Extended caution indices (ECI) specify the degree of confidence that can be placed in an individual's test score by analyzing patterns of item response. Among the most promising of such indices are the standardized ECIs. Contrary to the literature, several instances were found, in a previous study, of nonnormal distributions of ECIs with samples taken from four different grade levels. Several items from the study failed to fit the Rasch model. The current study examined whether ECI distributions approach normality if worst fitting items are dropped from analyses. The sample consisted of Comprehensive Tests of Basic Skills (CTBS) data from 700 third grade students and 737 seventh grade students. Seven out of eight distributions deviated from normality. Removal of the worst fitting items did not return the distributions to normal. The percentage of students at the extreme ends of the curves remained constant despite the inclusion or deletion of test items. Distributions were most stable at upper grade levels. Given the robustness of these indices, they may still prove useful in identifying mismeasured students. The results of this study posed more questions than answers. Further reesearch is needed, and replication using different measures of item misfit is recommended. (Author/MDE)