Are Sibship Characteristics Predictive of Same Sex Marriage? An Examination of Fraternal Birth Order and Female Fecundity Effects in Population-level Administrative Data from the Netherlands

ABSTRACT

Despite historical increases in the number of individuals engaging in same-sex relations and entering same-sex unions, the causes of sexual orientation remain an open question. Two biological processes that have received some degree of empirical validation are the fraternal birth-order effect (FBOE) and the female-fecundity effect (FFE). Respectively, these processes posit that having a greater number of older brothers and being part of larger sibships independently increase the odds of male homosexuality. Nevertheless, previous studies have relied on suboptimal data and methods, including underpowered and selected samples, and models that fail to fully disentangle the two processes. In addition, they have rarely analyzed samples of women. We address these limitations using high-quality, population-level linked register data from the Netherlands (n = 9,073,496). Applying a novel multivariable approach, we jointly examine the FBOE and FFE by comparing the sibship characteristics of men (n = 26,542) and women (n = 33,534) who entered a same-sex union against those who did not (n = 4,607,785 men and 4,405,635 women). Our analyses yield robust evidence of an FBOE on both male and female homosexuality, but no support for the FFE. Additionally, we find that individuals’ birth order affects the probability of entering a same-sex union, regardless of the sex of older siblings.

Acknowledgments

The authors would like to acknowledge excellent and insightful comments on the research by Prof. Cameron Parsell, Dr. Stefanie Plage, Dr. Ella Kuskoff and Dr. Andrew Clarke, as well as seminar and conference participants at the Melbourne Institute, the Society of Economics of the Household Meeting, the Annual Conference of the European Society for Population Economics, and the Young Demographers Meeting. Authors declare no competing interests. All authors contributed to the study equally. The data used in these analyses were provided by the Dutch national statistical office, Statistics Netherlands. Inquiries regarding the data access should be addressed to [email protected]. A replication package containing the code underlying our empirical analysis can be accessed at https://github.com/jankabatek/replication_FBOE_FFE

Disclosure Statement

The authors declare no competing interests for this study.

Supplementary Material

Supplemental data for this article can be accessed on the publisher’s website.

Notes

¹ We are cognizant of competing views on the appropriateness of the term ‘homosexual’, including the American Psychological Association’s stance against its use (American Psychological Association, 2020). In this paper, we use this term over ‘gay/lesbian’ – or other variations – to be consistent with the language used in previous studies of the FBOE (see e.g., Blanchard & Bogaert, 1996; Blanchard et al., 2020; Bogaert, 2006) and FFE (see e.g., Camperio-Ciani et al., 2004; Iemmola & Camperio Ciani, 2009). As such, our use of the term ‘homosexuality’ does not entail any value judgment about different sexualities, or the individuals who identify with them.

² Khovanova’s (2020) method for separating the FBOE from the FFE entails the calculation of three parameters: (i) p₁₁, the probability that the only boy in a one-son family is homosexual; (ii) p₁₂, the probability that the first son in a two-son family is homosexual; and (iii) p₂₂, the probability that the second son in a two-son family is homosexual. The FBOE occurs when p₂₂ > p₁₂, while the FFE occurs when p₁₂ > p₁₁. The existence of both the FBOE and the FFE is confirmed if p₂₂ > p₁₂ > p₁₁.

³ In fact, estimates of the FBOE in Blanchard’s (2018a) meta-analysis showed substantial variability, with 75% of such variability being driven by differences in the samples rather than by sampling error (Blanchard, 2018a). This indicates that sample-selection bias in these studies is more than likely, echoing concerns by Zietsch (2018) about the external validity of this pool of studies.

⁴ For example, Camperio-Ciani et al. (2004) recruited 98 homosexual men from three homosexual community groups and discotheques, while Gómez Jiménez et al. (2020b) recruited 115 transgender and 112 cisgender androphilic males through referrals among the Mexican muxe community.

⁵ This study was nevertheless critiqued by Blanchard (2007) and Blanchard and VanderLaan (2015) on three grounds: its focus on same-sex unions rather than self-identification as homosexual, measurement error in the identification of siblings, and suboptimal modeling choices. While our study also focused on same-sex unions, we were able to overcome the other two issues through the use of maternal identifiers to identify siblings and the introduction of a novel estimation method.

⁶ The birth year cutoffs were chosen based on a visual inspection of the distribution of birth years among individuals who entered a same-sex union (see Figure S1). The results are not meaningfully sensitive to the choice of cutoff years. As a robustness check, we also estimated a model using paternal siblings, yielding comparable results.

⁷ The introduction of registered partnerships for same-sex couples in the Netherlands took place in January 1998, and it was followed by the legalization of same-sex marriage in April 2001. We observe 18,155 individuals who entered same-sex partnerships and 41,921 who entered same-sex marriages. For 51,494 individuals, a same-sex union was the first formally recognized union, whereas 8,582 individuals entered the same-sex union after the dissolution of a previous different-sex union.

