Generalized Linear Mixed Models to Analyze Weekly Counts in a Split Plot Design.

This talk concerns the statistical methodology used to analyze count data that are not normally distributed and which were collected as part of an experiment on forage crops. An experiment on the effects of stubble height and mowing schedule on forage quality and yield was conducted in two different years. In each year, the design was a repeated measures split plot with mowing schedule as the whole plot factor and stubble height as the subplot factor. Plots were sampled for forage according to the mowing schedule over a three month period so the number of repeated samples for each level of the whole plot factor varies by level. Simultaneously, insect data were collected weekly in the same plots as well as in nearby edge plots not part of the forage experiment.  As a result, two additional factors, one related to the time since the last mowing and the other to do with edge effects, were incorporated in the analysis of the weekly insect data. We describe the use of generalized linear mixed models with count as either Poisson or Negative Binomially distributed for analyses of the insect data.