Poverty in US Lesbian and Gay Couple Households

ABSTRACT

Poverty is a widely researched topic in economics. However, despite growing research on the economic lives of lesbians and gay men in the United States since the mid 1990s, very little is known about poverty in same-sex couple households. This study uses American Community Survey data from 2010 to 2014 to calculate poverty rates for households headed by different-sex versus same-sex couples. Comparing households with similar characteristics, the results show that those headed by same-sex couples are more likely to be in poverty than those headed by different-sex married couples. Despite that overall disadvantage, a decomposition of the poverty risk shows that same-sex couples are protected from poverty by their higher levels of education and labor force participation, and their lower probability of having a child in the home. Lastly, the role of gender – above and beyond sexual orientation – is clear in the greater vulnerability to poverty for lesbian couples.

KEYWORDS:

• Poverty
• same-sex couples
• sexual orientation
• lesbian
• gay
• bisexual

JEL Codes:

• I32
• D31
• J16

Introduction

For good reason, poverty receives a great deal of attention in social science literature. The concept of poverty involves identifying the individuals, families, or households whose low incomes make it challenging or impossible to meet their health, nutritional, housing, educational, and other life needs. Understanding poverty and working to end it are at the core of national income support programs as well as international economic development programs, such as the United Nations’ Sustainable Development Goals. The fact that not all individuals and households are equally likely to be poor has also brought attention to the role of gender-related norms, policies, and economic outcomes that increase the risk of poverty for women and their families, in particular. This study links to that body of research and expands our understanding of the role of gender to consider how gender and sexual orientation intersect to shape the rates of poverty experienced by same-sex and different-sex couples in the United States. We also build on the growing knowledge about and interest in the economic well-being of lesbian, gay, bisexual, and transgender (LGBT) people and analyze data to reveal both the predictors and the protectors related to poverty in the lives of LGB people (similar data are not available for transgender people) in the US.

LGBT people living in poverty are often invisible in communities, in public policy, and in public discourse, largely because of stereotypes of LGBT affluence and consumer power that appear around the world. The tremendous surge in research on the economic lives of lesbians and gay men since M. V. Lee Badgett's (1995) groundbreaking study on the wages of LGB people has largely dispelled those stereotypes, but very little research exists on the effect of sexual orientation on poverty. Research on wage gaps has expanded beyond the US to many other countries (see Marieka M. Klawitter [2015] for an overview and meta-analysis).

Contributing to that invisibility is the difficulty of making direct comparisons of the probability of being poor (poverty rates) across different sexual orientations and gender identities, since such comparisons require data from representative samples of a population. While a growing number of countries are adding questions related to sexual orientation to surveys of probability samples, high-quality data on sexual orientation from such samples remain rare. In low-income countries, nonprobability samples provide evidence of vulnerability to poverty among LGBT people. For example, several nonprobability samples in India show high rates of poverty among LGBT people (Khan, Bondyopadhyay, and Mulji 2005; Newman et al. 2008; Masih, Singh, and Mishra 2012). A study of low-income LGBT people in Rio de Janeiro found high rates of discrimination and homelessness and low rates of employment (Itaborahy 2014). The only known direct comparisons of poverty rates from probability samples come from studies in the US, which find that LGB people (not including transgender people because of data constraints) are often more likely to be poor than are heterosexual people with similar characteristics. Those studies show higher poverty rates for lesbians and female same-sex couples, in particular, suggesting that this higher poverty burden falls disproportionately on women. These studies suggest the need for further research into the degree of poverty related to being LGBT, and they also show the important role of gender in adding to the vulnerability to poverty (Albelda et al. 2009; Badgett, Durso, and Schneebaum 2013).

In this study, we draw on five years of a probability sample of households in the US, the American Community Survey (ACS) from 2010–14, to compare the probability of poverty between same-sex and different-sex couples. Households headed by same-sex couples may have different poverty rates or chances of being in poverty than households headed by different-sex couples for two main reasons. First, same-sex couple households have different characteristics than different-sex couple households, and some of these characteristics are related to the probability of being in poverty. In particular, female same-sex couples would have a higher risk of poverty because the gender wage gap keeps earnings in households without male earners relatively low. Further, the greater racial and ethnic diversity in LGB couples (even though LGB individuals are slightly more likely to be white) may make them more likely to be poor since minority racial and ethnic groups have higher poverty rates. On the other hand, some factors push down the probability of same-sex couple households being in poverty. Higher levels of labor force participation, for example, serve as an important barrier to poverty. Since same-sex couples have higher levels of labor force participation than different-sex couples (at both the extensive and intensive margins), LGB households may be safer from poverty than heterosexual households. The differences in a couple's characteristics, which differ by the sexual orientation of the people in a couple, are therefore related to the couple's chances of being in poverty.

Second, LGB people and couples may be treated differently than heterosexual people and couples so that the same characteristics have different effects on the probability of being poor. Discrimination against LGB people may put the LGB group at a higher risk of poverty. For example, employer discrimination against LGB people can result in lower wages for LGB people, even if they have the same qualifications as heterosexual people.

In this study, we find that after controlling for a set of well-known poverty predictors, same-sex couples are more likely to be poor than are comparable different-sex married couples, and that lesbian couples are also more likely to be poor than unmarried different-sex couples. Using a decomposition of the role of differences in characteristics and differences in returns to characteristics on the probability of being in poverty, we provide a deeper analysis of sexual orientation differences. Same-sex couples’ poverty rates are held down by the fact that they have protective characteristics, particularly for male same-sex couples. However, both lesbian and gay couples face, overall, lower returns to their characteristics. This study therefore provides previously unknown information about gender and the economic situation of households headed by same-sex couples and may therefore be helpful in informing public policy aimed at protecting groups in a more vulnerable economic position.

Determining and measuring Poverty

The literature on poverty differentials across populations identifies several determinants or predictors of poverty. As noted above, the gender structure of a couple plays a particular role in this paper, given the focus on couples. Other factors predicting poverty include employment, education, race and ethnicity, household structure, age, disability, and area of residence. These individual- and household-level characteristics differ by sexual orientation, and therefore they provide first hints about how and why the chances of being in poverty might differ for same-sex versus different-sex couple households.

First, the gender of the householder is often cited as a strong determinant of poverty. Households headed by women – especially single-parent households – are more likely to be poor than those headed by men (Schiller 2004; Haughton and Khandker 2009). In our case, the gender of the householder and the spouse or partner is obviously relevant. On average, women earn less than men, but women in same-sex couples earn more than their same-gender counterparts in different-sex married couples. Although lesbians earn more than heterosexual women in many studies (Klawitter 2015), as well as in the 2010–14 ACS data, they still earn less than both gay and heterosexual (married) men.¹ Therefore, as a result of occupational segregation, discrimination, and other labor market disadvantages for women, we might expect lesbian couples to have the lowest combined income and therefore highest poverty rates, followed by heterosexual married couples, and finally by gay male couples to have the lowest poverty rates.

The second critical determinant of poverty – employment status – is perhaps the most obvious predictor of a person or family's chances of being in poverty, in that employment brings income, which is a direct opposing force to poverty. Although there are many “working poor” in the US – people who have jobs but do not earn enough at those jobs to escape poverty – employment generally serves as a bridge out of poverty (Schiller 2004). In more same-sex couples than different-sex married couples, both members of the couple work in the paid labor force, which protects same-sex couples from poverty (Black, Sanders, and Taylor 2007; Antecol and Steinberger 2013; Giddings et al. 2014).

