May 31, 2010

Books to Read While the Algae Grow in Your Fur, May 2010

Bruce Sterling, designed by Lorraine Wild, Shaping Things
I'll let the introduction speak for itself. — The graphic design is actually really very nice, though it doesn't call attention to itself.
John Stuart Mill, On Bentham and Coleridge
Finally read the old paperback volume of these two essays which I appear to have bought at a friends-of-the-library sale in 1997. (That had an introduction by F. R. Leavis, which I ignored.) Enjoyable, but it really did not succeed in convincing me to take Coleridge seriously. And of course when he wrote these essays Mill made his living as a functionary of the British Empire (in the form of the East India Company), and so had a personal stake in anti-democratic arguments — you can see him straining to find some way to evade the force of the idea that those who hold power ought to be accountable to those over whom it is wielded...
I realize that I shouldn't have laughed out loud reading Mill's contrast between the national characters of the Germans and the Italians, but come on.
S. N. Lahiri, Resampling Methods for Dependent Data
This is a thorough discussion of variants of the bootstrap for time series (chapters 2--11) and spatial random fields (chapter 12). Lahiri does not presume any previous knowledge of the bootstrap, though that would help; familiarity with theoretical statistics at the level of, say, Lehmann or Schervish is essential. A few proofs are referred to Lahiri's papers, otherwise this is self-contained, with an appendix reminding readers of the most essential results on stochastic processes.
Chapter 1 is an introduction to the idea of bootstrapping and a tour of the book. Chapter 2 discusses the forms of bootstrapping for time series, which are all based on the idea that rather than resampling individual observations, one needs to resampling whole blocks of consecutive data points. This preserves the dependence structure within each block, but messes it up at the transition between blocks; one therefore wants to let the length of the blocks grow as one gets more data. There are various ways of resampling blocks, with mostly but not entirely identical properties. Chapter 3 in particular discusses the estimation of sample means using these various block boostraps, assuming various sorts of strong mixing (most such results would carry over to cases with mere weak dependence). Chapter 4 shows how to reduce other problems to ones of sample means for transformations of the data. (I hadn't realized that a lot of the techniques for smooth functionals in statistics go back to von Mises.) Chapter 5 compares the first-order properties of the different bootstraps. Chapter 6 uses Edgeworth expansions to get at second-order properties; personally I found this chapter (unlike the others) nearly unreadable. (Since Edgeworth expansions are about series expansions for generating functions obeying certain combinatorial rules, it feels like it should be possible somehow to express them as Feynman diagrams, which would be a lot easier to grasp. If someone has done this, though, I can't find it.) Chapter 7 is about estimating how big the blocks should be, by more resampling. Chapters 8 and 9 are about alternatives to block bootstraps: either bootstrapping from parametric models (just linear ARMA models, chapter 8), or from the Fourier transform (chapter 9).
The previous theoretical results all rely on comparatively rapid decay of correlations and on the control of higher moments. Chapter 10 gives results on how to bootstrap when there is long-range dependence, and chapter 11 considers modifications for heavy-tailed data. (Also for maxima and minima, since the extreme values drawn from even light-tailed distributions tend to look heavy-tailed.) Interestingly, for both long-range dependence and heavy tails, one really needs the surrogate data to be, in an appropriate sense, smaller than the original --- trying to produce something just as large as the original time series turns out to lead to inconsistent estimates.
Finally, chapter 12 gives block bootstraps for spatial data, either on regular grids (in the limit of growing sampling regions of fixed shape) or irregularly spaced. The former is pretty straightforward, aside from annoyances about how to cover the edges of the sampling region. The latter is considerably more involved, but can handle both growing regions and increasingly dense sampling from a fixed region.
I am glad I read this, but I recommend it only for those with a serious interest in the theory of the bootstrap. (Even for them, chapter 6, oy.)
Laurence Gough, Silent Knives = Death on a No. 8 Hook and Hot Shots
Mind candy. Procedural series mystery set in Vancouver. I'd picked up a much later book in the series years ago (Heartbreaker) and loved it, but hadn't been able to lay hands on this more until I ran across this one, which I devoured instantly.
Quatermass and the Pit
Mind candy. I saw the 1950s BBC TV serial, rather than the 1967 movie remake. As the latter link indicates, this is very Lovecraftian science fiction: ancient alien strangeness implicated, in the most horrible way, in the oldest history of humanity. (Think "The Rats in the Walls" and At the Mountains of Madness) There is also I think some influence from Stapledon's great The Last and the First Men (ROT-13'd: fcrpvsvpnyyl, gur jnl uvf yngre fcrpvrf bs uhzna npuvrir gryrcnguvp cbjref guebhtu uloevqvmvat jvgu Znegvnaf).
