|
Activity Number:
|
458
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Wednesday, August 5, 2009 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Health Policy Statistics
|
| Abstract - #304549 |
|
Title:
|
A Simulation Study of Design Effect Approximations for Propensity-Score Weighted Data with Application to the MCAHPS Survey
|
|
Author(s):
|
Amelia M. Haviland*+ and Marc N. Elliott and Mary Slaughter
|
|
Companies:
|
RAND Corporation and RAND Corporation and RAND Corporation
|
|
Address:
|
4570 Fifth Avenue, Suite 600, Pittsburgh, PA, 15213,
|
|
Keywords:
|
Design effect ; propensity score ; observational data ; causal inference
|
|
Abstract:
|
Little and Vartivanian (2005) provide improved design effect approximation in the context of nonresponse weighting, showing that the well-known Kish approximation understates the improvements in MSE that can be achieved when nonresponse weights are predictive of the estimand of interest and may actually decrease variance and bias. We apply this work to propensity score weighted estimates of the Average Effect of Treatment on the Treated in observational data. Accurate estimates of effective sample size for propensity score weighting facilitate comparisons of efficiency with matching approaches. We describe a simulation parameterized in terms of the natural parameters for causal inference to provide heuristic guidance for such applications. An application infers the effects of Special Needs Plans, which are subject to notable selective enrollment, on CAHPS evaluations of plans.
|
