The Effect of Error in Item Parameter Estimates on the Test Response Function Method of Linking

Abstract

The effect of item parameter estimation error was examined for computation of the linking coefficients for the test response function (TRF) linking/equating method. A simulation study modeled a horizontal internal anchor linking situation by generating item parameter estimates for initial and target metrics from known sampling distributions. Four factors—size of standard error, number of common items, item response theory (IRT) model, and relation of standard error to the item parameter estimates—were fully crossed. The improvement ratio was introduced as a method for assessing the quality of linking. Results showed that linking was more accurate when there was less error in the item parameter estimates. Fifteen or 25 common items provided better results than 5 common items under both error conditions. Error in estimation of expected number correct was relatively constant within a given IRT model and parameter set. The TRF method was relatively robust in the presence of moderate error in item parameter estimates.