# Ordinal regression, categorical variables, and “step” function

I am doing an ordinal regression analysis using "polr" function. I got a result of the regression analysis and continued to use "step" function to find the final prediction model. As all my variables are categorical, the results I got after using "step" function is like this:

           Value Std. Error   t value      p value
Age_M2 0.1030895  0.5575150 0.1849090 0.8533004123
Age_M3 1.2548639  0.6122917 2.0494545 0.0404176893
Age_M4 0.8041738  0.5372956 1.4967064 0.1344696773
Age_M5 1.5132697  0.5516788 2.7430268 0.0060875712
Edu_M2 0.9388024  0.4126999 2.2747821 0.0229190073
0|1    0.5253434  0.5479828 0.9586859 0.3377170257
1|2    2.0368711  0.5999590 3.3950171 0.0006862431


I just would like to confirm that should the insignificant levels of categorical variables (like Age_M2, Age_M4) be removed from the final prediction model? So far I haven't seen any references regarding this as the variables are categorical, which might be a rare case.

Any help will be appreciated!

• Can you please fix up the formatting. It is very hard to read. You can do it by editting your question. – André.B Nov 11 '19 at 22:31
• Edited. Thanks! – Song Nov 12 '19 at 4:57