The Evaluation of Analytical Data
, Saunders College Publishing, PA, 1982, 25-61
Fair Use Act: Educational use only – excerpt is <10% of total pages (686)
Every physical measurement is subject to a degree of uncertainty that, at best, can only be decreased to an
acceptable level. The determination of the magnitude of this uncertainty is often difficult and requires
additional effort, ingenuity, and good judgment on the part of the observer. Nevertheless, evaluation of the
uncertainty in analytical data is a task that cannot be neglected because a measurement of totally unknown
reliability is worthless. On the other hand, a result that is not particularly accurate may be of great use if the
limits of the probable error affecting it can be set with a high degree of certainty. Unfortunately, there exist
no simple and generally applicable methods by which the quality of an experimental result can be assessed
with absolute certainty; indeed, the work expended in evaluating the reliability of data is frequently
comparable to the effort that went into obtaining them. Such an evaluation may take any of several courses,
including search of the chemical literature to profit from the experiences of others, performance of further
experiments that are designed to reveal potential sources of error, calibration of equipment used in the
measurements, and application of statistical tests to the data that have been obtained. It must be recognized,
however, that none of these recourses is infallible and that ultimately the scientist can only make a
judgment as to the probable accuracy of a measurement; such judgments tend to become more pessimistic
with increases in experience.
A direct relationship exists between the accuracy of an analytical result and the time required for its
acquisition. A tenfold increase in reliability may involve hours, days, or perhaps weeks of additional labor.
One of the first questions that must be considered at the outset of any analysis is the degree of reliability
that is required; this consideration will in large measure determine the amount of time and effort that will
be needed to perform the analysis.
It cannot be too strongly emphasized that a scientist cannot afford to
waste time in the indiscriminate pursuit of great reliability where it is not needed.
This chapter is devoted to a consideration of the errors that can affect an analysis, methods for their
recognition, and techniques for estimating and reporting their magnitude.
Definition of Terms
The chemist ordinarily performs a particular analysis on two to five samples. The individual results in such
a set of measurements will seldom be identical; it thus becomes necessary to select a central "best" value
for the set. Intuitively, the added effort of replication can be justified in two ways. First, the central value of
the set ought to be more reliable than any individual result; second, the variations among the results should
provide some measure of the reliability of the "best" value that has been chosen.