Chen H-F, Lin K-C, Wu C-Y, Chen C-L. Rasch validation and predictive validity of the Action Research Arm Test in patients receiving stroke rehabilitation. ObjectiveTo validate the internal construct and predictive validity of the Action Research Arm Test (ARAT). DesignSecondary study. SettingSeven medical centers. ParticipantsPatients with stroke (N=191). InterventionsNot applicable. Main Outcome MeasureThe internal construct validity of the ARAT score at pretreatment was examined using Rasch analysis. The predictive validity was examined by the correlations between performance on the ARAT before treatment and scores on the Wolf Motor Function Test, the Motor Activity Log, and the Stroke Impact Scale after treatment. ResultsThe 4-point ARAT scale had a disordered rating scale structure. Further Rasch modeling suggested revising the original 4-point scale into a 3-point scale. The 19 items measured 1 construct. The item difficulty hierarchy indicated that excluding the gross subtest, a score of 3 on the first item of any other subtest indicated the highest motor ability, and a score of 1 (the revised lowest rating) on the second item indicated the lowest motor ability. Tasks of “place hand behind head” and “place hand on top of head” showed poor item fit and item bias relevant to participants' ages. The ARAT items can reliably separate participants into 5.44 strata. Moderate to good correlations indicated good predictive validity. ConclusionsThe ARAT possesses good psychometric properties in stroke patients with mild to moderate motor severity and without severe cognitive impairment, and has evidence of unidimensionality, predictive validity, and reliability. The revised 3-point rating scale is recommended when the ARAT is administered on this population. The “place hand behind head” and “place hand on top of head” tasks misfit the Rasch model's expectations. Future studies are needed in the use of the ARAT on stroke patients with different levels of motor severity or with cognitive impairment.