It is a special case of matching: the distance metric is the (estimated) propensity score

1-to-n nearest neighbor matching is common when the control group is large compared to treatment group

**Pros: **

intuitive, robust, matched pairs, balance distributions in directions uncorrelated to estimated PS

**Cons:**

- much tuning: with or without replacement, 1-to-1 or 1-to-n, caliper, ties
- programming is hard
- difficult to extend to complex situations: sequential treatments, multi-valued treatments