I am running a multiple regression in which the dependent variable and both predictor variables are continuous, numeric and positive.
lm(DV ~ PV1 + PV2, mydata)
Both PV1 and PV2 are coming as significant using raw values and also after scaling:
lm(DV ~ scale(PV1) + scale(PV2), mydata)
I want to compare the role of PV1 vs PV2 in determining level of DV. The coefficients are 2 and -4 (using scaled values). I conclude that for every 1 sd increase in PV1, the DV increases by 2 and for every 1 sd increase in PV2, the DV reduces by 4.
However, in When conducting multiple regression, when should you center your predictor variables & when should you standardize them? , it is mentioned that this is not a valid approach to "interpret the betas as measures of importance" (see answer by @gung).
How can I determine relative contribution of predictor variables in determining dependent variable?
