Ergonomics interventions often focus on reducing exposure in those parts of the job having the highest exposure levels, while leaving other parts unattended. A successful intervention will thus change the form of the job exposure distribution. This disqualifies standard methods for assessing the ability of various exposure measurement strategies to correctly detect an intervention's effect on the overall job exposure of an individual worker, in particular for the safety or ergonomics practitioner who with limited resources can only collect a few measurements. This study used a non-parametric simulation procedure to evaluate the relationship between the number of measurements collected during a self-paced manufacturing job undergoing ergonomics interventions of varying effectiveness, and the probability of correctly determining whether and to which extent the interventions reduced the overall occurrence of pronounced trunk inclination, defined as an inclination of at least 20 degrees. Sixteen video-recordings taken at random times on multiple days for each of three workers were used to estimate the time distribution of each worker's exposure to pronounced trunk inclination. Nine hypothetical ergonomics intervention scenarios were simulated, in which the occurrence of pronounced trunk inclination in the upper 1/8, 1/4, and 1/2 of the job exposure distribution was reduced by 10%, 30% and 50%. Ten exposure measurement strategies were explored, collecting from one to ten pre- and post-intervention exposure samples from an individual worker. For each worker, intervention scenario and sampling strategy, data were bootstrapped from the measured (pre-intervention) and simulated (post-intervention) exposure distributions to generate empirical distributions of the estimated intervention effect. Results showed that for the one to three intervention scenarios that had the greatest effect on the overall occurrence of trunk inclination in the job, one to four pre- and post-intervention measurements, depending on worker, were sufficient to reach an 80% probability of detecting that the intervention did, indeed, have an effect. However, even for the intervention scenario that had the greatest effect on job exposure, seven or more samples were needed for two of the three workers to obtain a probability larger than 50% of estimating the magnitude of the intervention effect to within +/-50% of its true size. For almost all interventions affecting 1/8 or 1/4 of the job, limited exposure sampling led to low probabilities of detecting any intervention effect, let alone its correct size.
Analytical-methods; Ergonomics; Exposure-assessment; Exposure-levels; Exposure-methods; Injuries; Injury-prevention; Mathematical-models; Musculoskeletal-system; Safety-measures; Sampling-methods; Statistical-analysis; Work-analysis; Work-operations; Work-performance; Workplace-monitoring; Workplace-studies; Work-practices;
Author Keywords: Intervention effectiveness; Exposure Measurement; Bootstrapping
Svend Erik Mathiassen, Centre for Musculoskeletal Research, University of Gävle, Gävle, Sweden