Instrumental Variables: The Endogeneity Problem
You'll learn when standard regression breaks down for causal inference due to correlation between predictors and errors.
The Core Problem
In standard regression, we assume our predictor variables are independent of the error term. This is called the exogeneity assumption. When this assumption breaks down—when predictors and errors are correlated—we have endogeneity, and our causal estimates become biased.
Think of it like trying to measure how studying affects test scores, but motivation (hidden in the error term) influences both how much students study AND their test performance. The predictor (study time) is now tangled up with unmeasured factors in the error.
Why This Matters for Causality
Endogeneity destroys causal interpretation because:
- Omitted variables: Important confounders are missing from your model and hide in the error term
- Reverse causality: Y might actually cause X, not just the other way around
- Measurement error: Noise in your predictor correlates with the error
When endogeneity exists, your regression coefficient doesn't capture the true causal effect—it's contaminated by these hidden correlations. You might think increasing X by 1 causes Y to rise by β, but you're actually capturing spurious associations.
The Connection to Previous Methods
Unlike RCTs (where randomization breaks all correlations) or matching/DiD (which control for specific confounders), endogeneity represents a deeper structural problem that those methods can't always fix in observational data.
Key Takeaway: Endogeneity occurs when predictors correlate with the error term, violating the core assumption needed for causal inference in regression—setting the stage for instrumental variables as a solution.