## September 15, 2014

Posted by crshalizi at September 15, 2014 22:38 | permanent link

## August 29, 2014

### Rainfall, Data Structures, Sequences (Introduction to Statistical Computing)

In which we practice working with data frames, grapple with some of the subtleties of R's system of data types, and think about how to make sequences.

(Hidden agendas: data cleaning; practice using R Markdown; practice reading R help files)

Assignment, due at 11:59 pm on Thursday, 4 September 2014

Posted by crshalizi at August 29, 2014 11:30 | permanent link

### Lab: Exponentially More Fun (Introduction to Statistical Computing)

In which we play around with basic data structures and convince ourself that the laws of probability are, in fact, right. (Or perhaps that R's random number generator is pretty good.) Also, we learn to use R Markdown.

— Getting everyone randomly matched for pair programming with a deck of cards worked pretty well. It would have worked better if the university's IT office hadn't broken R on the lab computers.

Lab (and its R Markdown source)

Posted by crshalizi at August 29, 2014 10:30 | permanent link

## August 27, 2014

### Bigger Data Structures (Introduction to Statistical Computing)

Matrices as a special type of array; functions for matrix arithmetic and algebra: multiplication, transpose, determinant, inversion, solving linear systems. Using names to make calculations clearer and safer: resource-allocation mini-example. Lists for combining multiple types of values; access sub-lists, individual elements; ways of adding and removing parts of lists. Lists as key-value pairs. Data frames: the data structure for classic tabular data, one column per variable, one row per unit; data frames as hybrids of matrices and lists. Structures of structures: using lists recursively to creating complicated objects; example with eigen.

Posted by crshalizi at August 27, 2014 10:30 | permanent link

## August 25, 2014

### Introduction to the Course; Basic Data Types (Introduction to Statistical Computing)

Introduction to the course: statistical programming for autonomy, honesty, and clarity of thought. The functional programming idea: write code by building functions to transform input data into desired outputs. Basic data types: Booleans, integers, characters, floating-point numbers. Operators as basic functions. Variables and names. Related pieces of data are bundled into larger objects called data structures. Most basic data structures: vectors. Some vector manipulations. Functions of vectors. Naming of vectors. Our first regression. Subtleties of floating point numbers and of integers.

Posted by crshalizi at August 25, 2014 11:30 | permanent link

### Class Announcement: 36-350, Statistical Computing, Fall 2014

Fourth time is charm:

36-350, Statistical Computing
Instructors: Yours truly and Andrew Thomas
Description: Computational data analysis is an essential part of modern statistics. Competent statisticians must not just be able to run existing programs, but to understand the principles on which they work. They must also be able to read, modify and write code, so that they can assemble the computational tools needed to solve their data-analysis problems, rather than distorting problems to fit tools provided by others. This class is an introduction to programming, targeted at statistics majors with minimal programming knowledge, which will give them the skills to grasp how statistical software works, tweak it to suit their needs, recombine existing pieces of code, and when needed create their own programs.
Students will learn the core of ideas of programming — functions, objects, data structures, flow control, input and output, debugging, logical design and abstraction — through writing code to assist in numerical and graphical statistical analyses. Students will in particular learn how to write maintainable code, and to test code for correctness. They will then learn how to set up stochastic simulations, how to parallelize data analyses, how to employ numerical optimization algorithms and diagnose their limitations, and how to work with and filter large data sets. Since code is also an important form of communication among scientists, students will learn how to comment and organize code.
The class will be taught in the R language, use RStudio for labs, and R Markdown for assignments.
Pre-requisites: This is an introduction to programming for statistics students. Prior exposure to statistical thinking, to data analysis, and to basic probability concepts is essential, as is some prior acquaintance with statistical software. Previous programming experience is not assumed, but familiarity with the computing system is. Formally, the pre-requisites are "Computing at Carnegie Mellon" (or consent of instructor), plus one of either 36-202 or 36-208, with 36-225 as either a pre-requisite (preferable) or co-requisite (if need be).
The class may be unbearably redundant for those who already know a lot about programming. The class will be utterly incomprehensible for those who do not know statistics and probability.

Further details can be found at the class website. Teaching materials (lecture slides, homeworks, labs, etc.), will appear both there and here.

— The class is much bigger than in any previous year --- we currently have 50 students enrolled in two back-to-back lecture sections, and another twenty-odd on the waiting list, pending more space for labs. Most of the ideas tossed out in my last self-evaluation are going to be at least tried; I'm particularly excited about pair programming for the labs. Also, I at least am enjoying re-writing the lectures in R Markdown's presentation mode.

