Participants were 370 women aged 18–28 years (M_age = 20.54 years; SD_age = 2.28) recruited from a United States university community via an established subject pool, online announcements, and posted flyers. Most identified as White (74%) and non-Hispanic (93%), with some identifying as Asian (16%), Black/African American (8%), or multiracial (2%). They were selected for inclusion in the current analyses from the full sample of 473 women and 221 men. Men were excluded from analyses because the primary research question on the role of ovarian hormone milieu in the link between behavior and cognition only concerned women. Women were excluded from analyses for the following reasons: having a reproductive health or medical condition that could impact ovarian hormone milieu, including past pregnancy, irregular menstrual cycles, the use of hormone-containing medications other than OCs (n = 64), being inattentive during data collection, including sleeping (n = 14) or failing to follow directions on the mental rotations test (n = 19), or being statistical outliers (i.e., three standard deviations from the sample mean) on age (n = 6); outliers bias error terms and age is related to masculinity and especially to femininity (Barrett and White, 2002; Jones et al., 2011).

Women in the analytic sample were grouped according to whether they were naturally cycling (n = 141) or using OCs (n = 229). OC users had been using the same pill for at least 3 months, and NC women had not used any hormonal contraceptive for at least 3 months and had regular menstrual cycles. OC users and NC women did not differ significantly in age, t(368) = −1.36, p = 0.174, or ethnicity, χ²(1) = 0.25, p = 0.615, but there were disproportionately more Asian identified women in the NC (26%) vs. OC group (11%), χ²(4) = 21.02, p < 0.001.

The OCs used by women in the sample represented a wide range of formulations. All included ethinyl estradiol, but progestins ranged from the highly androgenic levonorgestrel (n = 15) and norethindrone acetate (n = 71) to the mildly androgenic norgestimate (n = 71) to the antiandrogenic drospirenone (n = 45), among a variety of others (n = 18). Importantly, however, OCs were not merely grouped according to progestin types, but rather, they were coded according to their estrogenic, progestational, and androgenic activity, which are based on hormone types and doses, particularly their biological (e.g., receptor) actions (Dickey, 2020). Each hormone was considered to have low, low-intermediate, intermediate-high, or high activity in each OC formulation based on assays conducted in rodent and human tissues; estrogenic activity was determined by mouse uterine assay, progestational activity was determined by human endometrial response, and androgenic activity was determined by rat prostate assay (detailed in Dickey, 2020). These levels were converted into a four-point Likert scale to facilitate their inclusion in quantitative analyses (1 = Low, 2 = Low-intermediate; 3 = Intermediate-high, and 4 = High). For example, the commonly used OC Loestrin FE 1/20 (ethinyl estradiol, norethindrone acetate) had activity codes of estrogenic = 2, progestational = 3, and androgenic = 4, the triphasic Ortho Tri-Cyclen (ethinyl estradiol, norgestimate) had activity codes of estrogenic = 4, progestational = 1, and androgenic = 2, and Yaz (ethinyl estradiol, drospirenone) had activity codes of estrogenic = 2, progestational = 4, and androgenic = 1.

Nine OC users had pill types that could not be coded because their specific formulations were not contained in the coding system (Dickey, 2020). For the remaining 220 OC users (which makes this study among the largest—if not the largest—in terms of OC users with defined pharmacokinetic pill properties; see Warren et al., 2014), average estrogenic activity was 2.67 (SD = 1.03), progestational activity was 2.66 (SD = 1.29), and androgenic activity was 2.50 (SD = 1.18), and all activity levels ranged from 1 to 4, suggesting notable variability. Moreover, the activity levels showed both overlap and distinction, with estrogenic and progestational activity correlated at r(218) = −0.64, p < 0.001, estrogenic and androgenic activity at r(218) = −0.36, p < 0.001, and progestational and androgenic activity at r(218) = 0.11, p = 0.114. Thus, estrogenic and progestational activity were most highly related in this sample, but their relation only resulted in 41% overlap between the codes.

