- Stephen L. Morgan and Christopher Winship, Counterfactuals and Causal Inference: Methods and Principles for Social Research
- The first reasonable introductory textbook on modern approaches to causal
inference I have seen. (Books
like Causation, Prediction and
Search, or Pearl's Causality,
are
*not*suitable as textbooks.) It alternates between talking about counterfactual random variables and using graphical models (being clear that the latter have at least as much expressive power as the former). After the introduction, which gives a very nice tour of the rest of the book, the first few chapters cover a simple example of the kind of effect estimation we want to do; how to use conditioning and Pearl's back-door condition to control for other variables; matching methods and propensity scores; and regression and why it is problematic. They then turn to methods which might be applicable when adequate conditioning is not, like instrumental variables (about which, soundly, they are very dubious), Pearl's front-door criterion, and longitudinal and regression-discontinuity designs. (Their discussion of the front-door criterion draws very interesting links to the literature on explanation-by-mechanisms, as in Elster, Tilly, or indeed DeLanda, which I need to think about more.) Manski's partial identification approach also gets looked at. The last chapter is a sort of victory lap. - The implied reader of this book is a social scientist who likes
quantitative data but is not very interested, and perhaps not very comfortable
with, mathematical derivations; everything has been brought to the level where
it can be followed, with a little work, by someone who remembers
*how*to do ordinary least squares regression, but is fuzzy about why \( \mathbf{X}^T \mathbf{X} \) controls the standard errors of the coefficient estimates*. Readers who know more statistical theory but not causal inference can, I think, just skip the worked numerical examples, and generally go faster through the book, but will still learn a lot, and not have to unlearn any of it later. (At no point did I notice any lies-told-to-children.) Non-social-scientists interested in what can be said about causal relationships from observational, non-experimental data will also find it useful. *Disclaimer*: Winship is an editor at a journal where I have a paper under review.- *: Because it's the generalization of the sum-of-squares for the independent variable in univariate regression; and the more points you have from a line, and the more widely spaced they are, the better you know the slope of the line.
- W. W. Tarn, The Greeks in Bactria and India
- Teeth-grindingly Eurocentric, and erects massive conjectures on what seem to me to be the most flimsy evidential foundations (e.g., those Seleucid princesses!), but did a monumental job in surveying the evidence from Greek literature about the Hellenistic presence in what is now Central Asia, Afghanistan, Pakistan and India; and also the coins, as they were known in the 1930s. (He tries to bring in Indian and Chinese literature as well, but doesn't know the languages and is self-conscious about relying on translations.) Someone should really try integrating this with what we now know from archaeology; maybe they have.
- Laura E. Reeve, Pathfinder
- Mind-candy. Previously: 1, 2.
- Sarah Vowell, The Wordy Shipmates
- Otto J. Maenchen-Helfen, The World of the Huns: Studies in Their History and Culture
- How on Earth do we know that
*any*of these archaeological finds belong to*Huns*? - Laurence Gough, The Goldfish Bowl
- First in the series. As hard-boiled as possible, under the circumstances.
- Sauna
- Creepy and moody historical horror movie. Not sure if some parts of it would be less weird if I were Finnish.

Books to Read While the Algae Grow in Your Fur; Writing for Antiquity; Afghanistan and Central Asia; Scientifiction and Fantastica; The Pleasures of Detection; The Beloved Republic; Enigmas of Chance

Posted at July 31, 2010 23:59 | permanent link