• Corpus ID: 220686442

Causal Discovery with Unobserved Confounding and non-Gaussian Data

@article{Wang2020CausalDW,
  title={Causal Discovery with Unobserved Confounding and non-Gaussian Data},
  author={Y. Samuel Wang and Mathias Drton},
  journal={arXiv: Methodology},
  year={2020}
}
We consider the problem of recovering causal structure from multivariate observational data. We assume that the data arise from a linear structural equation model (SEM) in which the idiosyncratic errors are allowed to be dependent in order to capture possible latent confounding. Each SEM can be represented by a graph where vertices represent observed variables, directed edges represent direct causal effects, and bidirected edges represent dependence among error terms. Specifically, we assume… 

Figures and Tables from this paper

Causal Discovery in Linear Structural Causal Models with Deterministic Relations

This paper focuses on a subclass of linear SCMs which do not have this property, i.e., models in which observed variables can be causally affected by any subset of the sources, and are allowed to be a deterministic function of other observed variables or latent confounders.

Differentiable Causal Discovery Under Unmeasured Confounding

This work derives differentiable algebraic constraints that fully characterize the space of ancestral acyclic directed mixed graphs, as well as more general classes of ADMGs, arid AD MGs and bow-free ADMGS, that capture all equality restrictions on the observed variables.

Causal discovery in heavy-tailed models

This work defines the causal tail coefficient that captures asymmetries in the extremal dependence of two random variables and proposes a computationally highly efficient algorithm that estimates the causal structure.

Learning linear non-Gaussian graphical models with multidirected edges

A new method to learn the underlying acyclic mixed graph of a linear non-Gaussian structural equation model given observational data is proposed and it is shown that one can augment the hidden variable structure of the recovered model by learning em multidirected edges rather than only directed and bidirected ones.

Equality Constraints in Linear Hawkes Processes

It is proved that Hawkes process models in a certain sense satisfy the equality constraints of linear structural equation models, which allows more refined constraint-based structure learning in this class of continuous-time stochastic processes.

Causal Discovery in Linear Latent Variable Models Subject to Measurement Error

It is shown that if both parts of the faithfulness assumption are imposed, for the latent variable model with unobserved parentless causes, the structure can be identified, and proposed causal structure learning methods for both models are proposed.

Causal Discovery with Heterogeneous Observational Data: Supplementary Materials

Proof techniques Because of the difference illustrated in the example above, to prove our causal identifiability results, existing proofs in the literature do not apply and significant efforts are

Identification of Linear Non-Gaussian Latent Hierarchical Structure

This paper defines a minimal latent hierarchical structure and develops a princi-pled method to identify the structure by testing for Generalized Independent Noise conditions in specific ways, and shows the effectiveness of the proposed approach on both synthetic and real-world data.

References

SHOWING 1-10 OF 76 REFERENCES

Learning Linear Non-Gaussian Causal Models in the Presence of Latent Variables

It can be shown that causal effects among observed variables cannot be identified uniquely even under the assumptions of faithfulness and non-Gaussianity of exogenous noises, but an efficient method to identify the set of all possible causal effects that are compatible with the observational data is proposed.

High-dimensional causal discovery under non-Gaussianity

This work considers graphical models based on a recursive system of linear structural equations and proposes an algorithm that yields consistent estimates of the graph also in high-dimensional settings in which thenumber of variables may grow at a faster rate than the number of observations, but inWhich the underlying causal structure features suitable sparsity.

A Linear Non-Gaussian Acyclic Model for Causal Discovery

This work shows how to discover the complete causal structure of continuous-valued data, under the assumptions that (a) the data generating process is linear, (b) there are no unobserved confounders, and (c) disturbance variables have non-Gaussian distributions of non-zero variances.

High-dimensional learning of linear causal networks via inverse covariance estimation

It is shown that when the error variances are known or estimated to close enough precision, the true DAG is the unique minimizer of the score computed using the reweighted squared l2-loss.

Linear Structural Equation Models with Non-Gaussian Errors: Estimation and Discovery

This thesis focuses on SEMs in which unobserved latent variables have been marginalized and only observed variables are explicitly modeled, and proposes several modifications to a naive procedure and shows that empirical likelihood can be an attractive alternative to existing methods when the data is non-Gaussian.

Causal discovery of linear non-Gaussian acyclic models in the presence of latent confounders

The results of experimental validation using simulated data and real-world data confirmed that RCD is effective in identifying latent confounders and causal directions between observed variables.

Ordering-Based Causal Structure Learning in the Presence of Latent Variables

It is proved that under assumptions weaker than faithfulness, any sparsest independence map (IMAP) of the distribution belongs to the Markov equivalence class of the true model, and a greedy algorithm is proposed over the space of posets for causal structure discovery in the presence of latent confounders.

Bayesian estimation of causal direction in acyclic structural equation models with individual-specific confounder variables and non-Gaussian distributions

An approach to determining the possible causal direction between two observed variables when latent confounding variables are present and a new linear non-Gaussian acyclic structural equation model with individual-specific effects that are sometimes the source of confounding is proposed.
...