Participants came to a university research laboratory for an hour-long test session. During the session, they provided informed consent, described their reproductive history in a brief interview, recorded information about their OC formulation from the pill packet they brought with them (if applicable), and completed a monitored online survey. They were compensated with either course credit or $15.

Participants responded to the questionnaire and completed the two cognitive tests described below as part of this study’s survey. All responses were provided on laboratory computers.

Analyses were conducted in three parts. First, a regression was used to examine the relation between self-perceived masculinity and 3D mental rotations skill for all women. Second, a moderation analysis was used to examine whether the masculinity-mental rotations relation differed between NC women (coded 0) and OC users (coded 1). Third, three separate moderation analyses in only OC users were used to examine whether the masculinity-mental rotations relation varied by pill estrogenic, progestational, or androgenic activity. Age and general cognition were covariates in all analyses, and all predictor variables were centered prior to analyses. Moderations were conducted using the PROCESS macro in SPSS (see Hayes, 2013), with follow-up simple slopes analyses conducted separately for each group (i.e., NC women vs. OC users) or for hormone activity and masculinity scores at the mean plus or minus one standard deviation. Type I error was set at 0.05 for all analyses.

The first inferential analysis examined the link between masculinity and spatial skills in all women. Results of the regression revealed a significant overall model, F(3,366) = 9.44, p < 0.001, R² = 0.07, due to the significant covariate of general cognition, b = 0.34, p < 0.001; age was not a significant covariate, b = 0.14, p = 0.431. Importantly, masculinity was not a significant predictor of 3D mental rotations, b = 0.59, p = 0.331.

The second inferential analysis was a moderation examining whether the link between masculinity and spatial skills differed for NC women and OC users; in other words, it examined whether the lack of an association in the full sample was due in part to women’s hormonal milieu. Results of the regression revealed a significant overall model, F(5, 364) = 6.76, p < 0.001, R² = 0.09. General cognition continued to be a significant covariate, b = 0.33, p < 0.001, and age was not, b = 0.17, p = 0.338. Although there was a significant main effect of masculinity, b = 2.35, p = 0.017 and not NC-OC group, b = 0.17, p = 0.819, these main effects must be considered in the context of the significant interaction between masculinity and group on 3D mental rotations, b = −2.80, p = 0.023. Follow-up simple slopes analyses revealed the expected significant positive association for NC women, b = 2.35, p = 0.017, but not for OC users, b = −0.44, p = 0.558. The nature of the interaction is also shown in the scatterplot in Figure 1. Individual women are shown by the gray circles (NC women) and black squares (OC users), with the zero-order linear relation for each group indicated by the gray (NC women) and black (OC users) lines.

The third inferential analysis was a moderation among only OC users examining whether the link between masculinity and spatial skills depended upon the pharmacokinetic properties of the pills. Similar to the previous moderation, it examined whether the lack of an association among OC users was due in part to the heterogeneity of OCs, specifically their estrogenic, progestational, and androgenic activity. Complete results are listed in Table 2 and plotted in Figure 2, and significant findings are discussed here.

For estrogenic activity (top third of Table 2), the overall model of the relation between masculinity and mental rotations skill was not significant, but the interaction of masculinity and estrogenic activity was significant. Estrogenic activity at one standard deviation below the mean (dotted line in Figure 2A) was positively related to masculinity and 3D mental rotations, b = 1.45, p = 0.201, whereas estrogenic activity one standard deviation above the mean (solid line in Figure 2A) was inversely related to masculinity, b = −1.69, p = 0.094. Thus, the masculinity-mental rotations relation became increasingly positive as levels of OC estrogenic activity decreased.

