Bolt, D.M., & Newton, J.R. (2011). Multiscale measurement of extreme response style. Educational and Psychological Measurement, 71, 814-833.
This paper has introduced a model designed specifically to account for the extreme response style (ERS), as a model-based approach to more accurate latent variable measurement. It showed removing ERS can help reveal higher correlations between main factors and other factors (p. 828). Under the authors’ proposed model, ERS is included as one of latent traits in multiple latent traits estimation. I was thinking: under this conceptualization of ERS, does it mean that the extent of ER of each person is a probabilistic function of each item, i.e. he may show different ER on different items?
In many cases ERS is one of sources of bias of measurement; however, there are conditions where ERS is essential to be included and should not be eliminated, e.g. measurement of individuals to tell the differences in their knowledge level of certain issues (differentiating experts from novices). But in most other cases, it is worth investigating as an empirical question to answer of the consequence of either accounting for or ignoring ERS on its effects to the measurement of the latent trait of interest.