Propensity score matching
Pair treated and untreated units with the same predicted chance of treatment.
What it means
Propensity score matching is an observational technique that reduces confounding by matching treated and control units that share a similar propensity score — the estimated probability of receiving treatment given their covariates. By balancing the distribution of measured covariates across groups, it mimics some features of randomization and simplifies high-dimensional adjustment into a single score. Its critical limitation is that it can only balance variables that were measured: unobserved confounders remain a threat, unlike in a true experiment. Related approaches use the score for weighting, stratification, or regression rather than one-to-one pairing.
Examples
To gauge a job program's effect, each participant is matched to a non-participant with the same likelihood of enrolling based on age, education, and prior wages.
To test whether a new rehab ward speeds recovery, analysts pair each admitted patient with an untreated one of the same modelled odds of admission, given age, diagnosis, and frailty.
A retailer judging its loyalty app compares members with non-members equally likely to sign up on spend and age — but it cannot match on how keen they already were.
First described in Rosenbaum & Rubin (1983).