For progestational activity (middle third of Table 2), the overall model was significant. General cognition was a significant, positive covariate, and the interaction of masculinity and progestational activity was significant, with effects in the opposite direction of estrogenic activity. Progestational activity at one standard deviation below the mean (dotted line in Figure 2B) was inversely related to masculinity and 3D mental rotations, b = −1.66, p = 0.093, whereas progestational activity one standard deviation above the mean (solid line in Figure 2B) was positively related to masculinity, b = 1.52, p = 0.169. Thus, the masculinity-mental rotations relation became increasingly positive as levels of OC progestational activity increased.

For androgenic activity (bottom third of Table 2), the overall model was not significant. The interaction of masculinity and androgenic activity was also not significant, but general cognition was a significant covariate (see Figure 2C).

Sensitivity analyses were conducted to determine whether length of OC use altered these relations; thus, moderations were repeated with length of OC use (in months) as a covariate. The pattern of results, including inferences about the significance of the interactions, did not change (estrogenic activity: b = −1.51, p = 0.039, progestational activity: b = 1.17, p = 0.035, androgenic activity: b = 0.64, p = 0.315).

Across the full sample (i.e., NC women and OC users combined), there was not a significant relation between masculinity and spatial skills. Even though a significant relation was expected according to the sex-role mediation hypothesis (Nash, 1979), the non-significance is not necessarily at odds with the extant literature. Past empirical studies have led to somewhat inconsistent results by statistically controlling for gender and even occasionally finding a stronger relation in men than women (Reilly and Neumann, 2013; Reilly et al., 2016). Importantly, results of moderation analyses qualify that non-significant masculinity-spatial skills relation in revealing that it depends upon ovarian hormonal milieu (marked by NC vs. OC status): Masculinity was significantly and positively related to mental rotations skill for NC women, but not for OC users. The robustness of this interaction effect is buttressed by the fact that NC women and OC users did not differ in mean levels in any study variables, including gender self-concept or mental rotations skill, consistent with past research in heterogenous samples (e.g., Beltz et al., 2015; Nielson and Beltz, 2021). Thus, the sex-role mediation hypothesis may only describe women with some hormonal milieus (e.g., those experiencing the typical fluctuations in endogenous ovarian hormones that characterize the menstrual cycle). As past studies on the sex-role mediation hypothesis did not report the hormonal milieu of study participants (e.g., whether all women were NC), it is possible that varying hormonal milieus contributed to their inconsistent findings. Additionally, varying hormonal milieus may even contribute to the inconsistent findings among men for whom the masculinity-spatial skills relation is not consistently found (e.g., Signorella and Jamison, 1986; Kelly and Beltz, under review), and who experience fluctuations in endogenous sex hormones levels depending upon season of the year, life experiences like fatherhood, and even daily experiences like winning a game (e.g., Zilioli and Watson, 2012; Smith et al., 2013; Grebe et al., 2019).

To directly investigate the extent to which ovarian hormonal milieus may have contributed to the null masculinity-spatial skills relation in OC users, the OCs of all users in the study were coded on a four-point Likert scale for their estrogenic, progestational, and androgenic activity; the codes were based on assays conducted in rodent models and human tissue response, largely reflecting the action of OCs at receptors in the context of other hormones (see Dickey, 2020). Moderation analyses then revealed that the relation between masculinity and 3D mental rotations depended upon both the estrogenic and progestational—but not significantly upon the androgenic—activity in OCs, although effects were clearly small (as seen in Figure 2) and require replication. Specifically, as OC estrogenic activity decreased but as progestational activity increased, there was an increasingly positive association between masculinity and spatial skills. Findings are consistent with the inverse association between OC estrogenic and progestational activity in this sample. Also, findings regarding estrogenic activity are generally consistent with past work among OC users in showing an inverse association between pill estradiol dose and spatial skills (Beltz et al., 2015), and are also consistent with past menstrual cycle studies showing an inverse relation between endogenous estradiol and spatial skills (Hausmann et al., 2000; Courvoisier et al., 2013; Hampson et al., 2014; Griksiene et al., 2018, 2019), as well as with studies that show improvements in spatial skills when estradiol levels are low (Maki et al., 2002; Courvoisier et al., 2013).

