Data considerations for Overlaid Contour Plot

To ensure that your results are valid, consider the following guidelines when you perform the analysis and interpret your results.

First, perform an analysis that fits and stores a model

Overlaid contour plots use the most recent stored model that you fit for a response. If you did not perform an analysis that automatically stores a model, you do not have a model to use. For more information, go to Stored model overview.

The model must contain at least two continuous variables

If the stored model does not contain at least two continuous variables, then this plot is not available. If you fit your model using Fit General Linear Model, then you can only include covariates on this plot. However, if you fit your model using either Analyze Factorial Design or Analyze Variability, then you can graph the continuous factors but not the covariates.

Note

A mixture design must have at least three component variables or at least two numeric process variables to create a plot.

Verify that the models meet the assumptions of the analysis

If the stored models do not meet the assumptions of the original analysis, the overlaid contour plot might be inaccurate. For more information, click the original analysis below, and then go to the "Data considerations" topic and the "Key results" topic.

• Analyze Factorial Design
• Analyze Binary Response for Factorial Design
• Analyze Definitive Screening Design
• Analyze Binary Response for Definitive Screening Design
• Analyze Variability
• Analyze Response Surface Design
• Analyze Binary Response for Response Surface Design
• Fit General Linear Model
• Fit Regression Model
• Fit Binary Logistic Model
• Fit Poisson Model
• Analyze Mixture Design

Variable settings should be similar to the data that you used to fit the model

If you use variable settings that are not within the range of the data that you used to fit the model, the fitted values can be misleading. Also, you should use combinations of variable settings that are similar to the combinations that you used to fit the model. If you identify points of interest, you can use Predict to determine whether these points are unusual compared to the data that you used to fit the model. Predict also includes prediction intervals in the output, which you can use to determine the precision of the predictions.

Note

To generate predictions for a mixture design, choose Stat > DOE > Mixture > Analyze Mixture Design > Prediction.