Posted by crshalizi at August 25, 2014 10:30 | permanent link

## July 31, 2014

### Books to Read While the Algae Grow in Your Fur, July 2014

Attention conservation notice: I have no taste.

Stephen King, Eyes of the Dragon
Mind candy. I really liked it when I was a boy, and on re-reading it's not been visited by the Suck Fairy, but I did come away with two thoughts. (1) I'd have been very interested to see what a writer with drier view of political power would have done with the story elements (the two princes, the evil magician, the exiled nobles) — Cherryh, say, or Elizabeth Bear. (2) Speaking of which, it's striking how strongly King's fantasy books (this one, The Dark Tower) buy into the idea of rightfully inherited authority, when his horror stories are often full of healthy distrust of government officials ("the Dallas police"). I don't think he'd say that being electorally accountable, rather than chosen by accident of birth, makes those in power less trustworthy...
Charles Tilly, Why?
Tilly's brief attempt to look at reason-giving as a social act, shaped by relations between the giver and receiver of reasons, and often part of establishing, maintaining, or repairing that relationship. He distinguished between reasons why involved cause-and-effect and those which use a logic of "appropriateness" instead, and those which require specialized knowledge and those which don't. "Conventions" are common-knowledge reasons which are invoke appropriateness, not causal accounts. (Think "Sorry I'm late, traffic was murder".) "Stories" give causal explanations which only invoke common knowledge. Tilly is (explicitly) pretty Aristotlean about stories: they involve the deeds of a small number of conscious agents, with unity of time, place, and action. Codes are about matching circumstances to the right specialized formulas and formalities --- are your papers in order? is the evidence admissible? Technical accounts, finally, purport to be full cause-effect explanations drawing on specialized knowledge.
The scheme has some plausibility, and Tilly has lots of interesting examples. But of course he has no argument that these two dimensions (generalist vs. specialist, causation vs. appropriateness) are the only two big ones, that everything (e.g.) the "codes" box really does act the same way, etc. So I'd say it's worth reading to chew over, rather than being deeply illuminating.
Elliott Kay, Rich Man's War
Sequel to Poor Man's Fight, continuing the same high standard of quality mind-candy. (No Powell's link because currently only available on Amazon.)
Alexis de Tocqueville, Democracy in America
Yet another deserved classic read only belatedly. Volume I is actually about de Tocqueville's observations on, and ideas about, democracy in America. This is interesting, mostly empirical, and full of intriguing accounts of social mechanisms. (I see why Jon Elster is so into him.) Volume II consists of his dictates about what democracy and social equality will do to customs and character in every society. This is speculative and often the only reference to America comes in the chapter titles. (I see why this would also appeal to Elster.)
I would dearly like to find a good "de Tocqueville in retrospect" volume. Some of his repeated themes are the weakness of the Federal government, the smallness of our military, the absence of serious wars, the relative equality of economic condition of the (white) population, the lack of big cities among us. So how have we managed to preserve as much democracy as we have? For that matter, how does the civil war and its outcomes even begin to make sense from his perspective?
&madash; Rhetorical observation: de Tocqueville was very fond of contrasts where democracy leads to less dispersion among people than does aristocracy, but around a higher average level. He either didn't have the vocabulary to say this concisely, or regarded using statistical terms as bad style. (I suspect the former, due to the time period.) He was also very fond of paradoxes, where he either inverted directions of causal arrows, or flipped their signs.
Maria Semple, Where'd You Go, Bernadette?
Literary fiction about Seattle, motherhood, marital collapse, aggressively eccentric architects, and Antarctica. Very funny and more than a bit touching.
Thomas Piketty, Capital in the Twenty-First Century [Online technical appendix, including extra notes, figures, and spreadsheets]
Yes, it's as good and important as everyone says. If by some chance you haven't read about this yet, I recommend Robert Solow, Branko Milanovic and Kathleen Geier for overviews; Suresh Naidu's take is the best I've seen on the strengths and weaknesses of the book, but doesn't summarize so much.
Some minor and scattered notes; I might write a proper review later. (Why not? Everybody else has.)
1. Perhaps it's the translation, but Piketty seems wordy and a bit repetitive; I think the same things could have been said more briskly. Perhaps relatedly, I got a little tired of the invocations of Austen, Balzac, and American television.
2. The book has given rise to the most perfect "I happen to have Marshall McLuhan right here" moment I ever hope to see.
3. Attempts to undermine his data have, unsurprisingly, blown up in his attackers' faces. Similarly, claims that Piketty ignores historical contigency, political factors and institutions are just bizarre.
