Home News Reduce the bias by using the inverse probability weighting (IPTW).

Reduce the bias by using the inverse probability weighting (IPTW).


The assignment of a person to a treatment group using observational data is non-random, so causal inference can be difficult. Propensity score weighting is a common way to address this issue. The propensity score measures the likelihood that someone will be assigned to the treatment arm based on their observable characteristics. The logistic regression of individual characteristics is used to estimate the propensity. This binary variable indicates whether or not the person received treatment. In order to calculate treatment effects, known confounders are removed from the propensity score.

A paper by Xu et. al. (2010) shows that IPTW may cause an overestimation in the pseudo-sample size, which can increase the risk of a type I mistake (i.e., rejecting a null hypothesis when it actually is true). These authors state that robust variance estimators can solve this problem but they are only able to work with large sample sizes. Instead, Xu with co-authors suggested that standardized weights be used in the IPTW. This is a simple and easy strategy to implement. Here’s how it works.

IPTW simply looks at the differences between the treated and untreated…

Continue reading…

Previous articleExecutives share 7 Life Sciences Predictions to Look Out For in 2023
Next articleShiftKey, a startup that provides healthcare staffing, raises $300M