I have two queries concerning ordinal regression (PLUM procedure) and its use.
1. Is there a multi-step multiple ordinal regression procedure that is parallel to hierarchical multiple linear regression? My copy of SPSS (Ver. 16.0) does not seem to offer any such thing. Is it due to inherent deficiency of this regression or just a missing feature in the program?
2. What sets the threshold, with regards to the number of ordinal values in survey items when used as dependent measures, from which linear regression is no longer a viable option and ordinal regression is mandatory? Obviously, nearly any survey item is in practice an ordinal measure that reflects a latent scale concept, but it is more than quite common to use linear regression with seven optional answers and even with five (Likert style items); however, what about four optional answers? And how about three? Can the fact that the optional answers create a perfect arithmetical series have an impact here? Your advice, as well as reference to relevant literature (preferably from social science), would be greeted with utmost gratefulness.
Sincerely,
Amir Hetsroni
|