Model Selection in Occupancy Models: Inference versus Prediction

1. Occupancy models are a vital tool for understanding the patterns and drivers of species occurrence, but their use requires a method for model selection – choosing between models with different sets of occupancy and detection covariates. The information-theoretic approach, which employs information criteria such as Akaike’s Information Criterion (AIC) is arguably the most popular approach for model selection in ecology and is often used for selecting occupancy models. However, the information-theoretic approach risks selecting models which produce inaccurate parameter estimates, due to a phenomenon called collider bias. 2. We aimed to investigate the consequences of collider bias (using an illustrative example called M-bias) in the occupancy and detection processes of an occupancy model, and to explore the implications for model selection using AIC and a common alternative, the Schwarz Criterion (or Bayesian Information Criterion, BIC). 3. We simulated datasets using known parameter values and fitted models with different combinations of covariates. We then compared the models on how accurately they estimated the effect of a focal covariate on the occupancy probability, the accuracy of their site-level occupancy predictions, and their level of support from AIC and BIC. 4. When M-bias was present in the occupancy process, AIC and BIC selected models which inaccurately estimated the effect of the focal occupancy covariate, while simultaneously producing more accurate predictions of site-level occupancy probability. In contrast, M-bias in the detection process did not impact the estimate; all models made accurate inferences, while the site-level predictions of the AIC/BIC-best model were slightly more accurate. 5. Our results emphasise the importance of distinguishing between parameter inference and prediction as separate tasks in ecological modelling. In the context of occupancy models, our findings suggest that information criteria can be used to select occupancy covariates if the purpose of the model is prediction, but not if the purpose is to understand how environmental variables affect occupancy. In contrast, detection covariates can usually be selected using information criteria regardless of the model’s purpose. More generally, our results support the view that information criteria should not be used to compare different biological hypotheses in observational studies. ### Competing Interest Statement The authors have declared no competing interest.