|
Activity Number:
|
306
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Tuesday, August 4, 2009 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Physical and Engineering Sciences
|
| Abstract - #303490 |
|
Title:
|
Blocked Experimental Designs for a Non-Normal Response
|
|
Author(s):
|
Dave Woods*+ and Peter van de Ven
|
|
Companies:
|
University of Southampton and TNO Quality of Life
|
|
Address:
|
School of Mathematics, Southampton, SO17 1BJ, United Kingdom
|
|
Keywords:
|
Binary response ; Block designs ; D-optimality ; Generalized Estimating Equations ; Generalized Linear Model ; Robust design
|
|
Abstract:
|
We discuss methods of obtaining efficient block designs for experiments with an exponential family response described by a marginal model fitted via Generalized Estimating Equations (GEEs). This design and modeling strategy is appropriate when the blocks are nuisance variables as, for example, may occur in industrial experiments. GEE models extend Generalized Linear Models to incorporate a correlation structure between experimental units in the same block. In common with designs for other nonlinear models, efficient designs for GEE models depend on the unknown model parameter values. Three strategies are investigated for finding exact designs robust to the values of the marginal model parameters. Designs found by these strategies are critically compared and assessed. The design strategies are motivated and illustrated throughout by an experiment from the aeronautics industry.
|
