Core Concepts

This guide provides a comprehensive overview of CausalBoundingEngine’s concepts, scenarios, and usage patterns.

Causal Bounding

Causal inference often faces the challenge of unmeasured confounding - variables that affect both treatment and outcome but are not observed. When identification of causal effects is impossible, causal bounding provides a principled approach to determine the range of possible causal effects compatible with the observed data and assumptions.

CausalBoundingEngine focuses on two key causal quantities:

Average Treatment Effect (ATE):

The difference in expected outcomes between treated and untreated states:

\[ATE = E[Y(1)] - E[Y(0)]\]

Probability of Necessity and Sufficiency (PNS):

The probability that treatment is both necessary and sufficient for a positive outcome:

\[PNS = P(Y(1)=1, Y(0)=0)\]

Scenarios

CausalBoundingEngine organizes algorithms by scenarios - different causal settings that determine which algorithms are applicable.