|
Activity Number:
|
433
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 5, 2009 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
ENAR
|
| Abstract - #305545 |
|
Title:
|
Instrumental Variables with Binary Data
|
|
Author(s):
|
Michael Baiocchi*+ and Paul R. Rosenbaum and Dylan Small
|
|
Companies:
|
University of Pennsylvania and University of Pennsylvania and University of Pennsylvania
|
|
Address:
|
3730 Walnut Street, Philadelphia, PA, 19104,
|
|
Keywords:
|
instrumental variables ; matching ; causal inference ; public policy ; public health ; endogeneity
|
|
Abstract:
|
Classic two-stage regressions do not produce consistent estimators of a binary treatment when the outcome is also binary. We have developed a nonparametric instrumental variable estimator which uses matching to produce a consistent estimator when the treatment and outcome are binary. The distribution of the estimator is known thus standard confidence intervals are calculable. By avoiding maximum likelihood, the computational requirements are also reduced. This method may be helpful for policy studies with endogeneity problems. We illustrate this method by estimating the effect of a premature baby receiving care at a regional center neonatal intensive care unit (NICU) versus a "low-level" NICU.
|
