Regression and Other Stories

by Andrew Gelman, Jennifer Hill, Aki Vehtari

Most textbooks on regression focus on theory and the simplest of examples. Real statistical problems, however, are complex and subtle. This is not a book about the theory of regression. It is about using regression to solve real problems of comparison, estimation, prediction, and causal inference. Unlike other books, it focuses on practical issues such as sample size and missing data and a wide range of goals and techniques. It jumps right in to methods and computer code you can use immediately. Real examples, real stories from the authors' experience demonstrate what regression can do and its limitations, with practical advice for understanding assumptions and implementing methods for experiments and observational studies. They make a smooth transition to logistic regression and GLM. The emphasis is on computation in R and Stan rather than derivations, with code available online. Graphics and presentation aid understanding of the models and model fitting.

Genres: Mathematics, Science, Textbooks, Technical, Nonfiction, Reference, Economics

552 pages, Paperback

Published July 23, 2020

Reviews

[Vysloczil]

One of a kind econometrics / applied stats (text)book! The level is extremely basic, there is essentially no linear algebra or stats background required - even though it is advisable obviously.

The reason why it is so unique compared to all its competitors is that it does not engage at all in statistical properties of estimators (consistency, efficiency, unbiasedness,...), but focuses solely on applied aspects and trains the readers intuition on how to do sound regression (and basic statistics). The focus is not just on describing data but also to predict and (in the last part of the book) to engage in (causal) inference.
From a philosophical point of view it follows very much what Andrew Gelman has been writing on his blog since years (decades?) and preaching in podcasts: questioning bland (frequentist) hypothesis testing, getting rid of significance levels and p-values, not engaging in too much causal language, questioning results, applying things from the Bayesian toolbox, ....
If I had to name an existing book that is closest to this one I would say Mastering Metrics by Angrist and Pischke (which is basically the undergrad version of their famous Mostly Harmless Econometrics text).

Aside from always aiming at explaining things in an intuitive way and pointing out common fallacies and flaws, the big plus of the book is that everything is always accompanied by simple R code that allows to replicate everything and play around on your own. In addition they emphasize heavily the importance of working with simulated (fake) data to train yourself - and they show you how to do it in very simple ways.

Compared to other entry level books their emphasis on Generalized Linear Models (GLM) is quite heavy. These are models that allow you to accommodate outcome variables that are binary, bounded, or discrete in different forms (logit, poisson regression,...).

[Mark · Rating 4 out of 5]

An excellent detailed but practical introduction to Regression from a largely Bayesian point of view with great examples and R code. There are many interesting asides, e.g. regression to the mean, and some key topics are explained in 2 or 3 different ways to aid your understanding. Also, by doing things in both a traditional frequentist - maximum likelihood way and then using stan_glm, the benefits of the Bayesian approach are seen.

[J.R. · Rating 3 out of 5]

If you can read a book on regression analysis and not commit suicide, then you'll probably learn some things from this book.

[Alexander Lyngstad · Rating 3 out of 5]

Not as accessible as some of you think it is
True pain was had

[Harlan · Rating 5 out of 5]

Highly, highly recommended for data scientists in academia or industry -- grad-level introduction to applied regression modeling from a practical, semi-Bayesian, post-Null-Hypothesis-Statistical-Testing viewpoint by several of the top experts in the world. Super-readable, lots of clear examples, code samples, and visualizations. Includes some really nice treatments of some advanced techniques such as causal analysis and imputation, without going into hierarchical or custom models too much.

[Henne · Rating 4 out of 5]

A very readable introduction to applied statistics by one of its leading practitioners, with just the right balance of classical and modern topics. If every practising social scientist had to pass an exam based on the topics in this book I can't help but think that the world would be a better place as a result.

[William Chiu · Rating 4 out of 5]

Confidence and power

Surprisingly, the chapter about power analysis and minimum sample size was the clearest exposition of power analysis that I have ever read. The last few chapters on causal inference was a little over my head.

[Plain Soup · Rating 5 out of 5]

The regression part is great, but it's the other stories that really make it.

Well explained, practical examples.

[Kevin Whitaker · Rating 4 out of 5]

Mostly review for me (especially as a regular reader of Gelman's blog) but a couple good tips or new ways of thinking about old topics. I can't really say how it would work as an introductory book.