Are there any tips as to how to diagnose the residuals of your regression model. For example, I have a residual plot that is both skewed and non-linear.
When you have many predictors, is there any quick way to pinpoint which variables need to be transformed or interaction terms to create so that you can try to make the residuals a little bit more normal and linear? I've tried transforming some but the shape of the residual plot looks the same, suggesting I am not really transforming the right variables.
I can produce correlation plot matrix - but not sure where to start looking? Or would partial residual plots do the trick? Now I know I can build another model, but I am hoping to stick with OLS before I go into something more complicated.
Unfortunately, this one doesn't have an answer: How to assess variable to transform in multiple regression?
