Speaking, as we were, of the Reformation, one point I neglected in my description of MacCulloch's book is the stress he lays, at several points, on the (veridical) perception that Latin Christendom was under attack by the Ottoman Turks, and they way the need to defend against the Ottomans forced the Catholic powers into compromises with Protestants, establishing at least some measure of toleration, much against everyone's will (since the Protestants would have preferred to rule, not be tolerated). Now, via Dani Rodrik, some statistical evidence:
Now, the statistics here are the kind of thing you'd expect to see from an economist (who is not an econometrcian), so I'm not entirely happy with it. I should make it clear that the things I'm about to complain about are things I could complain about for essentially any paper I read in quantitative social science. That being the case, I'll stick my grumblings down below, and just talk about why this matters. (What follows makes no pretense to originality, being merely warmed-over William McNeill, Ernest Gellner and Marshall Hodgson.)
The breakthrough to auto-catalytic intellectual, social and economic growth first took place in a fairly small part of northwest Europe, where the outcome of the wars of religion had been to establish a social space with some degree of secularism, mutual toleration and individual autonomy. Moreover, it took place in the context of intricate and intense rivalries between the different European states, where the new kinds of social power that the breakthrough made possible were eagerly seized upon as a strategic advantage, and where it was possible for unpopular or deviant thinkers (practical or theoretical) to move from one state to another, playing them off against each other. This was weird. If you look at other world civilizations, which were contemporary and certainly comparable, the trend was towards the formation of massive "gunpowder empires", which may have been more or less tolerant within certain limits, depending on the whim of the emperor (compare Akbar to Aurangzeb), but certainly did not encourage this kind of disruptive innovation. Had one of the Christian dynasties — e.g., the Hapsburgs — succeeded in establishing that kind of empire in Europe, it is hard to imagine modernity actually taking off. Had the Catholic powers at least been able to crush the Protestants, even if they remained dynastically divided, it's hard to see it being much more innovative than, say, the Muslim world, divided between the Ottoman, Safavid and Mughal empires, with assorted smaller principalties around them. Had the Protestants, by some miracle, been able to actually get what they wanted (perhaps under Gustavus Adolphus?), it would have been just as repressive. The situation in Amsterdam, London and their dependencies wasn't what anyone had set out to create; it was for the most part regarded as a regrettable realization that neither side could pound each other into submission, and so establish a proper Christian commonwealth. One of the reasons none of these scenarios took place, and so we enjoy our current approximations to open societies, was that all the Christians had to fight off the Ottomans, which in particular meant coming to those regrettable compromises which established toleration. This is actual quantitative evidence that the Ottoman threat really did have a substantialm impact, by way of making the Christians fight each other less and put up with each other more.
One could also imagine an alternative history where the Christians so completely failed to ignore the fact that the other side were damned dirty heretics who deserved to burn (in this world and the next) that the Ottomans were actually able to conquer much or most of Latin Christendom. This scenario also doesn't look good for the the great transformation, but it's worth remembering Ernest Gellner's remarks on a related counter-factual:
I like to imagine what would have happened had the Arabs won at Poitiers and gone on to conquer and Islamise Europe. No doubt we should all be admiring Ibn Weber's The Kharejite Ethic and the Spirit of Capitalism which would conclusively demonstrate how the modern rational spirit and its expression in business and bureaucratic organisation could only have arisen in consequence of the sixteenth-century neo-Kharejite puritanism in northern Europe. In particular, the work would demonstrate how modern economic and orgisational rationality could never have arisen had Europe stayed Christian, given the inveterate proclivity of that faith to a baroque, manipulative, patronage-ridden, quasi-animistic and disorderly vision of the world. A faith so given to seeing the cosmic order as bribable by pious works and donations could never have taught its adherents to rely on faith alone and to produce and accumulate in an orderly, systematic, and unwavering manner. Would they not always have blown their profits in purchasing tickets to eternal bliss, rather than going on to accumulate more and more?
A Muslim Europe would also have saved Hegel from the need to indulge in most painfully tortuous arguments in order to explain how an earlier faith, Christianity, nevertheless is more final and absolute than a chronologically later one, namely Islam. (In fact he did it by invoking the fact that Europe was only Christianised at the time of Charlemagne, who is at least suitably posterior to Muhammamed.) Had Islam, the later and by some plausible criteria purer faith, prevailed, no such problem would have arisen for a Muslim Hegel. There would have been no embarrassing boob in the welthistorischer timetable. Altogether, from the viewpoint of an elegant philosophy of history, which sees the story of mankind as a sustained build-up towards our condition, it would have been far more satisfactory if the Arabs had won. By various obvious criteria — universalism, scripturalism, spiritual egalitarianism, the extension of full participation in the sacred community not to one, or some, but to all, and the rational systematisation of social life — Islam is, of the three great Western monotheisms, the one closest to modernity. [Muslim Society, p. 7; my links]
Now, here are the promised grumblings about statistics:
Iyigun does everything by linear regression, so what he's doing is regressing various measures of how much the Christians were fighting among each other, in any given year, on several different measures of how much the Ottomans were fighting them that year, and a whole slew of control variables. (None of these include information on the weather or crops, oddly.) It seems to me that it would be more appropriate to use the propensity score framework devised by Don Rubin and co-authors here: basically, work out the likelihood of the Ottomans attacking the Christians in a given year, as a function of the controls, and then compare years where these "propensity scores" are similar, but the Ottomans did attack in one case and didn't in the other. While this isn't a solution to all causal inference problems, it seems like what's wanted here. (Rubin and Waterman have written a nice explanation of the method, and its advantages over regression for this sort of problem.)
Given that he's going to use linear regression, Iyigun certainly uses it perfectly well, and gets pretty much the same result from a whole bunch of different specifications, which is always comforting (and too often neglected). (This includes one set of results where he uses robust regression.) What's striking to me, though, is that while he gets a really big effect, with fighting the Ottomans reducing intra-Christian conflicts by anywhere between 22 and 45 percent, what he keeps talking about is whether or not his results are statistically significant. I am about as hard-core a frequentist statistician as they come, but this strikes me as just a little mis-placed... If you are going to do lots of significance testing in regression, I'd be happier if you did it by bootstrapping, rather than just t-tests; two ways come immediately to mind, but I'm sure there are others. (1: randomly permute the variable of interest, which here is Ottoman military activity, OTTOMAN, over the data points and re-run the regression, thus getting an estimate of the sample distribution of the coefficient on OTTOMAN, when that's independent of everything else, but the rest of the dependence structure is the same. 2: Omit OTTOMAN from the regression, estimate coefficients for all your control variables, then simulate from that model and fit your full regression, including OTTOMAN.) Maybe 200 observations is enough that this won't make much difference.
Update, 19 January 2010: I must've been having a brain-fault when I wrote the stuff about bootstrapping above. The obvious approach, of course, is to bootstrap the residuals, or the cases, to get a confidence interval for the coefficient on OTTOMAN, and see if it includes zero.
The sub-title comes from commenter "ajay" in this Crooked Timber thread, talking about something completely different.
Posted at July 01, 2007 23:00 | permanent link