The Impact of Instrumental Variables on Causal Inference
While being concerned with the presence of causality, empirically, causal relationships
are arguably among the toughest challenges to look for in practice. In fact, instrumental variables
have turned out to be, arguably, the most powerful econometric tool to address endogeneity
problems and to identify economic causality—a channel for the effect of a cause on an effect.
The potential of estimating causal effects has been substantially boosted through the use of
instrumental variables in applied econometrics, especially when it is impossible, or indeed
greatly unethical, to use random experiments.
Instrumental variables are precisely tailored to solve the grand problem of endogeneity,
which, in simple terms, means that problems such as explanatory variables of a model may find
themselves being correlated with the error term. Sources of this correlation could be many, for
example, omitted variable bias, measurement error, or reverse causality (Columbia University,
n.d.). The centerpiece of what one actually does within the IV framework is to find a variable—
an instrument—that is related to the endogenous explanatory variable but is uncorrelated with
the error term. For example, in research on returns to education, a common example of an
instrument for years of schooling is the distance to college. This instrument then affects the
likelihood of attending college (and thus years of education) but should be uncorrelated with the
innate ability or other unobservable factors that might affect earnings. Instrumental Variables
provide consistently estimated causal impacts of education on earnings.
While instrumental variables offer a powerful tool for causal inference, their application
should be made cautiously, with careful consideration and validation. The fundamental problem
is to find instruments that satisfy the exclusion restriction: the instrument changes the dependent
variable but only through the explanatory variable (Columbia University, n.d.). Of course, the
researcher has to base the argument and test the data on the strong theoretical validation of his
instruments. Furthermore, the interpretation of the IV estimates as LATE, not ATE, hinges on
the subtle distinctions within the population over which the specific causal effect is being
estimated. To be clear, in the example from the realm of education, the estimate derived from IV
would be an estimate of the effect of schooling on earnings for the person with a high-school
degree because there was a college nearby rather than the average effect for society.
In conclusion, instrumental variables have sharply and irrevocably changed the face of
applied econometrics through the construction of a firm route to estimate causal effects in
observational studies. They introduced new opportunities within the endogenous part of
problems in cases where the realization of the randomized experiment was infeasible. However,
this calls for caution in the use of the plausibly exogenous instrument and in the interpretation of
the results. As econometric tools continue to develop, it is likely that the use of instrumental
variables will remain for many years at the center of the methods used to offer causal inference
in economics, allowing researchers to keep uncovering important insights about economic
relationships and, therefore, to inform evidence-based policy.
Reference
Columbia University. (n.d.). Instrumental Variables | Columbia Public Health.
Www.publichealth.columbia.edu.
https://www.publichealth.columbia.edu/research/population-health-methods/
instrumental-variables