4. Granting that nobody has better point estimates, I wish he'd give margins of error as well. (A counter-argument: maybe he could calculate purely-statistical standard errors, but a lot of the time they could be swamped by nearly-impossible-to-estimate systematic errors, due to, e.g., tax evasion.)
5. His two "laws" of capitalism are an accounting identity (the share of capital in national income is the rate of return on capital times the ratio of capital to income, $\alpha = r \beta$ ), and a long-run equilibrium condition (the steady-state capital/income ratio is the savings rate divided by the economy-wide growth rate, $\beta = s/g$ ), the latter presuming that two quite variable quantities ($s$ and $g$) stay fixed forever. So the first can't help but be true, and the second is of limited relevance. (Why should people keep saving the same fraction of national income as their wealth and income change?) But I don't think this matters very much, except for the style. (However, Milanovic has an interesting defense of Piketty on this point.)
6. He gets the Cambridge Capital Controversy wrong, and while that matters for our understanding of capital as a factor of production, it's irrelevant for capital as a store of value, which is what Piketty is all about. Similarly, Piketty doesn't need to worry about declining marginal returns to capital in the economy's aggregate production function, which is good, because aggregate production functions make no sense even within orthodox neo-classical economics. (The fact that orthodox neo-classical economists continue to use them is a bit of an intellectual embarrassment; they should have more self-respect.)
7. The distinction between "income from labor" and "income from capital" is part of our legal system, and Piketty rests a lot of his work on it. But It seems to me an analytical mistake to describe the high compensation of a "super-manager" as income from labor. While it isn't coming from owning their corporation, it is coming from (partially) controlling it. In some ways, it's more like the income of an ancien regime tax farmer, or an Ottoman timariot, than the income of a roofer, nurse's aid, computer programmer, or even an architect. (Actually, the analogy with the timariot grows on me the more I think about it. The timariot didn't own his timar, he couldn't sell it or bequeath it, any more than a super-manager owns his company. Officially, income in both cases is compensation for services rendered to the actual owner, whether sultan or stock-holder.) It would be very nice to see someone try to separate income from labor and income from control, but I have no clue how to do it, statistically. (Though I do have a modest proposal for how to reduce the control income of super-managers.)
8. p. 654, n. 56: For "Claude Debreu", read "Gerard Debreu". (Speaking of economists' "passion for mathematics and for purely theoretical ... speculation"!)
ETA: Let me emphasize the point about production functions, marginal returns on capital, etc. It cannot be emphasized enough that capital, for Piketty, is the same as wealth, assets, stores of value which can be traded in the market. He does not mean non-human factors of production, "capital goods". (Cf.) Capital goods can work fine as assets, but a much more typical asset is a claim on part of the product achieved through putting capital goods and labor to use. Because he is looking at wealth rather than capital goods, the appropriate unit of measurement, and the one he uses, is monetary rather than physical. One consequence is that Piketty can legitimately add up monetary amounts to get the total wealth of a person, a class, or a country. (Whereas adding up capital goods is deeply problematic at best; I don't think even the dullest Gosplan functionary would've tried to get the total capital of the USSR by adding up the weight or volume of its capital goods.)
This also has implications for the "marginal product of capital" question. If a capital good is measured in physical units, it's not crazy to imagine diminishing marginal returns. If some line of work needs tools, equipment, a proper space, etc., to be carried out, then the first crude tools and the shack which allow it to get started increase output immensely, then having a bit more equipment and a decent space helps, and after a certain point one extra spanner or crucible, with no extra worker, does very little. (Not crazy, but also not obviously true: see the work of Richard A. Miller [i, ii], which I learned of from Seth Ackerman's piece on Piketty.) Some critics of Piketty's forecasts point to this, to argue that his vision of widening inequality will fail on these grounds. They equate the rate of rate on capital, Piketty's $r$, with the marginal product of capital, and, believing the latter must decline, think $r$ must shrink as well. We thus have the curious spectacle of apostles of capitalism claiming it will be saved by a falling rate of profit. (I believe Uncle Karl would have savored the irony.) This intuition, however, is based on physical units of capital --- spanners, crucibles, servers, square meters of buildings. What about in monetary units?

Posted by crshalizi at June 22, 2014 10:54 | permanent link

## May 31, 2014

### Books to Read While the Algae Grow in Your Fur, May 2014

Attention conservation notice: I have no taste.

