Mixed Effects Modeling in R

Earlier in the regression section, we mentioned that the Ultimatum Game dataset we were working with was actually a repeated design, which means that there was a correlated variance structure because all of the observations were nested within each subject. The standard solution to this problem in a GLM framework is to treat subject as a fixed effect and add (n-1) variables indicating a dummy code for each subject. This strategy will effectively model the mean for each subject removing inter-subject variability from the residual. However, this approach can be problematic because (1) it eats up your degrees of freedom, (2) it does not allow your model to generalize outside of your sample, and (3) it does not allow for modeling variations in individual subject coefficients. Mixed models provide an alternative approach that can address all of these limitations.