|
Activity Number:
|
197
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, July 30, 2007 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #309938 |
|
Title:
|
On Some Computational Issues in Marginal Latent Mixture Analysis
|
|
Author(s):
|
Yan Yang*+ and Douglas Simpson
|
|
Companies:
|
Arizona State University and University of Illinois
|
|
Address:
|
Department of Mathematics and Statistics, Tempe, AZ, 85287,
|
|
Keywords:
|
Correlated data ; EM ; Generalized estimating equation ; Quasi-Newton ; Zero-inflated
|
|
Abstract:
|
Data with bound-inflated responses are common in many areas of application. Often the data are bounded below by a real number (e.g., zero) with excess observations at the boundary value. We consider a general class of latent mixture models for inflated discrete and semi-continuous data that combines a degenerate distribution at the bound and a discrete or censored distribution. The latency resulting from not being able to identify which distribution has generated a boundary value leads to a pseudo-likelihood for correlated bounded data that cannot be factorized. We implement both the EM and Quasi-Newton algorithms to estimate the class of mixture models and compare the two methods. The asymptotic covariance matrix is adjusted by the sandwich estimator using the theory of generalized estimating equations. The methods are illustrated with an ultrasound safety study in laboratory animals.
|
