Causal inference
By R. Shanmugam. 2011
1,948 citations recorded by the source corpus.
Throughout POL572 and 573, we will learn how to use various statistical methods in order to make causal inference , which is a main goal of social science research. We all know the mantra “correlation is not causation.” The difficulty of inferring causality arises from the fact that we do not observe counterfactual outcomes, which are required to estimate causal effects. To formalize this intuition, we begin by describing the potential outcomes framework of causal inference.
Subfields: Artificial Intelligence, 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
- Percy Liang
- Frank Hutter
- Yejin Choi
- Tatsunori B. Hashimoto
- Mohit Bansal
- Noah A. Smith
- Quoc V. Le
- Yoshua Bengio
In Causal Inference & Discovery
- Percy Liang
- Tatsunori B. Hashimoto
- Bal'azs Galambosi
- Yann Dubois
- Jesse Clifton
- David Wuepper
- Koustuv Saha
- A. D'Amour
Citations in the corpus
- Length-Controlled AlpacaEval: A Simple Way to Debias Automatic Evaluators (2024, background): “…correlates – such as the length of the response – as undesirable mediators Vander-Weele (2015) in a causal graph and use regression-based causal inference Hernán & Robins (2010) techniques to provide simple adjustments to automated evaluations that control for any suspected spurious correlation.”
- How to make causal inferences using texts (2018, neutral): “aver, Benoit and Garry, 2003; Pennebaker, Mehl and Niederhoer, 2003; Quinn et al., 2010), with the growing literature on causal inference in the social sciences (Pearl, 2009; Imbens and Rubin, 2015; Hernan and Robins, 2018). The key to connecting the two traditions is recognizing the central role of”
- Principles of confounder selection (2019, background): “Such designs can allow researchers to examine the effects of time-varying exposures [40, 41], but even when assessing the effects of an exposure at a single point in time, such designs can help make more informed confounder selection decisions based on the temporal ordering of the data.”
- Estimating Treatment Effects with Causal Forests: An Application (2019, neutral): “We define causal effects via the potential outcomes model (Imbens and Rubin, 2015): For each sample i , we posit potential outcomes Y i (0) and Y i (1) corresponding to the outcome we would have observed had we assigned control or treatment to the i -th sample, and assume that we observe The average…”
- Making sense of sensitivity: extending omitted variable bias (2019, neutral): “Observational research often seeks to estimate causal effects under a “no unobserved confounding” or “ignorability” (conditional on observables) assumption (see e.g. Rosenbaum and Rubin 1983b; Pearl 2009; Imbens and Rubin 2015).”
- Does it matter who advises farmers? Pest management choices with public and private extension (2020, methodology): “…weighting observations according to their probability of receiving public or private extension advice (i.e. inverse probability weighting, see e.g. Hernan and Robins, 2019) and to allowing for combinations of advice and pest management choices, as well as the number of different extension sources.”
- A Survey of Learning Causality with Data (2018, background): “Historically, learning causality has been studied in a myriad of high-impact domains including education [36, 63, 65, 86], medical science [101, 109], economics [74], epidemiology [64, 99, 127], meteorology [38], and environmental health [94].”
- A Social Media Study on the Effects of Psychiatric Medication Use (2019, methodology): “Specifically, we adopt a causal inference framework based on matching, which simulates an RCT setting by control- ling for as many covariates as possible (Imbens and Rubin 2015).”
- Synthetic Difference In Differences (2018, neutral): “These policy changes are frequently not random—neither across units of analysis, nor across time periods—and even unconfoundedness given observed covariates may not be credible ( e.g., Imbens and Rubin 2015 .”
- Incremental causal effects (2019, background): “atments but not necessarily across all patients. Average treatment effects are not identifiable in this setting. Throughout the paper we assume that Stable Unit Treatment Value Assumption (Rubin, 1980; Hernan and Robins, 2010) holds, which says that the potential outcomes are well-defined and that ther”
- TabPFN: A Transformer That Solves Small Tabular Classification Problems in a Second (2022, methodology): “…works have applied causal reasoning to predict observations on unseen data by using causal inference, a method which seeks to identify causal relations between the components of a system by the use of interventions and observational data (Pearl, 2010; Pearl and Mackenzie, 2018; Lin et al., 2021).”
- Q-Learning: Theory and Applications (2020, background): “F e r s on 1 ) need not have mean zero when there is a positive probability of a null effect (Moodie estimate the sampling distribution of (cid:16) β (Van der Vaart 1991, Andrews 2000, Leeb & Poetscher 2003, Hirano & Porter 05 / 07 / ).”