GDA Models Results - Profiler Tab

Select the Profiler tab of the GDA Models Results dialog to access options to compute and display the predicted responses for each (dummy-coded) dependent variable (i.e., for each class specified in the dependent variable), as well as desirability profiles for combinations of multiple (dummy-) coded dependent variables. For a detailed explanation of the options on this tab, please see GDA - Response/Desirability Profiler. A detailed general discussion of desirability profiling and response optimization is provided in the context of the Experimental Design module, in the topic Profiling Predicted Responses and Response Desirability.

The implementation of these options in GDA are identical to those described in the Desirability topic (and also offered on the Profiler tab of the General Linear Models (GLM) and General Regression Models (GRM) modules), with the important exceptions that (1) the implementation in the Experimental Design module will also properly handle (constrained) mixture designs (in GDA, the Profiler tab is not available if there are mixture variables in the model), and (2) in GDA you can select to profile posterior classification probabilities for the classes (groups) specified in the categorical dependent variable (Using the options in the Values to profile box which is found on the larger GDA Models Results dialog - select the More results button if you are on the smaller GDA Models Results dialog).

Note: profiling regression-like predicted responses. As described in the Introductory Overview, the GDA module applies the general linear model to the discriminant function analysis problem, after (internally) recoding the class (group) memberships recorded in the categorical dependent variable into multiple (dummy-) coded dependent variables. There are as many dependent variables (created by the program) available for profiling as there are classes (or groups) specified in the categorical dependent variable. By selecting only a subset of those variables (i.e., classes or groups) for profiling, or by combining them in a particular manner into the desirability score, you can selectively search for the settings of the predictor variables and effects that maximize the scores for particular (coded) dependent variables, and hence the probability that a case belongs to one or more particular classes, as defined by the user.

Note: profiling posterior probabilities. The simple regression-like predicted responses assume that the dependent variable values of interest are continuous and unrestricted in range; that is, however, not the case. To reiterate, and as described in the Introductory Overview, the (internally) coded dependent variable values only have values of 1 and 0, depending on whether or not the respective case belongs to the respective class (group, represented by the respective coded dependent variable). Alternatively, you can choose the Posterior probabilities in the Values to profile box (available on the Profiler tab on the larger GDA Models Results dialog - select the More results button if you are on the smaller GDA Models Results dialog). These probabilities will always be in the range between 0 and 1, and thus are more readily interpretable (i.e., as probabilities). By choosing to profile posterior probabilities, and by including only certain (coded) variables (representing only certain classes or groups), or by combining posterior probabilities into desirability scores in a complex manner, STATISTICA General Discriminant Analysis (GDA) provides a unique and very powerful tool for finding the combination of predictor variable (or effect) values that maximize the probability of membership in a particular class (group) or set of classes.

For a detailed explanation of the options on this tab, see GDA - Response/Desirability Profiler. See also GDA - Index.