Analysis of Combined Designs

The primary difference in analyzing combined models versus normal models is in the Fit Summary tables. An example is given below:

This table provides a matrix of p-values values for all combinations of mixture and process models. Find the combination where both models are significant A p-value less than 0.05 is a good rule of thumb.

A Lack of Fit column will appear only if there are replicates. The lack of fit should be insignificant (p-value > .10 is desirable).

Several combinations of mixture and process models may be significant. In that case, use the follow-up columns, Adjusted and Predicted R-squared to make the selection.

Stat-Ease will try to identify and suggest a model combination that is significant in both the mixture and process order, insignificant in the lack of fit, with high adjusted and predicted R-squared values. This is a good starting model, but can be tweaked and modified as desired. If two models are suggested, Stat-Ease will default to the larger model.

For more information on interpreting these statistics, see the Fit Summary and Analysis of Variance sections in response surface and mixture designs.

Click on the Model tab next. Stat-Ease uses the suggested model as the default. You may change it if you wish.