On Multi-Cause Causal Inference with Unobserved Confounding: Counterexamples, Impossibility, and Alternatives
arXiv:1902.10286
Abstract
Unobserved confounding is a central barrier to drawing causal inferences from observational data. Several authors have recently proposed that this barrier can be overcome in the case where one attempts to infer the effects of several variables simultaneously. In this paper, we present two simple, analytical counterexamples that challenge the general claims that are central to these approaches. In addition, we show that nonparametric identification is impossible in this setting. We discuss practical implications, and suggest alternatives to the methods that have been proposed so far in this line of work: using proxy variables and shifting focus to sensitivity analysis.
Accepted to AISTATS 2019. Since last revision: corrected constant factors in linear gaussian example; fixed typos
References in corpus (7)
- Identifiability of parameters in latent structure models with many observed variables
- Causal Effect Inference with Deep Latent-Variable Models
- The Deconfounded Recommender: A Causal Inference Approach to Recommendation
- Identifying Causal Effects With Proxy Variables of an Unmeasured Confounder
- Multiple Causal Inference with Latent Confounding
- Implicit Causal Models for Genome-wide Association Studies
- Flexible sensitivity analysis for observational studies without observable implications
Cited by in corpus (7)
- Causal Intervention for Weakly-Supervised Semantic Segmentation
- Adversarial Visual Robustness by Causal Intervention
- Discussion of "The Blessings of Multiple Causes" by Wang and Blei
- Counterfactual Situation Testing: From Single to Multidimensional Discrimination
- Learning latent causal graphs via mixture oracles
- Identification of Latent Variables From Graphical Model Residuals
- Comment: Reflections on the Deconfounder