1
$\begingroup$

I'm planning to do logistic regression with my dependent variable as either with injury or no injury with one of my independent variables as average computer use.

I have attached a sample distribution of the average computer use - majority of data points are close in the .3-.5 hour range then another peaked in the 2.9-3.1 range.

Questions:

  1. Do I need to transform this data first before I run the logistic regression? I noticed that my other independent variables also exhibit this distribution shape.
  2. There seems to be a lot of outliers after the 6 hour range - how should I treat them?

Any suggestions or comments will be appreciated.

data distribution

$\endgroup$
1
$\begingroup$

You don't need to transform it for statistical reasons. Logistic regression does not make any assumptions about the distribution of independent variables (neither does linear regression).

Whether you ought to transform it is another matter and depends on what you are trying to find out. Categorizing continuous variables is almost always a bad idea. I suggest using a spline of the IV and seeing if there are nonlinearities in the relationship with the logit.

As to the outliers, you have to first figure out why you have those and what form of the relationship you are interested in. If you take the log of time then you change the relationship from additive to multiplicative. Is that what you want? Is that reasonable?

$\endgroup$
0
$\begingroup$

Having a normal distribution isn't a prior assumption when dealing with logistic regressions. So you don't necessarily have to transform anything.

And depending on how the outliers affect your results or assumptions it would be alright to drop them.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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