yellow brick road to stats heaven

~ a loose collection of statistical and quantitative research material for fun and enrichment ~

by roland b. stark

My Favorite Books on Statistics

Favorite Books - Word Cloud

"The best statistics books?" Of course, some of you may be laughing already, or very confused. Once a group of statisticians and I went out to lunch. The waitress, sensing that this was not exactly a family gathering, asked what sort of group we were. "We teach statistics" was the answer. She almost dropped her tray. When she finally could give voice to words, what came out was, "But... what is there to teach?"

"Would you and I even like the same books?" This list is NOT intended for mathematical statisticians who contend with the inner workings of maximum likelihood estimation or the Markov Chain Monte Carlo and read the likes of Andrew Gelman, Bradley Efron, and Alan Aggresti. Instead, it's meant for applied researchers or students of research - such as in the social sciences, education, or health care - interested in

new techniques to explore, or
deeper insights into methods and rationales.


The Tao of Statistics: A Path to Understanding (With No Math). Keller, Dana K. (2005), Sage Pub.

{Time to read: 1-2 hours.} This little confection combines haiku, illustrations, and easy, concise prose to lightly touch on the questions "Why would one use this or that statistical technique?" and "Which technique would help with this or that research question?" A pleasure, and one you'll want to share with friends.

Bayesian Statistics for Social Scientists . Phillips, Lawrence D. (1973), Thomas Crowell & Co.

{Time to read: 15-20 hours.} Exposure to Bayesian thinking just might revolutionize and inspire your approach to statistics. It happens to plenty of people. Now, many of them are quite advanced mathematically, and so most of the Bayesian literature is sadly unaccessible to researchers like me and (I'm guessing) you. Here is an exception.

Actually, much of the book is a fine but unremarkable introduction to probability and statistics. Where the fireworks pop is in the section on using a Bayesian structure to revise empirical findings in light of prior knowledge. You read me right: quantitatively bringing one's knowledge to bear on a research topic, rather than uncompromisingly excluding all but the data at hand (as most of us were taught to do). It's hard to describe the feeling that comes with learning this approach, other than saying you may start to feel...whole.

Summated Rating Scale Construction . Spector, Paul E. (1992), Sage.

{Time to read: 2 hours.} A friendly, focused guide to constructing quality survey items and then distilling them into a scale that will convey more reliable, condensed information. Very clear in introducing key concepts involving correlation, internal consistency (Cronbach's alpha), and factor analysis. I frequently reread this book during my first two or three years of survey work.

New Techniques to Explore

Handbook of Applied Multivariate Statistics and Mathematical Modeling . Tinsley, Howard E. A., and Brown, Steven D., Eds. (2000). Harcourt Brace.

{Time to read: many dozens of hours.} If you're the type who pokes through the menus of statistical software programs, gets curious about methods listed there, and wishes you had introductions more helpful than those offered in the Help files, ... you may well drool over this collection. It actually straddles the world of the applied researcher and mathematical statistician. It features many "star" authors responsible for inventing or popularizing particular methods over the past 30 years. Has tremendous variety: tools for assessing cross-sectional outcomes (regressions of various stripes -- linear and otherwise); longitudinal change (survival analysis, covariance structure modeling); how variables or subjects might cluster together (factor analysis, cluster analysis, multidimensional and circular scaling); and on and on and on. Just don't expect complete step-by-step instruction.

Biostatistics: The Bare Essentials . Norman, Geoffrey R., and Streiner, David L. (2000). BC Decker/pmph usa.

A refreshing contrast to most textbooks in that it uses humor throughout: good-natured, disarming, sometimes corny; a bit like Click and Clack's Car Talk show, if you listen to US National Public Radio. It also features relaxed visual displays and accessible practice exercises for each chapter. Neither "a breeze" nor completely rigorous, but gives practical, insightful coverage of some of the techniques most commonly encountered beyond intro courses. Applicable well beyond biostatistics.

Deeper Insights into Methods and Rationales

The Logic of Causal Order. Davis, James A. (1985). Sage.

{Time to read: 3-4 hours.} Unfortunately there is a tremendous allure to statistical packages that offer "ways to choose your most important variables for you." Stepwise and other algorithm-based regression methods are often taught, incorrectly, as easy ways to choose the best explanatory variables out of 10, or 20, or 200. Now, the statistical literature is replete with technical reasons why stepwise regression can be problematic, but what Davis explains is much more fundamental: that statistics, however sophisticated, cannot substitute for genuine thought about how things relate to one another. An event occurring later cannot cause one occurring earlier; subjective opinions and attitudes can scarcely cause objective conditions such as gender, age, or social class. Moreover (and this not so explicit in the book), the mere fact that regression incorporates control of different variables (showing relationships with some variables held constant) cannot in itself ensure that it will reveal causal relationships in their truest light.

