Course contentsShow
← Data Science
Lesson 1446 of 2,14536. Causal Inference BasicsFree lesson

Exact Matching

Learn to match units with identical covariate values and understand when this approach is practical or limiting.

Exact Matching

You'll learn how to pair treated and control units that share identical covariate values to estimate causal effects.

The Core Idea

Exact matching is the simplest form of matching in observational studies. The goal is to compare "apples to apples" by finding control units (untreated) that have exactly the same measured characteristics as treated units.

Imagine you're studying whether a training program (treatment) increases salaries. You have a 30-year-old employee with a bachelor's degree who took the training. With exact matching, you'd pair them only with 30-year-old, bachelor's-degree holders who didn't take the training. Every covariate must match perfectly.

When It Works

Exact matching is ideal when you have:

  • Few covariates to match on (2-4 variables)
  • Discrete or categorical variables (e.g., gender, education level, region)
  • Large datasets where finding exact twins is feasible

When It Fails

The method becomes impractical quickly due to the curse of dimensionality. With continuous variables (like age = 30.5 years) or many covariates, finding exact matches becomes rare or impossible. You might discard most of your data, losing statistical power.

If you need to match on age, income, education, experience, and location simultaneously with exact precision, you'll likely find few or no matches—making causal inference impossible.

The Takeaway

Exact matching creates perfectly balanced treatment and control groups by pairing units with identical covariate values, but it's only practical with few, discrete covariates and large sample sizes—otherwise you risk discarding too much data.

Key Takeaway: Exact matching eliminates confounding from measured covariates by requiring perfect covariate alignment, but quickly becomes impractical as dimensionality increases.