Robert Hughes, Rome: A Cultural, Visual, and Personal History
As the subtitle suggests, a bit of a grab-bag of Hughes talking about Rome, or Rome-related, subjects, seemingly as they caught his attention. Thus the chapter on ancient Rome contains a mix of recitals of the legends, archaeological findings, the military history of the Punic Wars (including a description of the corvus filled with what I recognize as school-boy enthusiasm), the rise of the Caesars — and then he gets to the art, especially the architecture, of the Augustan age, and takes off, before wandering back into political history (Diocletian--Constantine--Julian). The reader should, in other words, be prepared for a ramble.
Hughes is, unsurprisingly, at his best when talking about art. There is he knowledgeable, clear, sympathetic to a wide range of art but definitely unafraid of rendering judgment. If he doesn't always persuade (I remain completely immune to the charms of Baroque painting and sculpture), he definitely does his best to catalyze an appreciative reaction to the art in his reader, and one can hardly ask more of a critic
He's at his second best in the "personal" parts, conveying his impressions of Rome as he first found it in the early 1960s, and as he left it in the 2000s, to the detriment of the latter. (He's self-aware enough to reflect that some of that is the difference between being a young and an old man.) His ventures into the political and religious history of Rome are not as good — he has nothing new to say — but not bad.
Overall: no masterpiece, but always at least pleasant, and often informative and energizing.
R. A. Scotti, Basilica: The Splendor and the Scandal: Building St. Peter's
Mind candy: engaging-enough popular history, by a very-obviously Catholic writer. (My own first reaction to St. Peter's, on seeing it again for the first time as a grown-up, was that Luther had a point; my second and more charitable reaction was that there was an awesome space beneath the idolatrous and servile rubbish.)
Pacific Rim
Mind candy. While I like giant robots battling giant monsters, and I appreciate playing with elements of the epic (the warrior sulking in his tent; the catalog of ships), I'd have liked it better if the plot made more sense.
Sara Poole, Poisoner and The Borgia Betrayal
Mind candy: decent historical thrillers, though implausibly proto-feminist, philo-Semitic and proto-Enlightenment for the period.
Patrizia Castiglione, Massimo Falcioni, Annick Lesne and Angelo Vulpiani, Chaos and Coarse Graining in Statistical Mechanics
A good modern tour of key issues in what might be called the "Boltzmannian" tradition of work on the foundations of statistical mechanics, emphasizing the importance of understanding what happens in single, very large mechanical assemblages. Both "single" and "very large" here are important, and important by way of contrasts.
The emphasis on the dynamics of single assemblages contrasts with approaches (descending from Gibbs and from Jaynes) emphasizing "ensembles", or probability distributions over assemblages. (Ensembles are still used here, but as ways of simplifying calculations, not fundamental objects.) The entropy one wants to show (usually) grows over time is the Boltzmann entropy of the macrostate, not the Gibbs or Shannon entropy of the ensemble. (Thus studies of the dynamics of ensembles are, pace, e.g., Mackey, irrelevant to this question, whatever their other merits.) One wants to know that a typical microscopic trajectory will (usually) move the assemblage from a low-entropy (low-volume) macrostate to a high-volume macrostate, and moreover that once in the latter region, most trajectories that originated from the low-entropy macrostate will act like ones that began in the equilibrium macrostate. One wants, though I don't recall that they put it this way, a Markov property at the macroscopic level.
Randomizing behavior for macroscopic variables seems to require some amount of instability at the microscopic level, but not necessarily the very strict form of instability, of sensitive dependence on initial conditions, which we've come to call "chaos". Castiglione et al. present a very nice review of the definitions of chaos, the usual measures of chaos (Lyapunov exponents and Kolmogorov-Sinai entropy rate), and "finite-size" or non-asymptotic analogs, in the course of arguing that microscopic chaos is neither necessary nor sufficient for the applicability of statistical mechanics.
