Instrumental Variables

Slide 1 of 8

What is an Instrumental Variable?

An instrumental variable (IV) is a variable Z that satisfies three critical conditions enabling causal inference even when unmeasured confounding is present. This makes IVs one of the most powerful tools in the causal inference toolkit, particularly when randomization is not possible.

Definition: Instrumental Variable An instrumental variable Z is a variable that affects the treatment A but has no direct effect on the outcome Y except through its effect on A, and is independent of all unmeasured confounders of the treatment-outcome relationship.

The three core conditions for a valid instrument are:

1. Relevance (Instrument Strength): The instrument Z must be associated with the treatment A. Mathematically, we require $Cov(Z, A) \neq 0$ . The strength of this association directly affects the precision and bias of IV estimates.

2. Exclusion Restriction: The instrument Z affects the outcome Y ONLY through the treatment A. There is no direct path from Z to Y. This is the most critical and often most controversial assumption because it cannot be tested with data.

3. Independence (Exchangeability): The instrument Z is independent of unmeasured confounders U that affect both A and Y. Formally, $Z \perp\!\!\!\perp U$ . This ensures the instrument is "as good as randomly assigned."

The causal structure can be visualized as:

Z --> A --> Y
       ^   ^
        \ /
         U

Example: Mendelian Randomization in Cardiovascular Research Consider studying whether elevated LDL cholesterol causes coronary heart disease (CHD). Observational studies are confounded by diet, exercise, and socioeconomic factors. However, genetic variants affecting LDL metabolism (like PCSK9 variants) are: (1) Associated with LDL levels (relevance), (2) Affect CHD only through LDL changes (exclusion), (3) Randomly assigned at conception (independence). This approach has provided causal evidence supporting statin therapy.

Key Insight The power of instrumental variables lies in their ability to identify causal effects even when unmeasured confounding exists. This makes them invaluable in observational research where complete confounder adjustment is impossible.

Use ← → arrow keys to navigate