Pieter Geyl was a great historian … He distrusted generalizations and tested them against detail. Thus, Toynbee asserted that men were stimulated by adversity and instanced the response of the Dutch to the challenge of the threat from the sea. Geyl answered that Dutch civilization started in the lush inland areas of the Netherlands and spread from there to the sea-coast. What then, he asked, remained of “challenge and response”. To those who like sweeping generalizations, Geyl’s objection seemed trivial. What did one exception matter? To others, particularly to professional historians, it seemed decisive. —A.J.P. Taylor
As far as I know, this quote comes from a collection of essays by Taylor on the methodologies of historical research, although, ironically, this shows up in a book by an economist, Technology and Market Structure. (one of my favorites, incidentally, as it inspired much of my thinking about markets and tastes) This seemed especially relevant in light of a recent article in the Chronicle of Higher Education about economics of slavery.
The article characterizes the debate as between economists, who are quantitative and data-oriented, and historians, who are more “culturally” oriented. I think that’s BS. It’s the fight between Geyl and Toynbee again, squarely in the realm exclusively of historians, between who are interested in the details and the causal linkages and those who want to tell a big, simple story, never mind the specifics. Eric Foner is right to call the economic historians “champion nitpickers.” Pieter Geyl and A. J. P. Taylor would be proud to be given that label. The overlooked facts, even if they might seem trivial, are the decisive ones when they contradict the sweeping generalizations.
Numbers and data, especially when used properly in context of what statistics has always been–the study of patterns in deviations, i.e. of the exceptions to the generalization in the “details”–are a great tool for the “champion nitpickers.” It may be, to use my favorite example, 90% of the voters vote their party ID in House elections. But the remaining 10% cast their ballots overwhelmingly in favor of incumbents. The result is that, in terms of deciding the outcomes, the vast majority of the data is irrelevant in shaping the results. Paying attention to the “details,” on the other hand, offers a credible explanation for what is taking place as well as potential answers when underlying conditions change. In other words, this is exactly how science is supposed to work: allowing for evaluation of a “theory” against the facts, for refinement/update/change of that theory based on the contrary evidence, and retesting of the revised theory against more facts.
But a lot of modern day data folks are more like Toynbee than Geyl. They are not interested in a hard nosed look at the data focused on details that don’t fit. They want to tell a story, and are interested in using data and theories as props to shore up their big stories and generalizations. This is, of course, what I’d been calling wonkism, and why I think wonkism is the worst enemy of science, even if, exasperatingly, wonkism is constantly mistaken for science and a lot of criticism of “science” is really directed at wonkist abuse of science. The article is, in a sense, an example of this very point. The cultural historians are using data, economic data and a lot of it, too, in fact. They are also invoking a economics-inspired argument. But they are doing so to advance their ideological points, of condemning the morality of slavery. Now, one can have moral views about slavery, but does the morality depend on the facts and logic? If one resorts to fudging facts and logic, ignore or worse, proactively dismiss inconvenient but significant details with a handwave, and resorts of argument by assertion, then it does not matter how much data they cram into their claims. They are a fraud. I think a lot of data-intensive exercises, unfortunately, fall into this trap these days, as they rely on the data mainly as prop for their assertions without engaging in critical and nuanced evaluation of the contrary exceptions.
This is critical in light of the instructive little mistake in the article. The Chronicle article hilariously mischaracterizes Bob Fogel, as a “neoclassical economist.” Was it because he was at the University of Chicago and was a Nobel Prize winner? But the man was a committed communist, a member of the Communist Party, in fact, during the McCarthy Era, no less. He was, rare among ideological Marxists, a hardcore methodological Marxian as well–someone who believes in tight logical arguments and hard nosed look at the facts, not mushy philosophizing. He was arguing against mushy cultural historians of his day who believed that slavery was going to die a natural death because “Southerners wanted to stay part of Western civilization and Western civilization abhors involuntary servitude,” or some other nonsense. Fogel argued that the hard economic data showed that slavery was both increasingly more profitable and expanding, and that the hard profit motive was beating out the mushy cultural musings. In the end, though, he came to champion a decisive non-economic solution: the only force that could stop slavery was a cultural one, namely, the abolitionist sentiment in the north, backed up by the cold empirical fact of a million Union bayonets, not by some economic force of inevitability. Only if today’s wonks were like Bob Fogel and not those mushy cultural historians he argued against!
PS. One might say that, if it is the big story that you want to tell, you don’t really need data, or at least a lot of it. The basic principle of statistics says so: big points should be obvious enough even with a small sample size. If you need a huge sample size, the point cannot be very big. If you have a really big data, you can evaluate nuances, spot the inconsistent details, and so on that can make for richer, more complex, and less easily summarizable theory. But if the big data comes into the service of the big story, then you are using a big hammer to build a micromachine.