Constrained predictors


#1

Hi All,

I am in a situation where I have two predictors X_1 and X_2 which are quite related. i.e X_1=< X_2 by nature. Is there a way to incorporate this information in to my regression model?


#2

That should be reflected by conditioning on both X_1 and X_2.


#3

Can the effect of X_1 at higher and lower values interpreted straightforwardly?


#4

Yes, if X_2 is not a consequence of X_1.


#5

One approach is to rescale the predictors to X[1] and X[2] - X[1]. It changes the values you get for coefficients, so the priors mean slightly different things, but it can have a more natural interpretation.