Answers are grounded in this paper and cite their sources.
I congratulate the authors on the many advances to likelihood-based causal inference introduced in their contribution.I foresee its influence on many future theoretical and practical developments.This motivates understanding the ideas presented here under different perspectives, providing alternative ways of constructing models while facilitating extensions.Let us focus on structural causal models [SCMs, Pearl, 2009] and their relation to the contribution.Evans and Didelez suggest that the typical SCM construction, by conditional distributions, is at odds with marginal modeling.However, there is a long history of linear Gaussian SCMs applications, where error terms are commonly parameterized by marginal models [Richardson and Spirtes, 2002].Structural equations and marginal models go along, regardless of the challenges of encoding conditional independence constraints.Those can be helpful, but not always necessary, and they are not a defining characteristic of SCMs.Consider the following example: we would like to model the effect of X on Y when pre-treatment covariates Z also affect Y .Hidden background variables U x , U y and U z are postulated along the following structural equations:
No takes yet. Share an insight, caveat, or question.
Ricardo Silva (2024) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: