Books to Read While the Algae Grow in Your Fur, March 2023

Attention conservation notice: I have no taste, and no qualifications to opine on African-American political psychology, opinion-survey research, medieval Islamic Indology, or the history of the scientific revolution. Also, most of my reading this month was done while recovering from foot surgery and/or while bottle-feeding a baby, so I'm much less reliable and more cranky than usual.

Kel Symons et al., I Love Trouble

Cecil Castellucci and Marley Zarcone, Shade, the Changing Girl

Cullen Bunn et al., Harrow County, vols. 2, 3, 4, 5, 6, 7, 8

Philippe Thirault et al., Miss: Better Living Through Crime

Kurt Busiek et al., The Autumnlands, vols. 1 and 2

Ryan North et al., The Midas Flesh, vol. 2

Comic-book mind candy, assorted. Autumnlands reminds me a little of Zelazny from the 1960s or 1970s. --- Previously for Midas Flesh; previously for Harrow County.

Paul M. Sniderman and Thomas Piazza, Black Pride and Black Prejudice [JSTOR]

The central question here is whether, among African Americans in the greater Chicago area circa 2000, higher levels of racial pride lead to higher levels of prejudice against those not in the race, especially (but not exclusively) against Jews. The authors addressed this through opinion surveys, including some ingenious survey experiments *.

On substantive grounds I have little to say here. What troubles me about this though is that the authors (and their critics) seems content to take few-level ordinal data and run it through linear regression after linear regression, endlessly permuting which variable goes on the left hand side and which ones are on the right. The ideas about validating measurements are hopeless, along the lines of the Zeller and Carmines book which so disappointed me (unsurprisingly, since Sniderman and Carmines collaborated). They are also prone to the fallacy of confusing "this regression coefficient is not statistically significant" with "this relationship is unimportant" **, and they never once look at their residuals to check their a regression specification. To be clear, I have no doubt that the survey was done as well as humanly possible; it's the analysis of the results which drives me nuts.

At some point, I confess, I wanted to make them shut down their statistical software, hand over the data set, and run the whole thing through pcalg myself, using the chi-squared test for conditional independence that works for categorical variables. (This would assume all the systematically-important variables are measured, but then, so do their regressions.) I would then hand them back the inferred graphical causal model, and let them use it to address their substantive questions. (This is of course a fantasy, because pcalg didn't exist --- but not such a fantasy, because TETRAD was a thing in 2002.) The upshot of my fantasy would be a comprehensible, reliably-constructed guess at how all their different variables inter-relate, allowing one to draw real inferences. The way they actually proceeded instead gave them an uninterpretable mush --- or, rather, a mush which demands interpretation rather than supporting calculation. In all this, of course, they are