I am facing with an interesting problem here. I have a data set containing mixed nature of independent variables (some are nominal, some are ordinal, some are scale). The dependent variable is ordinal. I thought if I use CATREG in SPSS then I am going to get only one beta coefficient for each IV, not many beta coefficients for the different categories of the categorical variables as provided with an ordinal logistic regression. If an overall interpretation was possible, then that would be good for me. Now, when I am trying to run a CATREG with defining scaling levels other than 'nominal', the R-square value comes too small, about 0.202. But if I define nominal for all the dependent variables then the fit is improved and the R-square is about 0.60. Defining 'nominal' as a scaling level will make the analysis completely data oriented, as I know.
But when I divide the data into 2 parts to get 2 separate data sets for small firms and large firms (each set contains about 60 observations) and run CATREG with nominal level of scaling for all the IVs, then the R-square is almost 1 or exactly 1.
The coefficients table looks like this:
I don't know why this is happening. Can anyone tell me what to do? If I don't use nominal as the scaling level then for the two smaller data sets things may improve, but for the overall data set the R-square is too low and almost all the variables come insignificant.
Waiting for your kind suggestion. Can I use CATREG here? If I can, what should be the approach for this kind of problem?