Statistical analysis - Concrete Mix Design, Quality Control and Specification - page 215

Statistical analysis
Statistical analysis is not an exact science. However rigorous and elaborate
the statistical techniques used, the conclusions can be no more reliable than
the assumptions on which they are based. Where a limited amount of data
has been obtained from a one-off experiment or series of observations, it
can pay handsome dividends to apply very elaborate analysis techniques to
squeeze out the last drop of knowledge. However, quality control (QC) is
not a one-off experiment but a continuing flow of data. Furthermore it is
a field that is, or should be, rigidly governed by economic considerations.
The requirement is to ensure a given minimum quality of concrete in the
structure. This can be accomplished by using a higher average quality, at a
higher cost in materials, or by achieving a lower variability through higher
expenditure on control. The higher control expenditure itself can be in the
form of a large amount of rough testing with little analysis or in a smaller
amount of more carefully monitored testing and a more thorough analysis
of the results. A balance should be sought that yields the minimum overall
cost for a given required quality. The balance must take into account the
standard of personnel and equipment economically available. There is no
merit in devising a system that requires that every testing officer be a quali-
fied engineer and every team include a professional statistician, if the result
is a higher cost for a given minimum quality.
The concern should not be to apply elegant or rigorous statistics but only
to achieve accurate control of concrete quality. Relatively crude statistical
techniques can be used if their limitations are very clearly understood and
the controller must always be prepared to overrule or revise unrealistic con-
clusions produced mathematically. It is quite difficult to do this without
permitting bias to cloud judgment, but there are several factors that save it
from being almost impossible. One of these is that in QC work a conclusion
is usually provisional and subject to revision as further results are received,
thus a downturn in results may be dismissed as a chance variation or testing
error when first spotted, but if it is confirmed by subsequent results, it must
then be accepted. Another is that related variables such as slump, den-
sity, and concrete temperature can confirm or deny an unusual result by