Stroke patients commonly incorporate the use of compensatory strategies to aid arm transport to achieve better functions and positioning in the absence of distal voluntary movement. During assessment of voluntary movement, the presence of compensatory strategies may suggest that the patient has not attained a natural neurologic recovery and that the patient has acquired a pattern of learned non-use. Various research has been conducted to provide patients with more freedom and frequency to perform their rehabilitation program, such as game-based rehabilitation as well as rehabilitation robotics. Inline with the automation, the need for automated assessment is also prevalent to accommodate the minimal presence of therapist during the automated rehabilitation program. While several research has been done to automate the assessment of hand movement quality, less emphasis has been taken to automate the compensatory strategies largely provided by the torso movement during voluntary movement assessment. This study explores the use of a Torso Principal Component Analysis (PCA) Frame model to assess the compensatory torso movement and analyze the predictability in time-series performance data during various assessment tasks typically found in a clinical assessment context. While the idea has been suggested in the context of posture recognition, we believe that this is the first time such model is used for compensatory assessment after stroke. Our test results demonstrate that the Torso PCA frame is able to delineate between compensatory movement and normal movement within the context of assessment tasks performed and therefore deemed suitable as measures of torso compensation after stroke.