Statistical assessment of changes in ADL dependence: three-graded versus dichotomised scaling

The aim of this study was to investigate how dichotomising three-graded ADL Staircase data affects the possibility of detecting changes in ADL dependence between different assessment occasions. An authentic two-occasion data set was used as a basis for a simulation experiment. In all, we used four different data treatment principles, all utilising the matched pairing of the data. The first principle utilised a sum score technique, and the second within-person comparisons by means of item-by-item analysis of improvement or deterioration. The third principle used ADL ranks, a novel approach, while the fourth used within-item ranks. Independently of the data treatment principle used, the statistical power of all tests was reduced by 13-24% after dichotomisation, compared to when the three-graded scale was utilised. The results indicate that dichotomising ADL Staircase data results in information loss, and hence in reduced ability to detect changes. The need to consider the purpose of the ADL assessment before reducing the number of scale steps is highlighted. The knowledge generated in this study is useful for practitioners and researchers, aiming at evaluating rehabilitation interventions.