Findings regarding progestational activity are novel, as there is hardly any consideration of the effects of the progestins in OCs—beyond their androgenicity—in the extant literature (for an exception and null findings confounded by androgenicity, see Beltz et al., 2015). Furthermore, findings from past research on endogenous progesterone (e.g., from menstrual cycle studies that often have notable methodological limitations, such as small sample size) are mixed (reporting null, positive, and inverse associations; Hausmann et al., 2000; Maki et al., 2002; Griksiene and Ruksenas, 2011; Courvoisier et al., 2013; Hampson et al., 2014; Noreika et al., 2014; Pletzer et al., 2019; Shirazi et al., 2021). There does seem to be a potential indirect role for progesterone in spatial skills through attention, though. Progesterone has been linked to the local vs. global processing of visual stimuli (Pletzer et al., 2014), and similar global-local processing notions have been used to explain gender differences in spatial skills, especially mental rotations (Peña et al., 2008; Pletzer et al., 2014).

Therefore, this research is important not only for the empirical findings it uncovered, but also for introducing and illustrating the utility of a new coding scheme for OCs that disentangles the estrogenic, progestational, and androgenic properties of the pills. Past work has highlighted the importance of considering heterogeneity in ovarian hormonal milieus afforded by different OCs (Beltz and Moser, 2020), but most work has accomplished this by classifying users based on the types of hormones in their pills, such as androgenic or anti-androgenic progestins (e.g., Wharton et al., 2008; Griksiene and Ruksenas, 2011; Gurvich et al., 2020) or according to active ingredients specified by the FDA (e.g., Puts et al., 2010; Beltz et al., 2015, 2019). Some studies have also accomplished this by considering the exogenous dose of one hormone in the pills (e.g., ethinyl estradiol) without simultaneously considering other hormones (e.g., Beltz et al., 2015; Griksiene et al., 2018). The novel four-point coding scheme used in this study may be an improvement upon all of these approaches. It is based on the biological activity of the exogenous hormones in animal tissue, and it does not require the creation of inconsistent group classifications that could result in some participants being excluded from analyses (e.g., if they are using an OC that is not widely represented in the sample). Although this study did not have power to detect higher order interactions, this coding scheme permits the examination of interaction effects between different hormone activities in OCs in future studies with even larger samples (e.g., whether progestational and androgenic activity combine masculinity to influence outcomes).

Because OC androgenic activity did not moderate the masculinity-spatial skills relation, findings from this study may seem inconsistent with past work showing that OCs with highly androgenic progestins facilitate spatial skills, whereas those with anti-androgenic progestins reduce spatial skills (Beltz et al., 2015; Griksiene et al., 2018; Gurvich et al., 2020), and with work showing a positive association between endogenous testosterone and spatial skills (Hausmann et al., 2000; Pletzer et al., 2019). Not all studies find links between progestin androgenicity and spatial skills, though (e.g., Griksiene and Ruksenas, 2011), and no studies have investigated potential neuroendocrine modulation of the relation between masculinity and spatial skills. Moreover, and as noted above, the OC groupings used in past research may conflate androgenic and progestational activity by focusing on groups of OC users determined by progestin androgenicity. It could very well be that past work focused on androgenic properties of pills was actually reflecting (at least to some meaningful degree) progestational activity. For instance, OCs containing the anti-androgenic progestin drospirenone have among the highest progestational activity (Dickey, 2020). Such conflation is less likely in this study because of the small correlation between androgenic and progestational activity.

