Pitfalls arising from site selection bias in population monitoring defy simple heuristics

Abstract

Abstract Site selection bias can occur when researchers monitor animals at sites of great abundance, occupancy or quality. The prevailing heuristic is therefore that ecologists should never select sites using these criteria. There is thus concern that common monitoring schemes including surveys at colonies, migratory bottlenecks and artificial breeding sites intentionally monitor sites of great abundance. Whether such routine monitoring schemes likely succumb to site selection bias is unexamined. We simulate the dynamics of long‐lived vertebrate populations to test three potential pitfalls hypothesized to occur when sites are selected for monitoring based on abundance, occupancy or quality. (1) For regression to the mean, population levels might appear to decline at monitored sites because they were chosen during a peak of abundance and random fluctuations cause populations to decrease from that point. (2) Preferential sampling occurs when some sites are preferred by the focal organism and have an above‐average probability of being selected for monitoring. (3) The missing zero effect occurs when occupied sites are chosen more often for monitoring because of initial or historical occupancy. Our simulations demonstrate that regression to the mean should only occur when inter‐annual fluctuations in abundance are relatively large compared with the average difference between sites. If researchers monitor sites that are truly of great average abundance, regression to the mean is avoided. Preferential sampling can cause a delay in the detection of a decline because monitored sites are preferred by the focal species and thus should be the first to be occupied and the last to be abandoned. Finally, the missing zero effect can cause a perceived decline in a stable population because sites were chosen when they were initially occupied and thus can only be abandoned, whereas potential colonizations at initially unoccupied sites cannot be observed. Regression to the mean occurred during specific circumstances that do not seem biologically plausible for some study systems (e.g. colony surveys). It is therefore impractical to apply a single simple heuristic such as ‘never monitor animals at sites of great abundance’ across organisms of varying life histories.