Kathryn Roeder

I began my career as a biologist, but found that every question that interested me could only be answered by solving an even more intriguing statistical puzzle. Thus my career path veered into statistics. However, much of my work, both theoretical and applied, remains motivated by my scientific training.

The collaborative research I have enjoyed most is in the area of statistical genetics. Right now, an applied topic that interests me greatly is the use of statistical tools to understand the workings of the human genome and the nature of inherited diseases.

The connecting theme in my research is "mixture models," which I see as a way to model the heterogeneity in nature. Traditional statistical models attempt to explain the mean response of each individual in the study. Frequently, however, this is not possible because key variables may not be directly observable. For example, genetic differences among patients may lead them to respond differently to a drug therapy. In some instances one asks --- is there heterogeneity in the population or not? For example, is there a gene for high blood pressure that causes more patient-to-patient variability than one would expect from measured characteristics such as diet and fitness? In this example the discovery of an underlying variable, such as a gene, is the question of principal interest. In other problems, heterogeneity is merely a nuisance on the way to estimating other features of the population.

My results for mixture models have been used to better understand a broad range of scientific phenomena, from clusters and voids in the galaxies to DNA fingerprints. This, for me, is the most satisfying aspect of statistics, when methods you develop are applied to answer important scientific questions.

Some Related Publications

K. Roeder. (1994). "DNA Fingerprinting: a review of the controversy (with discussion)," Statistical Science , 9, pp. 222-278.

K. Roeder. (1994). "A graphical technique for detecting the number of components in a normal mixture," Journal of the American Statistical Association, 89, pp. 487-495.

K. Roeder, R.J. Carroll, B.G. Lindsay (1996). "A nonparametric maximum likelihood approach to case-control studies with errors in covariables. JASA, 91, pp.722-732.

B. Devlin, N. Risch, K. Roeder (1996). "Disequilibrium Mapping: Composite likelihood for pairwise disequilbrium" Genomics, 36, pp.1-16.