因果推斷三種分析框架及其應用綜述

馬忠貴; 徐曉晗; 劉雪兒

doi:10.13374/j.issn2095-9389.2021.07.04.002

因果推斷三種分析框架及其應用綜述

doi: 10.13374/j.issn2095-9389.2021.07.04.002

北京科技大學計算機與通信工程學院，北京 100083

基金項目: 中央高校基本科研業務費專項資金資助項目（FRF-DF-20-12, FRF-GF-18-017B）

詳細信息

通訊作者:
E-mail: g_runeko@163.com

中圖分類號: TG142.71
計量
- 文章訪問數: 2854
- HTML全文瀏覽量: 448
- PDF下載量: 593
- 被引次數: 0
出版歷程
- 收稿日期: 2021-07-04
- 網絡出版日期: 2021-10-19
- 刊出日期: 2022-07-01

Three analytical frameworks of causal inference and their applications

School of Computer and Communication Engineering, University of Science and Technology Beijing, Beijing 100083, China

More Information

Corresponding author: E-mail: g_runeko@163.com

摘要

摘要: 介紹因果推斷所涉及的基本概念及其三種分析框架：反事實框架、潛在結果模型和結構因果模型。首先，從反事實框架介紹因果效應的發端；然后，從基于反事實的兩個因果推斷分析框架：潛在結果模型和結構因果模型，來分別闡述兩個分析框架所涉及的關鍵理論和應用方法。其中，潛在結果模型使用數學和可計算的語言對因果理論進行闡述，是一種將假設、命題和結論清晰化表達的計算模型，其在原因和結果變量已知的前提下定量分析原因變量對結果變量的因果效應，并對缺失的潛在結果進行補齊，使觀察性研究的效果接近試驗性研究。結構因果模型則是一種基于圖論的因果推斷方法，它將事件分為觀察、干預和反事實三個層級，并通過do運算將干預和反事實層級的因果關系都降維成可以通過統計學手段解決的問題。最后，探討了現今多領域內因果推斷的應用場景，并總結了三種分析框架的異同點。
- 因果效應 /
- 因果推斷 /
- 反事實 /
- 潛在結果模型 /
- 結構因果模型
Abstract: Causality is a generic relationship between an effect and a cause that produces it. The causal relationship among things has been a research hotspot; however, the complexity of causality is sometimes far beyond our imagination. Although some causality problems seem easy to analyze, finding an exact answer may not be easy. Nevertheless, through the continuous innovation and development of empirical research methods in recent decades, we have had several clear analytical frameworks and effective methods on how to define and estimate causality. Exploring the causal effects among things is a promising research topic in many fields, such as statistics, computer science, and econometrics. With Joshua D. Angrist and Guido W. Imbens winning the Nobel Prize in economics for their methodological contributions to the analysis of causality in 2021, causal inference is expected to thrive in these fields. This paper briefly introduces the basic concepts involved in causal inference and its three analytical frameworks, namely, counterfactual framework (CF), potential outcome framework (POF), and structural causal model (SCM). Firstly, we introduce the origin of causal effects according to CF. Secondly, based on the counterfactual theory, two analysis frameworks are considered (POF and SCM), and we introduce the associated key theories and methods. The SCM explains the causal theory through mathematics and computable language, and it is a calculation model that clearly expresses hypotheses, propositions, and conclusions. It quantitatively analyzes the pair of cause variables under the premise that the cause and effect variables are known. The POF makes up for the missing potential results, such that the effect of the observational research is close to experimental research. The SCM is a causal inference method based on graph theory. It divides events into three levels: observation, intervention, and counterfactual. Through the “do” operation, the causal relationship at the intervention and counterfactual levels could be reduced to low-dimensional problems, which can be solved via statistical methods. Finally, the current application scenarios of causal inference in many fields are discussed in this paper, and the three analysis frameworks are compared.
- causal effect /
- causal inference /
- counterfactual /
- potential outcome model /
- structural causality model

HTML全文

參考文獻(66)

