Can you use logistic regression if you have more than two options for the dependent variable. Instead of just a yes/no option I could have more than 2 options. I have 20+ predictor variables.
Asked
Active
Viewed 178 times
1
-
2Yes you can! Google for multinomial logistic regression. – Hossein Mar 23 '17 at 20:06
-
@ilanman It's possible, but I expect that the OP meant exactly what the question said. – Glen_b Mar 23 '17 at 20:08
-
@Alex is your response ordered categories (e.g. high, medium, low) or nominal categories (red, blue, yellow)? – Glen_b Mar 23 '17 at 20:10
-
1Glen_b♦ my response would be nominal – Alex Mar 23 '17 at 20:19
1 Answers
3
If you have more then two categories it's not logistic regression, but multinomial logistic regression (also known as multinomial regression). While logistic regression model is defined in terms of conditional odds of one category as compared to another, in multinomial model you compare $K-1$ categories to the $K$th category that serves as a "default" category. If your categories are ordered, you will use ordinal regression, that is a similar model but for ordered categories.
See the Interpreting exp(B) in multinomial logistic regression thread for extended discussion of multinomial model.
-
Just to correct what Tim indicated, with more than two categorical variables it is actually Multinomial Logistic Regression not Multinomial Regression as confirmed by Tim's own reference at Wikipedia. – Sympa Mar 23 '17 at 20:22
-
1@Sympa the name "multinomial regression" is also used, you can easily google it. – Tim Mar 23 '17 at 20:34
-
Tim couldn't one advance that Logistic Regression works just fine with more than two variables. And, that in such situation it is called either Multinomial Regression or Multinomial Logistic Regression. – Sympa Mar 24 '17 at 05:36