Women’s Sexual Orientation and Occupational Tasks: Partners, Prejudice, and Motherhood

RAQUEL CARRASCO ANA NUEVO-CHIQUERO

Documento de Trabajo 2025/01

Febrero de 2025

fedea

Las opiniones recogidas en este documento son las de sus autores y no coinciden necesariamente con las de Fedea.

Orientación sexual y tareas profesionales de las mujeres: el papel de la maternidad,

Raquel Carrasco (UCM) y Ana- Nuevo-Chiquero (UAM)

Este estudio analiza cómo influye la orientación sexual en las elecciones ocupacionales de las mujeres, especialmente en relación con los tipos de tareas desempeñadas en el puesto de trabajo. Investigaciones previas han mostrado que las mujeres homosexuales tienden a percibir salarios más altos que las heterosexuales, mientras que en el caso de los hombres ocurre lo contrario. Una posible explicación es que las parejas del mismo sexo gestionan de manera diferente el equilibrio entre responsabilidades domésticas y laborales. Sin embargo, este estudio va más allá de la cuestión salarial y examina si las mujeres homosexuales y heterosexuales eligen ocupaciones con distintos niveles de cualificación. Para ello, clasifica las tareas laborales en abstractas, rutinarias y manuales y explora si las diferencias pueden estar relacionadas con las expectativas de fertilidad y con el impacto de posibles interrupciones en la carrera profesional.

Utilizando datos de la American Community Survey (2010-2018) para EE.UU., el estudio analiza si las mujeres homosexuales tienen mayor tendencia a desempeñar ocupaciones en las que las habilidades se deprecian más rápidamente con el tiempo. En un contexto en el que los avances tecnológicos han transformado los mercados laborales, las ocupaciones que requieren altos niveles de habilidades abstractas y analíticas suelen experimentar un mayor crecimiento salarial, mientras que aquellas centradas en tareas rutinarias tienden a estar peor remuneradas. No obstante, estas últimas suelen ser más compatibles con interrupciones laborales, como las derivadas de la maternidad.

Los resultados muestran que las mujeres homosexuales son más propensas a desempeñar empleos con un alto contenido de habilidades abstractas y manuales, evitando en mayor medida ocupaciones rutinarias. Estas diferencias persisten incluso tras tener en cuenta factores como la educación, las características de la pareja, la discriminación en el entorno laboral y la maternidad. Curiosamente, las mujeres homosexuales parecen inclinarse por estas trayectorias profesionales desde una edad temprana, mientras que las diferencias en la elección ocupacional se acentúan aún más a partir de los 25 años, una etapa en la que muchas mujeres ya han formado una familia. Dado que los trabajos rutinarios suelen facilitar interrupciones en la trayectoria profesional, es posible que las mujeres heterosexuales —quienes tienen una mayor probabilidad de anticipar la maternidad— se orienten hacia ellos, mientras que las mujeres homosexuales, con expectativas de menor o nula interrupción, opten por carreras que exigen un desarrollo continuo de habilidades.

Esta investigación sugiere que las expectativas de fertilidad desempeñan un papel clave en la trayectoria profesional de las mujeres, incluso antes de que éstas tomen decisiones concretas sobre planificación familiar. Aquellas que prevén ser madres pueden inclinarse por profesiones más flexibles, aunque con menores requisitos de cualificación y salarios más bajos. En cambio, quienes no esperan tener hijos con tanta probabilidad tienden a elegir carreras altamente cualificadas y mejor remuneradas, que demandan aprendizaje continuo pero ofrecen menos oportunidades de interrupción. Estos resultados cuestionan la idea tradicional de que las diferencias en la trayectoria profesional de mujeres homosexuales y heterosexuales se deben principalmente a la especialización en el hogar.

Desde una perspectiva política, el estudio resalta la importancia de fomentar el desarrollo profesional de las mujeres a través de intervenciones tempranas. Muchas decisiones clave sobre educación y trayectoria laboral se toman en la adolescencia, por lo que podrían ser beneficiosas políticas que expongan a las jóvenes a una mayor diversidad de opciones profesionales y les proporcionen modelos de referencia en campos no “tradicionales”. Además, medidas en el ámbito laboral, como la flexibilidad en los horarios, los permisos parentales y el acceso a servicios de guardería asequibles, podrían contribuir a reducir las interrupciones de la carrera profesional asociadas a la maternidad, permitiendo que más mujeres -independientemente de su orientación sexual- permanezcan en empleos altamente cualificados.

Raquel Carrasco^†

Ana Nuevo-Chiquero^‡

February 5, 2025

Abstract

This paper examines diferences in occupational task content among women based on their sexual orientation. Using data from the American Community Survey, we find that women in same-sex couples are more likely to be employed in occupations characterized by more abstract and manual tasks, and fewer routine components. These occupations are traditionally associated with greater flexibility, accommodating career interruptions, and minimizing skill depreciation. These diferences are not explained by individual or partner characteristics or by prejudice at the occupational level. Furthermore, our findings hold even after controlling for self-selection into the labor force. Heterogeneous efects by age and parental status suggest that these choices reflect long-term strategies rather than shortterm responses to childbearing. This points to a complex relationship between occupational choice and fertility, influenced by the probability of labour force exit and re-entry.

JEL classification: J15, J16, J71

Keywords: sexual orientation, occupation, intended fertility, gender wage gap

^∗We thank Antonia Diaz for her involvement in the early stages of this project, and Juanjo Dolado, Jen nifer Graves, Ricardo Mora, and Erik Plug for useful comments. We also thank seminar attendees at Universidad Autónoma de Madrid and University Loyola Andalucía. We gratefully acknowledge the support from MINECO/FEDER grant PID2022-13890OB-100, PID2022-141898OA-I00, CAM - EPUC3M11 (V PRICIT) and CEX2021-001181-M. The usual disclaimer applies.

