The risk of flawed inference in evolutionary studies when detectability is less than one.

Addressing evolutionary questions in the wild remains a challenge. It is best done by monitoring organisms from birth to death, which is very difficult in part because individuals may or may not be resighted or recaptured. Although the issue of uncertain detection has long been acknowledged in ecology and conservation biology, in evolutionary studies of wild populations it is often assumed that detectability is perfect. We argue that this assumption may lead to flawed inference. We demonstrate that the form of natural selection acting on body mass of sociable weavers is altered and that the rate of senescence of roe deer is underestimated when not accounting for a value of detectability that is less than one. Because mark-recapture models provide an explicit way to integrate and reliably model the detection process, we strongly recommend their use to address questions in evolutionary biology.