Violations of local stochastic independence exaggerate scalability in Mokken scaling analysis of the Chinese Mandarin SF-36

BACKGROUND Previous work using Mokken scaling analysis with the SF-36 has found subscales appearing to show excellent Mokken scaling properties. However, the values of scalability of the subscales are very large, raising the possibility that these are artificially high and this may result from violations of local stochastic independence between items. OBJECTIVES To analyse selected items from the Chinese Mandarin form of the SF-36 scale using Mokken scaling and to investigate if violations of local stochastic independence exaggerate scalability. METHODS Exploratory Mokken scaling analysis was run using the online public domain software R by entering 19 items from the Chinese Mandarin form of the SF-36 items into the analysis. The items in the resulting scales, judged by the size of Loevinger's coefficient, were analysed for violations of monotony, 95% confidence intervals and invariant item ordering, including inspection of item pair plots. RESULTS Two Mokken scales were obtained, one including items from the Physical Functioning subscale, and one including items from the Mental Health subscale of the Chinese Mandarin form of the SF-36. The Physical Functioning scale was very strong according to Loevinger's coefficient with high invariant item ordering; the Mental Health scale was moderately strong with weak invariant item ordering. CONCLUSION The strength of the Physical Functioning Mokken scale derived from the Chinese Mandarin form of the SF-36 is probably the result of an item chain and item overlap which violate local stochastic independence. This is due to the nature of the items in the Physical Functioning subscale, all of which relate to physical ability and some of which can only be achieved if previous items in the subscale have been achieved.