Chemistry Reference
In-Depth Information
mixture and the observed effect is due to the time differences not the temper-
ature effects. In this example, time is a confounding factor or a confounding
variable.
One approach to avoid these pitfalls is to use statistical experimental
design.
13.2 DESIGN OF EXPERIMENTS
Design of experiments, also called experimental design, is a plan to study the
various variables and their effect on the desired result, often yield or purity.
If planned and interpreted correctly, interactions and hidden variables can be
uncovered. An experimental design should include all relevant variables and
there should be enough experiments performed to determine important effects
and eliminate trivial variables.
Consider the situation of Sam. He is a junior process chemist associated
with a large volume process. Sam thinks that there may be a process advan-
tage obtained by increasing the agitation. He goes to the lab and runs two
experiments. The one with the current level of agitation gives a yield of 65%.
He then increases the stirring rate in the second experiment and obtains a yield
of 76%. He writes in his monthly progress report that he has obtained a yield
improvement of about 10%. The report is circulated throughout the business.
Everyone is excited about this breakthrough because a 10% yield improve-
ment will have a huge impact on the profitability of the business. A team is
assembled to pursue this lead. They run three more experiments with low agi-
tation and three with high. Including the original data, the percent yields with
low agitation are: 65, 72, 67, and 75 for an average of 69.75%. With high agi-
tation, the yields are 76, 65, 68, and 69 for an average of 69.5%. Comparing
the averages, there is no effect of agitation on yield. The team reports this
result. What do you think has happened to Sam's status and credibility within
the company?
Consider a second situation with Josephine at another company. She too is
working on a large volume process and does an experiment with the current
level of agitation, obtaining a yield of 65%, and the one with high agitation,
obtaining a yield of 66%. She reports that there is no effect of agitation on
yield. Had she done a total of four experiments for each condition, she would
see percent yields of 65, 64, 62, and 62 for the standard agitation and 66, 71,
69, and 72 for high agitation. The average yield for high agitation is 69.5%
versus 63.25% for standard agitation. By doing the comparison of the single
runs, Josephine missed seeing the real effect and the potential to improve yield
by varying a simple process variable. Had she compared averages, she would
not have missed this.
Search WWH ::
Custom Search