Reply to Falcon

We appreciate Falcon's [1] interest in our paper on male and female mating strategies in humans. However, she addresses a rather different issue to the one we were interested in. Her concern is with whether the psychological dispositions implied by the anatomical and sociological indices we used can best be described as representing distinct categories or are best characterized by a single underlying continuum. Our interest was founded on two motivating observations: (i) the fact that the literature on human reproductive strategies widely recognizes that males fall into two general classes (cads and dads) and (ii) the fact that, on a number of anatomical indices of mating system, humans always fall (and do so almost uniquely among mammals) midway between monogamous and polygynous species. Our concern was to determine the relative proportion of these two reproductive classes and ask whether these proportions explained the second of these facts. Our problem was how to determine the relative proportions of these alternative strategies, irrespective of whether they constitute genuine non-overlapping types (which, in fact, they do not), some kind of bimodal distribution or are simply the ends of a normal continuum. There is no easy way to do this, and taxometric analysis (which simply gives a qualitative estimate of whether a trait distribution is more or less likely to constitute discontinuous types) certainly does not provide the answer. Given the obvious fact that all but one of our six datasets were self-evidently bimodal, our decision was to opt for a rigorous method for decomposing mixed distributions. We first showed for five of our six datasets that the data are best described by two (and only two) normal distributions, and then used mixture modelling to estimate the means and proportions for these distributions. We would, of course, welcome alternative suggestions for improved methods to address these questions, because there is evidence that this system may not be an evolutionarily stable strategy (ESS), but may instead either be in a complex dynamic equilibrium or under directional selection [2] and more precise estimates of the proportions would help elucidate this.

That said, we feel obliged to point out several issues in Falcon's analyses. First, taxometric analysis is not as robust a tool as she claims. Aside from the fact that it only provides a qualitative estimate of whether or not a distribution is likely to consist of types [3], its developers are considerably more cautious in their interpretations of its results. Roscio [3] himself observes that ‘guidelines for interpreting the results of taxometric analyses may be better developed, but the evidence supporting them remains modest at best… The absence of evidence for taxonic [categorical] structure does not logically support an inference of dimensional structure because this alternative does not exhaust the possibilities’. Second, contrary to Falcon's claim, it has been shown that it is far from being the case that taxometric analysis yields more reliable results than mixture modelling, especially when, as in the present case, there is low class separation—not least because mixture modelling, unlike taxometric analysis, makes no unrealistic assumptions about the form of the data [4]. Third, Falcon uses the fact that the data in our samples are not normally distributed as grounds for an alternative approach. However, the fact that the data are not normally distributed is precisely our point: that is exactly what one would expect if the data are a composite of two or more normal distributions. Fourth, for her own analyses, Falcon uses a dataset pooled from studies mostly conducted on undergraduates that mix several different racial groups. Both sociosexual orientation inventory (SOI) [5] and 2D : 4D (second-to-fourth-digit ratio) ratios [6] vary significantly across racial and cultural groups; pooling heterogeneous datasets inevitably increases error variance. This being so, Falcon's results are hardly surprising. In addition, there are inevitably likely to be considerable differences in how late teenagers and adults respond on the SOI. For this reason, we explicitly used only adult Caucasians in our samples. Fifth, even so, Falcon's analyses do not show that her data cannot be decomposed into two subsets. Indeed, her simulated re-analyses of our data would, if anything, seem to confirm our conclusions. Sixth, Falcon argues that 2D : 4D and SOI do not correlate in her dataset. We explicitly distinguished between 2D : 4D and SOI, because they tell us slightly different things: one is biological and the other open to cultural influence. Whether or not they correlate may be less interesting than whether or not they display the same bimodal pattern and can tell us something about the extent to which culture can override biological predisposition. We note that her dataset used a version of the SOI that is considered less reliable than the revised SOI-R version that we used; SOI has high variation in internal consistency, as well as being highly skewed with low reliability response rates [7].

Competing interests

The authors declare they have no competing interests.