Ancestor regression in linear structural equation models

TL;DR

This paper introduces a new statistical testing method for causal discovery in linear structural equation models, capable of distinguishing ancestors from non-ancestors and estimating causal order with error control, even under Gaussianity.

Contribution

The paper presents a simple, statistically rigorous trick for causal discovery that extends to estimating causal order and provides error control guarantees, addressing limitations of existing methods.

Findings

Provides explicit asymptotic error control for false causal discovery.

Extends causal discovery to Gaussian models where other methods fail.

Includes a valid goodness of fit test for linear SEMs.

Abstract

We present a new method for causal discovery in linear structural equation models. We propose a simple ``trick'' based on statistical testing in linear models that can distinguish between ancestors and non-ancestors of any given variable. Naturally, this can then be extended to estimating the causal order among all variables. We provide explicit error control for false causal discovery, at least asymptotically. This holds true even under Gaussianity, where other methods fail due to non-identifiable structures. These type I error guarantees come at the cost of reduced empirical power. Additionally, we provide an asymptotically valid goodness of fit p-value to assess whether multivariate data stems from a linear structural equation model.