Department of Biostatistics Seminar/Workshop Series
Advances in Partition Modeling for Biomedical Research
Matthew S. Shotwell (Matt)
Graduate Student, Division of Biostatistics and Epidemiology, Department of Medicine, Medical University of South Carolina
Wednesday, October 27, 2010, 1:30-2:30pm, MRBIII Conference Room 1220
Intended Audience: Persons interested in applied statistics, statistical theory, epidemiology, health services research, clinical trials methodology, statistical computing, statistical graphics, R users or potential users
Partition modeling is used to account for heterogeneity in biomedicalphenomena by partitioning similar observations into clusters. The Dirichlet process mixture is a partition model that encodes prior preference for unbalanced partitions, making it suitable, for example, in outlier detection problems. A new class of partition models is presented that parameterizes the prior preference for more or less balanced partitions. A novel estimation method and its implementation are discussed. The methods are assessed in the context of facial image recognition, where the partition of images is known, and in a yeast microarray time series, where the partition of probes is unknown.