This paper introduced a model for the control of extreme response style (ERS). The model was extended by adding another latent trait, naming as the response style trait. Intended latent traits and the response style trait were estimated simultaneously. Results showed the estimation for all latent traits could be improved, especially when multiple scales were analyzed at the same time.
Q:
1. It is questionable whether this model could really separate the intended latent trait, and the response style trait, without external information about the intended latent trait. Similar to the situation discussed in the force choice model, for example, a questionnaire includes all five-point items, one respondent select five for all items. It is hard to say whether she/he is really very positive in the latent trait, or just prefer the extreme response patterns. In this study, two intended latent traits were estimated simultaneously, in addition to the ERS latent trait. The model was evidenced to gain some improvement. It seems the improvement was obtained by analyzing another intended latent trait, but not the ERS term. As the information to diagnose the respondent lied in the additional latent trait. If there is only one intended latent trait, effect of adding the ERS term to the model is unclear, and it is ambiguous to classify respondents with extreme response patterns. So the improvement in estimation is due to the second latent trait, or the ERS term?
2. Slope parameters for ERS are set at 1, -1, -1 and 1 for simulation. That is to say, if the absolute values are large for ERS, then it would bring larger effects to the response. If the ERS value is large at the negative direction, then the respondent would select the mediate option. If the ERS value is positive large, then the response would be extreme at either side. In other words, this model is sensitive to the response patterns with consistent selections in the category. Imagine that, there are ten 5-point items, a person select 1 for the first 5 items, and 5 for the last five items, does this model still work for this situation?