To Run the Analysis

Figure 1. Nonlinear Canonical Correlation Analysis dialog box

Recall the Nonlinear Canonical Correlation Analysis dialog box and navigate to the first set.
Select age and click Define Range and Scale.
Figure 2. Nonlinear Canonical Correlation Analysis Define Range and Scale dialog box
In the Define Range and Scale dialog box, select Single nominal as the scaling range.
Click Continue.
In the Nonlinear Canonical Correlation Analysis dialog box, click OK.

The eigenvalues for a two-dimensional solution are 0.806 and 0.757, respectively, with a total fit of 1.564.

Figure 3. Eigenvalues for the two-dimensional solution

The multiple fit and single fit tables show that Age in years is still a highly discriminating variable, as evidenced by the sum of the multiple fit values. In contrast to the earlier results, however, examination of the single fit values reveals the discrimination to be almost entirely along the second dimension.

Table with variables grouped by set in the rows and multiple fit, single fit, and single loss grouped by dimension in the columns — Figure 4. Partitioning fit and loss

Turn to the transformation plot for Age in years. The quantifications for a nominal variable are unrestricted, so the nondecreasing trend that was displayed when Age in years was treated ordinally is no longer present. There is a decreasing trend until the age of 40 and an increasing trend thereafter, corresponding to a U-shaped (quadratic) relationship. The two older categories still receive similar scores, and subsequent analyses may involve combining these categories.

Figure 5. Transformation plot for Age in years (nominal)

The transformation plot for Neighborhood preference is shown here. Treating Age in years as nominal does not affect the quantifications for Neighborhood preference to any significant degree. The middle category receives the smallest quantification, with the extreme categories receiving large positive values.

Figure 6. Transformation plot for Neighborhood preference (age nominal)

A change is found in the transformation plot for Newspaper read most often. Previously, an increasing trend was present in the quantifications, possibly suggesting an ordinal treatment for this variable. However, treating Age in years as nominal removes this trend from the news quantifications.

Figure 7. Transformation plot for Newspaper read most often (age nominal)

This plot is the centroid plot for Age in years. Notice that the categories do not fall in chronological order along the line joining the projected centroids. The 20–25 group is situated in the middle rather than at the end. The spread of the categories is much improved over the ordinal counterpart that was presented previously.

Sccatterplot of actual and projected centroids for age — Figure 8. Centroids and projected centroids for Age in years (nominal)

Interpretation of the younger age groups is now possible from the centroid plot. The Volkskrant and NRC categories are also further apart than in the previous analysis, allowing for separate interpretations of each. The groups between the ages of 26 and 45 read the Volkskrant and prefer country living. The 20–25 and 56–60 age groups read the NRC; the former group prefers to live in a town, and the latter group prefers country living. The oldest groups read the Telegraaf and prefer village living.

Interpretation of the other direction (Music preferred, Marital status, and Pets owned) is basically unchanged from the previous analysis. The only obvious difference is that people with a marital status of Other have either cats or no pets.

Scatterplot of centroids with Dimension 2 on the vertical axis and dimension 1 on the horizontal axis. Each point is the category of a variable. — Figure 9. Centroids labeled by variables (age nominal)