Unobserved Joint Distributions

Reinterpreting causal discovery as the task of predicting unobserved joint statistics

May, 2023

Abstract

If X,Y,Z denote sets of random variables, two different data sources may contain samples from PX,Y and PY,Z, respectively. We argue that causal discovery can help inferring properties of the `unobserved joint distributions' PX,Y,Z or PX,Z. The properties may be conditional independences (as in `integrative causal inference') or also quantitative statements about dependences.