Fixed or Random and the Effect of Zero Variance Components on Mixed Models Analyses
*George A. Milliken, Milliken Associates, Inc 

Keywords: mixed models, random effects, zero variance component estimate

The issue of an effect being designated as a fixed or random will be discussed in terms of types of factors and the inference space. The incorrect designation can result in making incorrect decisions or inferences. What happens with the analysis using proc mixed when the estimate of the variance component is zero? Several examples will be evaluated using simulation to determine the effect on the type I error when REML, REML with NOBOUND, and with/without KR degree of freedom option. RCB, Split-plot, and Repeated Measures designs are considered.