A Two-Step Method Based on lz* for Identifying Effortful Respondents

The likelihood-based person-fit statistic, lz*, is commonly used in educational assessments to distinguish between respondents who are putting in effort and those who are not. However, lz* depends on the estimated item parameters. Item parameter estimates based on data containing non-effortful respondents are biased, thereby undermining the strength of lz*. To address this issue, we propose a two-step method that leverages data mining techniques to obtain more accurate item parameter estimates and then uses them to compute lz*. The results show that the estimates based on the effortful group identified by K-means are more accurate, which improves the performance of lz* in terms of the precision of identifying effortful respondents when non-effort severity is high.