Mixture Surface Regression in GRM

Select Mixture surface regression on the General Regression Models Startup Panel - Quick tab to specify a mixture surface design (see also Ternary Plots, 3D). In the subsequent Quick specs dialog box, specify two or more continuous predictor variables, and one or more dependent variables.

Mixture designs are commonly used in industrial experimentation in cases when the predictor variable values are constrained to add to a constant sum; this would for example be the case when evaluating the relative amounts of various (mixture) components that make up a fish patty. By default STATISTICA will fit a second order polynomial mixture model to the data (see also Fitting options for ternary plots). Note that STATISTICA does not automatically check the validity of the mixture components, i.e., whether or not they actually sum to a constant value. Also, STATISTICA will not perform any rescaling of the original mixture components (e.g., to pseudo-components), but apply the respective surface model transformations to the raw data values of the continuous predictor variables. The analysis of mixture designs is also discussed in detail in Experimental Design - Mixture designs and triangular surfaces.

Combining mixture and process variables

STATISTICA General Regression Models (as well STATISTICA General Linear Models) also allows you to specify designs that include both mixture variables (which are constrained so that the values for each observation will sum to a constant) and process variables (which are not constrained). To analyze such designs, use option General linear models from the GRM Startup Panel - Quick tab. For example, you can analyze blocked mixture designs, or combine factorial designs with mixture surface designs.

For more information, refer to the Introductory Overview and GRM - Index.