Jean Orelien, MStat, Statistics and Public Health Research Division, Constella Group, 2605 Meridian Pkwy, Durham, NC 27713, (919) 313-7607, jorelien@asciences.com

In the generalized linear mixed model, few statistics are available for assessing the adequacy of the fit of the model. Traditional statistics such as the likelihood ratio test (LRT), the Akaike Information Criterion (AIC) or the Bayesian Information Criterion (BIC) have major limitations in that they require the fitting of 2 models. The LRT requires that the 2 models be nested. On the other hand, for the AIC and BIC in comparing two models, it is not clear how much of a difference is a significant one. In this paper, we review several other competing statistics such as the concordance correlation coefficient (CCC also denoted rc), Rsq1, proportion reduction in entropy and proportion reduction in deviance that have been proposed to assess goodness-of-fit of a model. While in theory, these statistics offer some advantages over traditional ones; we present results of simulation on their performance that shows that they tend to yield large values even in the absence of important terms from the model. Directions for future research are considered.

Learning Objectives: At the end of the session, participant will be able to

• Describe tools that have been used traditionally for assessing goodness-of-fit in the generalized linear mixed model
• Understand the limitations of these statistics
• List more recent methods that have been proposed for assessing goodness-of-fit in the generalized linear mixed model
• Assess the usefulness of the more recent tools that have been proposed

Keywords: Methodology, Statistics

Presenting author's disclosure statement:
Organization/institution whose products or services will be discussed: Constella Health Sciences
I do not have any significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.