|
Activity Number:
|
466
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Wednesday, August 1, 2007 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
WNAR
|
| Abstract - #309758 |
|
Title:
|
Non- and Semiparametric Subset and Model Selection for Multiple Categorical Predictors and Several Outcomes
|
|
Author(s):
|
A. G. DiRienzo*+
|
|
Companies:
|
Harvard School of Public Health
|
|
Address:
|
655 Huntington Avenue, Boston, MA, 02115,
|
|
Keywords:
|
Genomics ; Genotype sequence ; Multiple hypothesis testing ; Multiway combinations ; Simultaneous inference ; Sparse data
|
|
Abstract:
|
Consider selection of a semiparametric model for response given multiple categorical variables. It is wished that the model is easily interpretable and potentially includes higher-order predictor variable/level combinations. Presented is a two-stage approach for analysis. The first stage is a nonparametric subset selection technique that estimates the set of predictor combinations that have a nonredundant marginal association with response. The subset selection complexity parameters correspond to two error rates (e.g. (generalized) familywise error rate, proportion of false positives) and thus are easily interpretable and provide control over the number or proportion of false positives. The second stage constructs a simultaneous confidence band for the prediction accuracy of each member in a set of varying-sized candidate models. Application to a recent AIDS clinical trial is provided.
|
