Childhood gender-typed behaviour, sexual orientation, childhood abuse and post-traumatic stress disorder: a prospective birth-cohort study

Abstract ALSPAC birth-cohort data were analysed to assess prospective associations between childhood gender nonconformity (CGN), childhood/adolescent abuse, and adulthood PTSD symptoms. Structural equation models assessed whether abuse mediated the relationship between CGN and PTSD. Sex and sexual orientation differences were investigated. For females, higher parent-rated CGN at 30, 42 and 57-months was associated with mother-reported abuse, self-reported physical/psychological abuse, and/or self-reported sexual abuse. Higher CGN at 30-months was associated with more PTSD symptoms at 23 years. Self-rated CGN in males and females, and parent-rated CGN in males, were not associated with abuse or PTSD. Sexual minority identification was associated with higher CGN and abuse and for females, PTSD symptoms. In females, the relationship between greater CGN at 30-months and PTSD symptoms was separately mediated by each abuse variable. Self-reported sexual abuse was no longer a significant mediator after sexual orientation adjustment. Self-reported physical/psychological abuse significantly mediated the association alone when it was entered together with mother-reported abuse, even after sexual orientation adjustment. In conclusion, childhood gender nonconformity in females may increase the risk for adult PTSD symptoms, possibly mediated by childhood abuse. In females, mediation of the relationship between CGN and PTSD by sexual abuse may be particularly relevant for sexual minority individuals.

Gender nonconformity; childhood abuse; posttraumatic stress disorder; sexual orientation; prospective Childhood gender nonconformity and PTSD Childhood gender nonconformity (CGN) refers to the behaviour exhibited by children who do not follow the social norms expected of their assigned sex at birth. CGN has been found to be associated with post-traumatic stress disorder (PTSD) or trauma symptoms in some (e.g. D' Augelli et al., 2006;Roberts et al., 2012aRoberts et al., , 2012b, but not all (D'Augelli et al., 2002) retrospective or cross-sectional research.

Gender nonconformity and sexual orientation differences in abuse experiences
Individuals with a history of abuse in childhood have a much higher prevalence of PTSD than those without (19.8% vs. 2.1%; Cromer & Sachs-Ericsson, 2006). Recalled or current CGN has been found to be associated with reports of abuse in childhood and adolescence in retrospective and cross-sectional designs (e.g. Bos et al., 2019;Roberts et al., 2012a;Roberts et al., 2013). The only previous study using a prospective longitudinal design found that both males and females displaying more CGN had similarly higher odds of experiencing parental physical/emotional maltreatment in childhood (Xu et al., 2020). However, this study only used assessments of abuse that occurred before 6 years, which was perpetrated by the mother or her partner and was reported by the mother.
CGN is associated with an increased likelihood of identifying as a minority sexual orientation in young adulthood (Li et al., 2017). Minority sexual orientation has also been found to be associated with an increased risk of childhood abuse (Xu et al., 2020) and PTSD (Roberts et al., 2012b).

Childhood gender nonconformity, abuse and PTSD
It has been hypothesised that abuse may account for the associations between CGN and PTSD or trauma symptoms. Only one study has directly investigated this in a general population sample. Roberts et al. (2012a) found that increased exposure to retrospectively reported childhood abuse partially explained the increased risk of PTSD in those who retrospectively reported being the most gender nonconforming in childhood; this was still the case after adjusting for current sexual orientation. However, the study relied on retrospective self-reports of both CGN and childhood abuse. Furthermore, assessing causal relationships between these variables would require respecting their causal ordering, for example by fitting a mediation model, which this study did not do.

The present study
Previous research on associations between CGN, abuse and PTSD has relied on retrospective or crosssectional methods, both of which have methodological limitations. Additionally, few of the previous longitudinal CGN studies have investigated sex differences in associations; something which previous longitudinal research has identified as important, (Warren et al., 2019). Therefore, longitudinal studies are needed to assess the associations between CGN, abuse and PTSD, as well as sex differences in these relationships, using more appropriate statistical methods to assess the casual chain between these variables.

Study aims and hypotheses
The present birth-cohort study investigated prospectively whether higher levels of CGN, derived from parent-rated (2-5 years) and self-rated (8 years) measures of gender-typed behaviours, were associated with mother-and self-reported abuse in childhood/adolescence, and with self-reported symptoms of PTSD at 23 years. The study also investigated whether abuse mediated any association found between gender-typed childhood behaviour and later PTSD symptoms. Sexual orientation in adolescence was included as a covariate in the mediation model. The effects of interactions between sex and CGN were also investigated.
It was hypothesised that: 1. Higher CGN would be associated with greater abuse in childhood/adolescence. 2. Higher CGN would be associated with more selfreported PTSD symptoms at 23 years. 3. The relationship between CGN and later PTSD symptoms would be significantly mediated via abuse. 4. Minority sexual orientation at 15 years would be associated with greater abuse and PTSD. The mediation effects of abuse on the relationship between CGN and PTSD symptoms would hold when the models were adjusted for sexual orientation at 15 years.

Sample
The sample was taken from a longitudinal birth-cohort study, the Avon Longitudinal Study of Parents and Children (ALSPAC). Pregnant women resident in and around the city of Bristol in Avon, UK, with expected dates of delivery from 1st April 1991 to 31st December 1992 were invited to take part in the study. The initial number of pregnancies enrolled was 14,541 (for these at least one questionnaire has been returned or a "Children in Focus" clinic had been attended by 19/07/ 1999). Of these initial pregnancies, there was a total of 14,676 foetuses, resulting in 14,062 live births and 13,988 children who were alive at 1 year of age. Following further recruitment, the total numbers increased to 15,454 pregnancies and 15,589 foetuses. Of these, 14,901 children were alive at 1 year and took part in any data collection after 7 years of age Fraser et al., 2013). More details can be found on the study website, which includes a fully searchable data dictionary and variable search tool (http://www.bristol.ac.uk/alspac/researchers/our-data/). There were 7,148 females and 7,536 males who were alive at 1 year and who had data for at least one of the study variables. Most of the participants were white (95%) and most mothers were educated to O-level or above (30% below O-level, 35% O-level, and 35% Alevel or above). The present analyses were based on subsets of these participants. The number of participants varied between analyses due to missing data.

Ethical approval
Ethical approval for the study was obtained from the ALSPAC Ethics and Law Committee (see http://www. aslpac.bris.ac.uk) and for the analysis of secondary data by .

Measures
Parent-rated gender-typed behaviour Parents completed the Pre-school Activities Inventory (PSAI;  at 30, 42 and 57 months to rate their child's gender-typed behaviours (test-retest reliability of .66 for females and .62 for males). Scores were standardised according to the measure's instructions. To derive a measure in which both sexes could be analysed together, at each time point standardised Z-scores were calculated separately for males and females, and for males, these scores were subtracted from 1. Higher scores reflected more gender nonconforming behaviour.
Child-rated gender-typed behaviour At 8.5 years, children completed the Childhood Activities Inventory (CAI), a shortened 16-item version of the PSAI (split-half reliability .64 in both males and females; Golombok et al., 2008). Scores were standardised as above.

PTSD symptoms
At 23 years participants were asked whether they had ever experienced each of six 'stressful life events' (Weathers et al., 2013). Participants who answered 'yes' to any item completed the PTSD Checklist for the Diagnostic and Statistical Manual of Mental Disorders À5 (PCL-5 for DSM-5; Weathers et al., 2013). This measure has strong reliability (a ¼ .94) and convergent validity (r ¼ .74 to .85;Blevins et al., 2015).
Mother-reported abuse Mothers reported on their child's experiences of being 'physically hurt by someone' or 'sexually abused by someone' (yes/no) and on whether 'you/your partner were emotionally cruel' or 'you/your partner were physically cruel to your child' (yes/no) at various time-points between 8 months and 11 years.
Self-reported abuse Abuse experiences before 17 years were self-reported retrospectively at 22 years. Physical and psychological abuse experienced by an adult family member was assessed using eight items adapted from the Childhood Trauma Questionnaire . Sexual abuse experienced by any adult or older child was assessed using two items adapted from the Sexual Experiences Survey (Koss & Gidycz, 1985; see Appendix 1).
Deriving latent variables measuring abuse Latent variables were generated separately for the selfreported and mother-reported abuse items. For the four mother-reported abuse items, Exploratory Factor Analysis (EFA) suggested a single factor solution. Confirmatory factor analysis (CFA) achieved an adequate fit (RMSEA .05, CFI .97, TLI .90). Standardised factor loadings ranged from .27 to .67. For the 20 self-reported abuse items, EFA suggested a two-factor solution: 'self-reported psychological/physical abuse' (good fit; RMSEA .08, CFI .99, TLI .95; standardised factor loadings .59 to .80) and 'selfreported sexual abuse'. The latter demonstrated a very poor fit (RMSEA .45, CFI .68, TLI .04), likely due to the small number of positive responses (0.5-7.6%). Therefore, a binary variable was created indicating any vs no self-reported experience of sexual abuse.

