BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Towards a Complete Characterization of Target Law Identification i
 n Missing Data DAG Models - Razieh Nabi (Emory University)
DTSTART:20260303T094500Z
DTEND:20260303T103000Z
UID:TALK244369@talks.cam.ac.uk
DESCRIPTION:It is often said that the fundamental problem of causal infere
 nce is a missing data problem: comparisons of potential outcomes are diffi
 cult because only one response is observed for each unit. In this talk we 
 consider the converse perspective\, that missing data problems can be view
 ed&nbsp\;as causal inference problems\, but in an important sense harder o
 nes. Recovering the complete data law from the observed law can be interpr
 eted as identifying a joint distribution over counterfactual variables cor
 responding to values that would have been observed had measurement been po
 ssible. We study non-ignorable missingness (MNAR) models by imposing struc
 tural restrictions on the full data distribution\, consisting of an (un)re
 stricted target distribution together with a missingness mechanism that fa
 ctorizes according to a directed acyclic graph. This graphical formulation
  allows ideas from causal identification to be applied\, while also reveal
 ing gaps between causal and missing data identification. A key obstacle ar
 ises when missingness indicators are treated interventionally: sequences o
 f interventions can induce and propagate selection bias\, causing identifi
 cation to fail even in settings where familiar causal tools suggest succes
 s. The talk describes how these phenomena appear\, when they can be avoide
 d\, and when they fundamentally obstruct recovery of the target law. We di
 scuss partial solutions\, remaining failure modes\, and what structure a c
 omplete graphical identification theory for missing data DAG models would 
 need to capture.&nbsp\;
LOCATION:Seminar Room 1\, Newton Institute
END:VEVENT
END:VCALENDAR