Between- and within- observer variability in two methods for posture observation

摘要

This study determined between-observer and within-observer variability for two different video-based observational methods for assessing upper arm elevation. One method consisted in estimating postures from video stills while in the other, postures were judged from continuous two-minute sequences of video. The two observation methods addressed four different exposure variables: the mean elevation and the percentages of time spent with an arm elevation above 60°, above 90°, and below 15°. Three trained observers used both of these methods to analyze recordings from five hairdressers. Each hairdresser was recorded twice for 30 minutes each, and each observer analyzed all recordings twice with each observation method. The variance components in the data sets, both methodological and non-methodological, were calculated using a random effect model. The results showed that the methodological variance components contributed considerably to the total variance in data, in particular for the variable “below 15°”. Between-and within-observer variabilities were both prominent, within-observer variability being the larger in five of eight combinations of variable and observation approach. The continuous observation approach showed, in general, less methodologic variability than the work sampling approach. The findings suggest a potential for significant improvements in the precision of posture assessment by giving more emphasis to repeated observations than what is usually practised in observational studies. This point is illustrated by numerical examples.