PT - JOURNAL ARTICLE
AU - Conn, Paul B.
AU - Thorson, James T.
AU - Johnson, Devin S.
TI - Confronting preferential sampling in wildlife surveys: diagnosis and model-based triage<sup>†</sup>
AID - 10.1101/080879
DP - 2016 Jan 01
TA - bioRxiv
PG - 080879
4099 - http://biorxiv.org/content/early/2016/10/14/080879.short
4100 - http://biorxiv.org/content/early/2016/10/14/080879.full
AB - SummaryWildlife surveys are often used to estimate the density, abundance, or distribution of animal populations. Recently, model-based approaches to analyzing survey data have become popular because one can more readily accommodate departures from pre-planned survey routes and construct more detailed maps than one can with design-based procedures.Species distribution models fitted to wildlife survey data often make the implicit assumption that locations chosen for sampling and animal abundance at those locations are conditionally independent given modeled covariates. However, this assumption is likely violated in many cases when survey effort is non-randomized, leading to preferential sampling.We develop a hierarchical statistical modeling framework for detecting and alleviating the biasing effects of preferential sampling in species distribution models fitted to count data. The approach works by jointly modeling wildlife state variables and the locations selected for sampling, and specifying a dependent correlation structure between the two models.Using simulation, we show that moderate levels of preferential sampling can lead to large (e.g. 40%) bias in estimates of animal density, and that our modeling approach can considerably reduce this bias.We apply our approach to aerial survey counts of bearded seals (Erignathus barbatus) in the eastern Bering Sea. Models that included a preferential sampling effect led to lower estimates of abundance than models without, but the effect size of the preferential sampling parameter decreased in models that included explanatory environmental covariates.When wildlife surveys are conducted without a well-defined sampling frame, ecologists should recognize the potentially biasing effects of preferential sampling. Joint models, such as those described in this paper, can be used to test and correct for such biases. Predictive covariates are also useful for bias reduction, but ultimately the best way to avoid preferential sampling bias is to incorporate design-based principles such as randomization and/or systematic sampling into survey design.