^†Department of Economic Analysis and Quatitative Economics & ICAE, Universidad Complutense de Madrid, and FEDEA. E-mail: raquel.carrasco@ucm.es

^‡Corresponding author. Universidad Autónoma de Madrid, University of Edinburgh, and IZA. E-mail: ana.nuevo@uam.es. Address: Dpto. Análisis Económico: Teoría Económica e Historia Económica, Facultad de Ciencias Económicas y Empresariales. C/ Francisco Tomás y Valiente, 5, 28049 Madrid, Spain

1 Introduction

Homosexual women typically earn higher wages than heterosexual women, while the opposite is true for men (see, for instance, Jepsen and Jepsen (2022), among many others). Same-sex couples difer from their heterosexual counterparts in various ways, including specialization in household and market production, as highlighted by Orefice (2011) or Giddings et al. (2014).¹ This paper analyzes diferences in women’s occupational choices based on sexual orientation. We focus on the task content of occupations, categorized into abstract, routine, and manual components. We explore whether these diferences persist after controlling for potential drivers of occupational selection, including women’s and their partners’ characteristics, motherhood, workplace prejudice, and self-selection into the labor force.

Understanding these occupational diferences is important, as task content significantly influ ences wages and overall labor market outcomes. Recent research highlights how technological advances have changed skill requirements and reshaped wage structures across sectors (Ace moglu and Autor, 2011; Michaels et al., 2014). For example, occupations requiring higher levels of abstract and analytical skills have experienced stronger wage growth than routine tasks vul nerable to automation (Goos et al., 2014). Thus, analyzing variations in occupational task content across demographic groups, including sexual orientation, provides insights into how these groups experience diferent economic outcomes.

In addition to task content, fertility, both expected and realized, plays a crucial role in shaping women’s career decisions. Motherhood often leads to labor market interruptions and a shift toward more flexible but lower-paid jobs (Kleven et al., 2019; Budig and England, 2001). Researchers such as Polachek (1981) and Gronau (1988) have highlighted how anticipated career breaks afect occupational choices. Women may choose occupations that minimize the career costs of raising children, based on their fertility expectations. Skill accumulation stops and existing skills may depreciate during periods of labour force absence. As a result, women who expect to remain childless, potentially more common among homosexual women, are likely to choose occupations where the opportunity costs of raising children are high. This suggests a link between fertility expectations, occupational choices, and the risk of skill depreciation. Comparing women of diferent sexual orientations can thus ofer valuable insights into these dynamics.

¹For a recent and comprehensive review of the economic literature on sexual orientation, see Badgett et al. (2024).

Using data from the American Community Survey for the period 2010-2018, we assess whether homosexual women choose occupations with higher skill depreciation compared to heterosexual women and explore the underlying drivers of any observed diferences. We use the skill content of occupations as a predictor of skill depreciation, classifying occupations in terms of their skill requirements along the dimensions outlined by Autor and Dorn (2013): analytical, routine, or manual. Occupations with predominantly abstract tasks are more prone to changes in requirements than those with routine tasks. Manual occupations fall somewhere in between. Focusing on skills rather than occupations allows us to disentangle the efects of preferences, prejudices, or social norms, which may be more closely tied to occupational categories than skill content. We control for a range of factors, such as educational attainment, motherhood status, and partner characteristics, that may influence occupational choice by sexual orientation (Baumle, 2009). Additionally, we account for workplace prejudice, which previous research has identified as a key factor afecting the career trajectories of marginalized groups (Plug and Berkhout, 2004; Pager, 2007), and we address self-selection into the labor force based on unobserved preferences, skills, and constraints.

Our findings show that homosexual women are more likely to hold occupations with higher abstract and manual content and lower routine scores. These diferences persist after controlling for education, partner characteristics, motherhood status, workplace prejudice at the occupa tional level, and even when we control for self-selection into the labour force. Specifically, our results indicate that homosexual women tend to choose occupations with lower routine scores from a young age. The routine component has traditionally been associated with occupations that allow for career interruptions and with a lower skill depreciation. Diferences in abstract and manual components become more pronounced for women aged 25 and over when fertility is more likely to have been realized. This pattern suggests a complex relationship between oc cupational choices and fertility, influenced by labour market exit and re-entry as children grow older.

The paper is organized as follows. Section 2 reviews the relevant literature. Section 3 describes the data. Section 4 formulates the empirical strategy and discusses the results. Section 5 concludes.

2 Related literature

The economic literature has extensively documented disparities in the labour market based on sexual orientation. Empirical research has primarily focused on the relationship between sexual orientation and inequalities in earnings and income. Studies have found that homosexual men typically earn less than their heterosexual counterparts in similar positions, while lesbians tend to earn more than heterosexual women. For example, Badgett (1995) provides evidence that in the US, gay men earn 11% to 27% less than heterosexual men, even after accounting for education and occupation. Lesbians also experience earnings diferentials of 12% to 30%, though this gap diminishes when considering occupational selection bias. Antecol et al. (2008) found that lesbian women generally earn more than heterosexual women, regardless of marital status, while gay men earn less than married heterosexual men but more than cohabiting heterosexual men. These wage diferences are primarily attributed to variations in education levels.

Similar patterns have been observed in countries such as Canada (Carpenter, 2008), Australia (Carpenter, 2008), the UK (Aksoy et al., 2018), the Netherlands (Plug and Berkhout, 2004), Greece (Drydakis, 2011), and elsewhere. Temporal trends, as examined by Jepsen and Jepsen (2022), indicate limited improvement in wage parity for men in same-sex couples relative to married men in diferent-sex unions. For women in same-sex relationships, evidence of convergence with married women in diferent-sex unions is mixed.

