“Building on theories used to describe magnets, scientists have put together a model that captures something very different . . .”

There’s a story that (some) physicists and science reporters seem to like, which is the idea that some clever mathematician or physicist can derive universal laws of social behavior.

It’s time to tell you all: Hari Seldon never existed.

Here’s what I think of these stories of physicists who discover the laws of society. I think they’re comparable to the stories we used to hear about, of the eccentric inventor who built a car engine that got 200 mpg, or ran on water, but was then suppressed by GM because they couldn’t handle the competition. Or, if we want go Heinlein, some guy who builds a time machine in his garage out of random spare parts.

Or maybe I should put it in a way that everyone can understand. Look at Richard Feynman, master swordsman and physicist, the guy who was so much smarter and more charming than everyone else, the guy who bragged that when he took up biology as a hobby he could do better than the biologists, who bragged that when he took up painting as a hobby he was an awesome painter. Even Richard Feynman did not think he could discover laws of social behavior. Sure, maybe he had better things to do. But, just maybe, maybe . . . he realized it wasn’t gonna happen.

I was reminded of this recently when Peter Dodds pointed me to this news article by Nathan Collins, entitled “Physics Predicts U.S. Voting Patterns”:

Building on theories used to describe magnets, scientists have put together a model that captures something very different: voting patterns in U.S. presidential elections. . . . For example, a person who voted Republican in one election and lives in a politically neutral county but works in a heavily Democratic county would likely vote for a Democrat in the next election. . . . the researchers focused on the distribution of Republican margins of victory across U.S. counties as well as how correlations between two counties’ vote shares changed with the distance separating them, quantities more commonly used to describe the transition from a demagnetized block of iron to a magnetized one. . . . the researchers predicted a bell curve distribution of county-level margins of victory and surprisingly long-range correlations between counties; that suggests that some counties, at least, could feel the effects of social pressures in counties on the other side of the nation, they report this month in Physical Review Letters. What’s more, those patterns were a close match to the actual election data . . .

I disagree with this last claim. Their model seems to reproduce a normal distribution of county-level vote shares with spatial correlation. All the stuff about magnets and social interactions and a “noisy diffusive process” and “anysotropic coupling topology” are pretty much irrelevant.

The link above goes to an abstract only, but a quick search revealed the preprint, Is the Voter Model a model for voters?, by Juan Fernández-Gracia, Krzysztof Suchecki, José Ramasco, Maxi San Miguel, and Víctor Eguíluz. So you can read it yourself if you’d like (but I don’t recommend it).

Its a bit sad to see Science and Physical Review Letters falling for this sort of thing, but I guess the story is so appealing they just can’t help it. It’s a great narrative, that the laws of social nature are just out there, waiting to be figured out by some physicist.

The big picture

Maybe it would help to explain what I’m not saying here. I’m not saying that outsiders can’t or shouldn’t make contributions to social science, nor am I saying that complicated mathematical models from physical science shouldn’t be used to model social behavior, voting included.

Indeed, I myself am a former physicist, and in 2002 I published a paper (with Jonathan Katz and Francis Tuerlinckx) entitled, “The Mathematics and Statistics of Voting Power,” reproducing certain features of elections using stochastic processes defined on a tree of voters. In retrospect, I think maybe this paper could’ve been published in a physics journal! One of the models even has a phase transition—it’s based on the Ising model from quantum physics—and we have tons of math (see, for example, pages 430 and 432 of the linked paper). So it’s not that it can’t be done, but the work has to be evaluated on its own terms. The physics connection isn’t enough on its own.

To get back to my introduction to this post, there’s sometimes a pattern in the science media of naive beliefs about what mathematical modeling can do. We’ve seen it in the uncritical celebration by Malcolm Gladwell of the uncalibrated claims by John Gottman that he could use mathematical models to predict divorces. We even see it in some of the media treatment of Nate Silver, when people act as if he has some secret formula rather than (as Nate himself insists) simply using good data, sensible models, and hard work. And a few years earlier, the celebration of Steve Levitt as a guy who’s statistical sumo wrestling skills could be used to catch terrorists. Again, let me be clear: Nate Silver and Steve Levitt do excellent work; I’m not criticizing them, rather I’m criticizing the attitude that they’re doing something magic.

