1

Various people have advocated centering independent variables in regression, on the grounds that the intercept then refers to a sensible point (the mean of each IV) rather than the often impossible case that all the IVs are 0.

But does this really help?

First, the intercept is often not of interest.

Second, the mean of the variables depends on the particular data set.

Third, that means that, if you want to interpret the intercept, you need to know the mean of each IV.

So, I'd say that this reason for centering is not a good one. (There are some other reasons for centering which I am not discussing here).

Am I missing something?

Peter Flom
  • 94,055
  • 35
  • 143
  • 276
  • Re "the intercept is often not of interest." This is a curious statement because the question is predicated on the fact that "the intercept" is not a well-defined quantity: it depends on the origins of the units used to represent all the variables. In effect, I believe you are using "the intercept" in this statement to refer to the intercept *before* variables are centered but then you proceed to use it to criticize the intercept as computed *after* variables are centered! Your second objection applies *mutatis mutandis* to all statistical estimators and so is even more curious... . – whuber May 14 '19 at 14:28

0 Answers0