Background. Working in extreme postures has been identified as a risk factor for musculoskeletal symptoms. Directly measuring work postures is considered to be the most accurate approach for assessing these exposures, but it is often not feasible to directly measure posture due to time or budget constraints. Alternatively, direct measurements of postures can be predicted based on observations of workers’ postures. Since observers are known to differ in posture ratings, it may, however, be necessary to develop calibration procedures for each specific observer.

Methods. Arm and back postures of a random sample of 28 paper mill workers were measured via inclinometry and also were assessed by three observers from videos. Linear models with participant number and observer as random effects were resolved to assess whether or not observed postures were associated with the corresponding inclinometer values and if the effect of observer on slope and intercept was significant (p<0.05). The variance explained by these models was compared to the variance explained by corresponding linear models yet with observer entered as a fixed effect (i.e. allowing different slopes and intercepts for different observers).

Results. For all postures, the variance explained was similar when using observer as a fixed compared to a random effect (R-squared ranging from 0.41 to 0.56 for observer as fixed or random effect). Throughout, participant was the major source of variance.

Discussion. Our findings of similar amounts of variance explained when using observer as a fixed compared to a random effect for all postures indicates that calibration models developed for each individual observer may not necessarily perform better than a general calibration applying to any observer. Since posture observations explained only a small proportion of directly measured posture variance, observation may not be very useful in this setting