In factorial type experimental
designs the trials should be
carried out in random order.
It may be that factors that the tester
is unaware of are affecting the experiment.
If these so-called 'lurking variables' are
varying in a systematic manner the effect
response may not be apparent
from the analysis and lead to incorrect
results. Randomizing the order of the runs
destroys any systematic pattern and converts
it to random variation that can be detected