All simulators
Overlap & Trimming
Mirrored propensity-score histograms, extreme weights, effective sample size and what trimming costs.
Day 42: Unconfoundedness & Overlapnormalised difference in x
1.17
|·| > 0.25 is a warning
kept after trimming
100.0%
largest IPW weight
49.6
effective sample size
756 + 597
of 1492 treated + 1508 untreated
Who gets a new statin: treated patients (top) and untreated (bottom) by propensity score. Where one group is absent, no comparison exists and weights explode. Trimming to [α, 1 − α] buys stable weights but changes the population the estimate describes.