A third important determinant of poverty is one's educational attainment. Higher levels of education open opportunities for employment and better-paying jobs, and thus serve as protection against poverty. On average, people in same-sex couples have higher levels of educational attainment than those in different-sex couples (Black et al. 2007), which should protect them from poverty.

Family structure and presence of children are additional predictors of poverty. Children increase the need for financial resources but also limit the ability of parents to engage in paid work. Single-parent households are at the highest risk of poverty (Schiller 2004). More working-age adults in a household can serve as a barrier to poverty if they undertake employment and contribute to family resources. They might not contribute resources equally, though, so having more adults in the household does not always translate into more income for the household (Folbre 1986; Woolley and Marshall 1994).

Related, marriage is negatively correlated with poverty, partly because married couples include two potential earners rather than one and partly because married couples have lower poverty rates than unmarried couples. Unfortunately, for most of the time period for our data, it is not possible to identify which same-sex couples are married since most could not marry until the rapid expansion of states with marriage equality in 2013 and 2014 and the final Supreme Court decision granting a nationwide right to marry in 2015. However, many same-sex couples are rapidly marrying now that they have the opportunity in the US (45 percent had married by October 2015, only four months after the Supreme Court decision), suggesting that same-sex couples are more like married different-sex couples than unmarried different-sex couples (Gates and Brown 2015).

Demographic characteristics like race, ethnicity, age, and disability provide additional predictors of poverty. African Americans and Hispanics are much more likely to be in poverty than whites and non-Hispanics, for example, in large part because of discrimination in education, the labor market, and the housing market (Albelda, Drago, and Shulman 1994; Schiller 2004; Quillian 2012). Older people are at the highest risk of being in poverty; along with age comes decreased income as people enter retirement or face age-based discrimination (Rothenberg and Gardner 2011). Young adults, who may have taken jobs before completing their education or are in entry-level positions, are more likely to be in poverty than those age 25 or older (Schiller 2004). Disability, which can serve as a barrier to employment, can put a person or family at risk of poverty. Fluency in the English language also protects from poverty, as it increases educational and employment opportunities.

Finally, regional and community characteristics – such as access to public transport to commute to paid work, and well-developed infrastructure – can serve as a protector against poverty (Haughton and Khandker 2009); therefore, residing in a more urban area may be helpful in avoiding or escaping poverty. Residing in a city may protect LGB people because there may be more tolerance for sexual minorities in cities, thus reducing discrimination, unemployment, and social exclusion.

Four studies have looked at the poverty rates of LGB versus heterosexual families specifically and have taken into account these predictors. Randy Albelda et al. (2009) use 2000 Census data along with data from the National Survey of Family Growth and the California Health Interview Survey and find that both gay and lesbian couples were more likely to be in poverty than similar different-sex married couples. An update by M. V. Lee Badgett, Laura Durso, and Alyssa Schneebaum (2013) using newer data shows that LGB people continued to be more likely to be in poverty than heterosexuals. Anastasia H. Prokos and Jennifer Reid Keene (2010) use 2000 Census data to calculate poverty rates for different household types and find that both male and female same-sex couples have higher poverty rates than different-sex married couples but lower rates than unmarried different-sex couples. However, Prokos and Keene (2010) only include households with children in their analysis – a limited subsample if one wants to make conclusions about the broader same-sex couple population. A more recent analysis using data on couples with children in the Current Population Survey also shows a higher risk of poverty for female same-sex couples, but the higher risk was not statistically significant (Brown, Manning, and Payne 2016). The present study is the first journal article to offer an analysis of the poverty rates of the entire coupled LGB population in the US and to decompose the observed gaps into portions that reflect differences in couple characteristics and differences in returns to those characteristics.

Poverty is a complex concept, and as such, its measurement is not straightforward. Two measurement concepts dominate the literature.² First, poverty may be seen as a relational concept, as in Europe. The relational measure of poverty considers one's economic standing relative to that of others, so the poor are counted as anyone in the lowest fifth (or tenth, or third) of the income distribution. A second way of measuring poverty is in absolute terms. This approach employs the idea that there is some minimum level of goods and services that people need or should have. People or households who do not have the resources to access those goods and services are then considered poor. This is the official approach taken by the US government. The official poverty rate in the US is calculated as the percentage of people with an income below a certain level. The official US poverty thresholds – the minimum income that a family of a certain size would need to not be considered in poverty – were originally intended to estimate the minimum expenditure necessary for a nutritionally adequate diet for families of different compositions and sizes (meaning number and age of household members), and have been updated to keep up with inflation (Blank 2008). We use this measure in our analysis, but we also calculate poverty rates across sexual orientation using the Organisation for Economic Co-operation and Development (OECD) and European Union (EU) measures in the Supplemental Online Appendix.

Data and Measures

The present study pools cross-sectional data from the Integrated Public Use Microdata Series (IPUMS) sample of the ACS from 2010 to 2014 (Ruggles et al. 2015) to measure poverty in different household types.³ The ACS is an annual survey that sampled more than 15 million individuals living in the US in those years. The survey collects data on a vast range of demographic and economic information for individuals and households. Although there is no question on the ACS that asks directly about one's sexual orientation, researchers can identify people in same-sex couples. Each household has a “householder,” and this householder states their relationship to every other person in the household; when the householder has an “unmarried partner” (or, since 2013, a spouse) of the same sex, these people are in a same-sex couple and are said to be lesbian or gay.

To reduce the probability of having the sample of same-sex couples contaminated by different-sex couples who miscoded one partner's sex, this study follows the suggestions of Dan A. Black et al. (2007) and Gary J. Gates and Michael D. Steinberger (2009) and drops any observation for which the householder or the householder's spouse or partner has an imputed value for his or her marital status and who mailed in their completed survey (see Gates and Steinberger [2009] for details on locating same-sex couples in the Census and ACS data). This restriction reduces the number of same-sex couples in the 2010–14 sample by about 15 percent, but it is important because it substantially reduces the number of couples who are miscoded as same-sex but are actually different-sex.⁴

We use the ACS for several reasons. First, it is a large dataset with a sufficient sample size to study people from a small subgroup (those in same-sex couples) who have a rare characteristic (being poor). Second, the use of same-sex couples allows us to standardize the family composition across sexual orientations. Third, we have detailed information on both partners in the couple and can use characteristics of the couple to predict poverty rather than relying on a sample adult or the householder. The obvious disadvantage is that the sample does not allow for the identification of non-coupled LGB people, nor can we identify people who are transgender. No existing datasets provide all of these desirable qualities.

We use the official poverty thresholds set out and annually updated for inflation by the US Census Bureau.⁵ A person or household is considered to be in poverty if their household income falls below the Federal Poverty Line (FPL). The FPL is based on family size and the number of children in the household, as well as the age of the householder (65 and older, or not). The Census Bureau considers a “family” to be people who live in the same housing unit and who are related by blood, marriage, or adoption. Since not all same-sex couples could marry in the years that they were sampled, and since those who were married were not acknowledged as such in the ACS data until 2013, when the US Supreme Court ruled the Defense of Marriage Act unconstitutional, we amend the Census Bureau's definition of “family” for this study. In particular, we consider any cohabiting couple (married or unmarried partners) along with all children under age 18 living in the same household (regardless of their stated relationship to the householder) to be a family. Such an amendment to the definition of a family in measuring poverty to include unmarried but cohabiting partners as family members entered the literature in the 1990s (Ruggles 1990; Citro and Michael 1995; Carlson and Danziger 1999).