Another strand of the literature has examined the relationship between sexual orientation and occupational segregation, finding distinct patterns for gay and lesbian workers. Gay men are more likely to work in female-dominated occupations, while lesbian women tend to occupy higher-ranked positions in less female-dominated fields compared to heterosexual women. This trend is highlighted in studies by Black et al. (2007), Tebaldi and Elmslie (2006), and Anteco et al. (2008). Plug et al. (2014) further investigate whether gay and lesbian workers avoid prejudiced occupations. Their study examines the relationship between sexual orientation, sexual prejudice, and occupational segregation, accounting for selectivity efects that often complicate studies on discrimination. Their findings suggest that gay and lesbian workers prefer occupa tions with lower levels of prejudice, aligning with Becker’s model of employer and employee discrimination. This preference reduces their exposure to prejudiced colleagues and influences the degree of prejudice-based occupational segregation.

Regarding labor supply diferences between same-sex and diferent-sex couples, Antecol and

Steinberger (2013) find that within-couple income gaps are smaller in lesbian couples. Lesbian primary earners’ labor supply resembles that of heterosexual men, while secondary earners labor supply is similar to that of married heterosexual women. They note that the presence of children largely explains the labor supply diferences between secondary-earner lesbian partners and married heterosexual women. Giddings et al. (2014) observe that same-sex couples are less likely to show high levels of specialization, even when controlling for the presence of children, though this gap decreases across cohorts. Orefice (2011) show that gay and lesbian households respond to shifts in bargaining power similarly to heterosexual households, with labor supply influenced by factors such as age and diferences in non-labor income between partners.

The impact of parenthood on specialization is well-documented for heterosexual couples, where the presence of children often leads to increased specialization (Angelov et al., 2016). Without traditional gender norms, factors such as economic opportunities and individual preferences play a more important role in determining specialization. Previous studies have suggested that biological factors also influence childcare responsibilities within lesbian couples, with the biological mother typically dedicating more time to the child she gave birth to (Andresen and Nix, 2022). Nonetheless, the non-biological mother still spends more time with the child compared to heterosexual fathers. Overall, lesbian couples tend to share childcare responsibilities more equally compared to heterosexual couples (Badgett, 2003). Additionally, Goldberg (2010) found that lesbian parents often adopt a more egalitarian approach to childcare and household duties, further reinforcing the trend towards shared responsibilities.

In addition to the literature on sexual orientation, our research builds on the extensive literature examining occupational characteristics and their impact on labor market outcomes. Occupational selection significantly contributes to the gender gap, with skill intensity and depreciation being key factors in understanding these diferences (Gronau, 1988; Polachek, 1981). More recently, the task content of occupations has emerged as a critical determinant of employment and earnings, influenced by skill-biased technical change (e.g., Acemoglu and Autor (2011), Autor and Dorn (2013) or Cortes (2016)).²

²Another strand of the literature has emphasized the role of interpersonal or social skills for women in the labour market (see, for instance, Ngai and Petrongolo (2017) or Cortes et al. (2023)).

3 Data

The primary data used for this analysis is the American Community Survey (ACS), which will be complemented by the General Social Survey (GSS) and the Current Population Survey (CPS). The ACS, conducted annually by the US Census Bureau, collects demographic and socioeconomic information from a representative sample of the US population, providing timely data between censuses. We use data from waves 2010 to 2018. Identifying respondents’ sexual orientation is crucial for our analysis. Since the survey did not explicitly ask about sexual orientation until 2018, we infer this information from the marital or cohabiting status of household members. Our sample is then restricted to heads of household in couples and their partners. We will refer to “homosexual” as those women in same-sex couples and “heterosexual” as those in opposite-sex couples.³

Additionally, the ACS provides extensive background and labour market information, including age, education, marital status, state of residence, and the presence of children in the household. Due to the household nature of the ACS, this information is also recorded for partners. We use this information to categorize our sample into ‘main earners’ and ‘secondary earners’. In same-sex partnerships, both individuals appear as subjects of interest, whereas in heterosexual couples, only the woman is included. Thus, this classification allows us to examine an additional dimension of heterogeneity.⁴

Occupations reported in the ACS are matched with their task content compiled by Autor and Dorn (2013), who categorize tasks as abstract, routine, or manual. Each occupation is scored on these dimensions, ranging from 0 to 10. Notice that an occupation can have high scores in multiple task types; for example, surgeons have both high abstract and manual scores.

Prejudice against homosexual workers can also be an important driver of selection into diferent occupations. To account for this source of occupational segregation, we construct a measure of prejudice at the occupational level, according to Plug et al. (2014). We use data from the GSS, a comprehensive annual survey in the US that captures attitudes towards homosexuality. This includes respondents’ views on the morality of homosexuality and their stance on rights such as marriage and public roles for homosexuals.⁵ By aggregating these responses at the occupational level, we account for varying levels of prejudice across diferent job categories. Figure A1 presents the distribution of the share of individuals in the occupation that consider homosexuals should not marry by the main task of the occupation. While our analysis covers approximately 99.4% of sampled occupations with available prejudice data from the GSS, some occupational categories lack complete matching due to data limitations compared to the ACS.

³Our sample may be biased if homosexual and heterosexual women partner at diferent rates, as we exclude non-partnered individuals. The survey’s diferent levels of reporting spouses or partners also pose a potentia issue. However, individuals not acting upon their sexual orientation should not afect our estimation, provided their labor market choices align with their reported sexual orientation. Previous literature on sexual orientation using this dataset has recognized this issue (see, for instance, Antecol et al. (2008); Antecol and Steinberger (2013))

⁴We classify a woman as the main earner if she is the only one in the labour force or if her income is the highest in the couple when both individuals report wage income.

Finally, to address individual’s self-selection in the labor force, we construct an instrumental variable using data from the CPS March Supplement. This instrumental variable, akin to the “Bartik style” instrument, aggregates labor force participation across each state and year, encompassing four education groups (high school dropout, high school graduate, some college, college graduate or higher), and three age groups (under 24, 24 to 39, 40 to 55 years old). Due to data constraints in smaller states, not all education and age groups may be covered by this instrument, prompting us to limit our sample accordingly to ensure a robust analysis of participating women.

