"Counterintuitive" preferences in health-related quality-of-life measurement

The published preferences for scale steps in a health-related quality-of-life scale have been noted to be contrary to some prior assumptions about their rank ordering. The differences noted are actually statistically nonsignificant, and the observed ordering has a clear intuitive explanation. Several alternative explanations, including vagueness in the case descriptions, inaccuracy in the scaling method, the presence of interactions in the subjects' cognitive integration rules and chance inversions in the presence of the flat response surface characteristic of linear models, are all shown to be impossible or unlikely contributors to the empirical results. The implications of the "negative preferences" for other measurement approaches are discussed, as well as the role of separate attribute coefficients in health policy analyses.