Back to
NIST Results
Accuracy (z-score)
and Precision (p-score) Assessments
Accuracy Assessment (z-score)
Where x is the individual laboratory result, X is the "Exercise
Assigned Value," and s is the target value for standard
deviation.
As described in the IUPAC guidelines, the choice of s is
dependent upon data quality objectives of particular program.
It can be "fixed" and arrived at by perception,
prescription, or reference to validated methodology (e.g.,
s = 0.125 X, X is the analyte concentration) or it can be
an estimate of the actual variation (e.g., the calculated
s from the exercise data). The "fixed" performance
criterion is more useful in the comparison of a laboratory's
performance on different materials while the use of the
actual variation may be more useful within a given exercise,
for example, if the determination of a particular analyte
is more problematic than usual.
NIST has calculated and reported z-scores using both approaches
for each analyte for each laboratory. At a previous workshop,
it was decided to use "25% of the exercise assigned
value" as the fixed target value for standard deviation
for this program, at least for a few years. We also calculated
z-scores based on "one assigned-value standard deviation."
The z-scores calculated for these exercises can thus be
interpreted as shown in the following examples:
z-score (25% X):
+1 -> laboratory result is 25% higher than the assigned
value
-2 -> laboratory result is 50% lower than the assigned
value
z-score (s):
+ 1 -> laboratory result is one "exercise standard
deviation" higher than the assigned value
-2 -> laboratory result is "two exercise standard
deviations" lower than the assigned value
From a scientific point of view, IUPAC does not recommend
the classification of z-scores but allows that it is possible
to classify scores, e.g.:
| z | <= 2 Satisfactory
2 < | z | < 3 Questionable
| z | => 3 Unsatisfactory
The NIST report shows the calculated z-scores for each
laboratory for each reported analyte. These tables of the
results and performance include a summary of the number
of reported analytes that fall within each category for
each laboratory. Figures show the distribution of z-scores
(25%) by analyte.
Precision Assessment (p-score)
Since 1995, laboratories have been requested to process
each replicate in a different sample set for precision assessment.
For the calculation of p-scores for this program, the current
target CV for the three replicates is 15%. Tables show the
calculated p-scores for each laboratory for each reported
analyte.