Just pages 5-15 are worth the price you'll pay for this little book. The calculations that come in later may or may not apply to you, depending on your interest in quantifying relationships using particular methods such as path analysis. And to get the most out of Davis's book, you'll want to complement it with readings that deal with conducting regression in stages: what might be called sequential, focused, or hierarchical regression (not to be confused with the more technically complicated HLM). The key is to use well thought out, conscious, deliberate statistical control rather then letting the control happen according to some algorithm. Such readings are not exactly easy to find. For those who are truly motivated to master this technique I'd recommend Elazar Pedhazur's Multiple Regression in Behavioral Research.

Statistical inference: A commentary for the social and behavioural sciences. Oakes, Michael (1986). Wiley.

{Time to read: 8-12 hours.} I found this short, 182-page book to be dense, demanding, extraordinarily insightful, and extremely rewarding. It prodded me to learn more about many subjects related to the scientific method, inference in research, and statistical methods. One of the most important books I have ever read, and certainly one of the most important on quantitative research. I especially think of it as a terrific boon for any teacher of statistics.

Prerequisites would be about 3 courses in statistics, deep curiosity about statistics, and strong motivation to understand the concepts behind statistical inference. Oakes uses very little math; instead he uses rigorous, clever, incisive logic to delve into what statistical findings such as p-values really mean, how we should interpret statistical results, and what value significance tests might hold relative to statements about confidence intervals or effect sizes. In the process, he touches insightfully on a number of developments in the history of the field. (His sense of humor isn't bad, either.)

Oakes does an excellent job when, in the tradition of Amos Tversky, Daniel Kahneman, and Gerd Gigerenzer (as well as, in a more technical statistical way, Elazar Pedhazur), he exposes misconceptions about statistical inference that plague the work of so many, including, it seems, an awful lot of social scientists. You may find yourself doing double takes at his treatment of some of these misconceptions.

Oakes also gives a short, fascinating treatment of the polemics surrounding competing schools of statistical inference, including the methods termed Fisherian, Neyman-Pearson, Bayesian, and Likelihood. A dynamite finish. This book has no weak spots.

Quasi-experimentation: Design and analysis issues for field settings. Cook, Thomas D., and Campbell, Donald T. (1979). Houghton Mifflin.

{Time to read: dozens of hours.} I grant that it's not the most up-to-date, transparent, visually stimulating, or artfully written book. Even so, it's very rewarding if you're serious about expanding your toolkit and gaining a more professional grasp of challenging issues in research design--experimental, non experimental, or quasi-experimental. Material on threats to validity, especially, continues to be cited extremely widely. The discussion of time series modelling is intriguing, and they offer a particularly thought-provoking section on the intricacies of analyzing gain scores and the choice of ANOVA vs. ANCOVA.

Design and analysis: A researcher's handbook . Keppel, Geoffrey (1991), Prentice-Hall.

{Time to read: a few dozen hours.} On the one hand, this is largely a dense, equation-rich reference book on the fine points of analysis of variance and experimental design. On the other, it contains several sections, not always clearly demarcated, that are of rare value for their astute commentary on conceptual issues and on judgment calls one must make as a researcher/analyst. Better-than-usual explanations of statistical power, degrees of freedom, and interactions. The most thoughtful discussion of the multiple comparison problem that I've seen. If you are patient, adept at picking and choosing your sections, and undeterred by shifts from dry instruction to candid, reflective exploration, you should find this book valuable.

Multiple Regression in Behavioral Research: Explanation and Prediction (3rd edition). Pedhazur, Elazar J. (1997). Harcourt Brace.

{Time to read: the sky's the limit.} This man takes his work very, very seriously, much to our benefit. I would use the word "magisterial" if his self-importance weren't taken to such hilarious heights. But that comic relief is welcome, too.

The book plumbs the depths of the general linear model, providing extensive, rigorous, extremely well-considered treatment of statistical control, linear regression, analysis of covariance, logistic regression, and structural equation modeling. Includes important sections on topics such as nonlinear relationships and interactions. Has about a dozen sections that I reread every few months to a year because the ideas are so important to my understanding of analytic issues and so hard to come by anywhere else. It's not for beginners, but don't wait too long! You'll kick yourself for all the insights you could have developed, and all the conceptual and technical mistakes you could have avoided.

Two drawbacks. There is too much attention to SPSS, SAS, and other code for many people's taste. Also, it is weighted down by unnecessarily precise calculation, overemphasizing hypothesis testing and statistical significance at the expense of the size of effects. One of the more insidious statistical mistakes.

I welcome any comments you might like to add, publicly or privately.

copyright 2008 - 2018 by roland b. stark.

yellow brick stats homepage

yellowbrickstats home

my statistical and research consulting work