Causal inference by using invariant prediction: identification and confidence intervals
By Jan Peters, Peter Buhlmann, Nicolai Meinshausen. 2015
1,210 citations recorded by the source corpus.
What is the difference between a prediction that is made with a causal model and that with a non‐causal model? Suppose that we intervene on the predictor variables or change the whole environment. The predictions from a causal model will in general work as well under interventions as for observational data. In contrast, predictions from a non‐causal model can potentially be very wrong if we actively intervene on variables. Here, we propose to exploit this invariance of a prediction under a causal model for causal inference: given different experimental settings (e.g. various interventions) we collect all models that do show invariance in their predictive accuracy across settings and interventions. The causal model will be a member of this set of models with high probability. This approach yields valid confidence intervals for the causal relationships in quite general scenarios. We examine the example of structural equation models in more detail and provide sufficient assumptions under which the set of causal predictors becomes identifiable. We further investigate robustness properties of our approach under model misspecification and discuss possible extensions. The empirical properties are studied for various data sets, including large‐scale gene perturbation experiments.
Subfields: Artificial Intelligence, Molecular Biology, Statistics and Probability, Causal Inference & Discovery
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Yoshua Bengio
- Bernhard Schölkopf
- Sergey Levine
- Mohit Bansal
- Percy Liang
- Quoc V. Le
- Jianfeng Gao
- Ruslan Salakhutdinov
In Molecular Biology
- Bernhard Schölkopf
- Jörn-Henrik Jacobsen
- Christina Heinze-Deml
- Nicolai Meinshausen
- Yoshua Bengio
- Joris M. Mooij
- Marloes H. Maathuis
- Stefan Bauer
In Statistics and Probability
- Bernhard Schölkopf
- Peter Buhlmann
- Jan Peters
- Yoshua Bengio
- Kun Zhang
- Nicolai Meinshausen
- David Lopez-Paz
- Dominik Janzing
In Causal Inference & Discovery
- Bernhard Schölkopf
- Kun Zhang
- Jan Peters
- Dominik Janzing
- David Lopez-Paz
- Joris M. Mooij
- Peter Buhlmann
- Yoshua Bengio
Citations in the corpus
- Shortcut learning in deep neural networks (2020, background): “Under certain assumptions the intended (or even causal) solution can be learned from multiple domains and environments [131, 39, 111].”
- Underspecification Presents Challenges for Credibility in Modern Machine Learning (2020, background): “Several algorithmic approaches have also been proposed to overcome structural issues, including Peters et al. (2016); Heinze-Deml et al. (2018); Arjovsky et al. (2019); Magliacane et al. (2018).”
- Differentiable Causal Discovery from Interventional Data (2020, methodology): “Another type of constraint-based method exploits the invariance of causal mechanisms across interventional distributions, e.g., ICP [27, 14].”
- An Analysis of the Adaptation Speed of Causal Models (2020, background): “Following up on this idea, Peters et al. (2016) designed an algorithm able to identify robust causal features from heterogeneous data.”
- Toward Causal Representation Learning (2021, methodology): “To formalize this general intuition, one can consider a form of out-of-distribution generalization, which can be optimized by minimizing the empirical risk over a class of distributions induced by a causal model of the data [5], [169], [187], [204], [220].”
- Towards Causal Representation Learning (2021, background): “263, 243, 195, 152, 156, 153, 167, 204, 115], complementing the studies of Bareinboim and Pearl [12, 185], and it inspired a thread of work in the statistics community exploiting invariance for causal discovery and other tasks [189, 192, 105, 104, 115].”
- On the Binding Problem in Artificial Neural Networks (2020, background): “We note that the problem of inferring structure has also received considerable attention in the causality literature, often specifically focusing on cause-effect discovery (e.g. see Hoyer et al. (2009); Lopez-Paz et al. (2015); Peters et al. (2016) or Peters et al. (2017) for an overview).”
- Learning explanations that are hard to vary (2020, background): “…2012].14 Based on this insight, it was proposed that regression based on causal features should presents desirable invariance and robustness properties [Mooij et al., 2009, Schölkopf et al., 2012, Peters et al., 2016, Rojas-Carulla et al., 2018, Heinze-Deml et al., 2018, von Kügelgen et al., 2018].”
- Causal Models for Dynamical Systems (2020, methodology): “This finding adds to a recent debate suggesting to add invariance as a fitting criteria to data science methodology [Sch¨olkopf et al., 2012, Yu, 2013, Peters et al., 2016, Bareinboim and Pearl, 2016, Meinshausen et al., 2016, Yu and Kumbier, 2019].”
- Causality for Machine Learning (2019, background): “in Statistics at ETH, he carried on with the invariance idea, leading to a thread of work in the statistics community exploiting invariance for causal discovery and other tasks (Peters et al., 2016; Pfister et al., 2018a; Heinze-Deml et al., 2017; Heinze-Deml and Meinshausen, 2017).”
- Learning Neural Causal Models from Unknown Interventions (2019, methodology): “Table 1: Baseline comparisons: Structural Hamming Distance (SHD) (lower is better) for learned and ground-truth edges on various graphs from both synthetic and real datasets, compared to [37], [51], [18], [15] and [14].”
- Invariant Risk Minimization (2019, methodology): “Fourth, we could follow invariant causal prediction techniques [40].”
- Causal Regularization (2019, background): “It is believed (and can be proven subject to appropriate model assumptions) that finding statistical models that generalize well across different background conditions is closely linked to finding causal models [7, 8, 9, 10].”
- Causal Discovery from Heterogeneous/Nonstationary Data (2019, methodology): “The procedure in S 1 , which will be further discussed in Section 4.1, contains the methods proposed in (Hoover, 1990; Tian and Pearl, 2001; Peters et al., 2016) for causal discovery from changes as special cases.”
- A Meta-Transfer Objective for Learning to Disentangle Causal Mechanisms (2019, methodology): “This is especially relevant when the distributional change is due to actions by one or more agents, because agents intervene at a particular place and time, and this is reflected in the form of the interventions discussed in the causality literature (Pearl, 2009; Peters et al., 2016), where a single”
- The hardness of conditional independence testing and the generalised covariance measure (2018, background): “The recently introduced invariant prediction methodology [42, 27] aims to estimate for a specified target variable Y , the set of causal variables among potential covariates X1, .”
- Conditional variance penalties and domain shift robustness (2017, background): “Causal models thus have the defining advantage that the predictions will be valid even under arbitrarily large interventions on all predictor variables (Haavelmo, 1944; Aldrich, 1989; Pearl, 2009; Schölkopf et al., 2012; Peters et al., 2016; Zhang et al., 2013, 2015; Yu et al., 2017; Rojas-Carulla e”
- Causal Discovery from Nonstationary/Heterogeneous Data: Skeleton Estimation and Orientation Determination (2017, methodology): “There exist methods for causal discovery from changes of multiple data sets [Hoover, 1990; Tian and Pearl, 2001; Peters et al. , 2016] by exploiting the property of invariance of causal mechanisms.”
- Invariant Causal Prediction for Nonlinear Models (2017, methodology): “1 when using nonlinear ICP with invariant conditional quantile prediction (see Appendix B for details) as a conditional independence test: As the intersection of S 1 , . . . , S 3 is empty, we have ˆ S = ∅ .”