An Automatic Online Calibration Design in Adaptive Testing

Abstract

An accurately calibrated item bank is essential for a valid computerized adaptive test. However, in some settings, such as occupational testing, there is limited access to test takers for calibration. As a result of the limited access to possible test takers, collecting data to accurately calibrate an item bank in an occupational setting is usually difficult. In such a setting, the item bank can be calibrated online in an operational setting. This study explored three possible automatic online calibration strategies, with the intent of calibrating items accurately while estimating ability precisely and fairly. That is, the item bank is calibrated in a situation where test takers are processed and the scores they obtain have consequences. A simulation study was used to identify the optimal calibration strategy. The outcome measure was the mean absolute error of the ability estimates of the test takers participating in the calibration phase. Manipulated variables were the calibration strategy, the size of the calibration sample, the size of the item bank, and the item response model.