Russell Lenth - Colloquium Speaker
It is a common experience that a response transformation will often make the distribution of residuals more homogeneous, and/or make it possible to fit a more parsimonious model (e.g., by removing the need for including certain interactions). Transformations also play a key role as link functions in generalized linear models. The catch is that your client may or may not be comfortable discussing how some treatment affects, say, the cube root of an achievement score. This talk covers some ways of easing communication between consultant and client regarding transformations and link functions, with emphasis on features of the speaker’s R package, **emmeans**. We also will discuss some subtleties, technical traps, and tricks.