Advanced

Developer Tutorials: Using Generalized Linear Mixed Models to Model Experiment Designs that include Non-Gaussian Responses and Random Effects

Application Area:
Statistics, Predictive Modeling and Data Mining

This session is for JMP Pro users who understand basic predictive modeling principles and have used JMP Pro for predictive modeling.  There are also useful for those who understand mixed models or use SAS Proc GLIMMIX.

Many experiment designs include both non-Gaussian responses and random model effects, thus requiring a mixed model methodology. When the response is clearly non-Gaussian and there are no random effects, JMP Generalized Linear Models (GLM) or JMP Pro Generalized Regression models are appropriate. If there are random effects, Generalized Linear Mixed Models (GLMMs) are needed.

JMP Pro 17 Generalized Linear Mixed Model capability within Fit Model enables fitting of models with both non-Gaussian response variables and random design effects, such as blocking. Properly fitting a GLMM yields accurate hypothesis testing and/or estimation with appropriate standard errors for confidence intervals.

GLMMs are useful for researchers whose experiment designs include blocking or split-plot  features that should be random effects. It is also useful where the response is non-Gaussian, such as for count proportions (success/total) or binary counts (yes/no). The JMP GLMM capability fits the correct model to give researchers the power to test their hypotheses and accurately estimate model parameters.

The Key Developer of JMP GLMM capabilities demonstrates how to use the techniques, including how to fit Poisson or Binomial responses with fixed and/or random model effects.  She explains how to interpret the statistics and probabilities. She describes why and how the development team implemented GLMMs, including handling multiple comparisons, data scale estimates, standard errors and confidence intervals.

This JMP Developer Tutorial covers: Fit Model GLMM.addressing modern dilemmas around Bayesian methods and p-values.