Our results show that sexual attraction and CGT are influenced by genetic factors (accounting for 25% and 32% of the variance respectively). Genetic contributions as estimated in the univariate analyses had a much weaker impact on AGI (11%). The effect of non-shared environmental factors (including measurement error) on all traits was large. However, there was no effect of the shared family environment on any trait.
These findings are broadly consistent with previous population-level twin studies demonstrating a heritable basis to male and female sexual orientation. The heritability estimates reported here for female sexual attractions were larger than those reported by Bailey et al. 
for sexual attraction components (8%). For attraction, we found no effect of the shared environment in contrast to Bailey et al. 
who reported an estimate of 41%. Langstrom et al. 
reported shared environmental effects of same-sex sexual behavior of 17%. Kendler et al. 
did not separate their analysis by sex so we cannot compare the findings. Finally, our genetic estimates were lower than those reported by Kirk et al. 
who attempted to model two components of sexual orientation - sexual attraction and sexual experience - and reported estimates for females between 50 and 60%. Whilst Kirk et al. 
did use attractions in their study they supplemented these with the measures “attitudes to homosexual sex” and “lifetime same-sex partners” (from a range in an extensive sexual orientation questionnaire) which are not directly comparable to measure used here.
The effect of E on all traits was large. E includes phenotypic variation accounted for by non-shared environment and measurement error. A variety of sources can cause measurement error, including inadequate or imprecise assessment instruments and phenotype description, and a variety of response styles, specifically acquiescence, disacquiescence, extreme response, midpoint responding, and noncontingent responding 
. The twin modelling approach used in this study does not allow separation of the two sources, hence, quantification of the influence of measurement error is impossible. It is therefore likely that some inconsistency between our heritability estimates for sexual attraction compared to previous work might be due to different usage of Kinsey-type scales for measuring trait sexual orientation. Here we used a 5-item measure; Bailey et al. 
used the full 7-item Kinsey-scale; Langstrom et al. 
used number of same-sex partners; and Kendler et al. 
employed a single item with three response choices (heterosexual, bisexual, and homosexual in attractions). If we compare our data to Bailey et al. (both studies focused on attractions and using what approximate traditional Kinsey-type scales), it is possible that the relatively small difference in response options between the two studies (that is, a difference of 2) contributed to the differing heritability estimates for sexual attraction. Parameter estimates for sexual orientation might be unusually sensitive to the range of items used to assess the trait and thus future researchers should be mindful of the utilizing psychometrically robust scales.
Consistent with several studies the highest heritability was found for CGT (32%) 
. CGT seems to be the highest heritable correlate of sexual orientation reported thus far, and furthermore lies in the region of the h2
estimates generally reported for sexual orientation (measured both behaviorally and psychologically). This adds further support to the notion that CGT may be the main heritable component or endophenotype of sexual orientation 
. However, genetic effects for AGI were negligible compared to previous work and we are less confident about the validity of AGI as a robust correlate of sexual orientation 
. The inter-correlations with other measures of sexual orientation were low for AGI compared to CGT and showed little variability. Compared with CGT items which capture sex-typed behavior, interests and identity, AGI comprises identity items only and thus has restricted psychometric precision. Other documented indices of gender identity such as “gender diagnosticity” e.g. 
should be the focus of future twin studies if suitable short-form scales can be developed. As with attraction, the inconsistencies between the studies might be attributed to the number of items in the measures used. Our measure of CGT comprised four items and Bailey et al. 's 
five items used to check the reliability of self- vs. other-report of their 24-item measure . The relative comparability here (a difference between the two studies of only 1 item) provides some confidence for the validity of heritability estimates reported by both. However, their measure of AGI comprised seven items and ours only four (with relatively low internal consistency) which may explain the differences between studies for this particular measure.
The results from the multivariate analyses presented here provide evidence of a genetic overlap between CGT and sexual attraction but less for AGI. We detected a common latent phenotype with a heritability of 24% underlying sexual orientation, CGT and AGI as well as moderate phenotype-specific additive genetic factors and large phenotype-specific non-shared environmental factor loading on these traits. These data are supportive of those from Bailey et al. 
who also found that a common pathway ACE model best fitted the available data. Both studies support the notion that showed that genetic and non-shared environmental factors markedly contributed to the covariation among the measures, with all three measured variables (sexual attraction, CGT and AGI) being good indicators of an underlying latent factor. Overall, the present results support previous non-twin evidence for the existence of an intermediate phenotype for sexual orientation, such as for example “sex-atypicality” 
. A likely candidate for this latent phenotype is prenatal androgen exposure which shapes variations in gender nonconforming behavior and sexual orientation and the developmental coupling between them 
. Nevertheless, speculation about origins of this putative sex hormone-related phenotype is limited by two candidate gene studies of male sexual orientation both producing null results: one for the androgen receptor 
and another for aromatase 
. However, the absence of such associations in men does not imply a similar null result in women.
Insofar as our genetic estimates are additive, these data do not suggest a major role for epistatic or dominant allelic effects. Our results are also silent on the sources of non-shared environmental effects. However, what is clear from several twin studies, including the present, is that shared factors such as the home environment and parenting styles have little impact on human sexual orientation. Nevertheless, as we cannot be certain that our measures, particularly AGI, were robust (given the sizable loading of a common non-shared environmental factor on each trait) there is a necessary degree of imprecision to our parameter estimates and the model fitting results should be treated with caution. Future studies should be particularly careful in using measures of AGI. Several other limitations weaken overly strong conclusions from the present study. The response rate was lower than three other large twin studies (49%, compared to 53.8% in Bailey et al. 
60% in Kendler et al. 
; and 59.6% in Langstrom et al. 
) although it is not clear how this could systematically bias the parameter estimates reported 
. Also, the response rate here is in fact comparable to other epidemiological surveys of female sexual behavior 
. The representativeness of our twin sample also diminishes any putative selection biases, as shown by a large comparative study demonstrating that our twin population is very similar to singletons on a wide range of common health and lifestyle factors 
. A comparison of the sample characteristics in show that the MZ and DZ twins did not differ significantly on most demographic variables, arguing against the tendency for MZ to be more alike possibly due to shared upbringing. The comparably low internal consistencies for CGT and AGI may further reflect the heterogeneous nature of the constructs, suggesting that more items are needed to capture the range of manifestations of the constructs. We also used retrospective measures of CGT which may be influenced by recall biases. However, prospective studies confirm the predictive psychometric validity of measures of CGT that are comparable to the one used here as do studies of maternal reports of proband-recalled CGT and studies of childhood home videos 
Due to our considerably large sample size we had enough power to detect a rather small contribution of non-additive genetic factors, had it been present (1,800 twin pairs are needed to reject an AE model with a power of 80% when an ADE model is the true model, with respective contributions of additive genetics effects
0.50, dominant genetic effects
0.30 and non-shared environmental effects
. Nonetheless, there remained insufficient numbers of non-heterosexual participants to guarantee a high degree of statistical power in the genetic and environmental analyses. This is a well-known problem, as sexual orientation-related data are notoriously skewed 
In summary, we found genetic influences on female sexual orientation as measured via attractions and on CGT (a key developmental correlate of sexual orientation). A moderate effect of a common latent phenotype suggests that there are some overlapping mechanisms which may be responsible for sexual orientation. However, stronger conclusions are not warranted at this stage because of substantial measurement error. Future research efforts should focus on “sex-atypicality” as a possible intermediate phenotype for trait sexual orientation which may be more amenable to gene-mapping approaches.