Original post on stackoverflow: https://stackoverflow.com/questions/28773153/how-to-do-regression-model-selection-if-dummy-variables-are-involved

I am trying to do a logistic regression analysis in R with two continuous explanatory variables and six other explanatory categorical variables, and find a good regression model to do predictions. When I do step-wise model selections, there are always some levels of certain categorical variables identified as insignificant. I am just wondering how should I deal with this situation. Should I simply drop these levels, or I should force the program to keep all levels of the categorical variables and try to drop the relatively insignificant variables?

Thanks a lot!