Covariates
Sexual orientation At 15.5 years, participants were asked which description 'best fits how you think about yourself'; these were categorised into (Heterosexual (those responding '100% heterosexual') versus Non-heterosexual (all other responses); Jones et al., 2017).

Demographic characteristics
At 32 weeks gestation, mothers reported their highest educational level ('below O-level' v 'O-level' v 'A-level or above') and their child's ethnicity (white v non-white).

Statistical analyses
Analysis was performed using Stata version 15.0 (StataCorp LLC, College Station, TX). Preliminary correlations and t-tests were used to assess the bivariate relationships between key variables. Linear regression models were fitted using structural equation modelling (SEM) to assess whether gender-typed behaviour and sex (IVs) were associated with PTSD symptoms at 23 years (DV) and with self-reported physical/psychological abuse and mother-reported abuse (DVs). Standardised coefficients were reported to allow comparison between models throughout. Binary logistic regression models were fit to assess whether gender-typed behaviour and sex (IVs) were associated with self-rated sexual abuse (DV). Linear regression models also assessed whether each type of abuse report was associated with PTSD symptoms. Finally, models assessed whether sexual orientation was associated with gender-typed behaviour, abuse and PTSD. These models were all adjusted for the mother's education and the child's ethnicity. The same models were fitted a second time with a sex by gender-typed behaviour interaction term (or sex by sexual orientation in the sexual orientation models) to assess whether the relationships were different for males and females. Models were run with robust standard errors to address heteroskedasticity.

Missing data
Preliminary correlations were assessed using complete-case analysis. Full information maximum likelihood (FIML) was used to help account for missing data (see Appendix 2) in further analysis.

Mediation analysis
Mediation analysis was used to assess whether any associations found in the initial linear regressions between childhood gender-typed behaviour (predictor) and later PTSD (outcome) might be explained by abuse (mediator). With the exception of the univariate mediation model for self-reported sexual abuse (for which the 'paramed' programme in Stata was utilised to accommodate the binary mediator; Liu et al., 2014) all mediation models were fitted using SEM and robust standard errors to address heteroskedasticity. Standardised coefficients were used to allow model comparisons. 1,000 bootstrap replications were run to derive percentile 95% confidence intervals for the indirect, direct and total effects (Hayes & Rockwood, 2017). Self-reported sexual abuse was modelled as a single observed variable, while mother-reported abuse and self-reported physical/psychological abuse were modelled as latent variables.
Initially, separate univariate mediator models were run, each with a single abuse mediator to assess whether each type of abuse report mediated a significant proportion of the relationship between gendertyped behaviour and later PTSD. These were each run again adjusting for sexual orientation. Abuse measures that remained significant mediators were subsequently included together in a parallel mediation model. This model was fitted allowing for covariance between the two mediators and run again adjusting for sexual orientation.

Sample characteristics
Sample characteristics are shown in Appendix 3. Females were less likely to report being heterosexual than males (86.3% vs. 91.3%) and had higher PTSD scores.

Preliminary analyses
For associations between demographic variables, abuse, PTSD and CGN, for the whole sample and for males and females separately, see Appendix 4.
Associations between gender-typed behaviour, sex, abuse, and PTSD symptoms Gender-typed behaviour, sex and abuse Self-rated gender-typed behaviour was not significantly associated with mother-or self-reported abuse. More nonconforming parent-rated gender-typed behaviour at 30, 42 and 57 months and male sex were significantly associated with more mother-reported abuse (see Appendix 5). The addition of a sex by gender-typed behaviour interaction term revealed that the relationships between gender-typed behaviour and mother-reported abuse were significant for females only at all time points (see Table 1). There was a significant association between more nonconforming parent-rated gender-typed behaviour at 30 and 42 months and self-rated physical/psychological abuse for females only (see Table 1).
More nonconforming parent-rated gender-typed behaviour at 30, 42 and 57 months and female sex were significantly associated with more self-reported sexual abuse. The addition of a sex by gender-typed behaviour interaction term revealed that these relationships were significant for females only (see Table  1). These effect sizes were small.

Gender-typed behaviour, sex and PTSD symptoms
More nonconforming parent-rated gender-typed behaviour at 30 months (but not at 42 or 57 months) and female sex were significantly associated with more PTSD symptoms at 23 years (see Appendix 5). The addition of a sex by gender-typed behaviour interaction term revealed that the relationship between parent-rated gender-typed behaviour at 30 months and PTSD symptoms were significant for females only (see Table 1). Self-ratings of gendertyped behaviour were not associated with PTSD symptoms.

Associations between sexual orientation and abuse, PTSD symptoms and gender nonconformity
Regression models demonstrated that sexual minority identification was associated with higher PTSD scores, mother-reported abuse, self-reported physical/psychological abuse, self-reported sexual abuse and with more nonconforming parent-rated and self-rated gender-typed behaviour at all time points (see Appendices 6.1 and 6.2).
The addition of a sex by sexual orientation interaction term demonstrated that the association of sexual minority identification with higher PTSD scores and with self-rated physical/psychological abuse was significant for females only (see Table 2). However, the association between sexual minority identification and mother-reported abuse was significant for males only. For sexual abuse, the association with sexual minority identification was significant for both males and females (see Table 2). Minority sexual orientation was associated with parent-rated gender-typed behaviour for both males and females, while the relationship with self-rated gender-typed behaviour was significant for males only (see Table 3).

Discussion
Associations between gender-typed behaviour, abuse and PTSD for females For females, there was a small but significant association between more parent-rated masculine-typed .60 .57 .38 .85 .69 .71 .59 .21 .12 .12 .96 .97 .32 Gender-typed behaviour (30 months) Figure 1. SEM for the effect of gender-typed behaviour at 30 months on PTSD symptoms in females, mediated by self-reported physical/psychological abuse and mother-reported abuse. Indicators are shown on latent variables using dashed arrows. Curved arrows show covariances between errors. Path estimates are standardised beta coefficients. P-values are shown in brackets. behaviour at all timepoints and experiences of all types of childhood abuse reports, but there were no significant associations for self-rated gender-typed behaviours. Similarly, there was a small association between more parent-rated masculine-typed behaviour (at 30 months only) and self-reported PTSD symptoms at 23 years for females.
Multiple comparisons/chance findings could explain why only the 30-month CGN measure was significantly associated with PTSD symptoms in females. There are other possible reasons for this discrepancy. Children become more gender-typed as they grow older . The 30-month measure may be the assessment least affected by the child's attempts to conform to society's gender-related expectations. The parent-rated measure relies on behaviours that are observable to them; it could be that it is this observable behaviour that makes children most at risk for abuse and subsequent PTSD. The self-ratings were also made much later (at 8 years) than the mother ratings (30-57 months). It could be that the differences in findings also reflect changes in gender-typed behaviour over time or differences in willingness to report gender-atypical behaviours between mothers and children. In sexual and gender minorities, it has been suggested that victimisation may lead to the concealment of a minority identity (Meyer, 2003), so it is possible that children may conceal their gender nonconformity as they get older.
All three abuse variables significantly mediated the association in females between CGN at 30 months and PTSD at 23 years when analysed separately. When both mother-reported abuse and self-reported physical/psychological abuse were included as mediators in parallel, only the latter remained a significant mediator, suggesting that mother-reports added little further information. Parents will not necessarily be aware of all the abuse experienced by their children and may be less willing to report abuse that implicates themselves or a family member.

Associations between gender-typed behaviour, abuse and PTSD for males
For males, gender-typed behaviour was not associated with abuse or PTSD. This contrasts with previous research (e.g. D' Augelli et al., 2006;Roberts et al., 2012a). It may be that previous studies have only found associations in males due to methodological limitations such as retrospective reports of CGN or less rigorous analysis. However, there are other possible explanations. In contrast to the measure of gender-typed behaviour used in the present study, all previous retrospective/cross-sectional studies utilised measures of self-perceived gender atypicality as their indicator for CGN. It is possible that different aspects Gender-typed behaviour (30 months) Figure 2. SEM for the effect of gender-typed behaviour at 30 months on PTSD symptoms at 23 years in females, mediated by self-reported physical/psychological abuse and mother-reported abuse, adjusted for sexual orientation. Indicators are shown on latent variables using dashed arrows. Curved arrows show covariances between errors. Path estimates are standardised beta coefficients. P-values are shown in brackets.
of gender nonconformity are differentially associated with these outcomes. PTSD scores were significantly higher in females than males, and there were higher rates of missing data for PTSD and self-reported abuse measures for males than females. These factors may make it harder to detect associations for males. There may also be sex differences in reporting abuse or PTSD.
Unlike the present study, but using the same birth cohort, Xu et al. (2020) found a prospective association between more nonconforming parent-rated behaviour and mothers' reports of emotional and physical abuse up to 6 years of age in both males and females. The present study had multiple maternal ratings up to the age of 11 years and included sexual or physical abuse by anyone, in addition to parental physical or emotional abuse. Therefore, associations may depend on the types, timings, perpetrators, or reporters of abuse investigated.

