Question & Answer
Question
I am running the Ordinal Regression procedure (PLUM command) in SPSS/PASW Statistics. I would like to enter a block of predictors, such as a set of main effects, followed b y a second set of predictors, such as the interactions among the first set of predictors. The predictors in the first block would be contained in the second model, so that the first model would be nested in the second model. I would also like a test of whether the addition of the second block of predictors improves the fit of the model over the first block alone. This analysis would be similar to hierarchical linear regression, as described in Technote 1476749, in which multiple /METHOD ENTER subcommands are used to add blocks of new variables. Can such a hierarchical analysis be performed with the Ordinal Regression procedure?
Answer
In the Ordinal Regression procedure, the model is specified in the Location dialog or the /LOCATION subcommand. Unlike the /METHOD subcommand in Linear Regression, only one /LOCATION subcommand is allowed in the PLUM command and there is no mechanism in the subcommand or Location dialog to specify sequential blocks of predictors.
To test the improvement of the model fit by adding the second block, you need to run Ordinal Regression once for each block, adding the next block of predictors in each subsequent run. The -2 Log Likelihood (LL) is printed in the "Model Fitting Information" table in the output for each run. You could calculate the change in -2 LL and the change in degrees of freedom (DF) across subsequent models and compare that difference to a chi-square distribution with that DF to see if the change is larger than would be expected by chance. Here is an example of a pair of PLUM commands to test the improvement from adding the interactions among categorical predictors to the model.
PLUM chist BY numcred othnstal housng WITH age duration
/PRINT=FIT PARAMETER SUMMARY.
PLUM chist BY numcred othnstal housng WITH age duration
/LOCATION=numcred othnstal housng age duration housng*numcred housng*othnstal numcred*othnstal
/PRINT=FIT PARAMETER SUMMARY.
A feature enhancement request has been filed with SPSS Development to request that hierarchical modeling be allowed in the Ordinal Regression procedure.
Related Information
Historical Number
88910
Was this topic helpful?
Document Information
Modified date:
16 April 2020
UID
swg21486678