Skip to Content


Unsupervised and Supervised methods of Propensity Score Adjustment for Bias
Bob Obenchain <>
GPL (>= 2)
Unsupervised PS Methods define Local Treatment Differences (LTDs) within numerous Clusters of patients well-matched on their pre-treatment X-characteristics and display the resulting distribution of local effect-size estimates across Clusters. I now prefer to call this form of Nonparametric Preprocessing of observational outcomes Local Control; it uses patient blocking / matching concepts so as to rely only on a simple model (Nested ANOVA, treatment within cluster) that becomes more and more relastic as Clusters become small and numerous. In sharp contrast, the Supervised PS Methods provided here attempt to estimate unknow true Propensities with parametric models that can be quite wrong and unrealistic. PS estimates always need to be Validated; there is usually no guarantee that such estimatres actually block patients with similar X-characteristics together, like true propensities do.
Package Version Released
USPS 1.2-2 4 years 5 weeks ago
USPS 1.2-1 4 years 30 weeks ago
USPS 1.2-0
USPS 1.1-0
Your rating: None
Your rating: None