Grouping Effects on Jackknifed Variance Estimates of Item Response Theory Linking for Assessments With Multiple Forms

• In Event: 70.066 - Roundtable Session 28
  In Roundtable Session: 70.066-7 - Equating: Issues and Practices

Abstract

This study explored the effects of grouping approaches, random and non-random, on jackknife variance estimation of IRT linking for assessments with multiple forms. The same jackknifing approach was applied to the complete linking procedure, including item calibration, the Stocking-Lord test characteristic curve (S-L TCC) linking, and score equating for all the replicate samples. Although both methods yielded compatible jackknifed variances of the transformation constants (A and B) and the unrounded scale scores, the random method yielded larger variances than the non-random grouping methods. The results suggested that, for samples with strong clustering effects, the random method underestimated the jackknifed variances. Real data of the large scale assessments with multiple forms were applied in the analysis.

Authors

• Lin Wang, ETS

• Jiahe Qian, ETS

• Yi-Hsuan Lee, Educational Testing Service