Mixed Model Analysis of Research Data

Many, if not most, studies focusing upon treatment or group factors of direct interest also involve design structure factors whose levels or elements (for example: animals, days, locations, schools) are random. That is, the levels of these factors are typically assumed to be randomly selected from a conceptually large population for the particular study such that they are not likely to be consistently re-used for subsequent repetitions of the study at another time or place.  Furthermore, the typical intent is that the inference space is broad; that is, the investigator wishes to infer upon treatment effects within the context of a larger population from which those design structure elements are considered to be randomly drawn.  When classical linear model methods based on ordinary least squares are used to analyze such studies, the additional uncertainty due to these random effects is ignored.  Consequently, the reported treatment standard errors of means are typically understated in some cases whereas inferences on treatment effect can be less efficient in other cases. Mixed model analyses will be presented in this workshop as a way to appropriately account for the uncertainty due to random effects in order to obtain correct standard errors and efficient tests of hypotheses. In particular, the utility of mixed model analysis to discern true experimental replication from pseudo-replication will be demonstrated as a critical determinant of research reproducibility. Several applications based on the use of SAS PROC MIXED will be presented with corresponding R code based on the lme4 package also provided.

There is a $10 non-refundable registration fee for the MSU community. If you are not affiliated with MSU please contact cstatinf@msu.edu or call (517) 353- 9288