Talk:Research Methods/Mixed-model Design

This page should seriously be edited because it makes no attempt to explain the much more effective alternative approach to mixed designs: mixed-effect linear modeling. Repeated-measures ANOVA is a highly problematic statistical procedure due to the strict assumptions it requires, particularly sphericity of covariances. The example on the page has a Mauchley's test statistic that had a p-value of zero which is really bad. Yes, there are the HF corrections, but throwing a bunch of band-aids on a problematic procedure is way less ideal when there is a much more statistically powerful method available. With mixed-effect modeling, you don't even need to worry about sphericity. It is just like running a multiple regression with both the within and between subjects main effects/interactions as predictors, but there is one addition: a random intercept of within variables nested within each subject (which helps account for the problems that arise from repeated measures design where within subject variability can become an issue).