|
Activity Number:
|
260
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Tuesday, August 4, 2009 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Statistics in Epidemiology
|
| Abstract - #304948 |
|
Title:
|
Tests of Covariate Balance in Experiments and Observational Studies
|
|
Author(s):
|
Ben B. Hansen*+ and Jake Bowers and Jake Bowers
|
|
Companies:
|
University of Michigan and University of Illinois at Urbana-Champaign and University of Illinois at Urbana-Champaign
|
|
Address:
|
439 West Hall, Ann Arbor, 48109-1107,
|
|
Keywords:
|
covariate balance ; propensity score ; random assignment ; randomization-based inference ; subclassification ; matching
|
|
Abstract:
|
In randomized experiments, treatment and control groups should be roughly the same - balanced - in their distribution of pre-treatment variables. But how nearly so? Can descriptive comparisons meaningfully be paired with significance tests? An old argument says that they cannot, and recent contributions to the Journal of the Royal Statistical Society and to Statistics in Medicine say the same of such tests as applied to propensity-matched observational studies. While these arguments have some merit, they would be better directed at certain commonly occurring misinterpretations of the tests. We chart a path that leads to valid and important uses of tests for baseline imbalance, both in the design and analysis of randomized experiments and in propensity-score matching for observational studies.
|
