|Home | About | Journals | Submit | Contact Us | Français|
It is unclear how mating strategies are distributed in humans; do they vary along a continuum from promiscuous (multiple short-term relationships) to monogamous (long-term pair-bonded relationship) or do these represent alternate phenotypes? If the latter, how are these phenotypes distributed? Wlodarski et al.  addressed these questions empirically, performing mixture models on sociosexuality and second-to-fourth digit ratio (2D : 4D), from which they inferred that both variables reflect two underlying normal distributions, rather than a single underlying normal distribution. Given this apparent evidence of two phenotypes for both sexes, they then estimated their distributional parameters. Questions arise, however. Was their normality assumption warranted? Measures of sociosexuality are generally skewed, so assuming normality biases against selection of a single-component model. Further, are sociosexuality and 2D : 4D associated as they should be if both are indicators of mating strategy phenotype? In fact, the literature offers scant evidence of such a relationship in humans . Using a different modelling technique, and a sample with measures of both sociosexuality and 2D : 4D, I find no evidence that humans exhibit two sociosexuality phenotypes, or that sociosexuality and 2D : 4D are related.
The present sample (353 men, 693 women; approx. 50% white non-Hispanic, 32% Hispanic, 18% other or mixed) was drawn from five previous studies [3–7]. Participants completed the original sociosexual orientation inventory (SOI) , not the revised version  used by Wlodarski et al. Four items are the same across both inventories (after responses are re-scaled), and two more items are similar. Wlodarski et al. used only the attitudes and desires items. The present data include one desire, and three attitude items that are comparable across both inventories. Digit ratio was measured on a portion of the sample.
Taxometric analysis (TA) is a method for deciding between categorical and continuous models of psychological constructs. TA makes no distributional assumptions. Mixture models on skewed indicators tend to overestimate the number of latent categories . TA outperforms mixture modelling when the underlying indicator distributions are non-normal .
Instead of significance testing, TA uses consistency testing—multiple procedures converging on the same result. Two non-redundant taxometric procedures, maximum covariance (MAXCOV) and mean above minus below a cut (MAMBAC), were performed on the four attitude and desire items (for explanations of these procedures, see  and references therein).
Cases with missing data were excluded. Analyses were conducted using Ruscio's taxometric programs for R. Indicators were standardized, and all indicators served in all roles. Tied cases were randomly re-sorted for 20 replications. The MAXCOVs used 25 windows with 90% overlap. The MAMBACs used 50 cuts with 25 cases maintained at each end.
The comparison curve fit index (CCFI)—a relative fit statistic—was calculated for each analysis (see Ruscio  for details). CCFI values below 0.45 indicate a continuous structure, values above 0.55 indicate categories and values between 0.45 and 0.55 are ambiguous. See table 1 for results. Neither sex yielded a categorical decision, though the outcome for women was ambiguous.
Given that parametric methods are generally more powerful than non-parametric methods, this failure to detect categories could reflect a lack of power, rather than a lack of categories. To address this possibility, I simulated data consistent with Wlodarski et al.'s modelled phenotypes, which I analysed with the taxometric methods reported above.
Wlodarski et al.'s models based on their six item composite sociosexuality measure produced an approximate standardized difference between phenotypes of d = 2.42 giving each item an estimated d = 0.99. I created four items for each sample where 45.3% of cases were drawn from a standard normal distribution and the other 54.7% from a normal distribution 0.99 standard deviations higher than the first. See the electronic supplementary material for details. I analysed 100 random samples for each sample size (335 and 642). See table 2 for results. Using the consistency between procedures criteria, TA correctly identified these data's categorical structure in only 26.5% of the simulations. However, incorrect continuous decisions occurred in only 3% of the simulations. These results suggest that the continuous structure found for men would have been unlikely if the underlying structure was truly categorical.
Although TA lacks power to detect categorical structures under the present data conditions, it does not suffer from an inflated error rate. Is Wlodarski et al.'s method similarly robust to incorrect structural decisions, given deviations from normality? I simulated data consistent with my SOI data, which I analysed with finite mixture modelling assuming underlying normal distributions.
I fit beta distributions to both my male and female data, took 100 random samples from each distribution using Wlodarski et al.'s largest sample size for each sex (134 male, 187 female), and analysed them with the mixtools package for R using bootstrapped likelihood ratio tests comparing one- and two-component models (details in electronic supplementary material). As table 3 shows, mixture modelling incorrectly detected a categorical structure in the continuous, non-normal case a significant proportion of the time. Although correct most of the time for the male distribution, the false-positive rate of 24% was nonetheless unacceptably high. The false-positive rate for the more skewed female distribution was 98%. This method was not robust to non-normality.
I used correlations to investigate the relationship between digit ratio and sociosexuality. Across primates, lower ratios indicate promiscuity, which corresponds to higher sociosexuality [13,14]. Therefore, if digit ratio serves as a marker for mating strategy in humans, it should correlate negatively with SOI scores within each sex. I correlated 2D : 4D with the average of the attitude and desire items, and with the MAXCOV categorization as sociosexually restricted or unrestricted. See table 4 for results. None of the correlations were significant.
Analyses of the latent structure of sociosexuality that do not rely on problematic distributional assumptions do not support Wlodarski et al.'s conclusion that both men and women exhibit two mating strategy phenotypes. Simulations indicate that this lack of support, at least in men, is unlikely to be due to insufficient power, and that Wlodarski et al.'s findings could be an artefact of their normality assumption. Further, 2D : 4D was not significantly correlated with SOI, suggesting that it is unlikely to be a good indicator of mating strategy in humans.
Thanks to S. Gangestad, R. Thornhill, C. Garver-Apgar, J. Simpson, A. Cousins and P. Christensen for sharing their data, and to four anonymous referees for their helpful comments.
The accompanying reply can be viewed at http://dx.doi.org/10.1098/rsbl.2016.0213.
Research with human participants was approved by the University of New Mexico's Institutional Review Board. All participants provided informed consent. Project numbers are not available for four of the five studies included because of recent transitions in the IRB office. The fifth project number is HRRC 02-254.
These data cannot be shared publicly, because the participants gave consent for their data to be made public only in aggregate form. The data are available from the author by request.
I received no funding for this study.
I have no competing interests.