Evaluation of Generalized Linear Mixed Models for Analyzing Disease Incidence Data Obtained in Designed Experiments

Area of Science:

• Plant pathology
• Statistical modeling

Background:

• Disease incidence in nature often exhibits clustering, leading to higher variability than random patterns.
• Generalized linear mixed models (GLMMs) are suitable for analyzing binary disease data and accounting for multiple variation sources in experiments.

Purpose of the Study:

• To evaluate different GLMMs for analyzing Phomopsis leaf blight incidence in strawberry plants under fungicide treatments.
• To compare models accounting for heterogeneity via residual variance versus within-plot sampling variance.

Main Methods:

• Five experiments on strawberry Phomopsis leaf blight were analyzed using various GLMMs.
• Models differed in how they handled heterogeneity: residual variance (fixed, treatment-dependent, incidence-dependent) or sampling variance (constant, treatment-dependent).

Main Results:

• All tested GLMMs successfully fitted the experimental data.
• Weak evidence suggested residual-variance models might be slightly more appropriate than sampling-variance models.
• Model selection minimally impacted tests for treatment effects and mean differences.

Conclusions:

• The simplest GLMM, assuming fixed residual variance, is recommended for analyzing disease incidence data due to its ease of use and adequate performance.
• GLMMs offer a robust alternative to traditional methods like ANOVA for non-normally distributed disease data.