[1]	Sabine G H, Russell B. A history of western philosophy and its connection with political and social circumstances from the earliest times to the present day. Am Hist Rev, 1946, 51(3): 485 doi: 10.2307/1840112
[2]	Zhang D S, Sun G J. On the methodological characteristic and value of mill’s five canons. J Central China Norm Univ (Humanit Soc Sci), 2001, 40(6): 19 張大松, 孫國江. 論穆勒五法的方法論特征與價值. 華中師范大學學報(人文社會科學版), 2001, 40(6):19
[3]	Ying Q, Xu D S. Hume and the birth of analytical action philosophy—from the point of “causal explanation”. Tianjin Soc Sci, 2021, 12(2): 57 應奇, 徐東舜. 休謨與分析性行動哲學的誕生—從“因果解釋”視角看. 天津社會科學, 2021, 12(2):57
[4]	Granger C W J. Investigating causal relations by econometric models and cross-spectral methods. Econometrica, 1969, 37(3): 424 doi: 10.2307/1912791
[5]	Heckman J J. Econometric causality. Int Statistical Rev, 2008, 76(1): 1 doi: 10.1111/j.1751-5823.2007.00024.x
[6]	Cook T D, Campbell D, Shadish W. Experimental and Quasi-experimental Designs for Generalized Causal Inference. Boston: Houghton Mifflin, 2002
[7]	Lazarsfeld P. Problems in methodology. Sociology Today: Problems and Prospects. New York: Basic Books, 1959
[8]	Hume D. An Enquiry Concerning Human Understanding, and Selections from A Treatise of Human Nature. Chicago: Open Court Publishing, 1912
[9]	Lewis D. Causation. J Philos, 1973, 70(17): 556 doi: 10.2307/2025310
[10]	Imbens G W, Rubin D B. Causal Inference for Statistics, Social, and Biomedical Sciences. New York: Cambridge University Press, 2015
[11]	Rubin D B. Bayesian inference for causal effects: The role of randomization. Ann Statist, 1978, 6(1): 34
[12]	Morgan S L, Winship C. Counterfactuals and Causal Inference Methods and Principles for Social Research. 2nd Ed. Cambridge: Cambridge University Press, 2014
[13]	Cochran W G, Chambers S P. The planning of observational studies of human populations. J Royal Stat Soc Ser A, 1965, 128(2): 234 doi: 10.2307/2344179
[14]	Hume D, Beauchamp T L. An Enquiry Concerning Human Understanding. Oxford: Clarendon Press, 1999
[15]	Mill J S. A System of Logic: in Collected Works of John Stuart Mill. Toronto: University of Toronto Press, 1973
[16]	Fisher R A. Statistical methods and scientific inference. J Institute Actuaries, 1957, 83(1): 64 doi: 10.1017/S002026810005126X
[17]	Neyman J S, Dabrowska D M, Speed T P. On the application of probability theory to agricultural experiments. essay on principles. section 9. Statist Sci, 1990, 5(4): 465
[18]	Rubin D B. Estimating causal effects of treatments in randomized and nonrandomized studies. J Educ Psychol, 1974, 66(5): 688 doi: 10.1037/h0037350
[19]	Miao W, Liu C C, Geng Z. Statistical approaches for causal inference. Sci Sin (Math), 2018, 48(12): 1753 doi: 10.1360/N012018-00055 苗旺, 劉春辰, 耿直. 因果推斷的統計方法. 中國科學:數學, 2018, 48(12):1753 doi: 10.1360/N012018-00055
[20]	Rubin D B. Statistics and causal inference: Comment ifs have causal answers. J Am Stat Assoc, 1986, 81(396): 961
[21]	Sinclair B, McConnell M, Green D P. Detecting spillover effects: Design and analysis of multilevel experiments. Am J Political Sci, 2012, 56(4): 1055 doi: 10.1111/j.1540-5907.2012.00592.x
[22]	Bickel P J, Hammel E A, O’Connell J W. Sex bias in graduate admissions: Data from Berkeley. Science, 1975, 187(4175): 398 doi: 10.1126/science.187.4175.398
[23]	Greenland S, Pearl J, Robins J M. Confounding and collapsibility in causal inference. Statist Sci, 1999, 14(1): 29