Accordingly, we consider household income, which is used to compare against the FPL, to be the summation of the total pre-tax income of the two members of the householder couple. As defined by the Census Bureau, total income is the sum of income from all sources, such as earnings, pensions, social security, and interest.⁶ We do not include (adult) children's income, as in the standard US poverty definition, because of a lack of clarity about family interrelationships among members of a household in the ACS. In particular, we know the relationship of the householder to everyone else in the household, but we do not know the relationship between the partner and other people in the household. Thus, a household with a householder, partner, and child of the partner could appear as a couple with an “unrelated” child because the child may not be officially related by birth or adoption to the householder. To be consistent in our calculations, we define “family” as the householder, his or her partner or spouse, and all people under the age of 18 in the same housing unit. We then calculate the size of each family and apply the appropriate poverty income thresholds against the sum of the total income of the householder and his or her spouse or partner. While this definition excludes other adult family members (such as parents and siblings) that might be in the household, we have treated same-sex and different-sex couples comparably and consistently. Finally, we exclude households where the value of any component of total income for either one of the members of the couple was given a data quality flag by the Census Bureau.

We create household-level variables based on the poverty predictors discussed earlier to investigate the relationship between these characteristics and poverty: race, ethnicity, education level, employment status, age, disability status, and English fluency of both people in the couple, along with the geographic location of a household (region and metropolitan area size) and the number of children and adults living in the household. These variables capture household-level, not just person-level, information because the characteristics of both partners in the couple matter in determining poverty, not just the characteristics of one of the couple's members. For example, instead of looking at just one person's race, we code the couple's racial composition: both white, both African American, both Asian, both Native American, both “other” race, interracial couple with a white person, or interracial couple with neither person white.

All respondents who reported a positive number of paid work hours in a year are counted as working part time if they worked up to 1,500 hours, or full time if they worked more than 1,500 hours. An individual's number of paid work hours last year was calculated by multiplying the midpoint of the intervals used to measure “weeks worked last year” variable by the usual hours worked per week. People who report zero hours of paid work in a year can be identified as being either unemployed or not in the labor force, depending on the information provided about employment status at the time the survey was conducted. A person is considered disabled if they reported having cognitive, ambulatory, independent living, self-care, vision, or hearing difficulty. People cannot speak English fluently if they report that they cannot speak English or cannot speak it well. Because of the confidentiality requirements of the Census Bureau, almost 50 percent of cases in the 2010–14 data have an “unknown” or “not identifiable” metropolitan status. We therefore use a measure of a household's degree of urbanity/ruralness given by the so-called Beale scale, which ranges from 1 to 9 and increases with ruralness.⁷ Values of 1, 1–2, or 2–3 indicate a large, medium, or small metropolitan area, respectively, whereas a value greater than 3 denotes a non-metropolitan area. The means of these household characteristics for the different household types are given in Table 1.

