I am working with a dataset of 335 categorical variables. The dependent variable is also categorical variable, as following: How satisfied are you with your life: 1.unhappy ... to 10. very happy. I looked up in some papers, and it is said that Oprobit or Ologit could be used if dependent variable is categorical. But if all are categorical, could I also used these two types or logistic regression? Plus, do I need to convert all these categorical variables into dummy variables before doing regression?
$\begingroup$
$\endgroup$
1
-
1$\begingroup$ The part of the question about stata is off-topic, so please reformulate and concentrate on the other part. There are a lot of similar questions already, so try a search. $\endgroup$ – kjetil b halvorsen♦ Jan 26 '19 at 14:40
Add a comment
|
$\begingroup$
$\endgroup$
1
Yes, you can still use ordinal logistic or probit with categorical variables and any decent statistics program will create dummy codes for you (most offering a choice of how to parameterize them). I don't know Stata (and questions about coding are off topic) but I am sure it can do this.
The problem I foresee is that you have so many IVs. This will make model building tricky unless you have a very large data set. This topic has been discussed a lot here and elsewhere - if you want a book, I highly recommend Frank Harrell's Regression Modeling Strategies.
answered Jan 27 '19 at 13:25
-
1$\begingroup$ Indeed with 335 categorical predictors you will likely end up with 1,000+ dummies - Even with large amount of data it will be difficult to build a "meaningful" model - If all 335 variables seem relevant a priori, then you might want to take a look at variables selection procedures such as Group LASSO - First I would try to simplify the variables as much as possible by collapsing adjacent categories $\endgroup$ – Umka Feb 5 '19 at 23:14