The single-assemblage viewpoint on statistical mechanics has often emphasized ergodicity, but Castiglione et al. down-play it. The ergodic property, as that has come to be understood in dynamical systems theory, is both too weak and too strong to really be useful. It's too weak because it doesn't say anything about how quickly time averages converge on expectations. (If it's too slow, it's irrelevant to short-lived creatures like us, but if it's too fast, we should never be able to observe non-equilibrium behavior!) It's too strong in that it applies to all integrable functions of the microscopic state, not just physically relevant ones.
The focus on large assemblages contrasts with low-dimensional dynamical systems. Here the authors closely follow the pioneering work of Khinchin, pointing out that if one has a non-interacting assemblage of particles and considers macroscopic observables which add up over molecules (e.g., total energy), one can prove that they are very close to their expectation values with very high probability. (This is a concentration of measure result, though the authors do not draw connections to that literature.) This still holds even when one relaxes non-interaction to weak, short-range interaction, and from strictly additive observables to ones where each microscopic degree of freedom has only a small influence on the total. (Again, familiar ground for concentration of measure.) This is a distinctly high-dimensional phenomenon, not found in low-dimensional systems even if very chaotic.
Putting these two ingredients — some form of randomizing local instability and high-dimensional concentration of measure — it becomes reasonable to think that something like statistical mechanics, in the microcanonical ensemble, will work. Moreover, for special systems one can actually rigorously prove results like Boltzmann's H-theorem in suitable asymptotic limits. An interesting large-deviations argument by De Roeck et al. suggests that there will generally be an H-theorem when (i) the macroscopic variables evolve autonomously in the large-assemblage limit, and (ii) microscopic phase-space volume is conserved. This conclusion is very congenial to the perspective of this book, but unfortunately the work of De Roeck et al. is not discussed.
One feature which pushes this book beyond being just a careful and judicious defense of the Boltzmannian viewpoint on statistical mechanics is its treatment of how, generally, one might obtain macroscopic dynamics from microscopic physics. This begins with an interesting discussion of multi-scale methods for differential equations, as an alternative to the usual series expansions of perturbation theory. This is then used to give a new-to-me perspective on renormalization, and why differences in microscopic dynamics wash out when it comes to aggregated, macroscopic variables. I found this material intriguing, but not as fully persuasive as the earlier parts.
Conor Fitzgerald, The Dogs of Rome
Mind candy. Police procedural with local color for Rome. (Having an American protagonist seems like a cheat.)
Colin Crouch, Making Capitalism Fit for Society
A plea for an "assertive" rather than a "defensive" social democracy, on the grounds that social democracy has nothing to be defensive about, and that the taming of capitalism is urgently necessary. That neo-liberalism has proved to be a combination of a sham and a disaster, I agree; that lots of his policy proposals, about combining a stronger safety net with more microeconomic flexibility, are both desirable and possible, I agree. But what he seems to skip over, completely, is where the power for this assertion will come from.
Disclaimer: I've never met Prof. Crouch, but he's a friend of a friend.
Elliott Kay, Poor Man's Fight
Mind candy science fiction. Does a good job of presenting the villains sympathetically, from inside their own minds. Also, props for having the hero's step-mother make a prediction of how joining the navy will distort the hero's character, for it coming true, and for the hero realizing it, to his great discomfort.
Sequel.

