|
Activity Number:
|
84
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 4, 2003 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Business & Economics Statistics Section
|
| Abstract - #301463 |
|
Title:
|
Combining the Overidentification Test and Sensitivity Analysis for Instrumental Variables Regression Inference
|
|
Author(s):
|
Dylan S. Small*+
|
|
Companies:
|
University of Pennsylvania
|
|
Address:
|
3514 Lancaster Ave. Apt. 416, Philadelphia, PA, 19104-4922,
|
|
Keywords:
|
instrumental variables ; overidentification test ; sensitivity analysis ; observational study
|
|
Abstract:
|
The ability of instrumental variables (IV) regression to provide a consistent estimate of causal effects from observational data rests on the assumption that the instruments satisfy the exclusion restriction. In the presence of uncertainty about whether the exclusion restriction is exactly satisfied, two ways in which the weight of evidence provided by an IV regression can be assessed are the following: (i) If there are more instruments than endogenous variables, the assumption that the instruments satisfy the exclusion restriction can be tested via an "overidentification" test; (ii) A sensitivity analysis can be carried out to examine how much the results differ if plausible violations of the exclusion restriction are entertained. This paper develops a methodology for combining the information in the overidentification test and a sensitivity analysis. Our methodology sheds light on the circumstances under which failure to reject the exclusion restriction via the overidentification test increases confidence in the results from an IV regression. We apply our methodology to an empirical example of the impact of community colleges on educational attainment.
|
