| Jim Blum |
Inverse probability (IP) weighting is a broadly-applicable method for creating exchangeability between the
intervention and outcome, conditional on the confounders. At the most basic level, we estimate the probability
(propensity) for various intervention levels as a function of the confounders. These probabilities are used to reweight
the original data, creating a pseudo population where likelihood of treatment assignment is disconnected
(independent) of the confounders in the original population. Analysis on this pseudo population permits estimation of
the causal effect of the treatment on the outcome. Both IP weighting and stabilized IP weighting are considered, for dichotomous and more complex treatments. |