Object Scores
After examining the model summary, you should look at the object scores. You can specify one or more variables to label the object scores plot. Each labeling variable produces a separate plot labeled with the values of that variable. We’ll take a look at the plot of object scores labeled by the variable object. This is just a case-identification variable and was not used in any computations.
The distance from an object to the origin reflects variation from the “average” response pattern. This average response pattern corresponds to the most frequent category for each variable. Objects with many characteristics corresponding to the most frequent categories lie near the origin. In contrast, objects with unique characteristics are located far from the origin.
Examining the plot, you see that the first dimension (the horizontal axis) discriminates the screws and bolts (which have threads) from the nails and tacks (which don’t have threads). This is easily seen on the plot since screws and bolts are on one end of the horizontal axis and tacks and nails are on the other. To a lesser extent, the first dimension also separates the bolts (which have flat bottoms) from all the others (which have sharp bottoms).
The second dimension (the vertical axis) seems to separate SCREW1 and NAIL6 from all other objects. What SCREW1 and NAIL6 have in common are their values on variable length—they are the longest objects in the data. Moreover, SCREW1 lies much farther from the origin than the other objects, suggesting that, taken as a whole, many of the characteristics of this object are not shared by the other objects.
The object scores plot is particularly useful for spotting outliers. SCREW1 might be considered an outlier. Later, we’ll consider what happens if you drop this object.