Posted by crshalizi at May 31, 2014 23:59 | permanent link

## April 30, 2014

### Books to Read While the Algae Grow in Your Fur, April 2014

Attention conservation notice: I have no taste.

Chris Willrich, Scroll of Years
Mind candy fantasy. The blurbs are over the top, but it is fun and decidedly better-written than average. Mildly orientalist, though in a respectful mode.
Matthew Bogart, The Chairs' Hiatus
Kelly Sue DeConnick and Emma Rios, Pretty Deadly
Joe Harris, Great Pacific: 1, Trashed; 2, Nation Building
Brian K. Vaughan and Fiona Staples, Saga, vols. 2 and 3
L. Frank Weber, Bikini Cowboy
Terry LaBan, Muktuk Wolfsbreath, Hard Boiled Shaman: The Spirit of Boo
Comic book mind candy. Muktuk Wolfsbreath is especially notable for ethnographic accuracy, Pretty Deadly for the gorgeous art and genuinely-mythic weirdness, and Saga for general awesomeness. (Previously for Saga.)
Jeff VanderMeer, Annihilation
Mind candy, but incredible mind candy. The basic story is a familiar one for SF: an expedition into an unknown and hostile environment quickly goes spectacularly awry, as the explorers don't appreciate just how strange that environment really is. But from there it builds to a gripping story that combines science fiction about bizarre biology with genuinely creepy horror. It's Lovecraftian in the best sense, not because it uses the props of Cthulhiana, but because it gives the feeling of having encountered something truly, frighteningly alien. (In contrast.)
There are two sequels coming out later this year; I've ordered both.
Mind candy: a haunted house story, with a space-opera setting. (Self-presentation.)
S. Frederick Starr, Lost Enlightenment: Central Asia's Golden Age from the Arab Conquest to Tamerlane
Starr has been a historian of Central Asia throughout his long professional career, and like many such, he feels that the region doesn't get enough respect in world history. This is very much an effort in rectifying that, along the way depicting medieval Central Asia as a center of the Hellenistic rationalism which he sees as being the seed of modern science and enlightenment. (It's pretty unashamedly whiggish history.)
Starr's Central Asia is urban and mercantile. It should be understood as the historic network of towns in, very roughly, the basins of the Amu Darya and Syr Darya rivers, or Transoxiana plus Khorasan and Khwarezm. Starr argues that this region formed a fairly coherent cultural area from a very early period, characterized by intensive irrigation, the cultural and political dominance of urban elites, the importance of long-distance over-land trade (famously but not exclusively, the Silk Road), and so cross-exposure to ideas and religions developed in the better-known civilizations of the Mediterranean, the Fertile Crescent, Iran, India and China. One consequence of this, he suggests, was an interest in systematizing these traditions, e.g., compiling versions of the Buddhist canon.
With the coming of Islam, which he depicts as a very drawn-out process, some of these same traditions led to directions like compiling hadith. Beyond this, the coming of Islam exposed local intellectuals both to Mulsim religious concepts, to the works of Greek science and philosophy, and to Indian mathematics and science. (He gives a lot more emphasis to the Arab and Greek contributions than the Indian.) In his telling, it was the tension between these which led to the great contributions of the great figures of medieval Islamic intellectual history. Starr is at pains to claim as many of these figures for Central Asia as possible, whether by where they lived and worked, where their families were from, where they trained, or sometimes where their teachers were from. [0] He even, with some justice, depicts the rise of the 'Abbasid dynasty as a conquest of Islam by Khurasan.
Much of the book is accordingly devoted to the history of mathematics, natural science, philosophy, theology, and belles lettres in Central Asia, with glances at the fine arts (especially painting and architecture) and politics (especially royal patronage). This largely takes the form of capsule biographies of the most important scholars, and sketches of the cities in which they worked. These seem generally reliable, though there are some grounds for worry. One is that I can't tell whether Starr is just awkward at explaining what mathematicians did, or whether he doesn't understand it and is garbling his sources. The other is that there are places where he definitely over-reaches in claiming influence [1]. Even deducting for these exaggerations and defects, Starr makes a sound case that there was a long period of time --- as he says, from the Arab conquests to the coming of the Timurids --- when Central Asia was the home to much of the best intellectual activity of the old world. That this amounted to an "age of Enlightenment" comparable to 17th and 18th century Europe seems another over-fond exaggeration.
What Starr would have liked to produce is something as definitive, and as revelatory, as Joseph Needham's Science and Civilisation in China. (He's pretty up front about this.) He knows that he hasn't gotten there. He can't be blamed for this: even for so extraordinary a figure as Needham, it was the work of a lifetime, backed by a massive team. Still, one can hope that his book will help make such an effort more likely. In the meanwhile, it's a decently-written and mostly-accurate popular history about a time and place which were once quite important, and have since faded into undeserved obscurity.
What the book is doing with blurbs from various reactionary foreign-affairs pundits, up to and including Henry Kissinger, I couldn't say, though I have suspicions.
0: He also feels it necessary to make the elementary point that writing in Arabic didn't make these men "Arabs", any more than writing in Latin made contemporary European scholars "Romans". I will trust his judgment that there are still people who need to hear this.
1: E.g., on p. 421, it's baldly asserted that Hume found Ghazali's arguments against causality "congenial". Now, the similarity between the two men's arguments have often been pointed out, and the relevant book of Ghazali's, The Incoherence of the Philosophers, was known to the Scholastics in Latin translation. It's conceivable that Hume encountered a copy he could have read. Nonetheless, Ghazali's name does not appear, in any romanization, in Hume's Treatise of Hume Nature, Enquiry Concerning Human Understanding, Enquiry Concerning the Principles of Morals, Dialogues Concerning Natural Religion, or Essays, Moral, Political, and Literary. (I have not searched Hume's complete works.) No other writer on either philosopher, that I am aware of, suggests either a direct influence or even the transmission of a tradition, as opposed to a re-invention, and Starr provides no supporting citation or original evidence.
Mind candy, at the edge of being something greater. Humanity is visited by ridiculously advanced aliens, who leave behind artifacts which we understand no more than ants could comprehend the relics of the titular picnic. Owing to human greed, stupidity, and (it must be said) capitalism, this goes even worse for us than it would for the ants.
M. John Harrison, Nova Swing
Mind candy: noir science fiction, owing a massive debt to Roadside Picnic.
Elizabeth Bear, Steles of the Sky
Conclusion to the trilogy begun in Range of Ghosts and Shattered Pillars. It is, to my mind, magnificent; all the promise of the earlier books is fulfilled.
Felix Gilman, The Revolutions
Historical fantasy set in Edwardian London, and the outer spheres of the solar system, featuring under-employed young people with literary ambitions, dueling occult societies, interplanetary romances, and distributed Chinese rooms.
Gene Wolfe, The Claw of the Conciliator
My comments on The Shadow of the Torturer apply with even greater force.
Darwyn Cooke, Parker (1, 2, 3, 4)
Mind candy: comic book versions of the classic crime novels by Richard Stark. The pictures are a nice complement to the high-energy stories about characters with no morally redeeming qualities whatsoever.

