0

I have dataset at my disposal and I am suppose to run a logistic regression. What is the best way to determine which explanatory variables are useful and which are totally off? Is there any general rule or procedure to help me? Could you please share some guide for this? Is there any graphical way to determine this?

Thanks for help.

EDIT: I am looking for general and user-friendly way to do this, like some indications in plot of 2 variables. I am not looking for "technical" solution.

Lukas Tomek
  • 1
  • 1
  • 2
    Does this answer your question? [How to do logistic regression subset selection?](https://stats.stackexchange.com/questions/8303/how-to-do-logistic-regression-subset-selection) – Salih Dec 14 '19 at 12:36
  • I personally have found that the "leave-one-out" technique is helpful in weeding out the less useful predictors. This is where you iteratively re-run the regression, each time leaving out one predictor to determine the effect on the fitting results. If the "left-out" predictor seems to have little or no effect on the regression results, it can be considered as not useful. Sometimes this works well and can certainly be automated. – James Phillips Dec 14 '19 at 12:47
  • This is not really, what I am asking. I am more interested in less technical approach. For example, can we say something from graph of explanatory and explained variable if it is useful for regression? Or similar more general or demonstrative technique. – Lukas Tomek Dec 14 '19 at 13:42
  • 1
    [A lot](https://stats.stackexchange.com/search?q=logist*+regr*+featu*+answers%3A1+closed%3Ano+duplicate%3Ano+selec*) of posts here about this! – kjetil b halvorsen Dec 14 '19 at 14:28

0 Answers0