I am doing an analysis of interaction effects, using PLUM in SPSS.
My question is if the binary variables, that i have constructed to analyze the interaction effects, should be introduced as COVARIATE (i.e. using WITH) or not (i.e. using BY).
While I run the models with both, results are the same, except the polarity is inverted while inputing them as COVARIATEs
Any help will be very much appreciated
best regards
Rafael Lemaitre please see attached for clarity.
Good luck.
Hello Rafael Lemaitre. As you note, the model fit is the same either way, with only the sign of the coefficient differing. IIRC, PLUM does not allow you to specify which category is the reference category for a "factor" variable (i.e., variable that follows keyword BY). But you can fit the same ordinal logit model via GENLIN (using DISTRIBUTION=MULTINOMIAL LINK=CUMLOGIT on the MODEL sub-command). Another advantage of GENLIN is that you can use EMMEANS sub-commands to generate fitted values at desired combinations of the explanatory variables, which is useful when you're trying to illustrate the nature of an interaction. HTH.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS