Gathering Correlational and Experimental Validity Arguments for Identifying Careless Respondents With a Mixture IRT Model

The present study examined the validity of interpreting latent classes from a mixture modeldesigned to detect careless and/or insufficient effort responding (C/IER), using a preregisteredsurvey experiment. First, we hypothesized that C/IER prevalence would differ acrossexperimental conditions. The model identified more C/IER respondents in the evokingcondition than in the preventing condition, while the baseline condition did not differsignificantly from the other two conditions. Second, we expected model-based classificationsto align with established C/IER indicators. Our results supported this; participants who failedattention checks, performed poorly on content recognition tasks, or self-reported low effortwere more likely to be classified as careless. Exploratory analyses demonstrated therobustness of our findings across alternative model specifications. A comparison of C/IERdetection methods revealed that the mixture model and attention checks were most sensitiveto experimental manipulations. Overall, the results support the validity and robustness of themixture modeling approach for detecting C/IER