weighted grand mean with effects coded MLR? showing the last dummy (w/o collinearity)?
Mar 11, 2020 8:43 AM(126 views)
A. Is there relatively simply way to use a grand mean weighted by the sample size of each level of a categorical predictor (n=5), instead of using the unweighted grand mean? I would like to stay with the default, effect coding of the intercept and coefficients, but I have a feeling a reviewer will have a problem with my just stating "unweighted grand mean were used as the references..."
B. Second, is it OK to include the last, redundant level of the categorical predictor in my report if there does not appear to be an issue with collinearity? Is it possible to not fall for the "dummy trap" in doing so? If I can include the variable in my report, can I still report the original effects summary (i.e., R-square; adj; etc...) or would its inclusion affect this?