of results is robust to alternative approaches addressing multiple hypothesis testing as well as model and sample specification. We address our selection of controls, treatment of missingness and item non-response, and our estimation procedure. The results of these robustness exercises are contained in Tables 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30.

We begin with multiple hypothesis testing. Our estimation of differentials on a number of important outcomes means we may have incorrectly claimed statistical (in)significance. As the number of hypotheses tested increases, the likelihood of a Type 1 (incorrectly rejecting the null hypothesis) or Type 2 (incorrectly accepting the null hypothesis) error increases. There are several approaches to adjust the p-values and statistical significance for this possibility (see Jones et al., 2019; Peng et al., 2022). We do note, however, that multiple hypothesis testing does not present a concern regarding the magnitude of our estimates.

We implement three standard approaches. The first two, Bonferri–Holm and Bonferri–Sidak, use “step-down” approaches to calculate adjusted p-values where smaller p-values have more stringent adjustments than larger ones. The unadjusted p-values for each hypothesis tested are sorted from smallest to largest. The Bonferri–Holm adjustment multiplies the smallest p-value by the number of hypotheses tested (K), the second smallest by K − 1 and so on until the largest p-value is unadjusted. The Bonferri–Sidak adjustment is similar in spirit, though the p-value adjustment is exponential.Footnote 21 In both approaches, adjusted p-values are capped at 1. The Bonferri–Holm and Bonferri–Sidak approaches assume independence for all of the null hypotheses, which is likely violated in our case as the outcomes we investigate share a commonality: economic vulnerability. This violation may underpower our estimates. Therefore, our third approach controls for the family wise error rate. We follow Jones et al. (2019) and implement the free step-down resampling procedure outlined in Westfall and Young (1993) that leverages bootstrap** (we implement 10,000) to account for dependence across families of outcomes. We combine our estimates for employment outcomes (i.e. Table 3) into one family and markers of socioeconomic vulnerability (i.e. Table 4) in a second family.

The statistical significance of our pattern of results is largely robust to null hypothesis rejection based on p-values that are adjusted for multiple hypothesis testing, presented in Tables 16 and 17. The statistical (in)significance of our estimates did not change for any outcome among gay men. While nearly all our estimates are robust for lesbian women, the estimated differentials in experiencing a household job loss or being laid of due to the pandemic are not. We would not reject the null hypothesis of no differential likelihood of household job loss under conventional levels under our third adjustment approach that corrects for dependence across outcomes (Westfall & Young, 1993).Footnote 22 The estimated differential for being laid off is not significant under any adjustment approach. Finally, one of the many statistically significant differentials we estimate for bisexual men and women is not robust to p-value adjustment. With adjusted p-values the estimated differential in experiencing housing insecurity for bisexual men or women are statistically insignificant. We fail to reject the null of no significant difference in all three adjustment approaches.

Our results are also robust to alternative sets of control variables. (For employment outcomes, see Tables 18 for men and 19 for women. For vulnerability outcomes, see Table 20 for men and 21 for women.) In our baseline approach, we control for the presence of children, even though the path to parenthood, in many cases, differs by sexual orientation. However, excluding this control does not materially affect our results (Panel A). Similarly, income also differs by sexual orientation and family structure. However, controlling for income does not lead to attenuation in estimates of sexual orientation based differences in vulnerability (Panel B).Footnote 23 Relaxing our age restriction has similarly negligible effects (Panel C).

Second, our results are robust to our alternative treatments of item non-response. (For employment outcomes, see Tables 22 for men and 23 for women. For vulnerability outcomes, see Tables 24 for men and 25 for women.) Our baseline specifications includes a small percentage of individuals with some allocated demographic characteristics. Specifically, missing birth year, ethnicity, race, education, household size, and number of children present have been allocated by the Census Bureau using a simple hotdecking procedure (based on age and state).Footnote 24 Neither the Census Bureau nor the authors allocate values for any of our dependent values. Differences in missingness in these outcomes (not demographic characteristics) explains why the sample size in each column of Tables 3 and 4 vary. Limiting our sample to those individuals that responded to each item we utilize as a dependent variable (such that the estimation sample does not vary across specifications) yields point estimates that are very similar to those we present in baseline (see Panel A). Moreover, excluding individuals for whom the Census Bureau has allocated characteristics also produces point estimates that are also similar in size and statistical significance (Panel B). Finally, including individuals with missing sexual orientation via the inclusion of an indicator variable for “missing sexual orientation” does not effect our results (Panel C) nor does including an indicator variable for missing marital status (Panel D).

