What I propose is that each student run the chosen reaction at a unique combination of time and temperature and then each student's yield be plugged into a multiple regression program, including interaction terms, of course. Not only will preferred operating conditions be revealed, but also the concept of interaction and the fallacy of one-variable-at-a-time experimental design will become obvious.

If one wants to embellish the experience, one can recover starting material and teach the difference between yield and conversion.

One can use the data thus developed to grade the students by examining how far from the value predicted by the response surface each student's yield falls.

Variations on this theme could replace time and temperature with such continuous variables as reactant ratio, concentration, rate of addition, and agitator speed. One can also use discrete variables such as order of addition, solvent type, and workup method. Furthermore, one can use statistical design in studying purification methods where purity, by melting point for example, is used as the criterion of goodness.

Instead of suggesting specific experiments (I've been out of university too long to try that) I'd like to challenge teachers to try their hands at doing so and to share their experiences in these pages.

Literature Cited

1. Birney, D.; Starnes, S. J. Chem. Educ. 1999, 76, 1560.

See Author's Reply.