Weight of scaled vs unscaled features
Since we leave the dummy features unscaled, I wonder if their magnitude on the dependent varible is still comparable with features that have been scaled. Aren't they on different scales now (the dummy features vs the scaled features), and if so, wouldn't this affect the interpretability of the magnitude of these features ( so u can't say reason 1,2,3 and 4 are the most pronounced features, as their coefficients cant be compared to those of the scaled groups).
That's how I think. Or am I missing something? Looking forward to your answer.
Hi Kai N!
Thanks for reaching out.
Logistic regression models (such as the one we use in our analysis) are insenstive to the variables' magnitude. Then, in our case, we have transformed a categorical variable into several dummy variables. While this is important for the interpretability of the independent variables in the model, we don't need to scale the dummies. The reason is that the dummy variables are all at the same distance from the mean (geometrically speaking, for instance), so there's nothing to really standardize from this. That's why we have scaled the continuous variables and have not scaled the dummies.
Hope this helps.