[24]	Cochran W G, Rubin D B. Controlling bias in observational studies: A review. Sankhyā:The Indian Journal of Statistics,Series A, 1973, 35(4): 417
[25]	Rosenbaum P R, Rubin D B. The central role of the propensity score in observational studies for causal effects. Biometrika, 1983, 70(1): 41 doi: 10.1093/biomet/70.1.41
[26]	Abadie A, Imbens G W. Matching on the estimated propensity score. Econometrica, 2016, 84(2): 781 doi: 10.3982/ECTA11293
[27]	Bahadori M T, Chalupka K, Choi E, et al. Causal regularization[J/OL]. ArXiv Preprint (2017-2-23) [2021-6-11].https://arxiv.org/abs/1702.02604
[28]	Lee B K, Lessler J, Stuart E A. Improving propensity score weighting using machine learning. Stat Med, 2010, 29(3): 337 doi: 10.1002/sim.3782
[29]	Rosenbaum P R. Model-based direct adjustment. J Am Stat Assoc, 1987, 82(398): 387 doi: 10.1080/01621459.1987.10478441
[30]	Imbens G W. Nonparametric estimation of average treatment effects under exogeneity: A review. Rev Econ Stat, 2004, 86(1): 4 doi: 10.1162/003465304323023651
[31]	Robins J M, Rotnitzky A, Zhao L P. Estimation of regression coefficients when some regressors are not always observed. J Am stat Assoc, 1994, 89(427): 846 doi: 10.1080/01621459.1994.10476818
[32]	Hullsiek K H, Louis T A. Propensity score modeling strategies for the causal analysis of observational data. Biostatistics, 2002, 3(2): 179 doi: 10.1093/biostatistics/3.2.179
[33]	Pearl J. Causality: Models, Reasoning, and Inference, 2nd Ed. New York: Cambridge University Press, 2009
[34]	Pearl J. Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference. San Francisco: Morgan Kaufmann, 1988
[35]	Spiegelhalter D J, Dawid A P, Lauritzen S L, et al. Bayesian Analysis in Expert Systems. Statistical Science, 1993, 8(3): 219
[36]	Blyth C R. On Simpson’s paradox and the sure-thing principle. J Am Stat Assoc, 1972, 67(338): 364 doi: 10.1080/01621459.1972.10482387
[37]	Lauritzen S L, Spiegelhalter D J. Local computations with probabilities on graphical structures and their application to expert systems. J Royal Stat Soc:Ser B (Methodol), 1988, 50(2): 157
[38]	Pearl J. Causality. New York: Cambridge University Press, 2000
[39]	Pearl J. Causal diagrams for empirical research. Biometrika, 1995, 82(4): 702
[40]	Greenland S, Pearl J. Adjustments and their consequences—collapsibility analysis using graphical models. Int Stat Rev, 2011, 79(3): 401 doi: 10.1111/j.1751-5823.2011.00158.x
[41]	Greenland S, Pearl J, Robins J M. Causal diagrams for epidemiologic research. Epidemiol (Camb Mass), 1999, 10(1): 37 doi: 10.1097/00001648-199901000-00008
[42]	Zhao X L. Most Useful Science of Econometrics. Beijing: Peking University Press, 2017 趙西亮. 基本有用的計量經濟學. 北京: 北京大學出版社, 2017
[43]	Shimizu S, Hoyer P O, Hyvarinen A, et al. A linear non-gaussian acyclic model for causal discovery. J Mach Learn Res, 2006, 7(4): 2003
[44]	Hoyer P O, Shimizu S, Kerminen A J, et al. Estimation of causal effects using linear non-Gaussian causal models with hidden variables. Int J Approx Reason, 2008, 49(2): 362 doi: 10.1016/j.ijar.2008.02.006
[45]	Cai R C, Qiao J, Zhang Z J, et al. Self: Structural equational embedded likelihood framework for causality discovery // The Thirty-Second AAAI Conference on Artificial Intelligence(AAAI-18). New Orleans, 2018: 1787
[46]	Mai G Z, Peng S G, Hong Y H, et al. Causation inference based on combining additive noise model and conditional independence. Appl Res Comput, 2019, 36(6): 1688 麥桂珍, 彭世國, 洪英漢, 等. 混合加噪聲模型與條件獨立性檢測的因果方向推斷算法. 計算機應用研究, 2019, 36(6):1688
