Handling missing observations with multiple imputation

Missing observations are unavoidable in long-standing monitoring projects. Currently the Underhill (Underhill & Prys-Jones 1994) or birdSTATs indices (van der Meij 2013) are frequently used in ornithology. However both approaches underestimate the confidence intervals. Instead, we propose multiple imputation (Rubin 1987), the standard in medical and social science, but hardly known in ecology. Our analysis provides insight on the nature of the problems with the Underhill and birdSTATs indices and demonstrates the value of multiple imputation. Besides a short theoretical introduction, we will present our “multimput” package (Onkelinx et al 2016). This is a free and open-source R package (R Core Team 2016) which is available at https://github. com/inbo-BMK/multimput. The “multimput” package has several benefits: 1. It allows flexible modelling of both the counts at individual sites (required for the imputation) and of the population totals. 2. The models allow for covariates, hence seasonal effects resulting from multiple visits per site can be handled properly. 3. Several probability distributions (e. g. Poisson and negative binomial) are available.