Multiscale Measurement of Extreme Response Style
Response styles have the potential to undermine the validity of scale scores and this paper focus on extreme response style (ERS) which show that respondent tends to select the end points of a rating scale, such as 1 of strongly disagree or 7 of strongly agree. The authors used their general model proposed in 2009 to accommodate distinct, but possibly correlated, substantive traits, in the process of increasing the precision of ERS estimates. PISA data of two subscales for science enjoyment and science value were analyzed by three models: two dimensional NRM, three dimension NRM, and three dimensional NRM with ERS and constraints. Results of model comparison illustrated that the third model has the best fit for all fit indices. Also, parameters and covariance matrix of the third model were shown here. Finally, a series of simulation were conducted to evaluate the accuracy of the models aforementioned.
Questions:
1) I have difficulty to understand constrains setting of the proposed model, in particular setting of table 2.
2) Table 1 indicated that theta of ERS functioning on bias in total scores for theta ENJ. Negative value of theta ERS and positive value of theta ERS have decreasing and increasing tendency for theta ENJ on total scores bias, respectively. However, the latter one does not have property of monotonicity. Is it reasonable?