Our pattern of results is also robust to our estimation procedure. The differentials we estimate do not arise due to our use of least squares on binary outcomes. Estimating our binary variables via probit leads to similar marginal effects (Table 26 for employment outcomes and 27 economic vulnerability outcomes). Similarly, constructing our dependent variables expense difficulty, food insufficiency and housing insecurity by collapsing categorical outcomes into a single binary is non-consequential. If we replace linear probability models with an ordered probit estimation approach, sexual orientation based differentials in economic vulnerability persist (Tables 28, 29, 30). There are however, some differences. Differentials for lesbian women as well as bisexual men and women are larger at lower levels of food insufficiency (i.e. “sometimes not enough to eat” compared to “often not enough to eat”) and larger at higher levels of expense difficulty (“Very difficult” compared to “A little difficult”).

Exploring the role of the pandemic

These results confirm what many suspected at the pandemic’s onset: sexual minorities are uniquely economically vulnerable. However, it is difficult to determine the degree to which these negative outcomes reflect the disproportionate impact of the pandemic or underlying insecurity due to the generally worse labor market outcomes sexual minorities experience regardless of the state of the economy. We investigate the possible contribution of a disproportionate burden from the pandemic by measuring heterogeneity in outcomes by job loss (pandemic induced as well as otherwise), employment as an essential worker, educational attainment and variation in differentials over the duration of the pandemic. We also investigate the role of family structure, race, and ethnicity. Marriage since it has potential to shield against negative economic shocks, and the presence of children since the pandemic posed unique difficulties in the provisioning of care. Race and ethnicity, because stratification based on multiple marginalized identities placed Black and Hispanic sexual minorities in uniquely precarious positions. The results in this section indicate that the economical vulnerability of sexual minorities can not be fully explained by pandemic-related mechanisms.

Table 5 Role of pandemic and men’s economic vulnerability
Full size table

We begin by by investigating if the differential likelihoods vulnerability for sexual minorities reflect disproportionate impacts of the pandemic arising due to their economic position. That is, the likelihood and impact of job loss. The differentials discussed above are not due to pandemic related job loss (Panel A in Tables 5 and 6). The results come from an additional specification of equation 1 that also includes interactions between the sexual orientation indicators with having lost a job due to the pandemic. Not surprisingly having experienced a pandemic related job loss is associated with large increases in the likelihood of experiencing expense difficulty, food and housing insecurity and relying on debt, borrowing from friends or family, and government assistance. Gay men who lost their job due to the pandemic were 9 percentage points more likely to experience expense difficulty but less likely to take on debt.Footnote 25 However, even gay men who did not lose their job reported higher incidences of taking on debt, borrowing from friends or family and being on government assistance. Difficulties for bisexual men are not driven by those who lost their job either. A pandemic related job loss disproportionately increased the likelihood of food insufficiency for bisexual men, but even bisexual men who did not report a pandemic related job loss reported expense difficulty and food insufficiency all the while also exhibiting more debt, borrowing from friends, and government assistance.

The differentials estimated above for lesbian and bisexual women are not due to job loss either (Panel A of Table 6). Expense difficulty and food insufficiency are largely indistinguishable for lesbians who experienced a pandemic related job loss or not. Though, lesbians who did not experience a job loss were more likely, than those who did,Footnote 26 to borrow from friends and family. There are some differences for bisexual women and heterosexual women. A bisexual woman who lost her job to the pandemic was 11 percentage points less likely to report expense difficulty than a heterosexual women who lost her job to the pandemic (even though job loss still increases the likelihood of expense difficulty overall.) The expense difficulty result is surprising given that job loss is associated with more increased food insufficiency for bisexual women than for heterosexual women. The lower reports of expense difficulty could be related to bisexual women (and lesbians) being more likely to pursue co** strategies of taking on debt, borrowing from friends or family, and being on governmental assistance. These results are not unique to pandemic related job loss; they are similar if we investigate differentials by employment status—not working due to any reason.Footnote 27

Of course, not all workers were equally likely to experience job loss during the pandemic. We show that even though outcomes are better for essential workers in general, the buffer essential work status provides workers is smaller for gay and bisexual men and, more importantly, essential worker status does not explain the vulnerability sexual minorities experience.Footnote 28 We augment equation 1 with an addition indicator for being an essential worker as well as its interaction with the gay/lesbian and bisexual indicators. Essential workers were meaningfully less likely to experience household job loss, more likely to be employed and less likely to experience pandemic related job loss (Panel A of Table 7). However, the lower likelihood of experiencing a household job loss is 3 percentage points smaller for gay men than heterosexual men. Bisexual men in essential work had even higher likelihoods of being employed. Essential workers also had lower likelihoods of expense difficulty, food insufficiency, housing insecurity, as well as taking on debt, borrowing from friends and being on government assistance (Panel B of Table 5). While these benefits were similar for gay and heterosexual men, bisexual men who were essential workers remained 5 percentage points (9 percentage points minus 4 percentage points) more likely than heterosexual essential workers to experience expense difficulty. They also experienced increased likelihoods of taking on debt.