54 thoughts on ““Building on theories used to describe magnets, scientists have put together a model that captures something very different . . .”

  1. *cough* positivity ratio *cough*

    Seriously though, there is nothing new under the sun… in fact the real scandal here is how we keep repeating the same mistakes, over and over again. Sokal, Friedman, and I cited Stanislav Andreski’s “Social Sciences as Sorcery” (1972) in our rebuttal of the Losada nonsense. Here is another gem from the same book which tries to explain the attraction of this kind of thing:

    “…quantophrenic psychology gratifies the bureaucrats, big business, and the advertisers, by telling them what they want to hear. For it would make the job easier for the bureaucrat or any other kind of manipulator if everybody (except himself) resembled a push-button automaton, with perfectly predictable outputs following known inputs.”
    — Andreski, 1972, p. 139

    The term “quantophrenia” — defined by Wiktionary as “Excessive reliance on or use of facts and figures that can be derived using statistical or mathematical procedures” — was introduced by Pitirim Sorokin, the founder of sociology at Harvard, in “Fads and Foibles in Modern Sociology and Related Sciences”, 1956. So this kind of stuff was apparently a problem back then, even before we had the computing power to go on every possible fishing expedition in the universe simultaneously.

    • That paper was written by psychologists adapting math from quantum physics to model decision making. They argue that quantum probability allows them to account for particular patterns of behavior that standard probability theory cannot model without additional machinery. They’re not physicists insisting that physics is some kind of trump card.

      • I hope this paper does get a discussion here some time. Having spent most of the day yesterday performing physics demonstrations for hippies (http://eighteenthelephant.wordpress.com/2014/07/14/you-should-do-birthday-parties-year-2-at-the-oregon-country-fair/) — lots of fun, by the way — I’m even more appalled now than I was when I first saw this paper about the use of “quantum” in its terminology. Yes, it’s true that they’re only claiming to make use of math that is useful for quantum mechanics, not making nonsensical claims about quantum mechanics. But (i) things that are really “quantum” aren’t quantum because of the math, but because of the phenomena quantum mechanics observes and describes. I shouldn’t call a problem I apply linear algebra to “quantum” just because linear algebra is math that’s central to quantum physics. (ii) I think it’s simply irresponsible, given the large amount of pseudoscience out there invoking quantum mechanics, to deliberately use the terminology of this article.

        By the way: Is non-commutability of psychological measurements really such a surprise?

        • To expand on Feynman, in his famous “cargo cult” article (that’s you, political scientists–see http://neurotheory.columbia.edu/~ken/cargo_cult.html or google Feynman cargo cult” for the article), Feynman did talk about experimental methods and his interest in psychology. So Andrew is probably correct–Feynman almost certainly tried to do mathematical models of psychology (and probably other social sciences). He seems to have concluded that it wasn’t possible and that one should do very elementary replications to learn anything (an argument I’m more or less in agreement with).

          As far as Feynman’s reputation, it is oversized for his contributions to science–that is, there are roughly ten Noble prized winners in physics in the last century who contributed the same or more as Feynman but can you name them (I put Einstein above this group)? Why then the cult of Feynman? He had a talent for self-promotion but was also witty and an excellent speaker. This isn’t to say he wasn’t brilliant–one story I heard about him was that he was on a dissertation committee and hadn’t, as was typical, really read or attended to the thesis, but at the defense he noticed that something was amiss and stated something to the effect that “I don’t think this is correct because of such and such but I’ll vote to pass this if the rest of you agree”. The committee adjourned without a decision and the adviser and the student worked it out and Feynman was correct (the student submitted a corrected work about a year later).

        • “He had a talent for self-promotion but was also witty and an excellent speaker. ”

          And a good writer for popular audiences and an English speaker.

        • Okay, Feynman was one of the top 11 physicists of the 20th Century, and a delightful self-promoter, and an American. His fame seems pretty self-explanatory.