Table 1 Descriptive statistics by household type

	Sexual orientation/Marital status of head couple
	(1)	(2)	(3)	(4)	(5)
Variable	Different-sex married	Different-sex unmarried	Same-sex male	Same-sex female	All couple households
Percentage in poverty	5.9	14.6***	3.9***°°°	6.6**°°°	6.7
	(0.02)	(0.10)	(0.23)	(0.30)	(0.02)
Race of couple
Both white	80.4	72.2***	78.0***°°°	80.2°°°	79.6
	(0.03)	(0.13)	(0.47)	(0.47)	(0.03)
Both African American	5.4	7.9***	2.4***°°°	4.8**°°°	5.6
	(0.02)	(0.08)	(0.19)	(0.27)	(0.02)
Both Native American	0.2	0.5***	0.1°°°	0.2°°°	0.2
	(0.00)	(0.02)	(0.06)	(0.05)	(0.00)
Both Asian	4.8	1.4***	1.3***	1.0***°°°	4.5
	(0.02)	(0.03)	(0.13)	(0.11)	(0.02)
Both other race	3.1	5.5***	1.5***°°°	2.0***°°°	3.3
	(0.02)	(0.07)	(0.15)	(0.17)	(0.02)
Interracial – with white	5.5	10.9***	15.7***°°°	10.5***	6.1
	(0.02)	(0.09)	(0.41)	(0.36)	(0.02)
Interracial – without white	0.7	1.6***	1.0**°°°	1.3***°	0.8
	(0.01)	(0.04)	(0.12)	(0.14)	(0.01)
Ethnicity of couple
Neither Hispanic	85.7	76.8***	81.4***°°°	84.5***°°°	84.9
	(0.03)	(0.12)	(0.44)	(0.42)	(0.03)
One Hispanic	4.8	8.9***	13.9***°°°	9.5***°	5.2
	(0.02)	(0.08)	(0.39)	(0.34)	(0.02)
Both Hispanic	9.5	14.4***	4.7***°°°	6.0***°°°	9.9
	(0.03)	(0.10)	(0.25)	(0.29)	(0.03)
English fluency of couple
Both speak English	93.7	92.4***	97.8***°°°	98.5***°°°	93.6
	(0.02)	(0.08)	(0.18)	(0.15)	(0.02)
One speaks English	3.3	3.8***	1.5***°°°	1.0***°°°	3.3
	(0.02)	(0.06)	(0.15)	(0.11)	(0.02)
Neither speak English	3.0	3.8***	0.7***°°°	0.5***°°°	3.1
	(0.02)	(0.06)	(0.10)	(0.10)	(0.01)
Age of couple
Both 18–24	0.9	8.6***	1.6***°°°	3.2***°°°	1.6
	(0.01)	(0.08)	(0.15)	(0.22)	(0.01)
Both 25–34	10.8	25.7***	10.9°°°	12.5***°°°	12.2
	(0.03)	(0.12)	(0.37)	(0.37)	(0.03)
Both 35–49	24.7	18.1***	23.6**°°°	23.2***°°°	24.1
	(0.04)	(0.11)	(0.47)	(0.48)	(0.03)
Both 50–64	24.9	9.6***	17.5***°°°	18.6***°°°	23.4
	(0.03)	(0.07)	(0.40)	(0.40)	(0.03)
Both 65+	14.9	2.5***	4.8***°°°	4.5***°°°	13.6
	(0.03)	(0.04)	(0.21)	(0.20)	(0.02)
18–24/25–34	1.3	9.7***	2.8***°°°	4.0***°°°	2.1
	(0.01)	(0.09)	(0.21)	(0.24)	(0.01)
18–24/35–49	0.1	1.0***	1.2***	1.0***	0.2
	(0.00)	(0.03)	(0.14)	(0.12)	(0.00)
18–24/50–64	0.0	0.1***	0.3***°°°	0.5***°°°	0.0
	(0.00)	(0.01)	(0.06)	(0.10)	(0.00)
25–34/35–49	6.4	12.6***	10.0***°°°	10.8***°°°	7.0
	(0.02)	(0.10)	(0.34)	(0.37)	(0.02)
25–34/50–64	0.2	0.6***	1.5***°°°	0.9***°°°	0.2
	(0.00)	(0.02)	(0.14)	(0.11)	(0.00)
35–49/50–64	8.9	8.2***	17.6***°°°	14.3***°°°	8.9
	(0.02)	(0.07)	(0.41)	(0.38)	(0.02)
Under 49/65+	0.3	0.4***	1.9***°°°	0.9***°°°	0.3
	(0.00)	(0.01)	(0.15)	(0.10)	(0.00)
50–64/65+	6.6	2.9***	6.3°°°	5.6***°°°	6.2
	(0.02)	(0.04)	(0.24)	(0.23)	(0.02)
Education of couple
Both less than high school (LTHS)	4.9	6.8***	1.3***°°°	1.3***°°°	5.0
	(0.02)	(0.07)	(0.13)	(0.14)	(0.02)
Both high school degree (HS)	29.1	36.6***	19.7***°°°	22.4***°°°	29.7
	(0.04)	(0.14)	(0.45)	(0.48)	(0.04)
Both more than high school (MTHS)	31.7	18.4***	42.8***°°°	43.8***°°°	30.5
	(0.04)	(0.10)	(0.54)	(0.55)	(0.04)
LTHS/HS	7.7	13.0***	3.9***°°°	4.7***°°°	8.1
	(0.02)	(0.10)	(0.24)	(0.25)	(0.02)
MTHS/HS	25.2	23.3***	30.9***°°°	26.4**°°°	25.0
	(0.04)	(0.12)	(0.50)	(0.49)	(0.03)
MTHS/LTHS	1.5	2.0***	1.5°°°	1.2**°°°	1.6
	(0.01)	(0.04)	(0.13)	(0.13)	(0.01)
Employment of couple
Both unemployed	0.1	0.4***	0.2**°°°	0.3**°	0.2
	(0.00)	(0.02)	(0.05)	(0.07)	(0.00)
Both NILF	14.8	5.8***	7.4***°°°	6.9***°°°	13.8
	(0.03)	(0.06)	(0.27)	(0.26)	(0.02)
Both part time	2.4	4.6***	2.5°°°	3.3***°°°	2.6
	(0.01)	(0.06)	(0.17)	(0.20)	(0.01)
Both full time	35.8	43.5***	51.7***°°°	50.7***°°°	36.7
	(0.04)	(0.14)	(0.55)	(0.56)	(0.04)
Unemployed/NILF	0.5	1.0***	0.6°°°	0.5°°°	0.6
	(0.01)	(0.03)	(0.10)	(0.08)	(0.01)
Unemployed/Part time	0.5	1.4***	0.4°°°	0.6°°°	0.6
	(0.01)	(0.03)	(0.07)	(0.09)	(0.01)
Unemployed/Full time	2.0	3.8***	2.7***°°°	2.1°°°	2.2
	(0.01)	(0.05)	(0.18)	(0.15)	(0.01)
NILF/Part time	5.5	4.6***	3.9***°°°	4.1***°°	5.4
	(0.02)	(0.06)	(0.21)	(0.23)	(0.02)
NILF/Full time	21.0	14.1***	13.8***	13.1***°°	20.2
	(0.03)	(0.10)	(0.38)	(0.37)	(0.03)
Part time/Full time	17.3	20.7***	16.7°°°	18.5***°°°	17.6
	(0.03)	(0.11)	(0.41)	(0.43)	(0.03)
Disability of couple
Neither disabled	81.5	84.6***	83.2***°°°	81.2°°°	81.8
	(0.03)	(0.10)	(0.41)	(0.44)	(0.03)
One disabled	14.5	12.3***	13.6**°°°	15.0°°°	14.3
	(0.03)	(0.09)	(0.38)	(0.40)	(0.03)
Both disabled	4.0	3.1***	3.1***	3.8°°°	3.9
	(0.02)	(0.05)	(0.19)	(0.21)	(0.01)
Region
New England	4.8	5.4***	5.6***	7.0***°°°	4.9
	(0.02)	(0.06)	(0.26)	(0.28)	(0.02)
Mid-Atlantic	12.4	12.3	14.8***°°°	12.3	12.4
	(0.03)	(0.09)	(0.41)	(0.36)	(0.03)
East North Central	15.8	16.2***	12.4***°°°	13.6***°°°	15.8
	(0.03)	(0.10)	(0.37)	(0.39)	(0.03)
West North Central	7.6	7.5	5.4***°°°	6.6***°°°	7.6
	(0.02)	(0.08)	(0.28)	(0.31)	(0.02)
South Atlantic	18.9	17.8***	20.7***°°°	18.8°°	18.8
	(0.03)	(0.10)	(0.43)	(0.43)	(0.03)
East South Central	6.0	5.1***	3.7***°°°	4.5***°°	5.9
	(0.02)	(0.06)	(0.21)	(0.23)	(0.02)
West South Central	11.5	10.1***	8.8***°°°	10.0***	11.3
	(0.03)	(0.09)	(0.31)	(0.34)	(0.03)
Mountain	7.4	7.6***	7.0°°	8.6***°°°	7.4
	(0.02)	(0.07)	(0.28)	(0.31)	(0.02)
Pacific	15.5	17.9***	21.5***°°°	18.7***°	15.8
	(0.03)	(0.10)	(0.44)	(0.42)	(0.03)
Metropolitan size
Large	50.7	51.3***	69.0***°°°	57.4***°°°	50.9
	(0.04)	(0.14)	(0.51)	(0.55)	(0.04)
Medium	20.8	21.2***	17.5***°°°	21.7**	20.9
	(0.03)	(0.12)	(0.42)	(0.46)	(0.03)
Small	8.4	8.6**	5.1***°°°	7.5***°°°	8.4
	(0.02)	(0.08)	(0.25)	(0.31)	(0.02)
Non-metro	20.1	18.8***	8.4***°°°	13.4***°°°	19.9
	(0.03)	(0.11)	(0.31)	(0.37)	(0.03)
Household composition
Any children present	44.2	42.7***	9.3***°°°	26.7***°°°	43.9
	(0.04)	(0.14)	(0.35)	(0.52)	(0.04)
Number of children, when any	1.93	1.88***	1.86*	1.73***°°°	1.93
	(0.00)	(0.00)	(0.04)	(0.02)	(0.00)
Any children 18+ present	18.4	7.2***	3.4***°°°	7.6***	17.2
	(0.03)	(0.07)	(0.21)	(0.32)	(0.03)
Any other relatives 18+ present	4.3	4.5***	4.0°°	4.9**	4.3
	(0.02)	(0.06)	(0.23)	(0.25)	(0.02)
Any non-relatives 18+ present	1.2	5.2***	5.3***	4.7***°°	1.6
	(0.01)	(0.06)	(0.25)	(0.23)	(0.01)
Observations	2,349,733	207,625	12,002	12,068	2,581,428

Notes: Standard errors in parentheses. In columns 2, 3, and 4, ***, **, * indicate a statistically significant difference in means at the 1, 5, and 10 percent levels, respectively, relative to column 1. Additionally, in columns 3 and 4, °°°, °°, ° indicate a statistically significant difference in means at the 1, 5, and 10 percent levels, respectively, relative to column 2.

Source: Authors’ calculations based on ACS 2010–14.

