Applied statistics in field and semi-field studies with bees

Abstract

Field and semi-field studies are important tools in the ecotoxicological risk assessment of plant protection products for bees (honey bees, bumblebees and solitary bees). While these studies represent far more realistic conditions than laboratory tests, they also present a challenge for the analysis and interpretation due to the large and complex datasets. Therefore, in order to correctly answer the underlying ecotoxicological questions, it is crucial that these studies are not only thoroughly planned and conducted, it is also important that they are subjected to adequate statistical analysis. Our aim is to provide a better understanding on how to conduct and interpret statistical analyses in field and semi-field studies with bees made for regulatory purposes. An overview of how study design and statistics should be aligned with each other is given including the specific challenges of (semi-) field trials, as for instance how to address the problem of pseudoreplication if hives are regarded as experimental units. Different statistical tools are compared and their suitability for different data types and questions are discussed. Generalized Linear (Mixed) Models (GLMMs) are evaluated in more detail as they provide a flexible and robust tool for the analysis of honey bee (semi-) field data. Furthermore, some more light is shed on what p-values really tell us, how they can help to interpret data and how they should not be misinterpreted.