|
SPEAKER: |
|
|
TITLE: |
"Some Implications of the Latent Variable Regression Model - Inference on Regression Coefficients" |
|
DAY: |
Wednesday, February 10, 1999 |
|
TIME: |
3:30 p.m. [Coffee in BSB-202 at 3:00 p.m.] |
|
PLACE: |
BSB-108 |
The latent variable regression model is presented as a model that is appropriate for many situations where observational data have been collected. It includes errors and an underlying reduced rank for the predictor variables. Techniques based on these models are being widely used in both chemistry and chemical engineering. However, much of the statistical analysis done on these techniques rely on the standard regression model where the predictors are assumed to be error-free and full rank. In this presentation I show that ignoring the latent variable nature of the data leads to strange results when inference is done on the regression coefficients, B, from the prediction equation Yhat = XB.
Alison Burnham is a PhD graduate from McMaster (1997). She did her undergraduate degree in Applied Mathematics and Statistics at Waterloo (1987) and her masters in Statistics at Waterloo (1989). She has also worked as a statistician at Woodbridge Foam and at Ontario Hydro. Her research interests are industrial statistics with a concentration on latent variable models and methods applied to chemometrics (chemistry and chemical engineering) applications. She is currently working as a research scientist for the McMaster Advanced Control Consortium in the Department of Chemical Engineering at McMaster.
The following reference has been provided by Dr Burnham to be used as background for her talk. They are on reserve at Thode Library (STATS 770: Statistics Seminar).
[1] Burnham, AJ , MacGregor, JF & Viveros, R (1998). "Latent Variable Multivariate Regression Modelling," submitted to CHEMOMETRICS AND INTELLIGENT LABORATORY SYSTEMS.