Reply to Comment by Young et al.

Had our study been, as Young et al. imply, a data trawling exercise using hundreds of variables to look for a 'thread', then doubts about its validity might be justified. However, their account of our work bears little relationship to the methods, results or conclusions we report. For example, Young et al. claim that we used 396 tests to address our primary hypothesis. In fact, we used two. Young et al. fail to acknowledge that our work was based on a strong a priori hypothesis. Far from being 'genetically implausible', it has long been accepted that human sex ratios at birth are not 50 : 50 despite the equal production of 'male' and 'female' sperm, and that birth sex ratios vary between populations and across time. These facts have provoked debate about the likely role of the parental environment in influencing infant gender Our primary hypothesis, based on sound evolutionary principles and supported by considerable research on other mammal species, was that women with good nutrition at the time of conception would be more likely to bear sons. Even since the publication of our paper, two further papers on humans have appeared supporting this hypothesis (Bulik et al. 2008; Villamor et al. 2008). The 'main finding' of our study (notwithstanding Young et al.'s preoccupation with cereal) is a link between maternal nutritional status around conception and infant gender. Because intakes of different nutrients are inevitably correlated with one another (people eat food, not single nutrients), we used a standard method of data reduction—principal components analysis (PCA)—to summarize the patterns of nutrient intakes along new axes ('components'). PCA is widely advocated as a means of dealing with collinearity (the non-independence of predictors) that would otherwise violate one of the fundamental assumptions of regression analysis (see Its use made it unnecessary to conduct multiple tests on individual nutrients to examine the primary hypothesis of a link between maternal diet and infant gender: the first component gave a good description of women's nutritional intakes in a single variable. We showed that the relationship between the scores on this variable and offspring sex differed with time period, and then we presented a further test demonstrating a link for the preconception data. Young et al. maintain that because we measured women's diets at three time points, we inflated the chance of obtaining a positive result. Yet only one of these time points—the time around conception—was …