Notebooks

Causal Inference

17 May 2022 11:05

Spun off from Causality. Graphical causal models are, I think very strongly, the best way to approach this, and so they get their own notebook.

Something that puzzles me: Can we estimate the causal effects of common institutions? To give the concrete example that made this strike me: suppose we want to know whether offering new Ph.D.s "good" post-doctoral fellowships helps their later career. Any one post-doc program could try to estimate its effect in any of several well-established ways. (For instance, it might compare those who accept their fellowships to those just below the cut-off, or those to whom it made offers which were declined.) However, those control cases are very likely individuals who are going to get such fellowships, or similar treatment, anyway. The program is at most getting at its differential effect, compared to going into a different but similar program, in an institutional context where such programs are common (for the relevant population). It is entirely possible that each program has no differential effect, but that if they were all shut down, everyone would be much worse off. Is there any way to identify the effect of a pervasive institution, short of doing the experiment of, say, arbitrarily excluding half the relevant population? (This is obviously related to all sorts of questions, like paradoxes of composition, interference, etc.)

Recommended (current big picture):
• Clark Glymour
• The Mind's Arrows: Bayes Nets and Graphical Causal Models in Psychology [Mini-review]
• "What Went Wrong? Reflections on Science by Observation and The Bell Curve", Philosophy of Science 65 (1998): 1--32 [PDF reprint via Prof. Glymour]
• Sander Greenland, Judea Pearl and James M. Robins, "Causal Diagrams for Epidemiologic Research", Epidemiology 10 (1999): 37--48 [PDF via Prof. Pearl. Very much not just for epidemiologists.]
• Stephen L. Morgan and Christopher Winship, Counterfactuals and Causal Inference: Methods and Principles for Social Research [Mini-review]
• Judea Pearl
• Donald B. Rubin and Richard P. Waterman, "Estimating the Causal Effects of Marketing Interventions Using Propensity Score Methodology", math.ST/0609201 = Statistical Science 21 (2006): 206--222 [A good description of Rubin et al.'s methods for causal inference, adapted to the meanest understanding. I list this here rather than under "more specialized" because Rubin and Waterman do a very good job of explaining, in a clear and concrete problem, just how and why the newer techniques of causal inference are valuable, with just enough technical detail that it doesn't seem like magic. Rubin's paper-collection, Matched Sampling for Causal Effects, has much, much more if this appeals to you, though it is just a paper collection and not a proper book, so there's a lot of redundancy.]
• Peter Spirtes, Clark Glymour and Richard Scheines, Causation, Prediction and Search [Comments]
Recommended (more specialized):
• Kevin Arceneaux, Alan S. Gerber, Donald P. Green, "A Cautionary Note on the Use of Matching to Estimate Causal Effects: An Empirical Example Comparing Matching Estimates to an Experimental Benchmark", Sociological Methods and Research 39 (2010): 256--282 ["Cautionary" is not really strong enough.]
• Peter M. Aronow and Cyrus Samii, "Does Regression Produce Representative Estimates of Causal Effects?", ssrn/2224964
• Bryant Chen and Judea Pearl, "Regression and Causation: A Critical Examination of Econometrics Textbooks" [PDF preprint via Prof. Pearl]
• Tianjiao Chu and Clark Glymour, "Search for Additive Nonlinear Time Series Causal Models", Journal of Machine Learning Research 9 (2008): 967--991
• Diego Colombo, Marloes H. Maathuis, Markus Kalisch, Thomas S. Richardson, "Learning high-dimensional directed acyclic graphs with latent and selection variables", arxiv:1104.5617
• Allan Dafoe, "First Do No Harm: The Risks of Modeling Temporal Dependence" [PDF preprint]
• Angus Deaton, "Instruments, Randomization, and Learning about Development", Journal of Economic Literature 48 (2010): 424--455 [PDF reprint via Prof. Deaton]
• Vanessa Didelez, "Causal Reasoning for Events in Continuous Time: A Decision–Theoretic Approach", UAI 2015
• Vanessa Didelez, Sha Meng, Nuala A. Sheehan, "Assumptions of IV Methods for Observational Epidemiology", Statistical Science 25 (2010): 22--40, arxiv:1011.0595
• Felix Elwert and Nicholas A. Christakis, "Wives and Ex-Wives: A New Test for Homogamy Bias in the Widowhood Effect", Demography 45 (2008): 851--873 [PDF preprint courtesy of Prof. Elwert]
• Robin J. Evans and Vanessa Didelez, "Recovering from Selection Bias using Marginal Structure in Discrete Models", UAI 2015
• Franklin M. Fisher, "A Correspondence Principle for Simultaneous Equation Models", Econometrica 38 (1970): 73--92 [When are simultaneous systems of equations legitimate limits of models of time-evolution? And when does it make sense to calculate causal effects in simultaneous-equation models by "surgery"?]
• Carlos Fern&anacute;ndez-Loría, Foster Provost, "Causal Decision Making and Causal Effect Estimation Are Not the Same... and Why It Matters", arxiv:2104.04103 [Comments]
• Clive Granger, "Investigating Causal Relations by Econometric Models and Cross Spectral Methods", Econometrica 37 (1969): 424--439 [His original paper on what has come to be called "Granger causality". It's actually very interesting — I hadn't realized he got the idea from reading Norbert Wiener, but in retrospect that makes sense and explains why he formulated his test in the frequency domain — but I feel it's very much a dead end for actual causal inference.]
