Appendix B — Further reading
Books and papers referred to in the text, plus a few that are not but should be. Grouped by what you would be reaching for them for, rather than alphabetically, because the useful question is usually “where do I go next on this” rather than “who wrote that”.
B.1 If you want the Bayesian foundations
Richard McElreath, Statistical Rethinking: A Bayesian Course with Examples in R and Stan (2nd edition, Chapman & Hall/CRC, 2020).
If you are starting from scratch today, this is probably where I would send you. It builds the whole apparatus from first principles rather than presenting it as a set of recipes, and it takes causal reasoning seriously from early on instead of bolting it on at the end — the DAG material in Chapter 21 here is a short version of an argument he makes at length. The accompanying lecture course is free on his YouTube channel and covers much of the same ground; if you learn better by watching than by reading, you can do the course and treat the book as the reference.
Two practical notes. The book uses his own rethinking package rather than brms, so the code in it will not run against what we use here — Solomon Kurz’s free translation into brms and the tidyverse (https://bookdown.org/content/4857/) closes that gap chapter by chapter. And it is a long book. It rewards being worked through, not skimmed.
John K. Kruschke, Doing Bayesian Data Analysis: A Tutorial with R, JAGS, and Stan (2nd edition, Academic Press, 2015).
The puppy book — so called for the dogs on the cover, and the book I actually learnt Bayesian analysis from, in its first edition. It is patient in a way few statistics books are: it will spend a whole chapter on something another author would dispatch in a paragraph, and if you have bounced off Bayesian methods once already, that patience is the thing that gets you through. The first edition used BUGS; the second replaces that with JAGS and Stan, so the code is closer to current practice.
Where it differs from this book: Kruschke is a psychologist writing for researchers, so the worked problems are experimental rather than organisational, and he is a strong advocate for the ROPE-and-HDI approach to decisions, which I covered only briefly in Chapter 20.
Peter E. Rossi, Greg M. Allenby and Sanjog Misra, Bayesian Statistics and Marketing (2nd edition, Wiley, 2024).
My first Bayesian book. I had a handful of specific problems and this helped me solve them, so I have a lasting affection for it — but I would not send anyone here first. It assumes you already accept the Bayesian case and want the machinery, and the machinery is presented at a level that expects comfort with the mathematics. Come to it when you have a hierarchical problem that looks like a marketing problem — heterogeneous individuals, repeated choices, information to be pooled across units — which, if you squint, describes rather a lot of People Analytics.
Andrew Gelman, John Carlin, Hal Stern, David Dunson, Aki Vehtari and Donald Rubin, Bayesian Data Analysis (3rd edition, Chapman & Hall/CRC, 2013). Free PDF at http://www.stat.columbia.edu/~gelman/book/.
The reference work, usually just called BDA3. This is not a book to read through; it is a book to own and consult when you need to know what is actually going on underneath something. Considerably more mathematical than anything above.
B.2 If you want the regression frameworks in depth
Keith McNulty, Handbook of Regression Modeling in People Analytics: With Examples in R and Python (2nd edition, Chapman & Hall/CRC Press, 2026). Free online edition at https://peopleanalytics-regression-book.org/.
The reference work for regression in this field, and the book to go to when a chapter here has given you one worked example and you want the model family developed properly — the mathematics, the assumptions, the diagnostics, and exercises to practise on. It covers linear, logistic, multinomial, ordinal, count, hierarchical and survival models, each from first principles, in both R and Python.
Three specific reasons to reach for it alongside this one. It gives the classical version of most of what is here, which is what you need when a colleague asks why your answer differs from theirs. It covers structural equation modelling, which this book only points at (Chapter 22 in Chapter 19). And the Python implementations make it the better recommendation for a data science team that does not work in R.
The peopleanalyticsdata package that accompanies it supplies nearly all the data used here — see Appendix C.
B.3 If you want more on Empirical Bayes
David Robinson, Introduction to Empirical Bayes: Examples from Baseball Statistics (self-published, 2017). Announcement and links at http://varianceexplained.org/r/empirical-bayes-book/; the blog series it grew out of is free on the same site.
