7.4 Summary

To summarize, we have seen interesting results for contrasts in the context of \(2 \times 2\) designs, where depending on the contrast coding, the factors estimated nested effects (treatment contrasts) or main effects (sum contrasts). We also saw that it is possible to code contrasts for a \(2 \times 2\) design, by creating one large factor comprising all design cells, and by specifying all effects of interest in one large contrast matrix. In designs with one factor and one covariate it is possible to control group differences for differences in the covariate (ANCOVA), or to test whether regression slopes are parallel in different experimental conditions. Finally, contrast coding in generalized linear models works the same way as in the standard linear model.