          The only thing I’d add is that he was a regular guy in his cultural tastes, which helps make him a hero to say the upper 20% of the American public, not just the top 2%. My impression is that for other great physicists, being a regular guy in musical taste meant they prefered to Wagner to Schoenberg. For Feynman, it meant playing the bongos.

    • I had the same impression as Rahul during the buildup, at least until you got to the “couldn’t figure out social science”.

      I’d highly recommend the Feynman bio Genius by James Gleick. I didn’t get the impression Feynman was any better at bio than Michael Jordan was at baseball, which is to say very very good, but not good enough to play in the major leagues. But maybe I’m misremembering.

      [Update: Looks like my memory was correct:

      http://blogs.scientificamerican.com/oscillator/2013/07/27/feynman-on-biology/

      Feynman himself was quoted as saying, “It would have been a fantastic and vital discovery if I had been a good biologist. But I wasn’t a good biologist.”]

      • Even about painting (if memory serves me right) Feynman never claimed he became some Picasso but just someone who could manage to hold local exhibitions & sell some work.

        While that is not an average achievement, it definitely wasn’t unbelievable.

        These things are not a zero sum game. Very often, remarkable people tend to be simultaneously remarkable in many things all at once.

        • Andrew:

          You wrote:

          “But I think it’s pretty clear that a lot of other people didn’t think he [Feynman] was so great.”

          Out of ignorance, who were these guys? For a change I’d love to read reminiscences of contemporaries that didn’t like Feynman.

        • Read Genius. It’s a nice picture of the sociology of how Feynmann’s diagrams beat down the more purely symbolic approaches of his contemporaries. These guys of course all knew each other and saw each other pretty regularly (given the era) at conferences and meetings.

          (And please don’t use this as a launching point for DAGs vs. other ways of writing down models!)

        • Bob,
          Can you explain what difference you see between the way Feynmann’s diagrams won acceptance among physicists
          and the way DAGs are currently gaining acceptance among causal analysts?

        • Potentially more historically minded, on Feynman diagrams, is David Kaiser’s book, “Drawing Theories Apart: The Dispersion of Feynman Diagrams in Postwar Physics”.

          He, and colleagues, fit a handful of infection based models (SI, SIR, etc.) in a follow up paper [1] to capture more generally idea spread. It’s a good data set with a couple cute models.

          [1] The power of a good idea: Quantitative modeling of the spread of ideas from epidemiological models: doi.org/10.1016/j.physa.2005.08.083

        • Andrew Wahlen,
          This is a fascinating study, thanks for posting.
          There is an interesting comparison of the rate of adoption of Feynman diagrams in the USA and in the USSR. Fig. 1 says: “Adoption in the USSR occurred later because of scientific isolation from physicists in the West with the onset of the Cold War, and proceeded more slowly because of institutional resistance.” If one were to compare the rate of adoption of DAGs model, the findings would surely perplex historians of science; it took epidemiologists 5 years, Sociologists 10 years, economists 20 years, and statisticians X years, with X
          waiting to be estimated.

        • This is one example that I know. Perhaps there are more?

          Oral History Transcript — Dr. John F. Clauser
          http://www.aip.org/history/ohilist/25096.html

          “I didn’t particularly care for Feynman. I didn’t really like him very much. And frankly a lot of the stuff that he did and said was wrong, especially in foundations of quantum mechanics.”

        • At least there is a sort of self-depreciating end in the story “But is it art?” in “Surely, you’re joking…”:

          There was a guy there at the county art museum named Maurice Tuchman who really knew what he was talking about when it came to art. He knew that I had had this one-man show at Caltech. He said, “You know, you’re never going to draw again.”
          “What? That’s ridiculous! Why should I never.
          “Because you’ve had a one-man show, and you’re only an amateur.”
          Although I did draw after that, I never worked as hard, with the same energy and intensity, as I did before. I never sold a drawing after that, either. He was a smart fella, and I learned a lot from
          him. I could have learned a lot more, if I weren’t so stubborn!

