This is a good refresh of interaction tests. A few interesting points include
- Given a continous effect and a binary effect (say it has a coefficient of b) and their interaction, the interpretation of b is the difference of the intercept of the regression line in each level of the binary effect. As we scale / shift the continuous effect arbituarily, its value / interpretation changes accordingly
- multicolinearity introduced by the interaction team is necessary to reflect the model uncertainty
- an example to visualize the impact of the interaction between 2 continuous variables.
Some points that I am not sure about its value
- centering is useless statistically (but may be relevant for the optimization procedure)
- in sas area, we talked about contrast a lot, which address a lot author's concern.
