Evaluation of statistical methods and sampling designs for the assessment of microhabitat selection based on point data

Abstract

Information on resource selection by a species is essential for understanding the species’ ecology, distribution and requirements for survival. Research on habitat selection frequently relies on animal detection at point locations to determine which resource units are used. A variety of approaches and statistical tools can be employed for assessing selection based on habitat variables measured in those units. The aim of this work was to evaluate the reliability of common sampling designs and statistical methods in detecting habitat selection at fine scales based on point data We reviewed literature on microhabitat selection to determine characteristics of typical studies and analysed simulated small-mammal live-trapping data as a case study. We considered various scenarios differing in the number of sampled units and sampling duration. For each scenario, a set of simulated surveys was analysed through two univariate tests (Welch's t- and Mann–Whitney U-test), generalized linear models (GLMs), mixed-effect models (GLMMs) and occupancy models (OMs). Analysis of simulated data revealed that overall performance of all statistical methods improved with increased trapping effort. Univariate tests were especially sensitive to the number of sampling units, while modelling methods took also advantage of longer trapping sessions. Univariate tests and GLMs provided partially correct information in most cases, whereas GLMMs and OMs offered higher probabilities of fully describing simulated habitat preferences. With typical sampling efforts, appropriate statistical analysis of point data is able to provide a moderately accurate description of habitat selection at small scales, in spite of the violation of closure and independence assumptions of applied models. Modelling approaches are proliferating; we encourage using models that can deal with multiple sources of variability, such as GLMMs and OMs, when data are hierarchically structured. There is no a priori best survey design; it should be chosen according to the scope and goals of the study, environment heterogeneity, species characteristics and practical constraints. Researchers should realize that sampling design and statistical methods likely affect conclusions regarding habitat selection.