• Kevin D. Hoover, Causality in Macroeconomics
• Edward H. Kennedy, "Optimal doubly robust estimation of heterogeneous causal effects", arxiv:2004.14497
• Gary King and Richard Nielsen, "Why Propensity Scores Should Not Be Used for Matching" [Preprint via Prof. King. The point about trying to mimic an fully randomized rather than a blocked experiment is an interesting and I think correct one...]
• Samantha Kleinberg, An Algorithmic Enquiry Concerning Causality [Ph.D. thesis, NYU, 2010; PDF]
• Gustavo Lacerda, Peter Spirtes, Joseph Ramsey and Patrik O. Hoyer, "Discovering Cyclic Causal Models by using Independent Components Analysis" [PDF draft via Gustavo]
• Jan Lemeire, Dominik Janzing, "Replacing Causal Faithfulness with Algorithmic Independence of Conditionals", Minds and Machines 23 (2013): 227--249
• Judith Lok, Statistical Modeling of Causal Effects in Time (Ph.D. thesis, Vrije Universiteit Amsterdam, 2001)
• Marloes H. Maathuis, Markus Kalisch, Peter Bühlmann, "Estimating high-dimensional intervention effects from observational data", Annals of Statistics 37 (2009): 3133--31654, arxiv:0810.4214
• Milan Palus and Aneta Stefanovska, "Direction of coupling from phases of interacting oscillators: An information-theoretic approach", Physical Review E 67 (2003): 055201 [Thanks to Prof. Palus for a reprint. This is a kind of information-theoretic generalization of Granger causality.]
• Judea Pearl, "Linear Models: A Useful Microscope'' for Causal Analysis", Journal of Causal Inference 1 (2013): 155--170
• Tom Pepinsky, "OMFG Exogenous Variation! Or, Can You Find Good Nails When You Find an Indonesian Politics Hammer?" [Admittedly, less formal in presentation than many of the rest of these links]
• Maxim Raginsky, "Directed information and Pearl's causal calculus", arxiv:1110.0718
• J. D. Ramsey, S. J. Hanson, C. Hanson, Y. O. Halchenko, R. A. Poldrack and C. Glymour, "Six Problems for Causal Inference from fMRI", NeuroImage 49 (2010): 1545--1558 [PDF via Prof. Hanson; thanks to Prof. Glymour for having shared a preprint with me]
• James M. Robins, Richard Scheines, Peter Spirtes and Larry Wasserman, "Uniform Consistency in Causal Inference", Biometrika 90 (2003): 491--515 [CMU Statistics Tech Report 725, 2000]
• Mark R. Rosenzweig and Kenneth I. Wolpin, "Natural "Natural Experiments" in Economics", Journal of Economic Literature 38 (2000): 827--874
• Heather Sarsons, "Rainfall and Conflict" [From the Annals of Invalid Instruments... PDF preprint]
• Amit Sharma, Jake M. Hofman and Duncan J. Watts, "Estimating the Causal Impact of Recommendation Systems from Observational Data", pp. 453--470 in Michal Feldman, Michael Schwarz and Tim Roughgarden (eds.), Proceedings of the Sixteenth ACM Conference on Economics and Computation [EC '15], arxiv:1510.05569 [Recommendable in this context as a nice case study]
• Herbert Simon
• "Causal Ordering and Identifiability", in Studies in Econometric Method, 1953; reprinted as chapter 1 in Simon's Models of Man [PDF of the 1950 preprint version, as "The Causal Principle and the Identification Problem"]
• "Spurious Correlation: A Causal Interpretation", Journal of the American Statistical Association 49 (1954): 467-479 [PDF reprint]
• Bonnie Smith, Elizabeth L. Ogburn, Matt McGue, Saonli Basu, Daniel O. Scharfstein, "Causal Effects in Twin Studies: the Role of Interference", arxiv:2007.04511 [Comments under Heritability]
• Michael E. Sobel
• Peter Spirtes, "Limits on Causal Inference from Observational Data" [PostScript preprint; PDF]
• Bastian Steudel and Nihat Ay, "Information-theoretic inference of common ancestors", arxiv:1010.5720
• Elizabeth A. Stuart, "Matching Methods for Causal Inference: A Review and a Look Forward", Statistical Science 25 (2010): 1--21, arxiv:1010.5586
• C. Uhler, G. Raskutti, B. Yu, Peter Bühlmann, "Geometry of faithfulness assumption in causal inference", arxiv:1207.0547
• Halbert White and Karim Chalak, "Settable Systems: An Extension of Pearl's Causal Model with Optimization, Equilibrium, and Learning", Journal of Machine Learning Research 10 (2009): 1759--1799 [Thanks to Doug White for a preprint]
• Christopher Winship
Recommended (historical):
• Hubert M. Blalock, Causal Inferences in Nonexperimental Research [Comments]
• Jerzy Neyman, "On the Application of Probability Theory to Agricultural Experiments: Essay on Principles, Section 9", Statistical Science 5 (1990): 465--472 [Translation of part of Neyman's 1923 dissertation]
Not altogether recommended:
• Guido W. Imbens and Donald B. Rubin, Causal Inference for Statistics, Social, and Biomedical Sciences: An Introduction [for reasons given in my review for the Journal of the American Statistical Association]
• Yuta Saito, Shota Yasui, "Counterfactual Cross-Validation: Stable Model Selection Procedure for Causal Inference Models", arxiv:1909.05299 [My notes after reading were as follows: Their goal is a good one, but the fundamental issue is that we don't have observations of individual-level causal effects to cross-validate against. So they proxy that by a very standard doubly-robust estimator of said effects; at which point, why not just use that estimator? In any case, I want to see comparisons to the Naive Statistician's approach of just cross-validating for outcomes (rather than differences in potential outcomes). I should add that I've only read version 1 of the paper, from September 2019, and as I add this, in July 2020, they're up to version 5...]