The effect of autocorrelation on the estimation of workers' daily exposures

Daily 8-hr time-weighted average (TWA) measurements may not be independent since production rates, maintenance schedules, work practices, and ventilation can result in trends where consecutive values are correlated (autocorrelation). A sampling program which involves collection of measurements on consecutive days, therefore, can result in biased estimates of the mean and variance of the exposure distribution if a high degree of autocorrelation exists. Three simulated data sets were examined to assess the effects of autocorrelation on the estimation of exposure distributions. Results indicated that about 30% of estimated mean values from a highly-autocorrelated series were outside the 95% confidence interval observed for an uncorrelated series. Three data sets obtained from actual workplaces were found to have relatively little autocorrelation. This suggests that for workplaces such as those analyzed here, a random sampling program may not be necessary, and sequential sampling may produce accurate estimates of the parameters of the exposure distribution.