The descriptive statistics in Table 1 demonstrate the importance of controlling for these predictors to explain differences between same-sex and different-sex couples. Same-sex couples are more likely to have several characteristics that serve as protection against poverty, such as higher levels of education and employment, and having fewer children. Same-sex couples are more likely to be employed full time: 51.7 percent of male same-sex and 50.7 percent of female same-sex couples were composed of two full-time employed people, compared with just 43.5 percent of unmarried and 35.8 percent of married different-sex couples. Same-sex couples (both male and female) are more likely to have one or both partners with more than a high school degree: 43.8 (42.8) percent of female (male) same-sex couples are composed of two people whose formal education goes beyond a high school diploma, compared with just 31.7 (18.4) percent of married (unmarried) different-sex couples. On the other hand, some differences will increase the likelihood of poverty for same-sex couples relative to married or unmarried different-sex couples. Compared with married different-sex couples, same-sex couples are somewhat less likely to be composed of two white partners, and they are much younger.

Describing the sexual orientation poverty gap

The first rows in Table 1 give the overall household poverty rates by household type. Overall, 6.7 percent of all couple-led households were in poverty. This rate is much lower than the official poverty rate of 14.8–15.1 percent between 2010 and 2014 because these poverty rates are for people in couples only, excluding children, and because all couples have two potential earners. However, the rate of poverty among couples varies based on the sexual orientation and marital status of the couple leading the household. Households headed by unmarried different-sex couples had the highest poverty rate of all couple types: 14.6 percent of these households were in poverty, compared with 5.9 percent of married different-sex couple households. Same-sex male couple households had a lower poverty rate than different-sex married couple households (3.9 percent), but lesbian-led households had a higher rate (6.6 percent). The differences in these rates are statistically significant across all couple types.

Table A1 in the Supplemental Online Appendix shows the poverty rates for the couples with different characteristics. While these descriptive statistics are illuminating, they do not allow us to establish a clear relationship between sexual orientation, marital status, and poverty. To do that, we need to compare people of different sexual orientations and marital statuses in households with the same characteristics.

Multivariate analysis methods: Highlighting the role of sexual orientation and marital status

In the multivariate analysis, we compare households that differ in their sexual orientation or marital status but are identical in all other (observable) characteristics that could be related to poverty. For example, we compare a gay household and a different-sex married household whose leading couples have the same compositions of race and ethnicity, educational attainment, age, employment status, fluency in English, and disability, and who live in the same region, have a similar degree of ruralness/urbanity, and have the same number of children and adults. “Controlling” for these other characteristics allows us to highlight the role of sexual orientation/marital status, in particular, in predicting poverty.

Because the outcome variable of interest – being in poverty or not – is a dichotomous variable, we employ a probit regression. To find the marginal effect of any independent variable on Y, in this case being in poverty, we take the average of the marginal effect of X on Y for all households in the sample. In other words, we find the effect for each individual household and then show the average of those effects.

Along with finding the marginal effects of sexual orientation on the probability of being in poverty, we are also interested in seeing whether the other independent variables in Equation 1(1) $\left(\overline{Y_{DS}}\right)-\left(\overline{Y_{SS}}\right)=(\overline{X^{DS}}-\overline{X^{SS}}){\beta }^{DS}+\overline{X^{SS}}({\beta }^{DS}-{\beta }^{SS})+({\alpha }^{DS}-{\alpha }^{SS})$(1) have a different impact on the probability of being in poverty for households headed by different-sex versus same-sex couples. In a model whose underlying function is linear, the standard approach for decomposing the difference in an outcome across two groups into the portion explained by differences in the characteristics and differences in the returns to these characteristics is the Oaxaca–Blinder decomposition (Blinder 1973; Oaxaca 1973), modeled as: (1) $\left(\overline{Y_{DS}}\right)-\left(\overline{Y_{SS}}\right)=(\overline{X^{DS}}-\overline{X^{SS}}){\beta }^{DS}+\overline{X^{SS}}({\beta }^{DS}-{\beta }^{SS})+({\alpha }^{DS}-{\alpha }^{SS})$(1) for sexual orientation groups SS (same-sex) and DS (different-sex).

Myeong-Sun Yun (2004) provides an extension to the Oaxaca–Blinder decomposition methodology for nonlinear models, including the probit regression, showing that the decomposition for the binary probit model is essentially the same as for the ordinary least-squares (OLS) model. The decomposition for probit models is available as a standard option in the Stata package oaxaca. An example of the use of this decomposition to analyze the poverty incidence (in India) is given in Ira N. Gang, Kunal Sen, and Myeong-Su Yun (2008). We also follow Myeong-Su Yun's (2005) suggestion of dealing with the problem that the decomposition results for categorical variables depend on the choice of the omitted category by decomposing the effects in their normalized form, that is, expressing them as deviations from the grand mean.

Results: Exploring the sexual orientation gap in poverty rates

Table 2 shows the results of the multivariate probit models giving the contribution of being in a gay or lesbian couple on the probability of being in poverty. The results show that controlling for the race, ethnicity, employment, education, age, English fluency, and disability composition of the couple in a household, along with the number of adults and children in the home and the household's region and degree of ruralness (as well as fixed effects for years), both same-sex and different-sex unmarried couples are more likely than different-sex married couples to be in poverty. Lesbians are most likely to be in poverty, confirming the expectations laid out earlier in the paper; they are 2.4 percentage points more likely to be in poverty than different-sex married couples. Gay couples are 1 percentage point more likely to be in poverty than different-sex married couples. Unmarried different-sex couples are 2.1 percentage points more likely to be poor than married different-sex couples. These are meaningful differences in the risk of poverty, comparable in magnitude to racial differences in the probability of poverty and to the effects of education.

Table 2 Probit regression predicting being in poverty, average marginal effects

Sexual orientation (Omitted: Different-sex married)
Same-sex male	0.010***
	(0.003)
Same-sex female	0.024***
	(0.003)
Different-sex unmarried	0.021***
	(0.001)
Race of couple (Omitted: Both white)
Both African American	0.022***
	(0.001)
Both Native American	0.020***
	(0.002)
Both Asian	0.027***
	(0.001)
Both other race	0.005***
	(0.001)
Interracial – without white	0.012***
	(0.002)
Interracial – with white	0.006***
	(0.001)
Ethnicity of couple (Omitted: Neither Hispanic)
Both Hispanic	0.013***
	(0.001)
One Hispanic	0.000
	(0.001)
English fluency of couple (Omitted: Both speak English)
One speaks English	0.033***
	(0.001)
Neither speak English	0.061***
	(0.001)
Age of couple (Omitted: Both 50–64)
Both 18–24	0.094***
	(0.001)
Both 25–34	0.050***
	(0.001)
Both 35–49	0.024***
	(0.001)
Both 65+	−0.080***
	(0.001)
18–24/25–34	0.072***
	(0.001)
18–24/35–49	0.064***
	(0.003)
18–24/50–64	0.059***
	(0.008)
25–34/35–49	0.040***
	(0.001)
25–34/50–64	0.038***
	(0.003)
35–49/50–64	0.023***
	(0.001)
Under 49/65+	−0.015***
	(0.003)
50–64/65+	−0.046***
	(0.001)
Education of couple (Omitted: Both high school)
Both more than high school	−0.039***
	(0.001)
MTHS/HS	−0.022***
	(0.000)
MTHS/LTHS	0.004***
	(0.001)
LTHS/HS	0.021***
	(0.000)
Both less than high school	0.033***
	(0.001)
Employment of couple (Omitted: Both full time)
Both Unemployed	0.255***
	(0.003)
Both NILF	0.208***
	(0.001)
Both part time	0.157***
	(0.001)
Unemployed/NILF	0.241***
	(0.001)
Unemployed/Part time	0.209***
	(0.001)
Unemployed/Full time	0.108***
	(0.001)
NILF/Part time	0.182***
	(0.001)
NILF/ Full time	0.089***
	(0.001)
Part time/Full time	0.060***
	(0.001)
Disability of couple (Omitted: Neither disabled)
One disabled	0.007***
	(0.000)
Both disabled	0.011***
	(0.001)
Region (Omitted: West North Central)
New England	−0.009***
	(0.001)
Mid-Atlantic	−0.001
	(0.001)
East North Central	−0.001
	(0.001)
South Atlantic	0.005***
	(0.001)
East South Central	0.008***
	(0.001)
West South Central	0.004***
	(0.001)
Mountain	0.002*
	(0.001)
Pacific	−0.006***
	(0.001)
Degree of ruralness	0.003***
	(0.000)
Household composition
Number of children	0.021***
	(0.000)
Number of children 18+	0.003***
	(0.000)
Number of other relatives 18+	0.000
	(0.000)
Number of non-relatives 18+	0.004***
	(0.001)
Observations	2,581,428

Notes: Standard errors are in parentheses. Year fixed effects not shown. ***, **, * denote statistical significance at the 1, 5, and 10 percent levels, respectively.

