1

I am having a hard time to understand whether the independent variables in a logistic regression have to have some degrees of correlation. I came across a report that mentioned that in SEM the IVs are assumed to be correlated. Is it the case for logistic and linear regressions too?

I am asking this because I see that some authors show the correlation matrix of the IVs before a logistic regression while others do not. I use logistic regression in most of my studies so just want to be sure if I have to do so.

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
Ghose Bishwajit
  • 25
  • 4

1 Answers1

0

Logistic regression assumes that there is little to no multicollinearity between independent variables, which essentially means IVs shouldn't be too highly correlated with each other. I'd recommend checking out this thread on multicollinearity in multivariate regressors.

RCulkin
  • 9
  • 1
  • 3
    Welcome to CV. Could you provide some reference or authority for this assumption? I have not found it in standard textbooks like Hosmer & Lemeshow, for instance. Indeed, standard software handles perfect collinearity just fine. – whuber Aug 28 '18 at 22:24
  • 3
    High correlations between IVs surely make their interpretation harder, but it doesn't change the underlying model at all. You can whiten your IVs beforehand, and you'll get a virtually identical model (with coefficients appropriately rotated to match your transformation). So, in other words, you can keep all correlated IVs, it makes no difference. – Firebug Aug 28 '18 at 22:34
  • 1
    Thanks to everyone for the clarifications. I got the answer I was looking for and will greatly appreciate if you could provide a citable source to the assumption. Some reviewers are very strict about the source of citations. – Ghose Bishwajit Sep 01 '18 at 22:16