MaxDiff Platform Options

(Available only if Hierarchical Bayes is selected on the launch window.) Shows non-Firth maximum likelihood estimates and standard errors for the coefficients of model terms. These estimates are used as starting values for the Hierarchical Bayes algorithm.

Joint Factor Tests

(Not available if Hierarchical Bayes is selected on the launch window.) Tests each factor in the model by constructing a likelihood ratio test for all the effects involving that factor. For more information about Joint Factor Tests, see Joint Factor Tests in the Fitting Linear Models book.

Confidence Intervals

If Hierarchical Bayes was not selected, shows a confidence interval for each parameter in the Parameter Estimates report.

If you selected Hierarchical Bayes, the confidence intervals appear in the Bayesian Parameter Estimates report. The intervals are constructed assuming a normal distribution and are based on the Posterior Mean and Posterior Std Dev.

Correlation of Estimates

If Hierarchical Bayes was not selected, shows the correlations between the maximum likelihood parameter estimates.

If you selected Hierarchical Bayes, shows the correlation matrix for the posterior means of the parameter estimates. The correlations are calculated from the iterations after burn-in. The posterior means from each iteration after burn-in are treated as if they are columns in a data table. The Correlation of Estimates table is obtained by calculating the correlation matrix for these columns.

Comparisons

Performs comparisons between specific alternative choice profiles. Enables you to select factor values and the values that you want to compare. You can compare specific configurations, including comparing all settings on the left or right by selecting the Any check boxes. Using Any does not compare all combinations across features, but rather all combinations of comparisons, one feature at a time, using the left settings as the settings for the other factors. See Comparisons Report.

All Levels Comparison Report

Shows the All Levels Comparison Report, which contains a table with information on all pairwise comparisons of profiles. If you are modeling subject effects, you must specify a combination of subject effects and the table is specific to that combination of subject effects. Each cell of the table shows the difference in utilities for the row level and column level, the standard error of the difference, and a Wald p-value for a test of no difference.

Caution: The p-values are not corrected for multiple comparisons. Use these results as a guide.

The Wald p-values are colored. A saturated blue (respectively, red) color indicates that the Difference (Row - Column) is negative (respectively positive). The intensity of the red and blue coloring indicates the degree of significance, with a highly saturated red or blue meaning that the difference is highly significant.

Save Utility Formula

Creates a new data table that contains a formula column for utility. The new data table contains a row for each subject and profile combination, and columns for the profiles and the subject effects.

Save Gradients by Subject

Constructs a new table that has a row for each subject containing the average (Hessian-scaled-gradient) steps for the likelihood function on each parameter. This corresponds to using a Lagrangian multiplier test for separating that subject from the remaining subjects. These values can later be clustered, using the built-in-script, to indicate unique market segments represented in the data. For more details, see Example of Segmentation in Choice Models.

Save Subject Estimates