1

Im working on a linear regression with 3 independent variables (likert scales from 0-6) nothing out of the ordinary.

The IDs correlate from r=.3 to .58 with each other, which is a bit high although the VIF is still under 2. So usually I would assume that the multicollinieraty assumption would be ok here, but I mean-centered to IDs just to see what happens.

After centering, all IDs perfectly correlate with each other, which I dont know if thats an centering artifact or not. Also the VIF is now 1 which seems weird and the regression analysis throws out the 2nd and 3rd ID even when I forced SPSS (method enter) to consider them in the regression.

Any idea what I did wrong here?

Cheers

Sungod3k
  • 21
  • 2

1 Answers1

3

If you multiply random variables by constants and add other constants, this will not affect correlation between them. Below you can find a picture-proof for this claim. It comes from this answer by me that proves that the opposite is also not true.

enter image description here

So I guess, there must be some kind of bug in your code.

Tim
  • 108,699
  • 20
  • 212
  • 390
  • Yes, exactly. I checked a scatter plot all values are in a straight diagonal line, so there must be something wrong. The center operation is so simple that i cant figure out what wrong. Also the raw data still makes sense. Although im just noticing that my 2 of the 3 IDs have a disctly non zero mean... – Sungod3k Dec 22 '18 at 13:45
  • Got it. Was indeed a syntax error. It used only one ID to calculate all of the centered varibles. – Sungod3k Dec 22 '18 at 14:03
  • @Sungod3k so I guess you can mark it as solved. – Tim Dec 22 '18 at 14:14