Difference between revisions of "Instrumental Variables"
(Created page with "Instrumental variables (IV) estimation is quasi-experimental approach to data analysis and impact evaluation that overcome...") |
|||
| (One intermediate revision by one other user not shown) | |||
| Line 1: | Line 1: | ||
Instrumental variables (IV) estimation is [[Quasi-Experimental Methods | quasi-experimental]] approach | Instrumental variables (IV) estimation is a [[Quasi-Experimental Methods | quasi-experimental]] approach that overcomes endogeneity through the use of a valid instrument. IV estimation is a useful method in [[Data Analysis | data analysis]] to consistently estimate causal impact in the presence of omitted variables, measurement errors, or simultaneity between the outcome (Y) and the explanatory variable of interest (X). This page provides a general overview of IV estimation and assumptions. For more technical details on implementation, see [[Instrumental_Variables#Additional_Resources | Additional Resources]]. | ||
==Read First== | ==Read First== | ||
* | *To be valid, an instrumental variable must be correlated with the explanatory variable of interest (X) and uncorrelated with the error term (e). | ||
*<code>ivreg2</code> is a Stata command that conducts instrumental variable estimation and runs tests for over-identifying restrictions, under-identification, and weak instruments. | *<code>ivreg2</code> is a [[Stata Coding Practices|Stata]] command that conducts instrumental variable estimation and runs tests for over-identifying restrictions, under-identification, and weak instruments. | ||
==Overview== | ==Overview== | ||
Instrumental variables (IV) estimation is [[Quasi-Experimental Methods | quasi-experimental]] approach | Instrumental variables (IV) estimation is a [[Quasi-Experimental Methods | quasi-experimental]] approach that uses a valid instrument to overcome endogeneity (i.e. omitted variables, measurement error, or simultaneity). In cases of endogeneity, an OLS '''regression''' of the outcome variable (Y) on the explanatory variable of interest (X) yields an inconsistent estimate. While [[Randomized Control Trials | randomized control trials (RCTs)]] ensure exogeneity through [[Randomization in Stata | randomization]], they are not always logistically or ethically feasible. In these situations, '''quasi-experimental methods''', like IVs, prove useful for measuring causal impact under the [[Exogeneity Assumption|exogeneity assumption]]. | ||
A valid instrument must meet both the relevance and exogeneity conditions. The relevance condition states that the instrument is correlated with the explanatory variable of interest (X). The exogeneity condition states that the instrument is uncorrelated with the error term (e). In other words, the instrument affects the outcome (Y) only through X. To estimate causal impact with the instrumental variable, researchers can use two-stage least squares (2SLS), generalized method of moments (GMM) or k-estimators. For more information on implementing these methods, see [[Instrumental Variables#Additional | A valid instrument must meet both the relevance and '''exogeneity''' conditions. The relevance condition states that the instrument is correlated with the explanatory variable of interest (X). The '''exogeneity''' condition states that the instrument is uncorrelated with the error term (e). In other words, the instrument affects the outcome (Y) only through X. To estimate causal impact with the instrumental variable, researchers can use two-stage least squares (2SLS), generalized method of moments (GMM) or k-estimators. For more information on implementing these methods, see [[Instrumental Variables#Additional Resources | Additional Resources]]. | ||
<code>ivreg2</code>is a Stata command that implements IV estimation. It provides tests of over-identifying restrictions, under-identification, and weak instruments. To install the command, type <code>ssc install ivreg2</code> in Stata | <code>ivreg2</code>is a [[Stata Coding Practices|Stata]] command that implements IV estimation. It provides tests of over-identifying restrictions, under-identification, and weak instruments. To install the command, type <code>ssc install ivreg2</code> in '''Stata'''. For more information on the command and its options, type <code>help ivreg2</code>. | ||
==Back to Parent== | ==Back to Parent== | ||
Latest revision as of 16:05, 3 August 2023
Instrumental variables (IV) estimation is a quasi-experimental approach that overcomes endogeneity through the use of a valid instrument. IV estimation is a useful method in data analysis to consistently estimate causal impact in the presence of omitted variables, measurement errors, or simultaneity between the outcome (Y) and the explanatory variable of interest (X). This page provides a general overview of IV estimation and assumptions. For more technical details on implementation, see Additional Resources.
Read First
- To be valid, an instrumental variable must be correlated with the explanatory variable of interest (X) and uncorrelated with the error term (e).
ivreg2is a Stata command that conducts instrumental variable estimation and runs tests for over-identifying restrictions, under-identification, and weak instruments.
Overview
Instrumental variables (IV) estimation is a quasi-experimental approach that uses a valid instrument to overcome endogeneity (i.e. omitted variables, measurement error, or simultaneity). In cases of endogeneity, an OLS regression of the outcome variable (Y) on the explanatory variable of interest (X) yields an inconsistent estimate. While randomized control trials (RCTs) ensure exogeneity through randomization, they are not always logistically or ethically feasible. In these situations, quasi-experimental methods, like IVs, prove useful for measuring causal impact under the exogeneity assumption.
A valid instrument must meet both the relevance and exogeneity conditions. The relevance condition states that the instrument is correlated with the explanatory variable of interest (X). The exogeneity condition states that the instrument is uncorrelated with the error term (e). In other words, the instrument affects the outcome (Y) only through X. To estimate causal impact with the instrumental variable, researchers can use two-stage least squares (2SLS), generalized method of moments (GMM) or k-estimators. For more information on implementing these methods, see Additional Resources.
ivreg2is a Stata command that implements IV estimation. It provides tests of over-identifying restrictions, under-identification, and weak instruments. To install the command, type ssc install ivreg2 in Stata. For more information on the command and its options, type help ivreg2.
Back to Parent
This article is part of the topic Quasi-Experimental Methods
Additional Resources
- Cameron’s Instrumental Variables
- Wooldridge’s Chapter 15: Instrumental variables and two stage least squares
- David McKenzie’s Rethinking identification under the Bartik Shift-Share Instrument via the World Bank’s Development Impact blog
- Nagler’s Notes on Simultaneous Equations and Two Stage Least Squares Estimates
- Angrist and Pischke’s Mostly Harmless Econometrics
- Find an overview of three Stata commands for instrumental variable estimation here