Choosing Model Terms

Given the Unstructured correlation matrix, should the model include an interaction effect? You can compare the main-effects model to one with an interaction by comparing the QICC values obtained by each model.

  1. Recall the Generalized Estimating Equations dialog box.
    Figure 1. Generalized Estimating Equations Repeated tab
    Generalized Estimating Equations Repeated tab
  2. On the Repeated tab, select Unstructured from the Structure drop-down list in the Working Correlation Matrix group.
  3. Click the Model tab.
    Figure 2. Generalized Estimating Equations Model tab
    Generalized Estimating Equations Model tab
  4. Select Interaction as the type of term to build.
  5. Add age*smoker as a model term by selecting both age and smoker from the Factors and Covariates list.
  6. Click OK.
Figure 3. Goodness of fit for model with interaction
Goodness of fit for model with interaction

The QICC for the model with the interaction is 1831.805, which is larger than the QICC of 1827.127 for the model with just the main effects. The main effects model appears to be "better" than one with the interaction.

Next