hard20 pts · first answer countscausal-inference

Propensity scores near 0 and 1

You estimate the effect of a loyalty programme with propensity score weighting. Many treated users have propensity above 0.98 and many controls below 0.02. What is the main concern?

A.The model is well calibrated, so weighting is safeB.Poor overlap: few comparable users exist, so estimates rest on a handful of huge weights and extrapolationC.Propensity scores must be exactly 0.5 for all usersD.The outcome model must be a neural network

Discussion 0

Join free to post in the discussion.

Propensity scores near 0 and 1

Discussion 0

Related questions