|
Activity Number:
|
219
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 5, 2003 : 9:00 AM to 10:50 AM
|
|
Sponsor:
|
Section on Statistics in Epidemiology
|
| Abstract - #301205 |
|
Title:
|
Violations of Assumptions in Methods of Correcting for Selection Bias
|
|
Author(s):
|
Patrick Bradshaw*+ and Randy L. Carter and Alan D. Hutson
|
|
Companies:
|
University of Texas and University of Florida and University of Buffalo
|
|
Address:
|
7703 Floyd Curl Dr., San Antonio, TX, 78229-3901,
|
|
Keywords:
|
selection bias ; Monte Carlo study ; Heckman selection ; treatment effects model ; propensity score
|
|
Abstract:
|
Logistical and ethical problems that plague a randomized study often drive researchers to seek more convenient and ethnically palatable sources of data. However, other issues may surface such as selection bias-when the distribution of characteristics that may influence the outcome is not balanced between treatment and control groups. The treatment effect model from econometrics and covariance adjustment on the propensity score are methods often used to adjust for this. Our goal is to illustrate and compare the consequences of violating the assumptions each of these techniques relies upon. Our Monte Carlo studies show that both methods performed well under the appropriate model's assumptions and even when distributional assumptions were a bit relaxed. The treatment effect model proved robust yielding consistently unbiased estimates of the treatment effect under several scenarios including model misspecification. Propensity score adjustment worked well for a correctly specified model when treatment was independent of outcome however, either misspecification or treatment-outcome dependence showed notable bias in the estimates of the treatment effect.
|
