Chaughule, Thorat - Statistical Analysis/Design of Experiments
luated relative to the available resources. This may lead to a decision to forgo the expe-
rimentation.
The problem of experiment efficiency is most acute when several factors must be in-
vestigated in an experiment. When guided only by intuition, many different types of de-
signs could be proposed, each of which might lead to flawed conclusions. Some experi-
menters would choose to hold factors constant that could have important influences on
the response whereas; other experimenters may allow many unnecessary changes of
factors that are inexpensive to vary and few changes of critical factors that are costly to
vary. Efficiency is achieved in statistically designed experiments because each observa-
tion generally provides information on all the factors of interest in the experiment. If
each of the test factors had been investigated separately using the same number of test
runs then large number of test runs would have been needed. It is neither necessary nor
desirable to investigate a single factor at a time in order to economically conduct expe-
riments. Because there is a prevalent view that one-factor-at-a-time testing is appropri-
ate when there are several factors to be investigated (Mason, 2003).
8.7.4. Limitations of one-Factor-at-a-Time method
Most widely used experimental approach in many academic research laboratories is
one factor at a time. In studies when multiple factors affect a same response this tech-
nique is very useful. Consider an experimental setting in which the goal is to determine
the combinations of levels of several factors to optimize a response. The optimization
might be to minimize the loss of active nutrient in a food drying process. It might be to
maximize the color retention properties of the food material during drying. It might be
to maximize the rehydration properties of the dehydrated food materials. In each of
these examples the optimization is a function of several factors, which can be experi-
mentally investigated. Because of the complexity of simultaneously investigating the in-
fluence of several factors on a response, it is common practice to vary one factor at a
time in search for an optimum combination of levels of the factors.
The perceived advantages of one-factor-at-a-time testing are primarily two:- The number of test runs is believed to be close to the minimum that can be de-
vised to investigate several factors simultaneously. - One can readily assess the factor effects as the experiment progresses, because
only a single factor is being studied at any stage.
One-factor-at-a-time experimentation is not only used to determine an optimum
combination of factors. Often this type of testing is used merely to assess the importance
of the factors in influencing the response. This can be an impossible task with one-
factor-at-a-time designs if the factors jointly, not just individually, influence the re-
sponse.
The drawbacks of one factor at a time can be attributed to the single dimensional
approach which is laborious and time consuming, especially for large number of va-
riables, and frequently does not guarantee the determination of optimal conditions (Xu
et al., 2003). In fact the optimal factor combinations may not be obtained when only one
factor is varied at a time. Also, the combinations of levels that are tested do not neces-
sarily allow appropriate models to be fitted to the response variable. Additional test