Stabilized Weights
Stabilized weights are an alternative form of inverse probability weights that are sometimes used to reduce the variance of causal effect estimates. The impact of stabilized weights is primarily relevant when truncation is applied, because stabilization changes the scale of the weights and therefore affects which observations exceed truncation thresholds. However, when the treatment variable is binary and no truncation is applied, selecting the Stabilized Weights option does not change the fitted results. In this case, the marginal structural model is saturated, so stabilized and unstabilized weights produce identical estimates.
When the Stabilized Weights option is not selected, inverse probability weights are calculated by using the following equation:
w = 1/p(X|Z)
where
X is the binary treatment variable
Z is the set of pretreatment covariates that are specified in the treatment model
p(X|Z) is the probability of receiving treatment given covariates Z, calculated using the fitted treatment model.
Alternatively, stabilized weights are calculated by using the following equation:
wstabilized = p(X)/p(X|Z)
where p(X) is the marginal probability of receiving treatment, calculated by using a fitted intercept-only model.