The Effects Of Irregular Participant Group On IRT Anchor Test Equating

Test equating is the process of converting the participant ability level and item parameters on different test forms of the same psychological quality into a unified unit system so that they can be compared with each other.In equivalent design,the Non-Equivalent Anchor Test is currently the most widely used method.However,in the process of the actual anchor test designing,compiling and testing,there are often many uncontrollable factors that cause the participant ability distribution of the sample to be tested to have some irregularities,which in turn affects the quality of the anchor test,affects the test results of the anchor test,and thus affects the effects of the test equating.This study was based on two-parameter logistic model,used non-equivalent anchor test design,with the equivalent estimation accuracy as the evaluation index,simulated realistic situations through the Monte Carlo method,and investigated the influence of the irregularity of the participant ability distribution in the default situation on the IRT equivalent results,and the performance of different equivalent methods in this context was compared.This study results shows that:1?The equivalence coefficient A is less affected by the rule of the distribution pattern of the subjects ability group.Whether the distribution level of the two groups of participants to be equivalent is consistent or has a large symmetry effect,in overall,the stability is relatively high;The equivalence coefficient B is greatly affected by the distribution of irregular participants ability.With the degree of irregularity of the participants ability level and the difference of the ability levels of the two groups,the estimation accuracy is gradually reduced.2?The greater the deviation between the subject's ability distribution and the standard normal distribution,the larger the estimation error of the discrimination parameter and the difficulty parameter,and the lower the estimation accuracy,the worse the equivalent effect.3?In the situation of the test ability is irregularly distributed,the MS method generally performs better on the estimation accuracy of the equivalence coefficient B,difficulty parameter and the discriminant parameters in the case where the two groups of irregularly distributed subjects have a small ability level;the MM method and the SL method slightly dominate the estimation accuracy of the discriminant parameters when the two groups of irregularly distributed participant have different levels of ability;the SL method and the HB method are better than the estimated equivalence coefficient A,and there is no obvious advantage in other cases.