While the economic buffer associated with being an essential worker is similar for heterosexual women, there are some differences from men when examining sexual minorities. These differences suggest that the benefits of essential work are larger for women. Nonetheless, the experience of workers who are not essential do not drive those estimated above. The lower likelihood of job loss associated with essential work among lesbians nearly offsets the negative employment differential for them. Bisexual women who are essential workers in fact have a higher likelihood of employment than heterosexual women who are essential workers (Panel A of Table 8). There are no differences in the association between essential work and expense difficulty, food insufficiency, housing insecurity, taking on debt, borrowing or being on government assistance for lesbian and heterosexual women. However, bisexual women who were essential workers were more—not less—likely than heterosexual women who are not essential workers to experience expense difficulty. They were also more likely to take on debt, experience a smaller reduction in the likelihood of borrowing from friends or family and being on government assistance than heterosexual essential workers. Taken all together, the buffer that essential work may have played for heterosexual men and women did not extend to bisexual workers, so the differentials estimated above are not due to worse outcomes for LGB individuals who are not essential workers. We also note that there is little variation in vulnerability by educational attainment, which should be correlated with the ability to work virtually without interruption and be shielded from the worse affects of the economic downturn. The lack of variation suggests that our primary results are not driven by disproportionate burdens that LGB individuals with less education and potentially more precarious employment.Footnote 29

Just as employment and essential worker status do not drive the sexual orientation based differentials in economic vulnerability we observed above, the economic vulnerability we have estimated is not due to the source of the pandemic itself: contracting Covid-19. We investigate if the differentials we observe are driven by the impact of contracting Covid-19 by augmenting Eq. 1 with an indicator for reporting to have experienced a Covid-19 infection as well as its interaction with the gay/lesbian and bisexual indicators. Men who reported Covid-19 were more likely to experience a household job loss (and slightly more likely to report a pandemic related job loss, Panel B of Table 7). This increased likelihood of a household job loss was twice as large for gay men who reported a Covid-19 infection but not for bisexual men. There were no differential effects of a Covid-19 infection on economic vulnerability by sexual orientation for gay or bisexual men (Panel C of Table 5), suggesting that the disease in and of itself did not drive the patterns we have estimated above. The pattern of results is similar for women with the exception that the association between a Covid-19 infection and household job loss was larger for bisexual women but not lesbian women (Panel B of Table 8). Even though the negative effects of Covid-19 infections were larger for women than for men, there were no differential associations between contracting Covid-19 and economic vulnerability (Panel C of Table 6), suggesting again that the pattern of results we estimated above are not from differential incidences and impacts of the illness.

It also appears that the differentials we observe are not driven by the severity of the pandemic. In additional specifications that included interaction terms between the gay/lesbian and bisexual indicators with indicators for survey week, we found no meaningful variation in the relative outcomes for LGB individuals compared to their heterosexual counterparts at different times of the pandemic. Coefficient estimates (and differences in predicted probabilities) on interaction terms between sexual orientation and survey week were similar in size (and statistically indistinguishable) over the duration of our data.Footnote 30 Thus, even though the intensity of the pandemic—as well as the spread of infections and economic contractions—were uneven across time and could have been additive over all, this temporal variation did not appear to affect the sexual orientation based differences in relative outcomes.

The extent to which the individuals may have been vulnerable to the severity and burdens of the pandemic likely depend on family structure. Family structure, and marriage in particular, has long been associated with improved economic outcomes in good times and bad (Chauncey, 2009; Carpenter et al., 2021). Sexual minorities are less likely to cohabit and marry, as we also showed in Tables 1 and 2. Indeed, marriage is associated with better outcomes for heterosexual men (Panel C of Table 9) and women (Panel C of Table 10.) However, marriage does less to alleviate vulnerabilities for gay men. For example, while marriage is associated with a 10 percentage point increase in the likelihood of being employed, this relationship is 8 percentage points smaller for gay men (Panel C of Table 7, Column 2). The reduction in housing security and borrowing from friends and family associated with marriage is half as large for gay men than heterosexual men (Panel D of Table 5). Even though, the benefits to marriage are larger among women, they remain smaller for lesbians compared to heterosexual women. Marriage is associated with a 36 percent smaller (5 percentage points less than 14) reduction in the likelihood of experiencing expense difficulty for lesbian women (Panel D of Table 6). Bisexual women, as well as men, have associations with marriage that are largely similar magnitudes to their heterosexual counterparts [most bisexual individuals who cohabit do so with a member of a different-sex (Martell, 2021)]. However, the associations between marriage and expense difficulty, housing insecurity, and taking on debt are larger for bisexual than heterosexual women. Therefore, marriage is largely associated with improved outcomes. However, the improvements are smaller for LGB individuals. These individuals, married or not, are more economically vulnerable than heterosexuals. Therefore, we cannot attribute increased vulnerability we observe among sexual minorities to the pandemic and how its affects may have varied by household structure.

In addition to marriage, the presence of children, and the care they deserve and require, created unique difficulties for parents. While the path to parenthood can vary by sexual orientation, we explore differential association of parenthood with employment outcomes and economic vulnerability by including an additional interaction between the presence of children and the indicators for lesbian, gay, and bisexual sexual orientations. Not surprisingly the association between parenthood and outcomes varies by gender, with a larger relationship among women. Panel D of Tables 7 and 8 shows that the presence of children increased the likelihood of household job loss, having an employer shut down due to the pandemic, and decreased the likelihood of being employed for fathers relative to men without children. The same pattern of results holds for women, though the associations with motherhood are much larger (mothers are 8 percentage points less likely to be employed than women without children compared to 1.3 percentage point differential associated with fatherhood.) However, the association between parenthood and employment outcomes does not vary by sexual orientation.

The association between parenthood and economic vulnerability is also gendered. We show in Table 9 that parenthood is associated with an increase in the likelihood of experiencing expense difficulty, housing insecurity, borrowing from friends and family, and being on government assistance among men. It is associated with lower likelihoods of taking on debt, perhaps due to credit constraints. The associations between parenthood and economic vulnerability are of the same sign for women, but generally larger in absolute values (see Table 10). These associations with parenthood do not vary by sexual orientation with three exceptions. First, gay men and bisexual women who are parents do not have higher likelihoods of experiencing expense difficulty than their heterosexual counterparts. Second, parenthood is associated with a lower, not higher, likelihood of being on government assistance for gay men. Among women, the association is larger for bisexual, but not lesbian, parents. Third, parenthood is also associated with a reduced likelihood of experiencing housing insecurity for bisexual men, the opposite of that for heterosexual men, but a larger likelihood for bisexual women compared to heterosexual women. Once again, these data suggest that parenthood was particularly difficult during the pandemic regardless of sexual orientation. However, lesbian, gay, and bisexual individuals with and without children experienced worse employment outcomes and increased levels of economic vulnerability relative to their heterosexual counterparts. Therefore, the burdens we have uncovered here do not appear to be due to the pandemic or its differential effect on households with children and will likely persist in a post pandemic world.

Table 6 Role of pandemic and women’s economic vulnerability
Full size table

In addition to family structure, stratification along the lines of race and ethnicity also led to unequal vulnerabilities to the burdens imposed by the pandemic. These vulnerabilities appear to be formidable (Gould & Wilson, 2020; Hardy & Logan, 2020; Morales et al., 2021). We explore their intersection with sexual orientation by once again augmenting Eq. 1 with interaction terms of our indicators for sexual orientation with indicators for reporting being Black or African American and an Hispanic or Latinx ethnicity.Footnote 31 The results from this specification (see Table 33) replicate the disadvantages we document for sexual minorities above and underscores, once again and with only one exception, that racial and ethnic minorities—regardless of their sexual orientation—had worse outcomes than their white, non-Hispanic counterparts. On the employment margin, we do not observe a large role of intersectionality between race, ethnicity, and sexual orientation. There are two exceptions. First, bisexual Black men were less likely to have experienced a job loss in the household than heterosexual Black men.Footnote 32 Second, bisexual Hispanic women were less likely than heterosexual Hispanic women to have experienced a household job loss and more likely to be employed.Footnote 33 However, our results show that bisexual Hispanic women were more likely to be employed than heterosexual non-Hispanic white women. This particular finding requires a more in depth of analysis of employment type and effects of children in the home, which we reserve for future work.

