PIMS Lunchbox Lecture: Dr. Hua Shen
- Date: 12/06/2018
Dr. Hua Shen
Department of Mathematics and Statistics
University of Calgary
University of Calgary - Downtown Campus
Finite Mixture Models with Measurement Error and Misclassification
In many fields such as medicine, engineering, economics and psychology, we often encounter a pooled population and its observations while the subpopulation identification is of interest but unobserved. For example, primary care physicians want to know whether a patient is with or without a health condition and the phase or level of severity of a disease to provide appropriate care; engineers can be interested in identifying the state of the machine for predictive maintenance or early fault detection. Ignoring the heterogeneity results in biased estimates and invalid conclusions. However, the reference standard tests definitively identifying the subgroups can be expensive, time-consuming, invasive and less widely available. We are often compelled to use cheap and easily accessible subgroup-identification tools subject to misclassification to serve as the surrogates of the desired standard tests which often lead to sever consequences physically and financially. In other cases, we need to develop mixture models to investigate the properties of the subpopulations and ascribe postulated subpopulation identities to individuals based on the observed values. The mixture model-based clustering procedures have received increasing attention in recent years and have proven to be useful in modeling the heterogeneous data with a finite number of latent classes. However, they require accurate measurement of the observations which are sometimes unavailable. Furthermore, clustered data arise when the data comes from several different groups, referred to as clusters, giving multiple observations within each cluster. Such data is featured with hierarchical structure and imposes additional analytical challenges. This talk presents proposed methods to address these issues under the framework of latent variable models. Simulations and example data are used to illustrate the findings.
Where: Room 330, University of Calgary Downtown Campus, 906 8 Avenue SW
Date: Thursday, Dec 6, 2018
When: 12:00 - 1:00 pm
Sign up Deadline: Dec 3, 2018
The Pacific Institute for the Mathematical Sciences is grateful to Alberta Advanced Education and Technology, and the University of Calgary for their support of this series of lectures.