McCombs TODAY

Bayesian statistics is not, perhaps, the most accessible field of study, but Robert McCulloch, a professor of Information, Risk, and Operations Management at McCombs, opened up this important topic at a Faculty Research Presentation Oct. 20.

McCulloch, an internationally renowned statistician, illustrated how he uses BART (no relation to the Simpsons), a computer system that can analyze data to find the relationships between variables.

This process, called regression analysis, is used to predict trends and find patterns among sets of data. According to McCulloch, it is commonly used by companies with large amounts of consumer data or by financial analysts.

BART, which stands for Bayesian Additive Regression Trees, was a joint venture between McCulloch, Edward George from the University of Pennsylvania and Hugh Chipman from Acadia University in Nova Scotia.

McCulloch says that what makes BART so amazing is that it can interpret thousands of variables with only very few observations from which to pick out trends and consistently predict the correct trend function.

“Ten years ago I wouldn’t have believed that I could solve these kinds of problems,” said McCulloch.

The real-life example McCulloch offered was of a drug company who wanted to know how to make the most effective drug for a certain disease.

The company pinpointed 266 chemical characteristics of the drug and ran tests with 29,374 different combinations of those characteristics. For each test the researchers recorded whether or not the drug killed the bacteria.

All of this information was plugged into BART, and the program was able to tell the researchers that there were really only seven chemical characteristics that were important in the formulation of the drug. It also helped the researchers find the perfect dosages and combinations of those seven elements in order to manufacture the most effective drug.

To learn more about BART, download the full research paper (PDF).