2nd Annual Sobel Seminar: Elizabeth Thompson

Event Date: 

Wednesday, April 19, 2006 - 3:30pm

Event Date Details: 

Refreshments served at 3:15pm

took place in 2005

Event Location: 

  • Sobel Seminar Room; South Hall 5607F
  • Annual Sobel Lecture

2nd Annual Sobel Seminar


Elizabeth Thompson

Dr. Thompson is a Professor of Statistics and an Adjunct Professor of Genome Sciences at University of Washington. Her research interest is in the development of methods for inference from genetic data, and particularly from data observed on large and complex pedigree structures. Questions of interest range from analyses of long-term gene frequency differentiation in widely dispersed populations, to short-term extinction of genes in the small population of a highly endangered species; from inference of genealogical relationships among individuals to inference of the genetic basis of traits from data observed on members of a known pedigree; and from analyses of patterns of genome sharing in plants to modern methods for human linkage analysis. In recent years, several of these questions have been addressed using Monte Carlo likelihood


Fuzzy P-Values in the Detection of Genetic Linkage


Genetic linkage between two genome locations is dependence in the inheritance of genome at those locations. This dependence leads to individuals who share a trait of interest having increased probability of sharing genome at locations that are linked to genes that affect the trait. Tests for genetic linkage capitalize on detection of regions of the genome that exhibit this increased sharing among individuals of like trait phenotype. However, genome sharing is unobservable, and in standard testing approaches the uncertainty in latent genome sharing is confounded with strength of the evidence for linkage. Fuzzy p-values have recently been introduced by Geyer and Meeden (2005: in press) in a different context. In the current context, they provide a way to express both the strength of the evidence for linkage and the uncertainty in that evidence. The approach is applicable to any situation where the ideal measure of evidence is a function of unobservable latent variables, provided only realization of these variables conditional on observed data is possible.