diff --git a/looper.md b/looper.md new file mode 100644 index 0000000..5cce5f8 --- /dev/null +++ b/looper.md @@ -0,0 +1,175 @@ +# `A resource list for causality in statistics, data science and physics.` + +## Books + +Reverse chronological order, both technical and popular. + +* Pearl and Mackenzie + The Book of Why: The New Science of Cause and Effect (2018) + [amzn](https://www.amzn.com/dp/046509760X) + +* Hernán & Robins + Causal Inference (2018) + [online](http://bit.ly/2mSeeXI) + +* Rosenbaum + Observation and Experiment: An Introduction to Causal Inference (2017) + [amzn](https://www.amzn.com/dp/067497557X/) + +* Jonas Peters, Dominik Janzing and Bernhard Schoelkopf + Elements of Causal Ingerence: Foundations and Learning Algorithms (2017) + [mitpress](https://mitpress.mit.edu/books/elements-causal-inference) + +* Pearl, Glymour and Jewell, + Causal Inference in Statistics: A Primer (2016) + [amzn](https://www.amzn.com/dp/1119186846) + +* Morgan & Winship, + Counterfactuals and Causal Inference (2nd edition) (2015) + [amzn](https://www.amzn.com/dp/1107694167) + +* Causal Inference for Statistics, Social, and Biomedical Sciences: + An Introduction, Imbens & Rubin, (2015) + [amzn](https://www.amzn.com/dp/0521885884/) + +* Angrist & Pischke + Mostly Harmless Econometrics (2009) + [amzn](https://www.amzn.com/dp/0691120358/) + [princeton](https://press.princeton.edu/titles/8769.html) + +* Judea Perl + Causality: Models, Reasoning and Inference (2009) 2nd Edition + [amzn](https://www.amz.com/dp/052189560X) + +* Econometric Causality, + James J. Heckman + International Statistical Review (2008), 76, 1, 1–27 + [doi](http://dx.doi.org/10.1111/j.1751-5823.2007.00024.x) + +* Rosenbaum + Observational Studies (Springer Series in Statistics) 2nd Edition (2002) + [amzn](https://www.amzn.com/dp/0387989676) + + + +## Papers + +* Theoretical Impediments to Machine Learning With Seven Sparks + from the Causal Revolution, Judea Pearl + [arXiv:1801.04016](https://arxiv.org/abs/1801.04016) + +* Brodersen KH, Gallusser F, Koehler J, Remy N, Scott SL. + Inferring causal impact using Bayesian structural time-series models. + Annals of Applied Statistics, (2015), Vol. 9, No. 1, 247-274. + [link](http://research.google.com/pubs/pub41854.html) + +* Introduction to Causal Inference + Peter Spirtes + (2010) [jmlr](http://www.jmlr.org/papers/v11/spirtes10a.html) + +* Causal inference in statistics:An overview + Judea Pearl. (2009) [doi](http://dx.doi.org/10.1214/09-SS057) + +* Rubin, D. B. (1974). Estimating causal effects of treatments + in randomized and nonrandomized studies. + Journal of Educational Psychology, 66(5), 688-701. + [doi](http://dx.doi.org/10.1037/h0037350) + +* Haavelmo, T. (1943). + The statistical implications of a system of simultaneous equations. + Econometrica, 11, 1–12. + [jstor](http://links.jstor.org/sici?sici=0012-9682%28194301%2911%3A1%3C1%3ATSIOAS%3E2.0.CO%3B2-N) + +## Software + +* dagR: Directed Acyclic Graph with R + [CRAN](https://cran.r-project.org/web/packages/dagR/index.html)[doi](http://dx.doi.org/10.1097/EDE.0b013e3181e09112) + +* An R package for causal inference using Bayesian structural + time-series models + [CausalImpact](https://google.github.io/CausalImpact/CausalImpact.html) + [CRAN](https://cran.r-project.org/package=CausalImpact) + +* Python package [causalinference](https://github.com/laurencium/causalinference) [Vignette](https://github.com/laurencium/causalinference/blob/master/docs/tex/vignette.pdf) + +* Tetrad Project: Graphical Causal Models [homepage](http://www.phil.cmu.edu/tetrad/) + +* PyPhi: A toolbox for integrated information theory [arXiv](https://arxiv.org/abs/1712.09644) + +## MOOCs + +* A Crash Course in Causality: Inferring Causal Effects from Observational Data [link](https://www.coursera.org/learn/crash-course-in-causality) + +* Measuring Causal Effects in the Social Sciences [link](https://www.coursera.org/learn/causal-effects) + +## Quotes: Prediction and Causation + +* " Actually correlation lets you make predictions + in many cases, assuming you're making prediction + about the world as reflected in your data. + For example, correlations between photos and + their labels allows you usually to make predictions + about new photos. + + The problem gets difficult if you want to predict the + effects of actions taken in a different manner from + that which exists in your data. For example, if you + want to consider different treatment strategies for + cancer using data from past cancer patients, only + correlations will usually not suffice as you're + trying to predict counterfactual that might not exist in your data." + Uri Shalit, Technion (Forum Communication 01/2018) + +* Reply to Uri: + + "You don't have to have correlation to make a prediction in any case. + Chicken entrails used to suffice, and still do some places. + I would argue that chicken entrails might actually be better + than relying purely upon correlation if you know nothing + about any causality involved. + + Only if the correlation is the result of causality will + you be able to trust a prediction using correlation. + This is where the "science" in "data science" usually disappears, + as exemplified in your post." + + Recall, Fisher himself (while employed by the tobacco industry) + claimed that any link between lung cancer and smoking was mistaking + correlation for causality. Of course, Fisher, a life long smoker, + died from lung cancer also. Talk about causality bites, + predicting Fisher's means of death via the correlation would + have been trustable." + Mark Powell, Austin (Forum Communication 01/2018) + +## Quotes: Rubin vs. Pearl + +* " Rubin and Pearl are kind of "academic enemies". + Though neither completely dismisses the other, + they both make snide remarks about the other's work. + Pearl shows in his book exactly how Neyman-Rubin + potential outcomes can be derived from causal graphs. + As far as I know Rubin never really makes an + attempt to address Pearl's ideas directly. + However, Rubin, being a statistician, made + significant contributions to the practice of real-world + causal inference, which go beyond Pearl's interests. + Jamie Robins also made seminal contributions to this subject. + You can read some of the debate on Andrew Gelman's blog + [here](http://andrewgelman.com/2009/07/05/disputes_about/) + Pearl writes in the comment section and in that blog + post there are links to follow up posts. " + Uri Shalit, Technion (Forum Communication 01/2018) + +## Quotes: On the Pearl's Philosopy + +* "..while Pearl's work is foundational, and its importance + cannot be overstated, his published work is often + insufficient in addressing the real-world problems of + many data scientists. The reason is that Pearl is mostly + concerned with the problem of identification, i.e. which data + generating processes allow us to infer causation from observed data. + He is less concerned with the statistical problem of actually + inferring these purported causal relationships from data. + This is especially true if the data is high-dimensional + or noisy (Pearl usually considers a few binary or Gaussian variables)." + Uri Shalit, Technion (Forum Communication 01/2018)