Common Statistical Mistakes in Entomology:Blocking and Inference Space

摘要

The final article in this series on common statistical errors focuses on the practice of blocking in experiments, and its implications on inference space; i.e., the range of situations to which the results are applicable. 1 address the concept of inference space first by contrasting fixed-effect and mixed-effect models. A fixed-effect model is one in which all of the model effects are assigned or controlled by the investigator, with specific interest in their impacts on the response. One chooses thefactors investigated, and their levels, with that in mind. Importantly, there is no intent to draw inferences beyond the specific treatments and conditions that are observed. In a fixed-effects analysis, the statistical tests, variances, and standard errors are, strictly speaking, only applicable to the conditions and experimental units that were observed. In contrast, a mixed-effects model includes the fixed effects of interest, and one or more random effects that represent a sample from a larger population of physical entities or conditions.These entities (plots, blocks, fields, experimental repetitions) are not, in and of themselves, of interest; they simply represent a population of entities or conditions to which the results of the experiment are applicable. The inferences of a mixed-effects analysis are intended to extend to a larger population of experimental units and conditions similar to those that were observed. The test statistics and standard errors from a mixed-effects analysis reflect this broader inference space. Fixed-effectmodels are common in the current entomological literature. However, we usually intend our findings to be more broadly applicable, so the intended inference space is better aligned with a mixed-effects model. In practice, the breadth of the inference space depends to a great degree on the nature and arrangement of any blocking or repetition effects in the experiment.