|Friday, February 22|
|CS12 Theme 4: Software and Graphics #2||
Fri, Feb 22, 3:15 PM - 4:45 PM
Generalized Linear Mixed Model Based Power and Precision Analysis as a Tool for Planning Research Designs (302536)
Ramon Littell and Ralph O’Brien presented power and precision analyses using linear model theory via SAS® PROC GLM in the 1970s. Stroup (1999, 2002) extended the approach to linear mixed model theory using SAS PROC MIXED. The mixed model methods appear in Littell, et al. SAS for Mixed Models, 2nd edition. More recently, Stroup has extended these methods even further to include planning for studies that will use mixed models with non-normal data – i.e. generalized linear mixed model theory implemented with SAS PROC GLIMMIX. These methods appear in Gbur, et al. Analysis of Generalized Linear Mixed Models for Plant and Natural Resource Sciences (2012) and in Stroup’s forthcoming Generalized Linear Mixed Models – Modern Concepts, Theory and Applications (to appear September, 2012). Mixed model-based power and precision analysis allow planning research designs in the presence of serial correlation (e.g. for repeated measures designs), spatial correlation, and to compare alternative multi-level designs (e.g. split-plot, clustered or incomplete block designs). Most “conventional wisdom” concerning design presumes normally-distributed data. While the underlying concepts do not change for non-normal data, the way they play out, e,g, for binomial, categorical or count data, can be dramatically different – plans based on power analysis that does not take GLMM thinking into account can result in catastrophically inappropriate sample size assessment. This presentation will include GLMM inference basics relevant to power and precision analysis and work through examples of planning multi-level designs with Gaussian and non-Gaussian primary response variables.