RELOCATING NEST BOXES FROM POOR QUALITY SITES CAN BIAS INFERENCE INTO POPULATION DYNAMICS

Nest boxes are often used to monitor animals, and it is common practice to relocate nest boxes from unproductive sites into presumably better habitat. This relocation of nest boxes means that a given nest box program progressively monitors better sites over time. Ecological theory holds that occupancy and reproduction should generally increase with habitat quality. Thus, relocating nest boxes from poor quality sites might positively bias trends in occupancy and reproduction. These biased trends might cause researchers to be overly optimistic about the status of their focal populations. To demonstrate this potential pitfall, I built a stochastic model to simulate a nest box program that relocates the least productive 25% of nest boxes every 5 yr over a 25-yr study. The model assumed occupancy and reproduction levels for the entire population were stable throughout the study, so changes in occupancy and reproduction observed in nest boxes could only be due to relocation. I implemented this model under three settings: (1) Stable, where the same sites are monitored over the entire study; (2) Random, where the unproductive nest boxes are relocated to random sites; and (3) Learning, where the unproductive nest boxes are relocated to sites of better quality. For each of the 1000 simulations per setting, I performed logistic and Poisson regressions to determine whether there were statistically identifiable (P < 0.05) temporal trends in occupancy and number of young fledged from nest boxes. As expected, occupancy and number of offspring fledged from nest boxes were stable during the Stable simulations, and increased over the 25 yr during the Random and Learning simulations. Trends in occupancy were rarely identifiable during Stable simulations, were identifiable in 46% of simulations under Random settings, and in 97% of simulations under Learning settings. Trends in number of young fledged were identifiable in 18% of simulations under Stable settings, 91% of simulations under Random settings, and 100% of simulations under Learning settings. Such statistically significant trends, induced solely by relocating poorly performing nest boxes, represent a potential pitfall when interpreting vital rates measured using nest boxes. Potential solutions might include calculating occupancy using a subset of boxes that are never considered for relocation, or statistical models that account for preferential sampling.