Antonio Marini-Davis and Fernando J. Vonhoff.

Datasets in the field of neurobiology are becoming increasingly complex, and common statistical methods such as T-test and ANOVA are struggling to capture this complexity. Researchers often discard informative details of their data through use of improperly specified models or non-parametric tests to conclude treatment effects. I detail here the increasingly used method of generalized linear (mixed) models to make more precise statistical conclusions when data doesn’t fit the normal assumptions for traditionally used statistical tests. Using a real behavioral dataset generated using a two choice preference based assay using Drosophila Melanogaster, I delineate the process of recognizing when more complex modeling is necessary, and show the diagnostic tools to ensure proper conclusions are drawn when using model-based statistical inference. Proper model specification allows future researchers to retain valuable information in their data and make better informed conclusions about their experimental outcomes.

University of Maryland, Baltimore County. US