Summary of 6.4 - Propensity Scores and Inverse Probability Weighting (IPW)

00:00:00 - 00:10:00

The video discusses the concept of propensity scores and inverse probability weighting (IPW), and how they can be used to estimate the average treatment effect (ATE). IPW is a method that adjusts the weight of a confounder in the calculation of the ATE, and is based on the assumption that the probability of receiving a particular treatment is not affected by factors that are unrelated to the treatment itself. The video explains that IPW models different aspects of the ATE than the com estimation method, and provides a graphical intuition for how IPW works.

• 00:00:00 The propensity score is a single-dimensional scalar that describes your propensity for taking a particular treatment in the presence of relevant covariates. The propensity score theorem states that if w blocks all backdoor paths from treatment to outcome, then the propensity score equals e of w. This theorem has important implications for the positivity unconfoundedness trade-off, as the overlap between the measurements decreases with the dimensionality of the adjustment set.
• 00:05:00 Inverse probability weighting (IPW) is a method used to estimate causal effects when observational data is insufficient. It re-weights observational data by the inverse of the propensity score, which removes the effect of confounding variables from the causal estimate.
• 00:10:00 Inverse probability weighting is a method for estimating the average treatment effect (ATE), which is a measure of the difference in outcomes between people who receive a treatment compared to those who do not receive the treatment. This method is based on the assumption that the probability of receiving a particular treatment is not affected by factors that are unrelated to the treatment itself. The graphical intuition for inverse probability weighting is that it adjusts the weight of a confounder in the calculation of the ATE. This is important because it takes into account the fact that different factors might influence the probability of receiving a treatment differently. In contrast, com estimation assumes that the probability of receiving a treatment is the same for everyone, which can lead to incorrect estimates of the ATE. IPW also models different aspects of the ATE than com does. IPW models the average treatment effect across all possible doses of the treatment, whereas com only models the average treatment effect for the dose that was actually given. Additionally, IPW models the interaction between the treatment and the confounder, whereas com only models the treatment effect.