Table 1 presents descriptive statistics for our full ACS sample and various estimation samples, focusing on women in both heterosexual and homosexual couples. Approximately 1.5% of our sample identifies as being in a homosexual couple. Compared to women in heterosexual couples, these women have fewer children and generally have higher levels of education. While their income tends to be higher, total household income, reflective of a two-female household, tends to be lower. As expected, they exhibit higher rates of employment and lower rates of non participation in the labour force, and they are significantly more likely to be the main earner of the household (50% by construction compared to 30% of women in heterosexual couples). The bottom part of Table 1 details the occupational composition of women in the labour force. While diferences by sexual orientation are modest, they align with our expectations: women in homosexual couples tend to have occupations with a slightly higher abstract and manual component and a slightly lower routine component. Table A1 in the Appendix presents the descriptive statistics for the rest of the controls included in the analysis.

⁵http://gss.norc.org/About-The-GSS, last accessed 1 November 2021.

Table 1: Descriptive statistics

Note: Column (1) includes all women aged 18 to 55. Columns (2) and (3) restrict the sample to women in couples, with sexual orientation determined by the partner’s gender. Columns (5) and (6) further restrict the sample to women currently in the labor force. Columns (1) to (3) and (5) to (6) present the mean and standard deviation (in parentheses). Columns (4) and (7) display the diferences between the preceding two columns, along with the standard errors of these diferences.

4 Empirical strategy and results

4.1 Baseline model

We begin our analysis by examining the relationship between the task content of occupations and the sexual orientation of women, as defined by the gender of their reported partner. Using a sample of women in the labor force, we estimate the following model using Ordinary Least Squares (OLS):

\[T a s k _ {i} ^ {j} = \alpha^ {j} + \beta^ {j} {\bf 1} (\mathrm{homosexual} _ {i}) + X _ {i} ^ {\prime} \gamma^ {j} + v _ {i} ^ {j},\tag{1}\]

where j = {abstract, routine, manual} represents the task content scores for the occupation held by women i, and 1(homosexual ) is a binary variable that takes the value 1 if the woman is in a homosexual couple and 0 if she is in a heterosexual couple. includes individual controls added sequentially, and are the error terms. We run this regression separately for each score.⁶ These scores are not mutually exclusive, and occupations might score high in more than one category. Therefore, a positive correlation between individual characteristics and one particular score does not need to be compensated by a negative correlation with other scores.

Table 2 presents our baseline results for women in the labour force. Columns (1), (4), and (7) do not include any controls, while Columns (2), (5), and (8) include controls for age, race, state, and year of observation. Finally, Columns (3), (6), and (9) add controls for women’s education.

The top panel shows a consistent positive relationship between being in a same-sex couple and abstract and manual tasks, while the relationship with routine task scores is negative. These efects remain highly significant even after controlling for covariates such as age, race, education, and including state and year dummies. Given that our sample is restricted to women in couples, and only one woman per couple is included in the heterosexual subsample, but both women are included in each same-sex couple, the middle panel of Table 2 includes a control for women identified as the main earner in the household. The relationships reported in the top panel persist even after controlling for this additional variable.

Finally, the bottom panel further explores the efect of being in a same-sex couple by distinguishing between primary and secondary earners within the household. Our findings indicate that women in same-sex couples who are the primary earners tend to have higher abstract scores than women who are primary earners in heterosexual couples. Secondary earners in same-sex couples also show a positive relationship, though to a lesser extent. For example, when all controls are included, being the primary earner in a same-sex couple increases abstract scores by 0.217 points, which is a relevant increase as it accounts for 6% of the unconditional mean of abstract test scores. For secondary earners in same-sex couples, the abstract score increases by 0.155 points with respect to secondary earners in heterosexual couples. Similar results are obtained for routine and manual scores, with an impact of around 6% of the unconditional mean for each score, even for manual tasks, which have a substantially lower average in a sample of women. In these cases, the efects are not significantly diferent between primary and secondary earners in same-sex couples.

⁶We recognize that simultaneous equation estimation can improve the eficiency of the estimators. However, for simplicity, we have chosen to estimate the equations separately as our benchmark approach. This allows us to more efectively address potential self-selection issues in the labor force, which would complicate the estimation of the system.

Table 2: Task content of occupations and sexual orientation – OLS estimates for women in the labour force. Note: Sample restricted to heads of household in couples and their partners. Task scores range from 0 to 10. Estimating sample includes only women in the labour force. Robust standard errors clustered at the state level are reported in parentheses. * denotes significance at 1%, ** at 5% and *** at 1%.

Therefore, women in homosexual couples hold occupations with more abstract and manual task content and less routine content compared to those held by women in heterosexual couples, even after controlling for individual characteristics and accounting for the fact that we observe two women per homosexual couple. Notably, occupations with higher routine task scores tend to experience lower skill depreciation over time. Next, we examine whether these task content diferences help explain labor market outcomes within our sample of women in the labor force. Specifically, we proceed with estimating by OLS the following wage equation:

\[l n (w a g e _ {i}) = \phi_ {0} + \phi_ {1} {\bf 1} (\mathrm{homosexual} _ {i}) + X _ {i} ^ {\prime} \gamma + \sum_ {j = 1} ^ {3} \psi_ {j} t a s k _ {i} ^ {j} + u _ {i},\tag{2}\]

where ln(wage ) is the natural logarithm of the hourly wage for women i, are normally distributed error terms, and the rest of the terms are as defined above.

Table 3 presents these results. The odd-numbered columns control for the same variables as the last column of Table 2, while the even-numbered columns add abstract, routine, and manual scores at the occupation level as additional controls for the diferent specifications.

