Multiscale Measurement of Extreme Response Style
Daniel M. Bolt and Joseph R. Newton
Based on the results demonstrate by many studies--response styles tend to be stable respondent characteristics. The paper seek to demonstrate how such stability can be exploited to obtain better estimates of ERS through the simultaneous analysis of ratings from more than one rating scale and how the improved ERS estimates perform in estimating the original substantive trait. First, the procedure is demonstrated using real data subscales PISA2006 related to attitudes toward science. Then a simulation study was conducted to evaluate the accuracy of the approach.
Programme: Latent Gold
Part of what makes ERS an important response style to attend to is its established association with other respondent characteristics, including, for example, trait anxiety, income, education level, age, and ethnicity, among others.
Problem:
Interfering with cross-cultural comparisons
Leading to biased scale scores for certain respondents
Undermining attempts to study associations between the substantive trait and other constructs
Methods:
A: frequency counts of end point responses
B: the standard deviation of item scores within a respondent
A& B failure to account for the influence of the substantive trait
C: Added to a survey instrument uncorrelated with the substantive trait serving the sole purpose of identifying response style tendencies
For C:
Items can be difficult to find
Their inclusion on the survey instrument can undermine its validity
The new approach:
its use of a continuous representation of ERS as well as its attempt to psychometrically model how both substantive and ERS traits combine in affecting response category selection.
Comment:
If the text is so short, how can we make sure there is EPS other than it is the participants’ real responses？
In my opinion, EPS should have some correlation with the length of one test which can be used as a criterion for detecting whether we should add the parameter of EPS in our study.
The paper consider all items allowed to load only on their respective subscale’s factor, what will happen if one item measures two subscale or above？
More conditions of the effect of constrained category slopes can be considered in future study.