This guide provides instructions on analyzing fractional factorial experiments (experiments where not every possible treatment combination in a full factorial design is run) using the Fit Model platform.
Specify the Model and Analyze
- Experiments designed in JMP® will have a Model script saved to the data table. The model specification window will be populated with this model. To generate the model manually:
- Select Analyze > Fit Model.
- Click on the response under Select Columns, and click Y.
- Select the factors of interest, and click Add (under Construct Model Effects).
- To add specific interaction terms, select the variables under Select Columns and click Cross. Click Run to fit the specified model. In this experiment, all but two of the 2-way interactions are confounded with main effects.
- Feed Rate*Catalyst = Stir Rate*Temperature
- Feed Rate*Stir Rate = Catalyst*Temperature, and
One way to view the level of confounding of the factors is to examing the Color Map of Correlations in the Evaluate Design report. Here you’ll see the counding between the 2-way interations and main effects as well as the confounding
between the two 2-way interactions listed above.
- Click Run to fit your specified model
Note: If your design is saturated (there are no error degrees of freedom) as the case here, the Effects Summary, Parameter Estimates, and Effect Tests tables with not show test statitics and p=values. Select Estimates > Sorted Estimates from the top red triangle to see p-values based on Lenth Pseudo Standard Error (PSE, an estimate of residual standard error).
- Reduce the model (removing terms) as desired. Here we removed the 2-way interaction with the highest pseudo p-value (Feed Rate*Catalyst = Stir Rate*Temperature). Upon doing that, there is now a single degree of freedom to estimate experimental error. Test statistics and p-values will now appear in the Effect Summary table. Recall that all the main effects are counfounded with 2-way interactions and that the Feed Rate * Stir Rate interaction is confounded with Catalyst*Temperature. Thus it is unknown if it’s the main effects or the 2-way interactions they are each confounded with that have a significant effect on the response. Similarly, it is also unknown which of the two 2-way interactions are impacting the response.
Visit Design of Experiments Guide in JMP Help to learn more.