The estimates reflect the well-known lesbian wage premium in the labor market, even after controlling for individual characteristics. Column (5) shows evidence of the unequal impact of this lesbian premium within the couple. Specifically, there is a wage premium of 8.2% for women in same-sex couples who are the main earners. For secondary earners, we do not find a significant diference between homosexual and heterosexual women. Adding the scores for abstract, routine, and manual tasks reduces this gap by nearly 10%. Moreover, the returns to the diferent scores are as expected: the payof for abstract scores is higher than for routine scores, which in turn is higher than for manual scores. For example, in our more parsimonious specification with controls (Column (6)), a 1-point increase in the abstract score is associated with a 14.5% wage increase. Similarly, a 1-point increase in the routine score is associated with a 10.5% wage increase, and a 1-point increase in the manual score is associated with a 6.7% wage increase.⁷ Therefore, the behaviour observed in Table 2 is relevant for the labor income of women participating in the labor market and accounts to some extent for the gender orientation wage gap.

Table 3: Wage diferentials and task content of occupations (OLS)

Note: Sample restricted to heads of household in couples and their partners. Wages are measured as ln(hourly wage), and task scores range from 0 to 10. Estimating sample includes only women in the labour force. Robust standard errors clustered at the state level are reported in parentheses. * denotes significance at 1%, ** at 5% and *** at 1%.

⁷If we allow the payof of the scores to vary by sexual orientation (Table A2 in the appendix), we observe a lower payof for women in homosexual couples across all three scores.

4.2 Potential drivers: household and market factors

We now explore the potential factors driving the diferences in the occupational task content between women in homosexual and heterosexual couples. First, we examine the role of household characteristics, including those related to the partner and children. Diferences in labour market outcomes by gender might suggest that homosexual and heterosexual women have diferent expectations regarding their partner’s labour market outcomes and income, even before they are matched. As a result, they are likely to have considered these diferences in their decisionmaking. We include controls for the partner’s education (four categories) and total income to account for this. Columns (1), (4), and (7) of Table 4 present these results, mapping the results presented earlier. Compared to the last columns of Table 2, the point estimates for the diferences in abstract and routine scores between homosexual and heterosexual couples are slightly smaller, while for manual scores they are slightly larger; however, the impact remains highly significant. For example, being the main earner in a same-sex couple increases the task content of the occupations by 0.168 points, which accounts for 4.4% of the average of the abstract score. These results suggest that while women may take these partner diferences into account, they are not the main factors explaining the patterns observed.

Table 4: Task content of occupations and homosexuality – OLS estimates including family characteristics and prejudice.

Note: Sample restricted to heads of household in couples and their partners Task scores range from 0 to 10 Estimating sample includes only women in the labour force. Prejudice is proxied by the share of individuals in a given occupation that consider homosexuals should not be allowed to marry. Robust standard errors clustered at the state level are reported in parentheses. * denotes significance at 1%, ** at 5% and *** at 1%.

Another key household variable that might afect occupation and task selection is the pres ence of children. Homosexual and heterosexual women might anticipate diferences in their fertility, consistent with the lower fertility observed in the data. In our sample of working women, those in heterosexual couples have on average 1.25 children, while homosexual women have 0.58 children. Columns (2), (5), and (8) include controls for the presence of children in the household, and for the number of children above and below the age of 5.⁸ Surprisingly, diferences in the number and age of children present in the household do not significantly alter the relationship between task scores for women in homosexual and heterosexual couples. One plausible explanation would be that both homosexual and heterosexual women have been able to anticipate these diferences correctly, and thus, the actual presence of children does not alter their labour market choices. Alternatively, the impact of children might be mediated through labor force participation, an issue discussed in the following section.

Finally, Columns (3), (6), and (9) of Table 4 include controls for prejudice at the occupation level. Prejudice is potentially a key factor influencing diferences in occupational choices by sexual orientation. Research by Plug et al. (2014) and others highlight prejudice as a major driver of selection and wage gaps by sexual orientation. The ACS does not report experiences of prejudiced or the share of prejudiced individuals in specific occupations. Therefore, we use data from the GSS, where individuals are asked about their opinions on homosexuality and their occupations. We compute the share of individuals prejudiced against homosexuals in each occupation and include this as a control in our main specification. It is important to note that we can include controls at the occupation level because our outcome of interest is the task composition within the occupations, not the occupations themselves. Similar to partner and family composition characteristics, the estimated diferences between women in homosex ual and heterosexual couples remain highly significant even after controlling for prejudice at the occupational level. In summary, when controlling for partner, children, and prejudice as additional covariates, women in same-sex couples have an abstract score in their occupation that is 0.171 points higher for main earners and 0.107 points higher for secondary earners than their heterosexual counterparts. Similarly, they have a routine score of 0.249 points lower and a manual score of 0.080 points higher. These results indicate that controlling for these family and prejudice characteristics does not close the gap in the task content of the occupations.

⁸As mentioned above, the ACS does not contain information on the biological parentage of the children in the household.

4.3 The role of self-selection in labour force participation

Econometric specification

We examine the influence of labour force participation on estimating the main efect of interest. There is a significant diference in labour force participation rates between women in homosexual and heterosexual couples. Over 25% of women in heterosexual couples report being out of the labour force, compared to only 12.5% of women in homosexual couples. Moreover, there may be a potential self-selection problem, as our sample includes only women who choose to work, possibly difering in unmeasured ways from those who do not.

