Chaughule, Thorat - Statistical Analysis/Design of Experiments8.5.4. Randomization
Randomization of the sequence of test runs or the assignment of factor-level combi-
nations to experimental units protects against unmeasured sources of possible bias.
Randomization also helps validate the assumptions needed to apply certain statistical
techniques. The protection that randomization affords against unknown bias is easily
appreciated by considering the common problem of instrument drift or deviation. If dur-
ing an experiment, dryer drift builds over time, the later experimental measurements
will be biased because of the drift. If all tests involving one level of a factor are run first
and all tests involving the second level of a factor are run later, comparison of the factor
levels will be biased by the instrument drift and will not provide a true measure of the
effect of the factor. Randomization of the test runs cannot prevent instrument drift but
can help to ensure that all levels of a factor have an equal chance of being affected by the
drift. If so, differences in the responses for pairs of factor levels will likely reflect the ef-
fect of factor levels and not the effect of the drift. It should be noted that, the design cri-
teria discussed here may not be comprehensive. The discussion is presented as a guide
to some of the more important considerations that must be addressed in the planning
stages of most experiments. (Mason et al., 2003 ; Antony, 2003 & Montgomery, 2003).
8.6. DESIGNS FOR QUALITY IMPROVEMENT
Dehydration procedures in scientific and engineering experiments are frequently
guided by established protocol and subjective considerations of practicality. While such
experimental procedures may be viewed as economical in terms of the number of test
runs that must be conducted, the economy of effort can be deceiving for two reasons.
First, economy is often achieved by severely limiting the number of factors whose effects
are studied. Second, the sequence of tests may require that only one of the factors of in-
terest be varied at a time, thereby preventing the evaluation of any combined effect of
the experimental factors. The effective use of the statistical principles in design of expe-
riments ensures that experiments are designed economically, that they are efficient, and
that individual and joint factor effects can be evaluated.
Statistical experimental design is then motivated by an examination of problems that
frequently arise when statistical principles are not used in the design. Special emphasis
is placed on the investigation of the joint effects of two or more experimental factors on
a response. Statistical methodology for quality improvement can be divided into two
main categories: on-line and off-line statistical measurement procedures. In the past, on-
line statistical quality control techniques were used for assurance of product quality,
currently off-line investigations using engineering design techniques and statistical de-
sign of experiments used for quality improvement and increased productivity (Mason,
2003). Off-line experiments are performed in academic laboratories, pilot plants, and
preliminary dehydration runs, prior to the complete implementation of formulation or
dehydration process operations. Hence, it can be stated that statistical design of experi-
ments is an integral component of off-line quality-improvement studies. Once a target
has been determined, based on product quality, consumer preferences and manufactur-
ing capabilities, quality improvement relies on achieving the target value and on reduc-
ing variability.