|
Activity Number:
|
436
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 5, 2009 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Statistical Learning and Data Mining
|
| Abstract - #304902 |
|
Title:
|
Regression Trees for Group-Randomized Trials
|
|
Author(s):
|
Anne-Michelle Noone*+ and Rebecca Andridge
|
|
Companies:
|
Georgetown University and University of Michigan
|
|
Address:
|
4000 Reservoir Road, NW, Washington, DC, 20057,
|
|
Keywords:
|
Regression tree ; Cluster randomized trial ; Group randomized trial ; Intraclass correlation
|
|
Abstract:
|
Group-randomized trials (GRTs) are popular methods for assessing community-level interventions where randomization at the individual level is not feasible. Analysis of GRT data requires taking into account intraclass correlation between members of the same cluster. We focus here on data mining methods, specifically regression trees, and their applicability to GRT data. We briefly review regression tree methodology and existing extensions to correlated data and discuss their applicability to GRT data. The performance of traditional regression trees using GRT data with varying levels of intraclass correlation is evaluated through simulation. We suggest modifications to existing tree methods to accommodate the clustering and demonstrate with application to a school-based asthma intervention study.
|
