Good question and one I just encountered when I taught this class last semester! Let's say the true model is Y=b0 + bx + tau + E where x is the initial weight and b is the slope for the covariate. If we subtract x from both sides, as you suggest, we get (Y-x) = b0 + (b-1)x + tau + E. So the only way this removes the covariate effect is if b=1, which seems unlikely. The only reason way I can recommend doing an analysis on Y-x is if it is a meaningful quantity. Otherwise, it doesn't "remove" the covariate effect. It's better, in my opinion, to just work with the original Y value.
Very helpful and nice video, Thanks a lot
great tutorial, thanks
What is the difference of a covariate model compared to simply using the (End - Initial weight) as the dependent variable?
Good question and one I just encountered when I taught this class last semester! Let's say the true model is Y=b0 + bx + tau + E where x is the initial weight and b is the slope for the covariate. If we subtract x from both sides, as you suggest, we get (Y-x) = b0 + (b-1)x + tau + E. So the only way this removes the covariate effect is if b=1, which seems unlikely. The only reason way I can recommend doing an analysis on Y-x is if it is a meaningful quantity. Otherwise, it doesn't "remove" the covariate effect. It's better, in my opinion, to just work with the original Y value.
@@jonathanstallrich2143 OK, thank you so much for your reply, now I understand it better!