Table 7 Role of pandemic and men’s employment
Full size table
Table 8 Role of pandemic and women’s employment
Full size table
Table 9 Role of household structure and men’s economic vulnerability
Full size table
Table 10 Role of household structure and women’s economic vulnerability
Full size table

Similar to the employment margin, Black and Hispanic individuals were also more likely to experience economic vulnerability during the pandemic, see Table 34. We observe three intersectional impacts of sexual orientation and race or ethnicity for men and several for women. Many, but not all, of these intersectional effects are compounding. For example, lesbian Black and Hispanic women were more likely to be food insufficient and housing insecure than their heterosexual counterparts. This impact is in addition to increased vulnerability found for all Black and Hispanic women and the vulnerabilities discussed above for all lesbian and bisexual women. Some intersectional effects did not compound. Black lesbian women and bisexual Hispanic women were less likely to take on debt than their heterosexual counterparts, which likely relates to access to credit. Other non-compounding intersections are more difficult to understand. Bisexual Black and Hispanic women were less likely to report expense difficulty than heterosexual Black and Hispanic—but not white—women. Bisexual Black men were also less likely to report expense difficulty compared to heterosexual Black men. And gay Black men were less likely to report food insufficiency than heterosexual Black men, a finding that parallels their lower likelihood of taking on debt.Footnote 34 The mechanisms generating these non-compounding intersectional effects are not immediately clear. But, to be clear, the overall effects do not imply advantage. In both cases gay and bisexual black men were more likely to report vulnerability than heterosexual non-Hispanic white men. The intersecting impacts of sexual orientation with race and ethnicity are under studied. These results point to the vulnerabilities of each group highlight that the ways in which race, ethnicity, and sexual orientation jointly operate are complicated. The need for further study and depth analysis into the compounding impacts of belonging to multiple marginalized identity groups should be a priority.

Conclusion

Our findings document that, during the pandemic, LGB individuals experienced small differentials in the likelihood of being employed. However, they remain particularly vulnerable and are more likely to experience expense difficulty, food insufficiency, and housing insecurity. Alongside these disadvantages, sexual minorities are also more likely to be on governmental assistance and take on debt to meet their financial needs. This pattern of results is not driven by pandemic related job loss, essential worker status, incidences or impacts of Covid-19 infections, or household structure. Taken all together, this suggests that the differentials are in large part underlying differences that existed before, and will likely persist after, the pandemic.

The sexual orientation based differences in economic vulnerability we have documented here are concerning. However, we note that the design of the survey at hand, even though it is the best currently available, likely means our estimates are conservative. Lesbian, gay, and bisexual individuals experiencing the most precarity and vulnerability, such as poverty and homelessness [both of which are are more common among LGB populations (Badgett, 2018; Badgett et al., 2021)] are likely not represented in the data. They would have had the least access to the resources necessary (time and internet) to respond. Indeed, as has been acknowledged elsewhere (Bitler et al., 2020; Carpenter et al., 2022), the lower than typical response rate of the Pulse is consistent with this conjecture. This type of non-response is not fully corrected by our use of survey weights and should frame how we interpret our findings for sexual minorities. Understanding this uniquely vulnerable population, and how they experienced the pandemic, should be a priority for future work. Furthermore, future work should investigate in more detail the compounding vulnerability of being both a person of color and LGB during the pandemic.

Our results highlight that the recent expansion of legal access to marriage and protection from discrimination at the federal level have not led to social and economic equity for sexual minorities. As researchers, policymakers, and advocates promote agendas for progress, they should carefully consider the role of laws and policies that may uniquely benefit sexual minorities even if they were not designed with such intent. Our findings confirm earlier evidence that sexual minorities are more likely to be on government assistance, likely contributed in part—but not whole—to their lower earnings. The expansion of unemployment insurance, fiscal stimulus checks, government rental insurance, and the supplemental nutrition assistance program may have been served as a buffer against the worst of the pandemic for lesbian, gay, and bisexual individuals. Expanding the scope of LGB activists to consider these policies may be an important path to equity.

Future research should investigate the impact of these programs on the well-being of sexual minorities to contribute to the development of a framework to better understand sexual orientation sensitive policymaking. The extent to which these programs are effective is important to understanding current differences as well as the long-term impacts of economic vulnerability and debt, which remain under studied. Finally, future research should expand the scope of outcomes studied to understand the impact of the Covid pandemic on life satisfaction and mental health.