I am working on determining factors having significant impact on credit repayment performance of borrowers of a bank for which i wish to run a binary logistic regression model. Dependent variable is repayment status default (1) non default (0). But out of 14 variables only four are continuous while others are categorical. Will there be any problem having so many categorical variables?
One variable included in explanatory variables is education level of borrowers ED The responses were collected as number of borrowers under matriculate, matriculate, graduate, post graduate and others. When i wanted to convert them into categorical i made it like this
ÈD1=1 for under matriculate and 0 otherwise.
ED2=1 for matriculate and 0 otherwise.
ED 3 =1 for graduate and 0 otherwise.
Likewise for otherwise. So i got in place of one variable four new variables. I converted many such variables like this
Is it permitted?
Since you have many categorical variables, creating too many dummy variables can lead to multi-colinearity. So you might want to check the correlations once you finish creating all the dummies
All this would work if you have a large enough data set with a fair distribution of defaulters. Suppose, if you have a low event rate, apparently increasing the number of predictors may have some issues.