Thus, the coverage factor k is approximated as: ,
which is less than and close to 3.0, if the effective number of degrees of freedom and .
,
In summary, the bias uncertainty is pooled together with the uncertainty components in
.
It should be remembered, however, that only refers to quantities, which vary at each of the future measurements following the initial evaluation.
Note 3: For statisticians. Characterizing effects of uncorrected bias: If the systematic error (bias) is non-zero, confidence limits on the accuracy range A may be approximated as follows. The Smith-Satterthwaite approximation is generalized in approximating estimates in Eq 6 by:
in terms of a chi-square random variable
for the two cases
The effective number of degrees of freedom υ is determined by forcing the variance of
to reproduce the estimated variance of
or
in their
respective cases:
Calculation of or is generally straightforward and depends on specifics of the evaluation experiment and on significant influence parameters. The confidence limit A95% is then determined as in Eq 9:
This expression has been found [12] quite accurate, exhibiting negligible effects from the discontinuity: The chi-square approximation is expected to be worst when is large relative 3/15/03
224
NIOSH Manual of Analytical Methods
