Man’s curiosity has driven him to question and experiment for thousands of years. Through trial and error, standard research methods have evolved that are now an integral part of most science classes. However, some lesser-known techniques have also proven effective.
John Lawson, a faculty member in the Department of Statistics, has years of experience with statistically-designed research strategies and has recently written a textbook on this subject: Design and Analysis of Experiments with SAS. Although these methods were developed early in the second century and have found application in all areas of physical experimentation, they are still not emphasized in general science education.
One research method that has been emphasized in textbooks and curriculum for years, and one Lawson is seeking to diverge from, is the practice of isolating variables.
“This is a traditional approach they use in science classes,” Lawson explained. “By holding all the variables constant and varying one, you see how the output you’re interested in changes as a function of varying that one. And that’s good if you know the relationship exists and if you’re just trying to demonstrate it to students in a lab. But if you’re trying to do real research and you don’t know what relationship it is, then it’s very inefficient, time-consuming and subject to all kinds of problems.”
Factorial designed experiments eliminate these limitations. Using a design that joins many different factors into various combinations can be useful in marketing situations.
“If you’re making something—either a product or a service—and it’s composed of different attributes, you want to know what the customers would really like best,” Lawson explained. “You want to know if you should use more of one attribute, less of one, take one out, or put one in. You can tell what attributes affect their choice.”
Several other factorially designed research methods include: screening designs, response surface designs, designs for mixture experiments, designs for mechanistic models and designs for computer experiments.
Lawson teaches classes in statistically designed experiments and is anxious to collaborate with faculty and students who would like to explore the use of these methods in their research.