Watson R, Thompson DR, Wang W (2014) Violations of local stochastic independence exaggerate scalability in Mokken scaling analysis of the Chinese Mandarin SF-36 Health and Quality of Life Outcomes 12, 149
Roger says of the study: In assessing the psychometric properties of measurement scales - whether by factor analysis or item response theory - it is essential that items behave as they do - either forming factors or scales - due to their response to what is being measured (ie the latent trait). This property is called local stochastic independence (LSA). LSA is hard to assess, especially in polytomous items and we fear that some of us may have seen Mokkken scales (Mokken scaling is non-parametric item response theory) where none exist due to the violation of LSA. We investigated this by comparing two subscales of the SF-36: one which clearly violated LSA and one which probably did not and found that the scalability properties of the one violating LSA were very high and probably exaggerated. We discuss the consequences of this in the article.