Use of robust estimators of missing data in evaluating shiftwork performance

Abstract

Major problems apparent in studies evaluating the performance of employees across a large number of shifts often involve data missing due to a lack of employee participation in all possible shifts, and large variance around central tendency estimators owing to extreme responses on performance measures. These problems and possible solutions are considered here with reference to a data set collected from a number of shiftworkers in the steel industry. Data were collected from 28 employees over a morning shift and night-shift roster. The shift consisted of two cycles of MORNING-MORNING-NIGHT-NIGHT. The employees were male computer operators working a 12-h shift. The work required them to be constantly alert, and to read, monitor and respond to messages from multiple media channels. Included in the test battery were five performance indicators of cognitive speed and power. This battery was delivered using two IBM computers, which controlled the sequences and administration of the cognitive tasks. Administration of the battery was conducted in the work room and undertaken as close as possible to starting work, and as close as possible to completing work. The findings indicate that regression modelling was the most efficient way of estimating missing data. The use of M-estimators reduced the influence of extreme values on parameter estimation, and increased effect size over that observed using raw data.