A Cautionary Tale About Control Variables in IV Estimation
| DOI | http://doi.org/10.1111/obes.12177 |
| Date | 01 June 2017 |
| Published date | 01 June 2017 |
411
©2017 The Department of Economics, University of Oxford and JohnWiley & Sons Ltd.
OXFORD BULLETIN OF ECONOMICSAND STATISTICS, 79, 3 (2017) 0305–9049
doi: 10.1111/obes.12177
A Cautionary Tale About ControlVariables in IV
Estimation*
Eva Deuchert† and Martin Huber†
†Department of Economics, University of Fribourg, Bd. de P´erolles 90 1700, Fribourg,
Switzerland (e-mail: martin.huber@unifr.ch)
Abstract
Many instrumental variable (IV) regressions include control variables to justify (condi-
tional) independence of the instrument and the potential outcomes. The plausibility of
conditional IV independence crucially depends on the timing when the control variables
are determined. This paper worksthrough different IV models and discusses the (conditions
for the) satisfaction of conditional IV independence depending on whether the control vari-
ables are measured prior to or after instrument assignment. To illustrate the identification
issues, we consider the Vietnam War draft risk as instrument either for veteran status or
education to evaluate the effects of these variables on labour market and health outcomes.
Our empirical analysis based on the ‘Young Men in High School and Beyond’ survey sug-
gests that commonly used conditional IV strategies to estimate the impact of draft induced
military service or education may be severely biased due to the use of improper controls.
I. Introduction
The evaluation of causal effects in empirical economics frequently relies on instrumental
variable (IV) methods. Identification requires that the instrument is associated with the
endogenous treatment variable (IV relevance), but is not directly related with the outcome
or unobservables affecting the outcome (IV validity). In many empirical applications,
the latter assumption occurs only reasonable when controlling for further covariates, often
referred to as conditional IV independence. This paper systemicallyworks through different
IV models and discusses the conditions for the satisfaction of conditional IV independence
depending on whether the control variables are measured prior to or after instrument
assignment.
It is remarkable that in many IV studies, the discussion and justification of conditional
IV independence does not pay much attention to the time period in which the control
variables are measured, i.e. whether this happens prior to instrument assignment or at a
JEL Classification numbers: C26, J24
*Wehave benefitted from comments by Dionissi Aliprantis, PerJohansson and seminar par ticipants in Rotterdam,
Aarhus, Linz and Bludenz.
To continue reading
Request your trial