0

I'm running a linear regression with just one IV. When I run the regression with a constant using statsmodels I get a Multi-Collinearity warning. After searching on here I can see it could be a scaling issue. the coefficents are

constant:  14.0202      
IV:     -0.0123

My question is how do I deal with scaling, is it as simple as scaling the IV so for example scaling it up by 100

df['IV'] = df['IV'] * 1000

or would it be better no normalise the IV

(x-mean(x))/stdev(x)
em456
  • 359
  • 2
  • 11
  • Which scaling is used is not very important numerically, as long as values are of similar magnitudes as the constant. If we want to interpret coefficients, then it is often better to find a meaningful scaling, e.g. in 1000 or 1 Million Dollar units, or hours instead of seconds dependent on the data. – Josef Jun 26 '20 at 02:32
  • i tried both methods and using normalisation improves the rsquared – em456 Jun 26 '20 at 11:56

0 Answers0