We specify a selection equation for labour market entry as follows:

\[P _ {i} ^ {*} = \delta_ {0} + \delta_ {1} {\bf 1} (\mathrm{homosexual} _ {i}) + X _ {i} ^ {\prime} \psi + Z _ {i} ^ {\prime} \alpha + \varepsilon_ {i},\tag{3}\]

where represents the latent utility for woman i to enter the labor market, and is a vector of factors influencing this decision, excluded from the outcome equation. We use as exclusion restriction the state-education-age group level of female employment. is assumed to be jointly normally distributed with the unmeasured characteristics in the task equation,

To clarify the nature of the selection problem, it is important to note that it does not arise from women in homosexual couples being more likely to enter the labour market, but because these two groups might be diferent along unobservables (captured by . For instance, if , homosexual women with low values of the rest of the observed components will appear in our sample more frequently than similar heterosexual women, that would need a higher value of the error term to participate in the labour force. The range of the error term for women that we observe in the labour force would therefore be diferent for heterosexual and homosexual women. This results in a correlation between the homosexual dummy and unmeasured variables in the selected equation, potentially leading to an underestimation of the efect of homosexuality if the unmeasured variables lead to higher task scores.

Heckman (1979) shows that this selection bias problem is equivalent to an omitted variable bias problem in Equation 1. He shows that the problem can be solved by including the inverse Mill’s ratio in Equation 1 as an additional regressor, , where and are the pdf and cfd of a normal distribution, respectively.

We estimate the model in two steps: first, a probit model predicts the probability of sample selection; second, the outcome equation is estimated using OLS, incorporating the inverse Mills ratio to correct for sample selection bias. This approach is computationally simpler than the joint estimation of Equations 1 and 3 by maximum likelihood, and is widely used for linear outcome equations.

In this model, the coeficient does not necessarily coincide with the magnitude of interest. Our focus is on the average marginal efect of homosexuality on the observed sample of participating women, which is:

\[E (T a s k _ {i} ^ {j} \mid P _ {i} ^ {*} > 0, \mathbf {1} (\mathrm{homosexual} _ {i}) = 1) - E (T a s k _ {i} ^ {j} \mid P _ {i} ^ {*} > 0, \mathbf {1} (\mathrm{homosexual} _ {i}) = 0).\tag{4}\]

This marginal efect has two components: the direct efect, represented by , and an indirect efect, since 1(homosexual) also appears in the selection equation. Therefore, a change in 1(homosexual) not only changes the mean of but also afects the probability of an observation being included in the sample.

Results

Table 5 presents our results with all controls after implementing the two-stage Heckman correction as described above. We report the average marginal efect of each variable. Each column in this table can be compared to the last column for each score in Table 4, as all controls are included in the correction. The parameter estimates from Heckman’s model are detailed in Table A3 in the Appendix. In all cases, the estimates for the coeficient of the correction term, λ(·), are highly significant. Furthermore, Table A4 in the Appendix shows the relevance of the instrument in the first stage.

Selection into the labor force reduces the diferences in task scores by sexual orientation, but the diferences remain significant. The coeficient on abstract score decreases from 0.197 to 0.162 when the selection issue is considered (top panel), and from 0.141 to 0.119 when we include controls for whether the individual is the main earner in the household (mid panel). This indicates that including this control yields similar results with and without correcting for female labour force participation. Allowing the relationship to vary by main earner status reduces the coeficient for the secondary earner from 0.107 to 0.095 and for the main earner from 0.171 to 0.143, which are significantly diferent from each other at the level. Therefore, while labour force participation may be a driver of these diferences, the correction suggests that the overall efect on occupational choice is robust to women’s self-selection into the labour force.

Similarly, the relationship between being in a homosexual couple and the routine content of the occupation remains robust to correcting for selection in labour force participation. The efect decreases from -0.252 to -0.175 (top panel) and from -0.258 to -0.189 when controlling for the main earner (mid panel). The bottom panel shows that for the main earner, the efect decreases from -0.249 to -0.223, and for the secondary earner, from -0.277 to -0.158. Despite these significant reductions of up to 40%, the efect remains significant. Additionally, the diferences in manual scores decrease from 0.074 to 0.041 for the main earner and from 0.080 to 0.048 for the secondary earner, yet they remain statistically significant.

Table 5: Task content of occupations and homosexuality. Marginal efects from Heckman’s selection model.

Note: Robust standard errors clustered at the state level are reported in parentheses. * denotes significance at 10%, ** at 5% and *** at 1%. Task scores range from 0 to 10. Mid panel includes main earner controls, while the top panel does not. Heckman’s correction for labour force participation was estimated in two stages, with the female employment rate at the state/education/age group used as the exclusion restriction. The sample size increases by over 1 million observations as all women are included, regardless of their labor force status

Therefore, selection into the labour force driven by unobservable characteristics influences the occupational scores by sexual orientation. Despite this, notable unexplained diferences persist between these groups. To shed light on their causes, we examine heterogeneity across subgroups by fertility outcomes and age. It is crucial to emphasize that our previous analyses control for partner characteristics and income, thereby ruling out expected diferences attributable to partners’ gender-based income disparities.

4.4 Heterogeneous efects: children and age

The previous results control for a wide range of individual and household characteristics, including the presence of children in the household. Nevertheless, it is still plausible that the number of children modulates these diferences between homosexual and heterosexual couples, and the gap in task content is diferent by realized fertility. To investigate this issue, we divide our sample into three groups: women in households without children, households with children under 5 years old, and households with children over 5 years old. We then analyze whether the efect of homosexuality difers across these groups.

Table 6 presents these results, accounting for self-selection into the labor force. Regarding abstract scores, we find that homosexual women choose occupations with higher scores than heterosexual women across all groups, including those with young children. It is particu larly noteworthy that the diference is more pronounced for women with older children, which suggests that these women have to take the additional costs associated with children directly into account. There are also diferences between homosexual and heterosexual women, albeit somewhat smaller, for both heads of household and partners.

The reduction in the routine score persists even when comparing women of diferent sexual orientations who are childless, have small children or children over 5. Hence, if homosexual women opt for occupations with lower routine scores because of lower fertility expectations, they must be doing so with the expectation of lower fertility, they likely maintain this choice path once fertility is realized, indicating a degree of path dependence in occupational selection. Although we cannot control for the exact number and ages of children, the consistent presence of the efect across diferent groups suggests that fertility, if at all, is playing a role via expectations.

