DOI: 10.1080/16501970600803252

Objective: To validate the unidimensionality of the Action Research Arm Test (ARAT) using Mokken analysis and to examine whether scores of the ARAT can be transformed into interval scores using Rasch analysis. Subjects and methods: A total of 351 patients with stroke were recruited from 5 rehabilitation departments located in 4 regions of Taiwan. The 19-item Action ARAT was administered to all the subjects by a physical therapist. The data were analysed using item response theory by non-parametric Mokken analysis followed by Rasch analysis. Results: The results supported a unidimensional scale of the 19-item ARAT by Mokken analysis, with the scalability coefficient H = 0. 95. Except for the item “pinch ball bearing 3rd finger and thumb”, the remaining 18 items have a consistently hierarchical order along the upper extremity function's continuum. In contrast, the Rasch analysis, with a stepwise deletion of misfit items, showed that only 4 items (“grasp ball”, “grasp block 5 cm3”, “grasp block 2. 5 cm3”, and “grip tube 1 cm3”) fit the Rasch rating scale model's expectations. Conclusion: Our findings indicated that the 19-item ARAT constituted a unidimensional construct measuring upper-extremity function in stroke patients. However, the results did not support the premise that the raw sum scores of the ARAT can be transformed into interval Rasch scores. Thus, the raw sum scores of the ARAT can provide information only about order of patients on their upper extremity functional abilities, but not represent each patient's exact functioning.