to account for errors in both the measurement and the determination of the uncertainty components themselves. ISO GUM is somewhat sketchy about the coverage factor . Furthermore, the coverage factor can be interpreted in several ways. Most straightforward is the limited case where the uncertainty components can be re-evaluated each time the method is used (resulting in k proportional to a Student-t quantile). In this case, the covering intervals bracket the measurand for (for example) 95% of the measurements. Alternatively, the coverage factors based on the Student-t quantile specify intervals containing measurand values at levels of evaluation confidence in the mean (i.e., averaging over many method evaluations). In other words, for roughly 50% of method evaluations, intervals used at each measurement contain the measurand value greater than (for example) 95% of the time. The concept is consistent with the statistical theory of tolerance or prediction intervals. This approach is important to industrial hygiene since workplace air concentrations vary spatially and over time to such a degree that a method cannot be evaluated by simply taking replicate measurements [8]. However, industrial hygiene measurement methods have traditionally required confidence levels greater than 50% in the method evaluation. Generally, 95% confidence in a method validation is required. The different types of confidence levels are reflected simply in the numerical value and interpretation of the coverage factor. Of equal importance in the industrial hygiene field are details needed to handle systematic error (bias) relative to reference concentration measurements found during method evaluation. For example, the sampling rate of a given diffusive sampler for gases or vapors is generally measured once by the diffusive sampler manufacturer prior to use by multiple clients. As the samplers are not re-calibrated for each use, residual bias exists in the measurements due to uncertainty in sampling rates used [9]. (NIOSH methods typically do not cite performance for passive samplers because agreement among diffusive monitor manufacturers on test protocols has not yet been achieved, and a system of third party evaluation of diffusive monitor manufacturers sampling rates is not available.) Similarly, the calculation of desorption efficiencies may be performed only once or infrequently and can, therefore, introduce residual bias in measurements that use sorbent-captured samples, e.g., charcoal tubes. In aerosol sampling, detailed knowledge of the particle size-dependent bias of a sampler relative to a sampling convention, such as adopted by ISO/CEN/ACGIH/ASTM [10] for defining respirable dust, is often necessary to judge the usefulness of a given sampler. Each
3/15/03
209
NIOSH Manual of Analytical Methods