⁸ A cautionary note is due here. While we use same-sex marriage as a proxy for homosexuality in the manner implied in the FBOE and FFE literatures (i.e., experiencing same-sex attractions), we do not assume that all individuals in same-sex unions self-identify as ‘homosexual’ (or ‘gay/lesbian’). Some of these individuals may instead identify as ‘bisexual’, for example. Nevertheless, we expect the proportion to be quite low, as an overwhelming majority of partnered bisexual people are in different-sex rather than same-sex relations (Pew Research Center, 2019). This approach is also consistent with previous literature on the FBOE and FFE, which has tended to use ‘homosexuality’ as an umbrella term encompassing both homosexual and bisexual people (see e.g., Blanchard et al., 2006).

⁹ Birth-cohort fixed effects are crucial control variables in our model. This is because they capture the consequences of progressive fecundity decline (as documented in Figure S2) and any other dynamic social processes that have changed family structures over historical time. Given that the incidence of homosexuality increases over time while the average sibship size falls, failure to account for people’s birth cohorts would result in a spurious negative correlation between the two variables, thus invalidating any empirical test of the FFE. Besides using birth-cohort fixed effects, we can also account for this spurious correlation by matching the homosexual and heterosexual individuals on their birth years. We do this as a sensitivity check in the Supplementary Online Appendix.

¹⁰ The interpretation of the logistic-regression coefficients as measures of excess probabilities of same-sex union entry is only approximative. However, as explained in the results section, this approximation poses little risk in our application.

¹¹ A more powerful test of the FFE can be performed by averaging the excess probabilities associated with adding one younger sister and one younger brother to an existing sibship $\left(\frac{\left[{\beta }_1+\left({\beta }_1+{\beta }_4\right)\right]}{2}\right)$, and testing this composite coefficient against zero. Of note, averaging does not affect the statistical power of this test. An alternative approach would be to estimate a model specification that does not explicitly control for the number of younger brothers. In such a model, the coefficient β₁ would represent the influence of adding a younger sibling to an existing sibship, irrespective of their sex. We decided against using this model as our baseline specification because we wanted to assess whether or not the sex of younger siblings also influences the odds of same-sex union entry.

¹² This slight imbalance is consistent with the human sex ratio (Chahnazarian, 1988).

¹³ The need for measures of statistical uncertainty in population-level analyses – and their interpretation – has been the subject of debate. Detractors argue that these measures are intended for analyses of samples from an underlying population and uncertainty becomes irrelevant when sampling-frame coverage reaches 100% (Desbiens, 2007). Advocates argue that population-level confidence intervals and p-values express the chances of the observed differences manifesting in a super-population, which may consist of the same population observed under different states of the world, or the population of countries with comparable characteristics (Gelman, 2005; Graubard & Korn, 2002). Here, we follow the second approach.

¹⁴ The population shares of individuals who entered a same-sex union conditional on their sex, number of younger siblings, and number of younger brothers are shown in Figure S3. Unlike Figure 2, Figure S3 does not show any discernible pattern with regard to siblings’ sex.

¹⁵ Both the logistic-regression coefficients and the odds ratios approximate the same underlying concept: the relative increase in the probability of same-sex union entry associated with a one-unit increase in the variable of interest, all else being equal. To assess which measure is subject to a lesser approximation error, we produced direct estimates of excess probabilities of same-sex union entry (population-level average marginal effects), and compared them with the excess probabilities corresponding to the two approximate measures. Table S1 shows that the logistic-regression coefficients approximate population-level average marginal effects better than the odds ratios, with the values of excess probabilities being virtually identical at the presented level of numerical precision. That is why we base our discussion primarily on the regression coefficient estimates, and we include odds ratios mainly to facilitate the comparisons with earlier FBOE and FFE studies.

¹⁶ The same is true for the composite coefficient estimate averaging the influences of having one younger sister and one younger brother $\left(\frac{\left[{\beta }_1+\left({\beta }_1+{\beta }_4\right)\right]}{2}=-0.132,p<0.001\right).$

¹⁷ The estimates quantifying the influence of having an older brother (as opposed to a younger brother) can be obtained by summing together the coefficient estimates of β₂ and β₃ and subtracting the coefficient estimate of β₄. For men, this is 0.098 + 0.115 – 0.016 = 0.197. We note that this estimate is akin to the FBOE estimate recently used by Blanchard & Lippa (2021).

¹⁸ We note that the conventional FBOE estimate (0.066) can be fully reconstituted from our baseline coefficients (see Column 1 in Table 3). It combines the 12.5% increase in the probability of entering a same-sex union associated with having an older brother (instead of an older sister) with the 7.9% increase associated with moving one place down the birth order, and the 13.8% decrease associated with adding one sibling to the existing sibship: 0.125+0.079–0.138 = 0.066.

Additional information

Funding

This research was partially supported by the Australian Research Council Centre of Excellence for Children and Families over the Life Course (project number CE140100027) and an Australian Research Council Discovery Early Career Researcher Award for a project titled ‘Sexual Orientation and Life Chances in Contemporary Australia’ (2017–2020).