NI0SH's 0ccupational Exposure Sampling Strategy Manual (Manual) contains industrial hygiene and statistical guidelines for sampling occupational environments and analyzing exposure data. This proposal includes three areas of statistical investigation: (l) to examine alternative assumptions about the distribution of exposures by exploring a nonparametric procedure and by studying the fits of the Burr XII distribution to exposure measurements; (2) to represent the selection of high risk employees and maximum exposures by applying the theory of extreme values; and (3) to address the time series sampling of occupational environments by incorporating serial correlation into the analysis models. The concern for the distribution of exposure measurements will be addressed in two ways. First, a Bayesian nonparametric approach to compliance classification is provided and will be compared with parametric procedures. Second, an empirical feel for the relevance of normal, lognormal, Weibull and gamma distributions will be gained by plotting a variety of occupational exposure data on a moment-ratio diagram. Since the relevant portion of the skewness-kurtosis plane is covered by the Burr XII distribution, the fit of this distribution to a variety of exposure data will be examined. Applying the sampling theory of extreme values is the second objective. Specifically, the distribution of a maximum deals with the Manual's sampling of high risk employees and selection of the largest exposure measurement for evaluating ceiling standards. And, the basics of compliance sampling is to document the maximum exposures. The number of workers at lower exposures may be very large and this information could be utilized. As the third objective, the effect on compliance decisions of correlated exposures will be examined. Short, partial-shift samples collected on a maximum risk employee during a single shift are likely to exhibit higher serial correlation than a sequence of 8-hour samples. First-order autoregressive models will appropriately increase estimates of variance, for example of mean exposures. Consequently, not incorporating auto- correlation may result in anti-conservative inferences. The Manual's goals of simplicity, usefulness,and objectivity will be weighted carefully against the expected gains in this proposal. An alternative to the Manual's emphasis on confidence intervals will be examined. Using an extreme value distribution, the percentile rank of the standard, calculated for the sampling, plan, is a compliance probability upon which a decision can be made.