[47]	Fei N N, Yang Y L. Estimating linear causality in the presence of latent variables. Clust Comput, 2017, 20(2): 1025 doi: 10.1007/s10586-017-0824-5
[48]	He Y, Jia J, Yu B. Reversible MCMC on Markov equivalence classes of sparse directed ayclic graphs. Ann Statist, 2013, 41(4): 1742
[49]	Zhang H, Hao Z F, Cai R C, et al. High dimensional causality discovering based on mutual information. Appl Res Comput, 2015, 32(2): 382 doi: 10.3969/j.issn.1001-3695.2015.02.015 張浩, 郝志峰, 蔡瑞初, 等. 基于互信息的適用于高維數據的因果推斷算法. 計算機應用研究, 2015, 32(2):382 doi: 10.3969/j.issn.1001-3695.2015.02.015
[50]	Qin Q L, Li Q, Yan X X, et al. A comparative study on the four causal diagram models for causal inference in observation study. Chin J Heal Stat, 2020, 37(4): 496 覃青連, 李嶠, 顏星星, 等. 四種因果圖模型在觀察性研究因果推斷中的比較研究. 中國衛生統計, 2020, 37(4):496
[51]	Bareinboim E, Pearl J. Causal inference and the data-fusion problem. PNAS, 2016, 113(27): 7345 doi: 10.1073/pnas.1510507113
[52]	Abadie A, Cattaneo M D. Econometric methods for program evaluation. Annu Rev Econ, 2018, 10(1): 465 doi: 10.1146/annurev-economics-080217-053402
[53]	Wyss R, Ellis A R, Brookhart M A, et al. The role of prediction modeling in propensity score estimation: An evaluation of logistic regression, bCART, and the covariate-balancing propensity score. Am J Epidemiol, 2014, 180(6): 645 doi: 10.1093/aje/kwu181
[54]	Imai K, Ratkovic M. Covariate balancing propensity score. J R Stat Soc B, 2014, 76(1): 243 doi: 10.1111/rssb.12027
[55]	Bloniarz A, Liu H Z, Zhang C H, et al. Lasso adjustments of treatment effect in randomized experiments. PNAS, 2016, 113(27): 7383 doi: 10.1073/pnas.1510506113
[56]	Choudhury P, Allen R, Endres M. Developing theory using machine learning methods. SSRN Journal, 2018: 3251077
[57]	Zivich P N, Breskin A. Machine learning for causal inference: on the use of cross-fit estimators. Epidemiology, 2021, 32(3): 393 doi: 10.1097/EDE.0000000000001332
[58]	Wang T, Huang J Q, Zhang H W, et al. Visual commonsense representation learning via causal inference // 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Seattle, 2020: 1547
[59]	Chen P, Li Q, Zhang D Z, et al. A survey of multimodal machine learning. Chin J Eng, 2020, 42(5): 557 陳鵬, 李擎, 張德政, 等. 多模態學習方法綜述. 工程科學學報, 2020, 42(5):557
[60]	Niu Y, Tang K H, Zhang H W, et al. Counterfactual VQA: A cause-effect look at language bias // 2021 IEEE/CVF Computer Vision and Pattern Recognition. Virtual, 2021: 12695
[61]	Tang K H, Huang J Q, Zhang H W. Long-tailed classification by keeping the good and removing the bad momentum causal effect // Neural Information Processing Systems. Vancouver, 2020: 12991
[62]	Zhang D, Zhang H W, Tang J H, et al. Causal intervention for weakly-supervised semantic segmentation // Neural Information Processing Systems. Vancouver, 2020: 12547
[63]	Yue Z Q, Wang T, Zhang H W, et al. Counterfactual zero-shot and open-set visual recognition [J/OL]. arXiv preprint (2021-3-1) [2021-6-11].https://arxiv.org/abs/2103.00887v1
[64]	Yang X, Zhang H W, Qi G J, et al. Causal attention for vision-language tasks // Computer Vision and Pattern Recognition. Virtual, 2021: 9842
[65]	Bonner S, Vasile F. Causal embeddings for recommendation// Proceedings of the 12th ACM Conference on Recommender Systems. Vancouver, 2018: 104
[66]	Xu S Y, Li Y Q, Liu S C, et al. Learning post-hoc causal explanations for recommendation [J/OL]. arXiv preprint (2021-2-23) [2021-6-11].https://arxiv.org/abs/2006.16977