Background. Previous studies have shown that video-based observation of postures that are only partly visible leads to different daily summary values than when postures are fully visible. However, the source of these differences is unclear. The purpose of this study was to estimate the between-observer variance of trunk and arm posture estimates (relative to within- and between-worker variance), and to investigate the effect of visibility on this observer variability of trunk and arm postural exposure estimates.

Methods. Video recordings were made of 28 pulp mill workers for three full shifts each. Trunk and arm postures were then estimated by trained observers using a work sampling approach; posture images were also assessed as being “fully” or “partly” visible. REML techniques were used to estimate the between-worker, between-day and betweenobserver components of variance at different visibility levels; Wald-based confidence intervals and p-values were used to determine sources of variation.

Results. Estimates of partly visible postures (as agreed upon by all observers) were lower than fully visible postures. However, more than 90% of trunk observations and 85% of arm observations did not have full agreement on visibility between observers. Right upper arm posture showed smaller between-observer variance when all observers rated a posture to be fully visible, as compared to all observers agreeing it was only partly visible. This suggests partly visible data introduces more methodological (i.e. between-observer) variability when compared to fully visible data. However, no significant differences in between-observer variability were found for the trunk, suggesting that other factors explain the reported differences in estimated postures between fully and partly visible data in this case.

Discussion. Future studies involving concurrent direct measurement would determine whether there is a true difference in posture between partly visible and fully visible periods, or whether the difference between fully and partly visible periods are related to observer performance.