Instrumental Variables

Instrumental Variables (IV) are variables used in statistical and econometric models to provide consistent estimators when the explanatory variables are correlated with the error term, causing endogeneity bias. Endogeneity arises in regression analysis when an independent variable is correlated with the residuals, violating the classical assumption of exogeneity and leading to biased and inconsistent ordinary least squares (OLS) estimates. Instrumental Variables serve as tools to correct this problem by providing a source of variation in the endogenous regressors that is not correlated with the error term.

Definition and Purpose of Instrumental Variables

Instrumental Variables are variables that satisfy two key conditions:

1. Relevance: The instrument must be correlated with the endogenous explanatory variable. This ensures that the instrument can explain some variation in the problematic regressor.
2. Exogeneity: The instrument must be uncorrelated with the error term in the outcome equation. This condition ensures that the instrument affects the dependent variable only through its effect on the endogenous regressor, not directly or through omitted variables.

The primary purpose of using instrumental variables is to obtain consistent and unbiased estimates of causal effects in models where traditional regression techniques fail due to endogeneity.

The Instrumental Variables Estimation Framework

Model Setup

Consider the structural equation:

where:

• Y is the dependent variable,
• X is the endogenous explanatory variable,
• β is the parameter of interest,
• u is the error term correlated with X.

Because of the correlation between X and u, OLS estimation of β is biased.

Instrumental Variable Conditions

To address this, we introduce an instrument Z that satisfies:

• Cov(Z, X) ≠ 0 (Relevance)
• Cov(Z, u) = 0 (Exogeneity)

Under these conditions, the variation in X explained by Z is free from the endogeneity bias, allowing consistent estimation.

Two-Stage Least Squares (2SLS)

The most common estimation method using instrumental variables is Two-Stage Least Squares:

• First Stage: Regress the endogenous variable X on the instrument Z:

  $$\backslash mrow>\; X\; =\; \backslash pi\; Z\; +\; v$$

  where π is the first-stage coefficient, and v is the first-stage error term.

• Second Stage: Use the predicted values of X (denoted as Ŷ from the first stage) to estimate the structural equation:

  $$\backslash mrow>\; Y\; =\; \backslash beta\; \backslash hat\{X\}\; +\; \backslash epsilon$$

This process isolates the variation in X that is exogenous, yielding consistent estimates of β.

Properties and Identification

Identification

For the IV estimator to be identified, the number of valid instruments must be at least equal to the number of endogenous regressors. In the just-identified case, the number of instruments equals the number of endogenous variables. If there are more instruments than endogenous regressors, the model is overidentified, allowing for tests of instrument validity.

Instrument Strength

The strength of an instrument refers to how strongly it is correlated with the endogenous regressor. Weak instruments, with low correlation, lead to biased IV estimates and large standard errors. Tests such as the F-statistic in the first stage are used to assess instrument strength; values above 10 are typically considered indicative of sufficiently strong instruments.

Testing Instrument Validity

Overidentification Tests

When there are multiple instruments, overidentification tests like the Sargan or Hansen J test are applied to assess whether the instruments are valid, i.e., uncorrelated with the error term. Failure to reject the null hypothesis suggests instruments are valid.

Endogeneity Tests

Tests such as the Durbin-Wu-Hausman test compare IV estimates to OLS estimates to detect the presence of endogeneity. Significant differences indicate that OLS estimates are biased and IV estimation is necessary.

Applications and Challenges

Applications

Instrumental variables are widely used in economics and related fields to estimate causal relationships where controlled experiments are not feasible. Examples include estimating returns to education, effects of health interventions, or price elasticity when prices are endogenous.

Challenges

• Finding Valid Instruments: Identifying instruments that satisfy both relevance and exogeneity is often difficult.
• Weak Instruments: Using weak instruments can produce unreliable estimates.
• Interpretation: IV estimates identify local average treatment effects (LATE), meaning the estimated effect pertains to the subpopulation affected by the instrument, which may limit generalizability.

Summary of the Instrumental Variables Approach

Instrumental Variables provide a powerful method to address endogeneity by leveraging external sources of variation that affect the explanatory variables but not the outcome directly. The methodology relies on strong theoretical justification and empirical testing of instrument validity and strength to deliver consistent and reliable causal estimates in observational data contexts where standard regression techniques fail.