Posted by crshalizi at April 30, 2014 23:59 | permanent link

## April 01, 2014

### "Proper scoring rules and linear estimating equations in exponential families" (Next Week at the Statistics Seminar)

Attention conservation notice: Only of interest if you (1) care about estimating complicated statistical models, and (2) will be in Pittsburgh on Monday.

Steffen Lauritzen, "Proper scoring rules and linear estimating equations in exponential families"
Abstract: In models of high complexity, the computational burden involved in calculating the maximum likelihood estimator can be forbidding. Proper scoring rules (Brier 1950, Good 1952, Bregman 1967, de Finetti 1975) such as the logarithmic score, the Brier score, and others, induce natural unbiased estimating equations that generally lead to consistent estimation of unknown parameters. The logarithmic score corresponds to maximum likelihood estimation whereas a score function introduced by Hyvärinen (2005) leads to linear estimation equations for exponential families.
We shall briefly review the facts about proper scoring rules and their associated divergences, entropy measures, and estimating equations. We show how Hyvärinen's rule leads to particularly simple estimating equations for Gaussian graphical models, including Gaussian graphical models with symmetry.
The lecture is based on joint work with Philip Dawid, Matthew Parry, and Peter Forbes. For a recent reference see: P. G. M. Forbes and S. Lauritzen (2013), "Linear Estimating Equations for Exponential Families with Application to Gaussian Linear Concentration Models", arXiv:1311.0662.
Time and place: 4--5 pm on Monday, 7 April 2014, in 125 Scaife Hall.

Much of what I know about graphical models I learned from Prof. Lauritzen's book. His work on sufficienct statistics and extremal models, and their connections to symmetry and prediction, has shaped how I think about big chunks of statistics, including stochastic processes and networks. I am really looking forward to this.

(To add some commentary purely of my own: I sometimes encounter the idea that frequentist statistics is somehow completely committed to maximum likelihood, and has nothing to offer when that fails, as it sometimes does [1]. While I can't of course speak for every frequentist statistician, this seems silly. Frequentism is a family of ideas about when probability makes sense, and it leads to some ideas about how to evaluate statistical models and methods, namely, by their error properties. What justifies maximum likelihood estimation, from this perspective, is not the intrinsic inalienable rightness of taking that function and making it big. Rather, it's that in many situations maximum likelihood converges to the right answer (consistency), and in a somewhat narrower range will converge as fast as anything else (efficiency). When those fail, so much the worse for maximum likelihood; use something else that is consistent. In situations where maximizing the likelihood has nice mathematical properties but is computationally intractable, so much the worse for maximum likelihood; use something else that's consistent and tractable. Estimation by minimizing a well-behaved objective function has many nice features, so when we give up on likelihood it's reasonable to try minimizing some other proper scoring function, but again, there's nothing which says we must.)

[1]: It's not worth my time today to link to particular examples; I'll just say that from my own reading and conversation, this opinion is not totally confined to the kind of website which proves that rule 34 applies even to Bayes's theorem. ^

Posted by crshalizi at April 01, 2014 10:45 | permanent link

## March 31, 2014

### Books to Read While the Algae Grow in Your Fur, March 2014

Attention conservation notice: I have no taste.