The Objective and Red Sands
Mind candy. These are the only American movies (except arguably Iron Man) I've run across about the American experience in Afghanistan since 2001; both are horror movies, which probably means something. Some people would deduce from these that, by venturing into Afghanistan, we fear we are entangling ourselves with something very old and dangerous and better left alone; but I think that kind of criticism is B.S., and anyway that ship sailed long ago. (Pointers to other movies, especially ones not in this genre, would be appreciated.) ROT-13'd spoilers: Gur Bowrpgvir fhssref sebz gur snpg gung vs V gnxr frevbhfyl gur yrnq npgbe'f nccrnenapr va aneengvir-2002, ur jbhyq unir whfg orra fgnegvat gb funir jura gur PVN jnf jbexvat jvgu gur zhwnuvqrra. Zber shaqnzragnyyl, cybg fhssref sebz vgf vafvfgrapr ba fraqvat n zna gb qb n HNI'f wbo, gb fnl abguvat bs gur fho-iba-Qnavxra qrhf rk gevnathyb ng gur raq. Erq Fnaqf jnf na nygbtrgure fhcrevbe ubeebe zbivr, ohg znqr gur zvfgnxr bs fubjvat gur zbafgre ng gur raq; cergraq gung qvqa'g unccra.
Ta-Nehisi Coates, The Beautiful Struggle: A Father, Two Sons, and an Unlikely Road to Manhood
Coates grew up a year younger than me, and forty-odd miles north; this is a wonderfully-written memoir of what it was like to grow up a dreamy, head-in-the-clouds boy into fantasy stories and role-playing games in Maryland in the '80s, and it makes me remember that life very well. (Even some of what he says about his father struck a chord.) But in so, so many ways he and I might as well have grown up in different worlds, and that makes me angry on his behalf.
It's a beautiful book; read it.
Charles Tilly, Explaining Social Processes
A collection of Tilly's papers on the methods and objects of the social sciences and social history. I find them agreeable and sensible: Tilly stresses the importance of causal explanation of social phenomena by concatenating robust mechanisms; of path dependence; of networks, relations, and the aggregation of recurrent interactions into durable social structures; and the uselessness of thinking about "societies", or some kind of invariant pattern of "social change". His substantive research — on migration chains, "contentious" politics, and the development of the modern state — appears repeatedly, but appropriately, as examples. The essays are unedited and so overlap with each other (and even cite each other in the journal versions), but not so badly that I felt put upon.
That said, I was a bit disappointed in this collection, because, I guess, I was expecting a more systematic statement of Tilly's views on how social phenomena are put together and how they should be investigated. (He rightly insists that the two questions are linked.) In particular, methodological individualism seems to me to have a lot more going for it than Tilly allowed; even explanation by invariants is in better shape. As every school-child knows, when you put agents with invariant decision-making mechanisms in an environment which largely consists of other agents and let them go, their macroscopic behavior is generally path-dependent, and many models of small-scale behavior will only work as local approximations. (See, for instance.) Would Tilly say that this was nonetheless missing something? That assuming an individualistic and invariant infrastructure to explain relational and transient phenomena violates Occam's Razor? Distinguish this from what he meant by "methodological individualism" and "invariance"? I wish I knew; sadly, there will be no chance to find out.
— Many, but not all, of the essays reprinted in this book are available online.
Jack Vance, The Dying Earth
Re-read for the first time in more than a decade, it's still just as good as it always was. (Actually, I got this on CD to listen to while exercising, which worked surprisingly well, and then re-read it immediately after finishing the disc.)
Thomas Barfield, Afghanistan: A Cultural and Political History
The best synoptic history I know of (at least in English), running from the 17th century to the present. Barfield is an anthropologist who did ethnographic fieldwork in Afghanistan in the 1970s, so "culture" here really means social organization and widely-shared ideas about political legitimacy, rather than literature, music, etc.; pleasingly, one of his touchstones is ibn Khaldun. A full review will appear Any Day Now. In the meanwhile: strongly, strongly recommended if you have any reason to care about Afghanistan.
Kathleen George, Afterimage
Mind candy. Police procedural with really good characters, plus local color for Pittsburgh. (I think I can identify every restaurant, except I can't think of anywhere in Shadyside where someone could be watched from the second floor and get a gourmet pizza.) Third (?) book in a series; I'll be tracking down the others. — Update: Prequels and sequels.
Jack Campbell, Victorious
Mind candy. A fitting and triumphal conclusion to the long anabasis. — Update: sequel!

Books to Read While the Algae Grow in Your Fur; Scientifiction and Fantastica; Pleasures of Detection, Portraits of Crime; Writing for Antiquity; Progressive Forces; Enigmas of Chance; Afghanistan and Central Asia; Commit a Social Science; The Continuing Crises; Philosophy; Cthulhiana

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

Three-Toed Sloth