Online Program

Comparing Linear Mixed Models Between Statistical Software (303004)

Keywords: Mixed models, software, SAS, Stata, R, SPSS

Linear mixed effects models are commonly used in the analysis of correlated observations. For those new to mixed models or a software package, it can be easy to use the provided defaults, but the default model assumptions and analysis options vary by statistical package. This poster will outline the defaults and options available for linear mixed models in SAS, Stata, R, and SPSS. Variations in data format, model estimation, method for determining the denominator degrees of freedom, and covariance structure for random effects and residuals are reviewed. Several scenarios will be explored, including random coefficient models, multilevel models, and residual structures. Examples will explore differences in the results and how to achieve the same results in all packages. Variation in default model assumptions can produce different results depending on the statistical package used, and variations in the available options also can affect the reproducibility of results in another statistical package.