The confirmatory mixture IRT model for inattentive responses

Inattentive responses, which can threaten measurement quality, are common in rating- or Likert-scale data. In this study, we developed a confirmatory mixture IRT model for inattentive responses to distinguish inattentive responses from normal responses so that test validity could be ascertained. Parameters of the new model could be estimated by the Bayesian methods implemented in the freeware WinBUGS. Findings from simulationsindicated that item parameters and latent group membership of the new model were recovered fairly well; ignoring inattentive responses by fitting standard IRT models yielded biased parameter estimates; fitting the new model to data without inattentive responses would not yield severely biased estimates. Two empirical examples were provided to demonstrate applications of the new model.