8
$\begingroup$

I am using multilevel modeling in lme4 package in R. With the warning of rescaling variables, how can I find out WHICH variables need rescaling? All variables are centered. Dummy variables are coded as -1 and 1, the variable age has range of -34 and 31 but with square terms. Is it because of the square terms?

$\endgroup$
1
  • $\begingroup$ Yes, rescale all of them except the dummies. You could still rescale dummies too but it is not useful. $\endgroup$
    – Tom
    Feb 10, 2019 at 16:04

1 Answer 1

5
$\begingroup$

A reasonable approach is to do a "summary" of the dataset and look at the means and medians of the numerical variables, and also possibly their variances. If you see any that are orders of magnitude different from others, that will give you a clue.

You could just simply standardise all the numeric variables first and see if the model converges. If it does then iteratively de-scale them to see which one(s) is/are causing the problem(s).

$\endgroup$
1
  • $\begingroup$ are the coefficients of the model fitted with the rescaled variables "ready to use"? Must the new independent variables for prediction of new values (by the fitted model) be scaled as well? $\endgroup$ Nov 10, 2021 at 13:31

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.