Source: Authors’ calculations based on ACS 2010–14.

The results in Table 2 also allow us to compare the marginal probability of being in poverty for same-sex couples versus different-sex unmarried couples. Subtracting one effect from the other, we see that same-sex male couples are 1.1 percentage points less likely to be in poverty than different-sex unmarried couples. Lesbian couples, on the other hand, are 0.3 percentage points more likely to be in poverty than different-sex unmarried couples.

These results show the influence of gender and marital status. First, although a same-sex male couple household appears to benefit from the advantage of having two male incomes, the negative effects of sexual orientation drive the chances of being in poverty above that of different-sex married couples after controlling for other predictors of poverty. However, the findings for female same-sex couples flip the gender and sexual orientation contributions to average incomes. As discussed earlier, two women's incomes would likely be smaller than a man's income added to a woman's income. But with respect to sexual orientation differences, the literature often finds a so-called “lesbian wage premium,” in which the average lesbian earns more than an average heterosexual woman with the same characteristics (Klawitter 2015). But even if that difference favors individual lesbians on average, this study shows that a same-sex female couple can still end up being more likely to be poor. In a sense, the gender effect appears to outweigh the lesbian wage premium. However, other research has found that within many lesbian couples only one of the members of the couple has a wage premium while her partner faces a wage penalty compared to heterosexual women (Schneebaum 2013), which strengthens the reasons to expect that the household-level earnings of same-sex women is clearly lower than it is for different-sex married couples.

Second, in terms of marital status, marriage matters for the risk of poverty for heterosexual couples, since unmarried heterosexual couples are clearly at a high risk of poverty, particularly when looking at the unconditional poverty rates in Table 1. However, that excess risk falls dramatically after controlling for other factors, and in Table 2, unmarried heterosexuals fall in between male couples and female couples. The remaining difference might well be related to other unobserved characteristics of unmarried couples that differ from married couples.

The risk of poverty shown in Table 2 gives a very different picture from the simple poverty rates of Table 1: after controlling for other factors, lesbians fare worse compared with both kinds of different-sex couples, and gay men fare worse than married heterosexual couples. We can reconcile these two comparisons of poverty by decomposing the difference in poverty rates by sexual orientation. Are the differences across couple type due mainly to the couples’ characteristics, or is it the returns to those characteristics that play a larger role? For gay male couples, the decomposition will explain why their unconditional poverty rates are lower than for married and unmarried heterosexual couples. For lesbian couples, we are explaining why their poverty rates are higher than for married heterosexual women but lower than for women in unmarried heterosexual couples. Table 3 decomposes the difference in the likelihood of being in poverty for gay couples compared to married and unmarried heterosexual couples; Table 4 does the same for lesbians. The underlying probit regressions for each subpopulation can be found in Table A2 in the Supplemental Online Appendix.

Table 3 Oaxaca–Blinder decomposition for poverty rates, gay versus straight

	Different-sex married versus gay		Different-sex unmarried versus gay
	0.020***		0.107***
Total poverty rate gap	(0.002)		(0.002)
Portion due to …	Characteristics	Returns	Characteristics	Returns
Total	0.025***	−0.005***	0.099***	0.007***
	(0.001)	(0.002)	(0.001)	(0.002)
Race	0.001***	−0.006	0.002***	0.096
	(0.000)	(0.004)	(0.000)	(0.687)
Ethnicity	0.000***	−0.004	0.001***	0.060
	(0.000)	(0.003)	(0.000)	(0.430)
English	0.001***	0.001	0.003***	−0.119
	(0.000)	(0.004)	(0.000)	(0.864)
Age	−0.010***	0.0003	0.033***	−0.045
	(0.000)	(0.002)	(0.001)	(0.326)
Education	0.005***	0.001	0.021***	−0.027
	(0.000)	(0.002)	(0.001)	(0.199)
Employment	0.016***	−0.001	0.015***	0.091
	(0.001)	(0.002)	(0.001)	(0.623)
Disability	0.000***	−0.002	−0.000	0.002
	(0.000)	(0.002)	(0.000)	(0.038)
Region	0.000***	0.001	0.001***	−0.026
	(0.000)	(0.001)	(0.000)	(0.181)
Degree of ruralness	0.001***	0.001	0.002***	−0.017
	(0.000)	(0.001)	(0.000)	(0.122)
Number of children 18+	0.000***	0.000	0.000*	−0.001
	(0.000)	(0.000)	(0.000)	(0.009)
Number of other relatives 18+	0.000	−0.000	−0.000	0.004
	(0.000)	(0.000)	(0.000)	(0.027)
Number of non-relatives 18+	−0.000***	−0.000	−0.000	0.001
	(0.000)	(0.000)	(0.000)	(0.007)
Number of children	0.010***	0.000	0.023***	−0.015
	(0.000)	(0.000)	(0.001)	(0.108)
Year	−0.000***	−0.000	−0.001***	0.004
	(0.000)	(0.000)	(0.000)	(0.030)
Different-sex	2,349,733		207,625
Same-sex	12,002		12,002

Notes: Standard errors are in parentheses. ***, **, * denote statistical significance at the 1, 5, and 10 percent levels, respectively.

Source: Authors’ calculations based on ACS 2010–14.

Table 4 Oaxaca–Blinder decomposition for poverty rates, lesbian versus straight

	Different-sex married versus lesbian		Different-sex unmarried versus lesbian
	−0.007***		0.080***
Total poverty rate gap	(0.002)		(0.003)
Portion due to …	Characteristics	Returns	Characteristics	Returns
Total	0.012***	−0.019***	0.081***	−0.001
	(0.001)	(0.002)	(0.001)	(0.002)
Race	0.000***	−0.009	0.001***	−0.001
	(0.000)	(0.007)	(0.000)	(0.003)
Ethnicity	0.000***	−0.006	0.001***	−0.001
	(0.000)	(0.005)	(0.000)	(0.002)
English	0.001***	−0.000	0.003***	0.004
	(0.000)	(0.009)	(0.000)	(0.008)
Age	−0.008***	0.005	0.028***	0.003
	(0.000)	(0.004)	(0.001)	(0.006)
Education	0.003***	0.004	0.020***	0.002
	(0.000)	(0.004)	(0.001)	(0.004)
Employment	0.011***	0.009	0.014***	−0.000
	(0.000)	(0.008)	(0.001)	(0.002)
Disability	−0.000	−0.005	−0.000	−0.000
	(0.000)	(0.004)	(0.000)	(0.001)
Region	0.000***	0.000	0.001***	0.000
	(0.000)	(0.001)	(0.000)	(0.001)
Degree of ruralness	0.000***	−0.011**	0.001***	−0.003
	(0.000)	(0.005)	(0.000)	(0.006)
Number of children 18+	0.000***	0.001	−0.000	0.000
	(0.000)	(0.001)	(0.000)	(0.001)
Number of other relatives 18+	−0.000	0.000	−0.000	0.000
	(0.000)	(0.000)	(0.000)	(0.000)
Number of non-relatives 18+	−0.000***	−0.001	0.000*	−0.000
	(0.000)	(0.000)	(0.000)	(0.000)
Number of children	0.004***	−0.000	0.013***	0.000
	(0.000)	(0.002)	(0.001)	(0.001)
Year	−0.000***	0.000	−0.001***	0.000
	(0.000)	(0.001)	(0.000)	(0.000)
Different-sex	2,349,733		207,625
Same-sex	12,068		12,068

