Propensity-score clipping and trimming

Create extreme inverse-probability weights, constrain them, and compare that operation with excluding people outside common support.

How are clipping and trimming different?

Clipping moves fitted propensity scores away from zero and one when weights are calculated, retaining every person but modifying the estimator. Trimming excludes observations under a stated support rule, so the result targets the retained population rather than the original full population.

Try it in the experiment

  1. Increase treatment selection to create poorer overlap.
  2. Raise the clipping threshold and compare the weights and point estimates.
  3. Open the trimming experiment and inspect who is excluded.

Neither operation repairs a structural positivity violation. Clipping trades sensitivity to extreme weights for bias; trimming changes both the sample and the population described by the estimate. Report the rule and examine sensitivity rather than presenting either as automatic.