Dawn Woodard crunches numbers to deduce the probability of future outcomes from past events, as well as to infer the characteristics of a population. 

Woodard uses Bayesian statistics, one of two camps statisticians divide themselves into, the other being frequentist statistics. “Its theoretical foundations are older than those of frequentist statistics but, due to advances in computation, it has become popular as an applied tool only in the last two decades,” says Woodard.

Wherever there’s data, Woodard can bring her skills to bear on a problem. She has worked on issues as diverse as groundwater pollution and breast cancer screening. “People come to me with problems sometimes,” she says. “Often another statistician in the department has a project they’re interested in but don’t have the time to approach it.”

While a graduate student at Duke, Woodard was asked by a large consortium of breast cancer groups to look at its large data set of mammograms to determine the accuracy of individual radiologists. It might seem like a simple task. Radiologists were deemed to have read a mammogram incorrectly if a patient diagnosed as healthy was found to have breast cancer within a year. But it’s more complicated than that.

Although there were about half a million cases in the data set, some radiologists had seen relatively few patients. Were differences in accuracy due to skill, or simply chance? And some doctors see more older patients than others. How much of the difference was attributable to demographics like age?

While the percentage of breast cancer cases that were correctly detected did not vary significantly, according to Woodard’s analysis, the percentage of unnecessary false detections did. “We hope that these results can help radiologists refine their approach so as to reduce unnecessary and costly follow up,” she says.

Bayesian computations involving large data sets or complex models can take a long time and aren’t always reliable, so Woodard also does theoretical research of the algorithms used in statistics. Through mathematical analyses, she tries to determine which algorithms are the most efficient and which are the most accurate. “My degree is in statistics, but my approach is somewhat of an engineering approach,” she says. “I pay a lot of attention to the computational efficiency of the algorithms I use.”