Notes: Standard errors are in parentheses. ***, **, * denote statistical significance at the 1, 5, and 10 percent levels, respectively.

Source: Authors’ calculations based on ACS 2010–14.

Table 3 shows that the difference in the simple poverty rate of gay male couples and different-sex married couples (which is 2 percentage points higher for the heterosexual couples, as in Table 1, thus the positive sign) is explained predominantly by various characteristics the gay male couples have that protect them from poverty. The “total” row of the table shows that the “portion due to” differences in characteristics between same-sex male couples and married different-sex couples would alone generate a 2.5 percentage point lower poverty rate for men in same-sex couples, which is bigger than the actual difference of 2.0 percentage points. Differences in effects of the predictors of poverty (that is, differences in “returns” in the language of the Oaxaca–Blinder decomposition) push the probability of being poor in the other direction, toward a 0.5 percentage point lower poverty rate for married different-sex couples. In other words, if the characteristics of gay and married heterosexual couples had been the same, the difference in returns to those characteristics would have resulted in gay couples having a higher poverty rate than heterosexual couples. Summing the two effects gives the overall poverty gap of –2.0 percentage points, the top row. Put differently, the differences in characteristics over-determine gay couples’ lower poverty rates.

The largest driver of gay couples’ relative protection from poverty is their employment characteristics; these explain more than 80 percent (0.016/0.020) of the poverty advantage (that is, lower poverty rates) for gay couples over heterosexual married couples. Gay couples’ higher education levels and their lower numbers of children in the household are the other main factors keeping the poverty rates of gay couples low. Their youth is the main characteristic that puts same-sex male couples at higher risk of being in poverty than different-sex married couples.⁸

The returns to the characteristics of same-sex male and different-sex married couples overall differ in a statistically significant way, but none of the individual characteristics are statistically significant.

Comparing gay couples versus different-sex unmarried couples, a somewhat similar story emerges. Again, gay male couples have a lower poverty rate than the unmarried different-sex couples, a 10.7 percentage point difference. More education, having fewer children, and being less likely to be very young are the main drivers of gay couples’ lower poverty rates. In this comparison, though, the overall difference in the returns to the characteristics faced by gay couples and heterosexual unmarried couples is small but statistically significant, and also favors gay couples. No single category of covariates is statistically significant, however.

Overall, the probit models for gay male couples show that while the unconditional poverty rate for gay couples is lower than for heterosexual married couples, those gay male couples face higher vulnerability to poverty when we hold constant the characteristics of the couples (Table 2). The Oaxaca–Blinder decompositions in Table 3 reveal that it is mainly differences in levels of employment, education, and the number of children in the household that keep gay couples out of poverty, compared with heterosexual couples.

The Oaxaca–Blinder decomposition of the difference in poverty rates between same-sex female and different-sex couples is given in Table 4. Compared with married different-sex couples, same-sex female couples have a 0.7 percentage point higher poverty rate, and a 8.0 percentage point lower poverty rate than different-sex unmarried couples.

Starting with the comparison with married heterosexual women, the second row presenting the total decomposition results shows that lesbian couples have characteristics that protect them from being in poverty. The characteristics of same-sex female couples and different-sex married couples alone would generate a 1.2 percentage point lower poverty rate for women in same-sex couples because their characteristics protect from poverty. As was the case for men in same-sex couples, lesbian couples benefit from their higher education levels, higher employment rates, and from having fewer children in the household. On the other hand, lesbian couples’ relative youth appears to hurt them.

However, female same-sex couples get lower returns than heterosexual married couples; that is, those beneficial characteristics have less of a poverty-reducing protective effect, pushing their poverty rate 1.9 percentage points above the rate for heterosexual married couples. Although each individual estimate for the returns of various characteristics presented in Table 4 (except for the degree of ruralness) lacks statistical significance, some of those components are sizable in the context of poverty rates, so as a whole it is these returns that push the lesbian poverty rate higher than that of married heterosexual couples.

Decomposing the poverty rate difference for lesbian couples and different-sex unmarried couples, we see that lesbian couples’ characteristics over-predict their lower poverty rates slightly. Lesbian couples’ race, ethnicity, English proficiency, education, age, employment, region, degree of ruralness, and lower number of children protect them from poverty vis-à-vis unmarried different-sex couples. Compared with different-sex unmarried couples, though, lesbians do not suffer from less positive returns to these beneficial characteristics. Lesbian couples have slightly better returns on their English proficiency, age, education, region, and number of children, but they face more negative returns on their employment status and living in more rural areas, although none of those effects are statistically significant, and they net out to account for a very small contribution to the poverty gap.

Overall, the Oaxaca–Blinder results show that the primary reason that gay and lesbian couples have lower unconditional poverty rates than unmarried different-sex couples – and for gay men, lower than for married different-sex couples – is that on average, the same-sex couples have characteristics that protect them from poverty, such as higher employment rates, more education, and fewer children. In addition, although lesbian couples have a higher poverty rate compared to married heterosexual couples, lesbians’ characteristics alone would have reversed that pattern. In that case, the lesbians’ disadvantage in unconditional poverty rates was explained by the less protective effect, or lower returns, to those factors.

Indeed, while the contribution of differences in returns to the poverty predictors seems small in comparison to the effect of differences in characteristics in Tables 3 and 4, differences in returns explain the patterns that we see in Table 2. The differential returns are negative for female same-sex couples when compared to either married or unmarried different-sex couples, driving the higher poverty risk for lesbians in Table 2. The differential returns are negative for male same-sex couples when compared to marriage couples, explaining gay couples’ higher risk in Table 2, but the differential returns are positive when compared with unmarried different-sex couples. Thus, the decomposition results reconcile why same-sex couples appear to do better than different-sex couples in the unconditional poverty rate gap shown in Table 1 but (mostly) worse in the conditional gap given in Table 2.

Discussion and Conclusions

This paper's analysis finds that, despite a common myth of gay affluence (Badgett 2001), there are many American families headed by same-sex couples with incomes below the poverty line. However, whether there is a relative disadvantage for same-sex couples depends on the analytical perspective. Looking just at simple (unconditional) poverty rates suggests that male same-sex couples are the least likely to be poor, but lesbians are in a somewhat worse position: gay male couples had the lowest poverty rates, followed by married different-sex couples, lesbian couples, and unmarried different-sex couples. In contrast, when we compared couples with the same observable predictors of poverty, we found that lesbian couples were more likely to be poor than either married or unmarried different-sex couples, and gay male couples were more likely to be poor than married different-sex couples (but not more or less likely than unmarried couples). The decompositions revealed that male and female same-sex couples have characteristics that keep their simple poverty rates low, even though they are more likely to be poor after controlling for those characteristics. Here, we discuss the implications of these sets of findings for public policy and future research.