Associations between minority sexual orientation, abuse and PTSD
In line with previous research, the present study found significant associations between minority sexual orientation and abuse and PSTD. In females, significant mediation of the relationship between CGN and PTSD still held after adjusting for sexual orientation for mother-reported abuse and self-reported psychological/physical abuse. However, the proportion mediated was slightly reduced, similar to Roberts et al. (2012a). The mediation by sexual abuse was no longer significant when adjusting for sexual orientation. Caution is needed in interpreting the extent of attenuation in relation to sexual orientation, in part due to the likely under-reporting of minority sexual orientation (Berg & Lien, 2006) and the relatively young age (15 years) of assessment. Gender nonconforming children may also be at elevated risk for abuse partly due to others' perceptions of associations with minority sexual orientation, regardless of whether the individual themselves subsequently has a sexual minority identity.

Strengths and limitations
The study's strengths include the large sample, prospective design and the inclusion of both parent-rated and self-rated gender-typed behaviour and abuse across multiple timepoints. Future research should investigate the mediation effects of different types of abuse using more diverse longitudinal samples with validated abuse measures. Effect sizes in this study were generally small, and caution should be taken in interpreting these results. However, given that this was a 21-year longitudinal study in a large representative sample, and that experiences of abuse and PTSD are likely to be under-reported, the relationships between gender-typed behaviour, sexual orientation, abuse and PTSD warrant further research.

Study implications
More research is needed into why CGN may put females specifically at risk for PTSD, investigating associations with a range of CGN and abuse measures at multiple time points. Those supporting gendernonconforming females should be aware of their increased risk and any possible signs of abuse. As previous studies have found associations for males, those working with gender nonconforming males should also be vigilant for signs of abuse. Gender nonconforming young people may be at increased risk for victimisation due to perceptions of sexual minority identification (e.g. van Beusekom et al., 2016). Sexual minority orientation was also associated with abuse and PTSD in and of itself in the present study. Therefore, those working with young people should also be aware of the abuse and PTSD risk in sexual minority adolescents.

Disclosure statement
No potential conflict of interest was reported by the author(s).  Before 11 years, n (%) Were you touched in a sexual way by an adult or older child or were you forced to touch an adult or older child/teenager in a sexual way No 2,417 (95.4%) 1,323 (98.5%) Yes, once 66 (2.6%) 9 (0.7%) Yes, more than once 50 (2.0%) 11 (0.8%) Did an older child/teenager force you or attempt to force you into any sexual activity by threatening you or holding you down or hurting you in some way when you did not want to .31) Ã p < .05 ÃÃ p < .01 ÃÃÃ p < .001. Binary logistic regressions with gender-typed behaviour as the predictor, controlling for sex, mother's education and child's ethnicity. Run with robust standard errors. Model 2 includes the addition of the interaction between sex and gender-typed behaviour. (Sexual Orientation: 0 ¼ heterosexual, 1 ¼ sexual minority; Sex 0 ¼ Female, 1 ¼ Male). Higher scores on gender-typed behaviour measures reflect more nonconforming behaviour. Separate effects for males and females are presented in Table 3.

PTSD symptoms 23 years a Abuse b
Motherreported a Self-reported physical/psychological a OR .94) Ã p < .05 ÃÃ p < .01 ÃÃÃ p < .001. Linear ( a ) and binary logistic ( b ) regressions with sexual orientation as the predictor, controlling for mother's education and child's ethnicity. Run with robust standard errors. Model 2 includes the addition of the interaction between sex and sexual orientation. (Sexual Orientation: 0 ¼ heterosexual, 1 ¼ sexual minority; Sex 0 ¼ Female, 1 ¼ Male). Numbers in brackets refer to 95% Confidence Intervals. Separate effects for males and females are presented in Table 2. years in females, medicated by mother-reported abuse Appendix 8. SEM for the effect of gender-typed behaviour at 30 months on PTSD scores at 23 years in females, mediated by self-reported physical/psychological abuse