Activity Number:
|
236
- Causal Modeling Methods in Epidemiology
|
Type:
|
Contributed
|
Date/Time:
|
Monday, July 31, 2017 : 2:00 PM to 3:50 PM
|
Sponsor:
|
Section on Statistics in Epidemiology
|
Abstract #324127
|
View Presentation
|
Title:
|
Sensitivity Analysis and Power in the Presence of Many Weak Instruments with Application to the Effect of Incarceration on Future Earnings
|
Author(s):
|
Ashkan Ertefaie* and Dylan Small
|
Companies:
|
University of Rochester and University of Pennsylvania
|
Keywords:
|
Exclusion restriction assumption ;
Instrumental variables ;
Test statistics ;
Two-stage least squares ;
Unmeasured confounders
|
Abstract:
|
Instrumental variable (IV) methods provide a consistent treatment effect estimate in the presence of unmeasured confounders under certain assumptions. Using IVs that are weak predictors of the treatment can lead to effect estimates that are invariably sensitive even to small departure from the assumptions. This becomes increasingly problematic when there are many weak IVs. We develop an inferential procedure that is specifically designed for settings with many weak IVs. It examines the sensitivity of inferences for the parameter of interest to possible violations of IV assumptions. A power formula for sensitivity analysis is also provided. We study the effect of imprisonment on earnings using data on all individuals sentenced for felony in Michigan in the years 2003-2006.
|
Authors who are presenting talks have a * after their name.