We next examine the heterogeneity in diferences between women in homosexual and heterosexual couples across diferent age groups. It is important to note that age is strongly correlated with the cohort, given that our data spans over 10 years. Therefore, the results should be interpreted with caution. Table 7 shows that the diference in abstract scores, once individual and household characteristics well as self-selection into labour force participation is taken into account, is primarily driven by older women. Women under 25 do not appear to behave differently, and those aged 25-39 show much smaller diferences than those aged 40-55. When we allow the efect to vary by household head status, women in homosexual couples tend to report occupations with lower abstract scores.

On the other hand, the lower incidence of routine tasks in occupations held by women in homosexual couples is evident for all age groups, although it also increases with age. Therefore, the choice driving this diference may be made early in their careers. Similarly, small diferences in the manual content of occupations also appear over the lifetime.

Table 6: Task content of occupations and homosexuality. Marginal efects from Heckman’s selection model by number of children.

See notes Table 5.

The heterogeneous efects by children and age therefore suggest that the higher presence of abstract tasks and the lower presence of routine tasks in occupations held by women in homosexual couples may be a long-term choice rather than a response to immediate circumstances such as childbearing. While we cannot rule out the possibility that this is driven by other unobservable occupational characteristics correlated with task distribution (or by preferences), at least part of the observed diference might be driven by choices made early in life to accommodate future diferences in fertility.

Table 7: Task content of occupations and homosexuality. Marginal efects from Heckman’s selection model by age

See notes Table 5.

5 Conclusions

This paper examines diferences in the skill composition of occupations between homosexual and heterosexual women using data from the American Community Survey. Our analysis re veals that women in homosexual relationships are more likely to hold occupations requiring a higher degree of abstract and/or manual skills, even after accounting for other well-known dif ferences by sexual orientation, such as education and partner characteristics. We also control for occupational-level prejudice against homosexuals by exploiting the variation in prejudice within occupations of similar skill levels. We interpret these occupational diferences as being driven by fertility expectations, with homosexual women generally anticipating lower fertility than heterosexual women. This perspective contrasts with the idea that diferences in occupational choices are primarily due to household specialization.

These diferences persist even when we control for selection into the labor force using a Bartik-style instrument. Homosexual women tend to hold occupations with lower routine scores from a young age, but diferences in other skill scores become more pronounced for women aged 25 and over when fertility is more likely to have occurred. This suggests a complex relationship between occupational choice and fertility, influenced by periods of exiting and re-entering the labor force as children grow older.

Understanding the role of occupational choice in women’s labor market outcomes and wages has been a long-standing topic in economic literature. Our study adds to this discussion by high lighting how varying fertility expectations and realizations among diferent groups of women influence the skill composition of their occupations. Specifically, we find that homosexual women, who generally have lower expected fertility, are more likely to choose occupations requiring higher levels of abstract and manual skills. This finding contrasts with heterosexual women, whose occupational choices might be influenced more by anticipated breaks in their careers for child-rearing. This insight is crucial because it shows that fertility expectations can shape career paths even before any actual fertility decisions are made. Women anticipating fewer or no children might invest more in careers that demand continuous skill development and ofer higher wages, whereas women expecting to have children might opt for occupations that ofer greater flexibility but potentially lower wages and skill requirements.

Moreover, the timing of policies aimed at addressing gender gaps in the labor market is crucial. Educational and occupational decisions are often made early in a woman’s career, sometimes even during adolescence. Therefore, policies targeting these decisions from the outset are more likely to be efective. For instance, educational interventions can expose girls to a variety of careers and provide them with role models in non-traditional fields. Additionally, workplace policies that create flexible work arrangements, parental leave, and afordable child care can mitigate the career disruptions often associated with childbearing and rearing. This support can enable women to stay in or return to high-skill occupations more easily.

