Realism and robustness require increased sample size when studying both sexes

Abstract

Female subjects have been historically excluded from biomedicine and other related areas of study [1]. Such exclusion has disadvantaged females and prevented a fuller understanding of biology. Therefore, in 2016, the National Institute of Health (NIH) mandated that all NIH-funded animal and human studies consider sex as a biological variable (e.g., [2]). Yet, sex as a biological variable has not been welcomed with open arms, most likely because many researchers believe they need to increase the overall sample size with 2 sexes compared to using only 1 sex (e.g., [2]). Recently, Philips and colleagues published a PLoS Biology article titled "Statistical simulations show that scientists need not increase overall sample size by default when including both sexes in in vivo studies" [3]. As indicated in their title, the authors have concluded and recommended no increase in sample size with both sexes, which was based on a set of simulations exploring a simple but-as they claim-likely biological scenario. Their conclusion is great news for researchers who feared coping with increased experiment sizes and costs.However, Philips and colleagues have assumed homoscedasticity between the 2 sexes, meaning variances or standard deviations of the sexes are the same throughout their simulations. Here, we first explain why such an assumption is biologically unrealistic and why heteroscedasticity between 2 sexes should be the norm rather than the exception by pointing out a wealth of empirical evidence and evolutionary arguments. We then show the results from a simulation study expanding Philips and colleagues' work by incorporating heteroscedasticity. Our results clearly indicate that we need to increase the overall sample size to have robust statistical inference. Further, we provide statistical recommendations to deal with heteroscedasticity. We also briefly touch on what funding agencies and ethics committees can do, given our results.