1

In a follow up to one of the business problem i have discussed here,
for one of the logistic regression model, i have 19 predictor variables of which around 8 are categorical with multiple categories.

Instead of including all in the model where one could correlate to another,
i am thinking of using a chi square test of independence at a given p value.
With this i would come to realization if there is any correlation among the categorical predictors.

If so i would include the correlated variables one by one in the model
and would used the significant one.
I have researched on this on SE where they have suggested to use chi-
square but unsure if my approach to drop these correlated var is justified.

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
Shivi Bhatia
  • 23
  • 1
  • 7
  • I would suggest adding the formula for the model you are using and more description of your data, perhaps even adding part of your data. What is your response variable? Are you testing independence of the predictor variables or are you interested in their effect on the response? If you describe more clearly what you are doing and include your research question this might be more answerable. – RAND Aug 15 '16 at 14:47
  • @ Please see the formula (fullmod= glm(survey ~ rep_score + prd+ und+ region, data = training, family = "binomial")). This are only 4 variables i have added so far and then continuously measuring the AIC, deviance and the null deviance i will add the other vars. here dep var is survey rating 0 or 1. I did search on how to add a snippet of my data here but could not find any help. Plz suggest as to how i can add data here so that you can relate to the data i am dealing with. I am as a comparison also using glm2 & BRGLM package. – Shivi Bhatia Aug 15 '16 at 16:36
  • To me this doesn't seem quite a duplicate, since it is asking not just whether we can correlate categorical variables, but goes further by asking whether we can use that for variable selection in regression. – Silverfish Jul 31 '17 at 19:11

0 Answers0