Treatment of Outliers in Cognitive and Psychomotor Test Data

摘要

Many statistical tests and review articles have pointed out the possible adverseeffects that outliers can have on model parameter estimates, and have suggested several methods for detecting and treating outliers. In the present study, the effects of two different methods for treating outliers in aptitude tests (data deletion and data transformation) were investigated at the item and total-score level on the internal consistency and criterion-related validity of six computerized tests being evaluated by the U.S. Air Force. Over 2,000 pilot training candidates were tested. Results indicated that neither outlier treatment method at either level of analysis had significant effects on tests' psychometric characteristics. Possible reasons for these findings include the rarity with which outliers actually occur, and the robustness of linear modeling methods. Computer-based tests, Test reliability, Outliers analysis, Test validity.