Dear all,
I am running Ordinal Regression with ordered response score (DV) of 1, 2 and 3. As I ran the ordinal regression, I got the WARNINGS. Therefore, I reduced to three level score from five level score (DV). But, again I got the WARNING message however, this time small cases (say 32%).
I tried to reduce the variables categories (both Dependent and Independent) into smaller sizes (for example education level reduced from four categories to two categories etc.) so that the response are high and that is how I managed to have less cells with zero frequencies. I got 32% as having zero frequencies from that of 50%.
I also eliminated the cases with missing information so that I have now about 800 cases. I have 10 independent variables in the model. Is it better idea to reduce the independent variables to run into the model? I did but again not success in fixing the WARNINGS and violated the ASSUMPTION OF PROPORTIONAL ODDS.
PLEASE suggest how to fix this WARNINGS. I have also found in the literature that violation of proportional odds assumption is common when the sample size is large (how large? mine is abt 800!) and explanatory variables is large (mine is 10). Is it possible to report the RESULT ignoring the WARNINGS and/or if this assumption is not met. Any suggestion and guidance is much appreciated.
Thanks,
- Badges
- Science topic
- Similar topics
- Geoscience
- Asia
- India