Collider bias (aka sample selection bias) in observational studies: why the effects of hyperandrogenism in elite women’s sport are likely underestimated

The medical condition of hyperandrogenism has entered common parlance because the International Association of Athletics Federation (IAAF) decided to ban women with the condition from competing against women who have ‘normal’ levels of testosterone (see box 1). This paper argues that the two articles substantiating the IAAF’s regulation1 2 fail to prove that elevated levels of natural testosterone are causally linked to better sporting performance. By studying elite athletes both papers induce a collider stratification bias that attenuates the competitive advantage of testosterone—a little-known bias that often plagues studies of highly selective samples.