Really approachable, and short. We used Empirical Bayes extensively in OrganizationView’s text analysis — for example which themes genuinely distinguish one group’s comments from another’s, when some themes appear five times and others five thousand — and this book was the inspiration for it. The examples are all baseball batting averages, which sounds like a barrier and is not: a batting average is a proportion estimated from a small and uneven number of attempts, and so is almost every rate you will meet in People Analytics. Read it after Chapter 14 if you want the method developed further than we take it, or before Chapter 14 if you would rather meet the idea somewhere with no organisational politics attached.
It is also where I first met the technique in Chapter 5 for comparing two posteriors by subtracting their draws. His chapter on Bayesian A/B testing gives four ways to answer “is player A genuinely better than player B” — simulation, numerical integration, an exact closed form derived by Evan Miller, and a normal approximation — and is worth reading for the argument that the crude simulation answer is usually the one to reach for first.
B.4 If you want the causal toolkit
Joshua D. Angrist and Jörn-Steffen Pischke, Mostly Harmless Econometrics: An Empiricist’s Companion (Princeton University Press, 2009).
A book I go back to time after time. It is the standard reference for the design-based tradition that Chapter 22 draws on — difference-in- differences, instrumental variables, regression discontinuity — and its real contribution is less the techniques than the attitude: start from the question of where your comparison actually comes from, and only then worry about the estimator. It is classical rather than Bayesian throughout, which matters less than you might expect, because identification is not a Bayesian or frequentist question.
Joshua D. Angrist and Jörn-Steffen Pischke, Mastering ’Metrics: The Path from Cause to Effect (Princeton University Press, 2014).
The easier of the two — the same ideas with the mathematics stripped back and built around a handful of famous studies. Mostly Harmless is far from a difficult book, but if you have no econometrics background this is the sensible entry point, and it is short enough to read in a week.
Nick Huntington-Klein, The Effect: An Introduction to Research Design and Causality (2nd edition, Chapman & Hall/CRC, 2025). Free online edition at https://theeffectbook.net/.
The best single place to go for the DAG material in Chapter 21. Where Angrist and Pischke give you the designs, Huntington-Klein gives you the step before them — how to state what you believe about the causal structure, and how that statement determines which variables belong in the model. Code is given in R, Stata and Python. Free online, which makes it an easy thing to point a colleague at.
B.5 On measurement and judgement
Douglas W. Hubbard, How to Measure Anything: Finding the Value of Intangibles in Business (3rd edition, Wiley, 2014).
This should be on every People Analyst’s desk. It is crammed with techniques for doing an analysis in the real world, where the objection you get is not “your posterior is misspecified” but “you can’t measure that”. Hubbard’s answer — that a measurement is anything that reduces uncertainty, and that you almost always know more before you start than you think you do — is Bayesian-first, whether or not the reader notices, and his material on calibrating your own estimates is the practical groundwork for the elicitation workflow in Chapter 23.
Howard Wainer, Truth or Truthiness: Distinguishing Fact from Fiction by Learning to Think Like a Data Scientist (Cambridge University Press, 2016).
Short essays on causal claims that do not survive contact with the data. This is where I met Rubin’s manipulability idea in the form it takes in Chapter 13 — that a variable you cannot intervene on cannot be a cause in any useful sense — and it is much more entertaining than that description makes it sound.
Daniel Kahneman, Olivier Sibony and Cass R. Sunstein, Noise: A Flaw in Human Judgment (Little, Brown Spark, 2021).
The noise-versus-bias distinction that Chapter 19 is built on: two assessors disagreeing at random is a different problem from every assessor being wrong in the same direction, it is usually the larger of the two, and organisations almost never measure it.
B.6 If you want the history
Sharon Bertsch McGrayne, The Theory That Would Not Die (Yale University Press, 2011).
The AF447 story in the Welcome, plus the U-boat hunts and lost-submarine searches that used the same logic decades earlier. Worth reading for the long stretch in the middle of the twentieth century when the method was professionally disreputable and being used anyway, by people who needed an answer rather than an argument.