In Example 3, the group means and the population base rates are equivalent, but Group 1’s variability is 10 times larger than Group 2’s. The researcher determined that correct classification of Group 1 individuals was five times more useful than correct classification of Group 2 individuals (utility values of +5 and +1, respectively). The researcher weighted incorrect classification of Group 1 and Group 2 as equivalent and not very serious (both -1). There are two cut-points such that scores greater than 8.44 and less than 11.56 should be classified as Group 2; otherwise Group 1. Using these criteria, 88% of the Group 1 cases and 88% of the Group 2 cases are expected to be correctly classified. Because of the relatively very small variance for Group 2, cases that are very near the mean for Group 2 are highly likely to be from Group 2, while cases farther from the mean in either direction are much less likely to be from Group 2. Thus, overall classification accuracy is 88% even though the two population means are identical.
Contact: Michael Healy or Dale Berger
Last Modified: 9/24/2011
Return to the WISE homepage