References

1. Acemoglu, Daron and David Autor (2011) “Skills, tasks and technologies: Implications for employment and earnings,” in Handbook of Labor Economics, Vol. 4: Elsevier, pp. 1043– 1171.
2. Aksoy, Cevat G, Christopher S Carpenter, and Jef Frank (2018) “Sexual orientation and earnings: new evidence from the United Kingdom,” ILR Review, Vol. 71, No. 1, pp. 242–272.
3. Andresen, Martin Eckhof and Emily Nix (2022) “What causes the child penalty? Evidence from adopting and same-sex couples,” Journal of Labor Economics, Vol. 40, No. 4, pp. 971–1004.
4. Angelov, Nikolay, Per Johansson, and Erica Lindahl (2016) “Parenthood and the gender gap in pay,” Journal of Labor Economics, Vol. 34, No. 3, pp. 545–579.
5. Antecol, Heather, Anneke Jong, and Michael Steinberger (2008) “The sexual orientation wage gap: The role of occupational sorting and human capital,” ILR Review, Vol. 61, No. 4, pp. 518–543.
6. Antecol, Heather and Michael D Steinberger (2013) “Labor supply diferences between married heterosexual women and partnered lesbians: A semi-parametric decomposition approach,” Economic Inquiry, Vol. 51, No. 1, pp. 783–805.
7. Autor, David H and David Dorn (2013) “The growth of low-skill service jobs and the polarization of the US labor market,” American Economic Review, Vol. 103, No. 5, pp. 1553–97.
8. Badgett, MV Lee (1995) “The wage efects of sexual orientation discrimination,” ILR Review, Vol. 48, No. 4, pp. 726–739.
9. (2003) Money, myths, and change: The economic lives of lesbians and gay men: University of Chicago Press.
10. Badgett, MV Lee, Christopher S Carpenter, Maxine J Lee, and Dario Sansone (2024) “A review of the economics of sexual orientation and gender identity,” Journal of Economic Literature, Vol. 62, No. 3, pp. 948–994.
11. Baumle, Amanda K (2009) “The cost of parenthood: Unraveling the efects of sexual orientation and gender on income,” Social Science Quarterly, Vol. 90, No. 4, pp. 983–1002.
12. Black, Dan A, Seth G Sanders, and Lowell J Taylor (2007) “The economics of lesbian and gay families,” Journal of Economic Perspectives, Vol. 21, No. 2, pp. 53–70.
13. Budig, Michelle J and Paula England (2001) “The wage penalty for motherhood,” American Sociological Review, Vol. 66, No. 2, pp. 204–225.
14. Carpenter, Christopher (2008) “Sexual orientation, income, and non-pecuniary economic outcomes: New evidence from young lesbians in Australia,” Review of Economics of the Household, Vol. 6, pp. 391–408.
15. Carpenter, Christopher S (2008) “Sexual orientation, work, and income in Canada,” Canadian Journal of Economics/Revue Canadienne d’Economique ^´ , Vol. 41, No. 4, pp. 1239–1261.
16. Cortes, Guido Matias (2016) “Where have the middle-wage workers gone? A study of polarization using panel data,” Journal of Labor Economics, Vol. 34, No. 1, pp. 63–105.
17. Cortes, Guido Matias, Nir Jaimovich, and Henry E Siu (2023) “The growing importance of social tasks in high-paying occupations: Implications for sorting,” Journal of Human Resources, Vol. 58, No. 5, pp. 1429–1451.
18. Drydakis, Nick (2011) “Women’s sexual orientation and labor market outcomes in Greece,” Feminist Economics, Vol. 17, No. 1, pp. 89–117.
19. Giddings, Lisa, John M Nunley, Alyssa Schneebaum, and Joachim Zietz (2014) “Birth cohort and the specialization gap between same-sex and diferent-sex couples,” Demography, Vol. 51, No. 2, pp. 509–534.
20. Goldberg, Abbie E (2010) Lesbian and gay parents and their children: Research on the family life cycle.: American Psychological Association.
21. Goos, Maarten, Alan Manning, and Anna Salomons (2014) “Explaining job polarization: Routine-biased technological change and ofshoring,” American Economic Review, Vol. 104, No. 8, pp. 2509–2526.
22. Gronau, Reuben (1988) “Sex-related wage diferentials and women’s interrupted labor careers-The chicken or the egg,” Journal of Labor Economics, Vol. 6, No. 3, pp. 277–301.
23. Heckman, James J (1979) “Sample selection bias as a specification error,” Econometrica: Journal of the Econometric Society, pp. 153–161.
24. Jepsen, Christopher and Lisa Jepsen (2022) “Convergence over time or not? US wages by sexual orientation, 2000–2019,” Labour Economics, Vol. 74, p. 102086.
25. Kleven, Henrik, Camille Landais, and Jakob Egholt Søgaard (2019) “Children and gender inequality: Evidence from Denmark,” American Economic Journal: Applied Economics, Vol. 11, No. 4, pp. 181–209.
26. Michaels, Guy, Ashwini Natraj, and John Van Reenen (2014) “Has ICT polarized skill demand? Evidence from eleven countries over twenty-five years,” Review of Economics and Statistics, Vol. 96, No. 1, pp. 60–77.
27. Ngai, L Rachel and Barbara Petrongolo (2017) “Gender gaps and the rise of the service economy,” American Economic Journal: Macroeconomics, Vol. 9, No. 4, pp. 1–44.
28. Orefice, Sonia (2011) “Sexual orientation and household decision making.: Same-sex couples balance of power and labor supply choices.,” Labour Economics, Vol. 18, No. 2, pp. 145–158.
29. Pager, Devah (2007) “The use of field experiments for studies of employment discrimination: Contributions, critiques, and directions for the future,” The Annals of the American Academy of Political and Social Science, Vol. 609, No. 1, pp. 104–133.
30. Plug, Erik and Peter Berkhout (2004) “Efects of sexual preferences on earnings in the Netherlands,” Journal of Population Economics, Vol. 17, pp. 117–131.
31. Plug, Erik, Dinand Webbink, and Nick Martin (2014) “Sexual orientation, prejudice, and seg regation,” Journal of Labor Economics, Vol. 32, No. 1, pp. 123–159.
32. Polachek, Solomon William (1981) “Occupational self-selection: A human capital approach to sex diferences in occupational structure,” The Review of Economics and Statistics, pp. 60–69.
33. Tebaldi, Edinaldo and Bruce Elmslie (2006) “Sexual orientation and labour supply,” Applied Economics, Vol. 38, No. 5, pp. 549–562.

Appendix

A.1 Distribution of prejudice variables

Figure A1: Distribution of prejudice by the main task of the occupation

A.2 Descriptive statistics - additional variables

Table A1: Descriptive statistics – additional variables

A.3 Additional results - wage equation

Table A2: Wage diferentials and task content of occupations

Note: Robust standard errors clustered at the state level are reported in parentheses. * denotes significance at 10%, ** at 5% and *** at 1%.

A.4 Additional results - Heckman’s correction for labor force participation

Table A3: Heckman’s correction coeficients - baseline results (two-step)

Note: Robust standard errors clustered at the state level are reported in parentheses. * denotes significance at 10%, ** at 5% and *** at 1%.

Table A4: Heckman’s correction – first stage

Note: Robust standard errors clustered at the state level are reported in parentheses. * denotes significance at 10%, ** at 5% and *** at 1%.

Table A5: Heckman’s correction coeficients- by number of children in the household

Note: Robust standard errors clustered at the state level are reported in parentheses. * denotes significance at 10%, ** at 5% and *** at 1%.

Table A6: Heckman’s correction coeficients – by age Note: Robust standard errors clustered at the state level are reported in parentheses. * denotes significance at 10%, ** at 5% and *** at 1%.