Support for low-income LGBT people: This study makes visible LGBT people who are living in poverty, a group that has rarely been studied. Regardless of whether they are over- or underrepresented, the mere existence of LGBT people living in poverty should raise questions about how well public policies designed to reduce poverty are working for LGBT families. A recent federal study pointed out the absence of evidence on how well income support programs and human services serve the needs of low-income LGBT people (Burwick et al. 2014). Prejudice against LGBT people among social workers, administrative agency staff, or other clients might generate barriers to accessing benefits for low-income LGBT people, as would rules related to adoption or legal relationships that are a poor fit for same-sex couples’ families. In addition, LGBT people might have particular needs or combinations of needs (such as health, housing, legal, and so on) that would be better served by programs designed specifically for LGBT people. New programs that train human services practitioners in cultural competency could reduce barriers. Local, state, and federal agencies could commission research and initiate data collection that would lead to a better understanding of these needs and challenges.

Also, employer bias and discrimination might make it harder for low-income LGBT people to get and keep jobs that allow them to exit poverty. Public policies that ban employment discrimination based on sexual orientation and gender identity would help to level the playing field for LGBT people seeking paid work. Today, only twenty-two out of fifty states include sexual orientation in their nondiscrimination statutes, and only twenty of those include gender identity as well. Federal protection is currently dependent on an agency interpretation of sexual orientation and gender identity discrimination as constituting illegal sex discrimination. Therefore, explicit federal statutory protection and more widespread state coverage could make it easier for low-income LGBT workers to find paid work at a wage that pulls them out of poverty.

Increasing the benefits of characteristics that reduce poverty: The decompositions show that same-sex couples are more likely to be poor in the basic predictive model because they do not get as much protective value, or returns, to characteristics that reduce poverty. The one clear factor that hurts lesbians, in particular, is living in more rural areas, where people may have more negative attitudes toward LGBT people. Other likely suspects include employer decisions that influence the returns to education and the stability of jobs, as well as potential disadvantages of intersectionality for LGBT people of color or those with disabilities. Further research might help identify more precisely why the returns to these characteristics are negative for same-sex couples, and nondiscrimination policies might also play a role in mitigating those disadvantages.

Policies that reduce poverty for all: Even though this paper focused on gaps in poverty rates across sexual orientation, reducing the gap might be less important if policies were to reduce poverty for people in all groups. A series of policy simulations shows that raising the minimum wage, reducing the gender pay gap, and reducing racial pay gaps would reduce poverty for all couple types in the ACS. These policies would reduce the overall poverty gap between different-sex and same-sex couples slightly, but would generate a much larger reduction in the gaps for lesbians and for people of color in same-sex couples (Badgett and Schneebaum 2014, 2015, 2016). Other policies such as childcare and paid family leave programs to remove barriers to parents’ employment could help same-sex as well as different-sex couples. Of course, other poverty reduction strategies might not have such widespread positive effects, so it is important to assess which groups gain to make sure strategies are broadly inclusive.

Finally, we note the need for better data and more research to study parts of the LGBT community not identifiable in the ACS: single LGB people and transgender people. Getting a fuller picture of which parts of the LGBT community are most vulnerable to poverty, and why, will greatly help our ability to answer important questions and to move policy and research forward.

ACKNOWLEDGMENTS

We thank Nancy Folbre for useful comments on an earlier version of this work. Nina Schubert provided excellent research assistance.

SUPPLEMENTAL DATA

Supplemental data for this article can be accessed at https://doi.org/10.1080/13545701.2018.1441533.

Additional information

Notes on contributors

Alyssa Schneebaum

Alyssa Schneebaum holds a PhD in Economics and a graduate certificate in Advanced Feminist Studies from the University of Massachusetts Amherst. She is currently Postdoctoral Researcher at the Vienna University of Economics and Business in the Department of Economics and the “Economics of Inequality” research institute. Her research focuses on gendered economic inequality, particularly in labor markets, in the household, and in educational opportunity and attainment.

M. V. Lee Badgett

M. V. Lee Badgett is Professor of Economics at the University of Massachusetts Amherst and Williams Distinguished Scholar at UCLA's Williams Institute. Her current research focuses on poverty in the LGBT community, employment discrimination against LGBT people, and the economic cost of LGBT exclusion.

Notes

1 Average yearly earnings are US$39,268 for lesbians and only US$23,575 (US$22,387) for heterosexual married (unmarried) women; gay men earn US$50,480 and heterosexual married (unmarried) men earn US$46,745 (US$30,538).

2 A conceptualization of poverty that looks beyond income or expenditure as a measure of well-being is “multidimensional poverty,” which relates traditional income or expenditure poverty to substandard social (economic, political, and civic or cultural) inclusion, economic well-being, and capability (such as educational attainment, health, and self-respect). There are quantitative measures of multidimensional poverty in the literature (Kakwani and Silber 2008; Wagle 2008). None discuss sexual orientation explicitly.

3 We use the nominal incomes of households in the year they were surveyed and compare it with that year's poverty thresholds.

4 The vast majority of dropped households were in the 2010–12 samples. Due to Census Bureau coding and editing changes beginning in 2013, this issue is much less of a concern after that point. Up through 2012, about 25–33 percent of same-sex households were affected. In 2013 and 2014, it was 3–4 percent.

5 In a following discussion, we consider other ways of calculating poverty.

6 While the FPL is updated for inflation each calendar year, the ACS collects information on income received during the past twelve months expressed in contemporary US dollars from respondents who are surveyed throughout the year. Since there are no month-specific adjustment factors available in the ACS, amounts do not reflect calendar year dollars.

7 The county typology codes are based on the status of the county as measured in the 2010 census. The designations are as follows: 1 = County in metropolitan area with a population of at least 1 million; 2 = County in metropolitan area with a population between 250,000 and 1 million; 3 = County in metropolitan area with a population less than 250,000, 4 = Non-metropolitan county with an urban population of 20,000+, adjacent to a metropolitan area; 5 = Non-metropolitan county with urban population of 20,000+, not adjacent to a metropolitan area; 6 = Non-metropolitan county with urban population of 2,500–19,999, adjacent to a metropolitan area; 7 = Non-metropolitan county with urban population of 2,500–19,999, not adjacent to a metropolitan area; 8 = Non-metropolitan county completely rural or less than 2,500 urban population, adjacent to metro area; 9 = Non-metropolitan county completely rural or less than 2,500 urban population, not adjacent to a metropolitan area. The codes we use are a weighted average (based on population) of the Beale Code for each PUMA. Thanks go to Gary Gates for providing the idea on how to measure ruralness.

8 For both male and female same-sex couples, the contribution of the year dummies is negative, serving to increase relative poverty for same-sex couples. We suspect this effect stems from several factors: (1) poverty was falling over the 2010–14 time period; (2) our sample adjustment described earlier might have dropped more low-income people in the earlier years if they were more likely to make errors on the survey (perhaps because of lower levels of education and English competency). It is a fairly small effect and does not change the general finding for either men or women.