Brian Michael Bendis, Michael Avon Oeming, et al. Powers (1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14)
Comic book mind candy, at the super-hero/police-procedural interface.
Martha Wells, The Fall of Ile-Rien (The Wizard Hunters, The Ships of Air, The Gate of Gods)
Mind candy. I read these when they were first published, a decade ago, and re-visited them when they came out as audiobooks. They remain unusually smart fantasy with refreshingly human protagonists. (Though I think the narrator for the recordings paused too often, in odd places, to be really effective.)
Graydon Saunders, The March North
Mind candy. Readers know I am a huge fan of science fiction and fantasy (look around you), but there's no denying both genres have a big soft spot for authoritarianism and feudalism. I picked this up because it promised to be an egalitarian fantasy novel, and because (back in the Late Bronze Age of Internet time) I used to like Saunders's posts on rec.arts.sf.written . I am glad I did: it's not the best-written novel ever, but it's more than competent, it scratches the genre itch, and Saunders has thought through how people-power could work in a world where part of the normal life cycle of a wizard would ordinarily be ruling as a Dark Lord for centuries. (I suspect his solution owes something to the pike square.) The set-up calls out for sequels, which I would eat up with a spoon.
Marie Brennan, The Tropic of Serpents: A Memoir by Lady Trent
Mind candy: further adventures in the natural history of dragons.
Franklin M. Fisher, Disequilibrium Foundations of Equilibrium Economics
This is a detailed, clear and innovative treatment of what was known in 1983 about the stability of economic equilibrium. (It begins with an argument, which I find entirely convincing, that this is an important question, especially for economists who only want to reason about equilibria.) The last half of the book is a very detailed treatment of a dis-equilibrium model of interacting rational agents, and the conditions under which it will approach a Walrasian (price-supported) equilibrium. (These conditions involve non-zero transaction costs, each agent setting its own prices, and something Fisher calls "No Favorable Surprise", the idea that unexpected changes never make things better.) Remarkably, Fisher's model recovers such obvious features of the real world as (i) money existing and being useful, and (ii) agents continue to trade for as long as they live, rather than going through a spurt of exchange at the dawn of time and then never trading again. It's a tour de force, especially because of the clarity of the writing. I wish I'd read it long ago.
Fisher has a 2010 paper, reflecting on the state of the art a few years ago: "The Stability of General Equilibrium --- What Do We Know and Why Is It Important?".
— One disappointment with this approach: Fisher doesn't consider the possibility that aggregated variables might be in equilibrium, even though individuals are not in "personal equilibrium". E.g., prevailing prices and quantities are stable around some equilibrium values (up to small fluctuations), even though each individual is constantly perpetually alternating between seizing on arbitrage opportunities and being frustrated in their plans. This is more or less the gap that is being filled by Herb Gintis and collaborators in their recent work (1, 2, 3, 4). Gintis et al. also emphasize the importance of agents setting their own prices, rather than having a centralized auctioneer decree a single binding price vector.
The Borgias (1, 2, 3)
Mind candy. Fun in its own way, but I'm disappointed that they made the politics less treacherous and back-stabbing than it actually was. (Also, there's a chunk of anachronistic yearning for equal rights and repulsion over corruption.)
ObLinkage: Comparative appreciation by a historian of the Renaissance. (I haven't seen the rival series.)
House of Cards
Mind candy. Rather to my surprise, I enjoyed this at least as much as the original. (I will not comment on the economics of the second season.)
Homeland (1, 2, 3)
Mind candy. Well-acted, but I find the politics very dubious. In particular, "let the CIA do whatever it wants in Iran" is a big part of how we got into this mess.
Dexter (5, 6, 7 and 8)
Mind candy. Not as good as the first four seasons, but then, very little is. I refuse to believe the ending. Previously: 2, 3, 4.
Allie Brosh, Hyperbole and a Half: Unfortunate Situations, Flawed Coping Mechanisms, Mayhem, and Other Things That Happened
As a sentient being with a working Internet connection, you are aware that Brosh is one of this age's greatest writers on moral psychology (and dogs). This is a collection of some of her best pieces.
John Dollard, Caste and Class in a Southern Town
The easy way to read this ethnography of "Southerntown" (= Indianola, Miss.) in 1936 would be as a horror story, with a complacent "thankfully we're not like that" air. (At least, it would be easy for me to read it so.) Better, I think, to realize both how horrifying this was, and to reflect on what injustices are similarly woven into my own way of life...
Seanan McGuire, Half-Off Ragnarok
Mind candy: the travails of a mild-manner Ohio cryptozoologist. (I think it could be enjoyed without the earlier books in the series.) Previously: 1, 2.

Posted by crshalizi at March 31, 2014 23:59 | permanent link

Three-Toed Sloth:   Hosted, but not endorsed, by the Center for the Study of Complex Systems