The validity of visual observation to assess posture in a laboratory-simulated, manual material handling task

A wide range of methods to evaluate posture in work situations relies on simple, unaided visual observation. In the present study the validity of visual observation to assess posture in a dynamic job was evaluated. Postural aspects were observed and recorded during a manual materials handling job simulated in a laboratory. The results from these observations, concerning gross body posture, torso flexion, arms and legs position, and load to be handled, were compared to the results obtained by direct opto-electronic recording. The agreement on a sample-to-sample basis (expressed by Cohen's kappa) was poor for the variables torso flexion (mean value for kappa = 0.38), position of arms (kappa = 0.43) and legs (kappa = 0.46) and load to be handled (kappa = 0.50) and acceptable only for the gross body posture (kappa = 0.79). Moreover, for each variable except gross body posture, the crude distributions of all observations and recordings across categories (irrespective of time) were significantly different. The results demonstrate that the observations are not valid. It is concluded that dynamic work situations require less simple, more time consuming methods (e.g., analyzing film or video records of the job) than a posture registration method based on direct visual observation.