Study findings and presumed methodological advances should be considered in light of other characteristics of the sample and approach. Regarding the sample, it is likely not representative of all OC users, as most participants were White, from a Midwestern university community, and six women who were deemed statistical age outliers were excluded from analyses. Women who identified as Asian were also disproportionately underrepresented among OC users (compared to NC women). It is vital that future research on ovarian hormone links to the brain and behavior include increasingly diverse and longitudinal samples to best inform women’s health across the lifespan. Hopefully, such research will be facilitated by the advances in OC-related research methods proposed in this and other recent work (reviewed in Beltz and Moser, 2020; Hampson, 2020).

Regarding the study approach, a common, validated, and reliable measure of spatial skills, specifically the Vandenberg and Kuse (1978) 3D mental rotations test, was used. This test shows an established gender difference (Voyer, 2011; Halpern, 2013; Beltz et al., 2020), making it ideal for this study’s goal of detecting gendered links (via masculinity) to mental rotations performance. From these findings, however, it is not clear what aspect of 3D mental rotations skill is linked to masculinity. Although general cognitive ability can be largely discounted (as it was a covariate in all analyses), other possibilities are not easily parsed, including actual mental visualization, strategy use including global vs. local processing, or working memory interference caused by time constraints (see Peters, 2005; Pletzer, 2014; Boone and Hegarty, 2017). It is also not clear if the pattern of findings would be the same if a different mental rotations test was used, particularly a test that shows smaller average gender differences than does the Vandenberg and Kuse (1978) measure; this includes 2D tests and the Shepard and Metzler (1971) 3D test. Thus, future work aimed at replicating and potentially decomposing mental rotations skill and its links to masculinity in the context of varying hormonal milieus is needed.

Also, the focus of this study was on the overarching hormonal milieu afforded by having a natural menstrual cycle or using OCs with particular pharmacokinetic formulations, both of which holistically reflect neuroendocrine function. This leads to some unique considerations for NC women and OC users. Regarding NC women, we did not include menstrual cycle phases in analyses, as methods for determining them are error-ridden without many repeated assessments (Hampson, 2020; Gloe et al., under review). Nonetheless, study findings are broadly consistent with menstrual cycle research on spatial skills, suggesting small roles for estradiol and progesterone. Regarding OC users, we similarly did not assess active vs. placebo pill phase, as placebo phases and adherence to them vary greatly across users. Moreover, we did not assess time of pill ingestion; it is linked to spikes in bloodstream hormone concentrations, but the spikes have unclear implications for concentrations across days, and the temporal association between bloodstream concentrations and central nervous system function is largely unknown (Jusko, 2017; Hampson, 2020). Thus, future research using valid and reliable measures of cycle phase and investigating the pharmacokinetic properties of OCs on neurally mediated processes is sorely needed.

Moreover, this study—like all complex biopsychosocial research on OCs—has some unique considerations. For instance, this study was novel in leveraging a coding scheme for estrogenic, androgenic, and progestational activity of OCs, there are undoubtedly individual differences in the biopsychosocial impacts and correlates of these exogenous hormones, as humans are likely more complex than the animal models upon which the coding scheme is based. Thus, there is a pressing need for future work that considers individual women and their daily OC use across time (e.g., intensive longitudinal studies, such as Weigard et al., 2021; Gloe et al., under review).

Finally, the vast majority of the extant research on the sex-role mediation hypothesis (and on gender, more broadly) has operationalized masculinity in terms of explicitly defined gendered personality characteristics (e.g., independent vs. patient), which are to some extent culturally determined and time-varying (for a meta-analysis, see Donnelly and Twenge, 2017). In the present study and consistent with established research (e.g., Egan and Perry, 2001; Beltz et al., 2021; Nielson and Beltz, 2021; Kelly and Beltz, under review), masculinity was operationalized by gender self-expression, with its meaning (and the meaning of femininity) being implicitly defined by the participant. Gendered personality qualities and self-expressions are certainly related, but they are distinct constructs with distinct outcomes (Eagly and Wood, 2017; Hyde et al., 2019). Thus, future biopsychosocial research will likely benefit from having multidisciplinary teams who grapple with psychosocial measurement alongside biological indices of hormone activity.