The publication of the Feature Article in Nature about my research on American political violence elicited a wave of comments on the Web. The expression ‘feeding frenzy’ comes to mind. I’ve had a lot of fun reading those comments that I came across (and thanks to various people who sent me links). Partly its sheer vanity (hey, I am as human as the next guy), but it is also interesting in other, less trivial ways.
Because 99.9 percent of those who comment on cliodynamics didn’t bother to consult any of my academic articles (no surprise there) or even popular writings, one can trace readily how the ‘signal’ gets degraded in the transmission chain (an interesting study of cultural evolution here). So Laura Spinney’s article in Nature is pretty good because she extensively interviewed me in Frankfurt (but I did not see her text before the publication, I don’t agree with all of her emphases and interpretations). The articles based on her Nature piece are substantially worse, as their authors misunderstand more points and fill in the gaps with their own musings. And so on. After a few more steps the information content degrades to the level of white noise (or, to put it bluntly, garbage).
But it doesn’t prevent the commenters from passing their judgment on cliodynamics. As far as I can see, most judgments fall into two broad (and mutually contradictory) categories. One is that cliodynamics is crap because it is obviously wrong. The other one is that cliodynamics is crap because what it says is obvious without a need for any sophisticated analytical techniques.
No news here, we all know that the blogosphere is mostly – what would be a polite way of saying it, a garbage heap? Although I must admit, it could be worse. In the Russian blogosphere I would be immediately accused of being a paid agent of – the Liberals, or the Conservatives, or the KGB and personally Putin, or the CIA, or the oil interests, you name it. The closest I came of being accused of pushing the political agenda is by Maria Konnikova of Scientific American, who wrote that “in history … quantification and precise explanation is … so politically useful.” I wonder, to whom? The truth is, some results of my analysis will be unpalatable to the Republicans, others to the Democrats, and certain ones to me personally, but that’s a separate story.
While we certainly don’t expect the typical denizens of the blogosphere to bother with consulting the sources, might we expect better of the scientists? Apparently not. I am sorry to single out Massimo Pigliucci, who is a dear colleague, but here’s what he said, according to the Yahoo News piece by Natalie Wolchover:
Pigliucci isn’t convinced that the 50-year cycle of violence Turchin has identified in U.S. history reflects more than just a random fluctuation. “The database is too short: the entire study covers the period 1780-2010, a mere 230 years,” Pigliucci wrote in an email. “You can fit at most four 50-year peaks and two secular ones. I just don’t see how one could reasonably exclude that the observed pattern is random. But of course we would have to wait a lot longer to collect new data and find out.”
This sounds like I used data-mining techniques on the 230-year time series to extract cyclical patterns from it. If that was what I did, Massimo would be completely right. But please, I’ve cut my teeth on the analysis of time-series data (just see my CV if you don’t want to take my word for it).
Had Massimo read my paper in JPR, he would have seen that I use the US political violence database as an ‘out-of-sample dataset.’ Stripping it of technical jargon, this means that I developed the hypothesis about what we should observe on one set of data, and then I used a completely different data set to test the hypothesis. This approach allows us to avoid circularity. The well known problem in statistical inference is that if you fit a complex enough model to data, you can ‘explain’ it perfectly without any error. For this reason, when we really want to find out how good our model is, we use it to predict data that was not used in developing the model. And that is what I did.
In the beginning of the paper I describe the predicted pattern (the long secular wave with 50-year cycles superimposed on it). I illustrate this pattern with data from the Roman and French history. Then I test the predictions with a novel dataset on American history.
This is a true scientific prediction, even though it is not about the future – it’s about the pattern for 1780–2010. When I developed the hypothesis (it was published in my 2003 book Historical Dynamics) I had no knowledge of what the US pattern would be, because there was not a database that could tell me this. Later, in 2008 I started compiling the sources for the study that would eventually result in the database. When I analyzed the resulting database I was actually surprised by how ‘neat’ the pattern was (in previous analyses the 50-year cycle was messier, with its period ranging between 40 and 60 years).
This is all clearly laid out in my JPR article. And it has more – it’s not just the pattern in the political violence data; a multitude of other ‘structural-demographic’ variables changed systematically in a way that supports the predictions of the theory.
I know, I know – I am backlsiding into the ‘scientist mode’ again. So to end on a lighter note, here is the best indictment of my work that I have seen so far (it is one of the comments on the Nature piece; I quote it in full):
I think this whole article is filled with useless information, and not even a few good guesses. However, the subject is facinating. For a more scholarly approach to the same issue, try analysis of the Mayan calendar. I am aware that the Mayans were able to work out mathematical predictions for periods of upheaval ;and when I have seen how closely the Mayan predictions correlate to what we know, it is downright eerie. I am not an expert but I know two people who have studied the Mayan calendar intensely. We (all) do not hold superstitious beliefs but it is apparent the Mayan elite were able to obtain some insights that we are clearly missing.