IMS
Session Slot: 8:30-10:20 Monday
Estimated Audience Size: 100
AudioVisual Request: Overhead Projector
Session Title: Difficult Likelihoods and Simple Solutions
Theme Session: No
Applied Session: No
Session Organizer: Lele, Subhash The Johns Hopkins University
Address: Department of Biostatistics, 615 N.Wolfe St., Baltimore, MD 21205.
Phone: (410) 955 3505
Fax: (410) 955 0958
Email: slele@welchlink.welch.jhu.edu
Session Timing: 110 minutes total (Sorry about format):
Opening Remarks by Chair - 2 minutes First Speaker - 25 minutes Second Speaker - 25 minutes Third Speaker - 25 minutes Discussant - 10 minutes Floor Discusion - 5 minutes
Session Chair: Lele, Subhash The Johns Hopkins University
Address: Department of Biostatistics, 615 N. Wolfe St., Baltimore, MD 21205.
Phone: (410) 955 3505
Fax: (410) 955 0958
Email: slele@welchlink.welch.jhu.edu
1. A Composite Likelihood Approach to (Co)variance components estimation
Taper, Mark, The Montana State University
Address: Department of Biology, Montana State University, Bozeman, MT 59715.
Phone: (406) 994 2332
Fax:
Email: taper@rivers.oscs.montana.edu
Lele, Subhash, The Johns Hopkins University
Abstract: Variance and covariance components estimation has significant practical applications in animal breeding and evolutionary biology. The use of Maximum likelihood and Restricted maximum likelihood is difficult for very large data sets that commonly occur in animal breeding. This is mainly due to the necessity of inverting large matrices and the non-uniqueness of the solutions. In this paper we suggest a new approach that is applicable as generally as the maximum likelihood, involves no inversion of matrices and is guaranteed to have a unique solution. We present the consistency and efficiency properties of this method of estimation. We also present an analysis of a dataset that involves covariance components estimation in the presence of correlated random effects.
2. Weighted likelihood equations: The case of mixture models
Markatou, Marianthi, Columbia University
Address: Department of Statistics, 615 Mathematics Building, New York, NY 10027.
Phone: (212) 854 3969
Fax:
Email: markat@stat.columbia.edu
Abstract: Assume that the operating model is finite mixture of K components. We will discuss a computationally simple approach of obtaining estimates for the parameters of the mixture model which are consistent, efficient, asymptotically normal and robust in the presence of the outliers. We will discuss goodness of fit and address a number of issues associated with the algorithm such as starting values, stopping rule and rate of convergence of the algorithm.
3. Pseudo-likelihood is interesting and easy to use
Seymour, Lynne, University of Georgia
Address: Department of Statistics, 204 Statistics Building, Athens, GA 30602.
Phone: (706) 542 3307
Fax:
Email: seymour@rolf.stat.uga.edu
Abstract: In cases in which the joint distribution is intractable but conditional probabilities are known, the pseudo-likelihood provides a quick and dirty method for parameter estimation. The maximum psedo-likelihood parameter estimate has some very nice properties and is a promising alternative to the lengthy MCMC estimation schemes.
Discussant: Natarajan, Nandini Brown University
Address:
Phone:
Fax:
Email: ranjin@stat.brown.edu
List of speakers who are nonmembers: Taper, Mark