Empirical Estimation of Standard Errors of Compensatory MIRT Model ParametersObtained from the NOHARM Estimation Program

摘要

Two studies were carried out to evaluate the quality of multidimensional itemresponse theory (MIRT) model parameter estimates obtained from the computer program NOHARM. The purpose of the first study was to compute empirical estimates of the standard errors of the parameters. In addition, the parameter estimates were evaluated for bias and the effects of using different starting values and anchor items. The second study was included to compare the performance of NOHARM with the findings of an earlier simulation study which evaluated other MIRT estimation programs. Results were generally good, with fairly small standard errors for most parameter estimates and little indication of bias. Although the estimation procedure appeared to be robust under different starting values, the specific choice of items used to anchor the solution appears to have important effects on the magnitude of the estimated standard errors. The comparison of NOHARM with other programs was very favorable and supports the use of NOHARM for practical MIRT applications.