4.
UNCERTAINTY AND ANALYTICAL LAB PROCEDURES Interest in measurement uncertainty and ISO GUM is currently finding its way into the criteria for the accreditation of analytical labs [4]. The result will no doubt be high confidence in understanding one component of the combined or expanded uncertainty—namely the analytical component. Several general approaches to controlling and characterizing analytical uncertainty in routine lab practices seem reasonable. a.
Validated Method Adoption
One possibility is for a lab to adopt a published, evaluated method. Such an adoption would require an initial establishment of the method within the lab’s capabilities. Equivalence to the published method would be established during this initial phase. Thereafter, the method’s uncertainty as documented in the original publication would be claimed for the lab results. Ongoing analysis of a limited number of quality control samples would provide evidence that the method as implemented in the lab remains stable. An example of this approach is the current practice in some labs that handle sorbent tubes to analyze about 4 lab blanks per set of field samples analyzed. The variability in the blank results are then continually compared to past lab performance so as to detect problems which may occur in analysis. Though the small number of degrees of freedom (= 3) does not give a tight figure on the uncertainty, it nevertheless gives assurance that the method is stable. As a specific example of method evaluation data and documentation of an uncertainty budget, data from n = 16 exposures of diffusive samplers in a controlled environment are shown in Table 2. The evaluation is somewhat simplified for this example; a more comprehensive evaluation would also measure effects of wind velocity, humidity, temperature, and concentration time-dependence (potentially significant to diffusive monitoring). Analysis of these data can be handled by an ordinary calculator capable of computing means and standard deviations. Note that the uncertainty ( , where n = 16 is the number of measurements) in the bias is the value that accounts for residual bias due to imperfect correction. Very similarly, the uncertainty in the reference concentration is pooled to arrive at a combined uncertainty. Interestingly, neither of these two contributions corresponds to quantities that vary during sampler application subsequent to its initial evaluation. The background for documenting residual (uncorrectable) bias can be seen in Note 2 at the end of this chapter.
3/15/03
215
NIOSH Manual of Analytical Methods
