Propensity Score: Concept and Estimation
You'll learn: What a propensity score is and how to estimate the probability that an observation receives treatment based on its characteristics.
The Core Idea
In observational studies, treatment assignment isn't random—people self-select or are selected based on their characteristics. A propensity score is the probability that an individual receives treatment given their observed covariates (characteristics). Instead of matching on many variables individually (as in exact matching), we can summarize all those characteristics into a single score between 0 and 1.
Think of it like a credit score: rather than lenders comparing your income, debt, employment history, and age separately, they use one number that captures all those factors. Similarly, the propensity score captures how "likely" someone is to be treated based on everything we know about them.
Why Propensity Scores Matter
When treatment assignment is based on observed characteristics (confounders), comparing treated and control groups directly gives biased estimates. If we can balance groups that have similar propensity scores, we're comparing apples to apples—people who were equally likely to receive treatment but some did and some didn't.
Estimating Propensity Scores
The most common method is logistic regression:
- Your outcome variable: Treatment status (1 = treated, 0 = control)
- Your predictors: All relevant covariates that might influence treatment assignment
- The fitted probability from this model is the propensity score
Each observation gets a score estimating their probability of treatment. These scores can then be used for matching, weighting, or stratification to balance groups and estimate causal effects.
Key Takeaway: The propensity score reduces the dimensionality problem in matching by summarizing all covariates into a single probability of treatment assignment, typically estimated using logistic regression.