Regression Identifiability and Edge Interventions in Linear Structural Equation Models

TL;DR

This paper introduces regression identifiability, a new criterion for determining when edges in linear structural equation models can be uniquely identified, with graphical conditions and applications to interventional data analysis.

Contribution

The paper develops a novel regression identifiability criterion with necessary and sufficient graphical conditions for edge and covariance matrix identifiability in linear SEMs.

Findings

Established necessary and sufficient conditions for covariance matrix identifiability.

Provided graphical criteria for edge identifiability.

Discussed applications in interventional data analysis and constraint discovery.

Abstract

In this paper, we introduce a new identifiability criteria for linear structural equation models, which we call regression identifiability. We provide necessary and sufficient graphical conditions for a directed edge to be regression identifiable. Suppose $\Sigma^*$ corresponds to the covariance matrix of the graphical model $G^*$ obtained by performing an edge intervention to $G$ with corresponding covariance matrix $\Sigma$. We first obtain necessary and sufficient conditions for $\Sigma^*$ to be identifiable given $\Sigma$. Using regression identifiability, we obtain necessary graphical conditions for $\Sigma^*$ to be identifiable given $\Sigma$. We also identify what would happen to an individual data point if there were such an intervention. Finally, we provide some statistical problems where our methods could be used, such as finding constraints and simulating interventional data from observational data.