6

I am curious that since we don't have normality assumption of the independent variable in logistic regression, why do I see people using log transformation for independent variables in logistic regression?

user12331
  • 669
  • 1
  • 6
  • 10
  • This has come up before [here](http://stats.stackexchange.com/q/147612/7071). Also, we usually make assumptions about the error terms, not the dependent or independent variables. – dimitriy Jun 30 '15 at 06:27

1 Answers1

8

The reason for such transformations have nothing to do with their distribution. Instead, the reason has to do with the functional form of the effect. Say we want to know the effect of the number of publications on the probability of getting tenure. It is reasonable to believe that getting an extra publication when one has only 1 publication has more impact compared with getting an extra publication when one has already published 50 articles. The log transformation is one way to capture such a (testable) assumption of diminishing returns.

Nick Cox
  • 48,377
  • 8
  • 110
  • 156
Maarten Buis
  • 19,189
  • 29
  • 59