Abstract:

Testing programs often reuse items due mainly to the difficulty and expense of creating new items. This poses potential problems to test item security if some or all test-takers have knowledge of the items prior to taking the test. Simulated data are used to assess the effect of preknowledge on item response theory true and observed score equating. Root mean square error and bias were used for the recovery of equated scores and linking coefficients for two scaling methods. Results of this study indicated that item preknowledge has a large effect on equated scores and linking coefficients. Furthermore, as the mean ability distribution of the group difference, the number of exposed items, and the number of examinees with item preknowledge increase, the bias and RMSE for equated scores and linking coefficients also increase. Additionally, IRT true score equating results in a larger bias and RMSE than does IRT observed score equating. These findings suggest that item preknowledge has the potential to inflate equated scores, putting the validity of the test scores at risk.

Keywords: