Instrumental variables (IV) Instrumental Variableare statistical tools that help address confounding and identify causal effects in observational studies, especially when randomized trials are impractical or unethical. They work by isolating exogenous variation in the treatment variable, allowing researchers to estimate causal relationships reliably. The Hidden Bias: Machine learning models can be severely biased by unobserved confounders, variables that influence both the input features and the outcome but are not included in the training data. IVs as a Solution: Instrumental variables [1] provide a way to estimate causal effects even when such unobserved confounders are present, offering a more robust approach to causal inference in machine learning. Analogy: The Lever: An instrumental variable is like a lever that we can use to move the treatment variable without directly affecting the outcome, allowing us to isolate the treatment’s true causal effect.