  2. But I’d say the science reporters are more to blame than anyone else, aren’t they? I mean, it’s hard to picture respectable physicists actually saying they’re “deriving the universal laws of social behavior” or anything like that. On the other hand, by using statistical mechanics techniques to model social systems they might provide some useful insight on how these systems behave.

    It’s true that these techniques are more and more part of the standard statistics toolbox, so you might as well say they’re simply fitting a statistical model. But I think it’s still valid at this point to publish such findings in Physics journals, since these techniques were originally developed in physical contexts. The models are usually evaluated on their own – compared to real data, for instance – and while the Physics connection might be emphasized for publicity, I’d be sad to hear there are physicists who believe that people behave like magnets.

    • I disagree there somewhat. PRL shouldn’t be accepting such papers. Even when using a borrowed technique I think the best place for such work is the discipline where the application lies.

      I can’t help thinking PRL is doing this for cheap publicity. A paper on presidential elections gets far more popular press than physicists making some incremental advance on magnets.

      • One problem with this idea is that most journals in social sciences will reject such papers just because the math is complicated, as most social scientists don’t have the skills necessary to understand it (although they have many other skills that physicists don’t). A possible solution is submitting to interdisciplinary journals, but they are few, and mostly not very respectable.

        I think it’s interesting for other physicists to know that Physics techniques may provide insight into such matters. It should be also interesting for social scientists, it’s true, but in practice it’s not that simple: it takes some convincing, and the physicists’ reputation doesn’t help with it.

        • I wonder, do such things get refereed entirely by physicists too? Or does PRL keep some social scientists on its referee roster?

          Alas, we’ll probably never know the particulars!

      • It makes a lot of sense that PRL published this. In the realm of opinion-dynamics on graphs, PRL and Phys. Rev. E are the two or the primary places where papers get published (RoSoc’s Interface is the other high-impact one).

        A lot of the studies are purely theoretic and simulation based. Their connection to reality tends to be… tenuous.

        I could see that something like this could be quite interesting to a broad swath of their audience who might be thinking about how to fit their models to actual data, regardless of whether or not the paper actually tells us something about reality. Publishing it where people would be interesting in it seems pretty fair to me, both from the perspective of the authors, and also from the perspective of PRL.

        • I’d agree with that if the editors were conscientious enough to mention prominently stuff like that the work’s connection to reality is tenuous.

        • I’m not convinced that burden should be placed on the editors. It seems more fitting to be placed on the authors (or as Andre suggests, the reporters). For many of the vote-model-esque theoretical papers the lack of a direct connection seems pretty obvious. Maybe less so here when they actually try and fit a model to data it’s not quite as obvious.

          And, to be fair, the authors of this PRL paper make no claim that they have discovered the underlying physical rules that govern elections. All they say is that a simple model of opinion diffusion (which happens to be called the Voter Model) can recreate some of the course-grained statistical patterns seen in real world election data. As per their abstract:

          “the model captures… the stationary vote share fluctuations across counties and the long range spatial correlations that decay logarithmically with distance.”

          If I read their Figure 3 (f & g) right, the model predicts get the direction in which the vote share of a county actually changes correct only about as often as not. So it seems to me all the authors are saying is: “Hey, look! Our model *surprisingly* captures some coarse grained statistical patterns of reality.” Which is something different from saying that “reality is captured by our simple model.”

          Having read through the paper now, it feels to me that it falls very neatly in PRL’s (or Phys.Rev E’s) scope of being interested in dynamical systems on complex topologies (generally networks). Perhaps the issue is more the press surrounding it than the paper itself?

      • Is PRL even a respectable physics journal (to physicists)? It seems they have an ongoing problem with overstepping their expertise. Joshua Goodman wanted to rebut (is that a word?) a paper on language modeling they published, but Goodman couldn’t get his counter-arguments published in PRL itself, so he just put them on arXiv. The last sentence of Goodman’s arXiv abstract: “I conclude by suggesting that Physical Review Letters should not publish Letters unrelated to physics.” Andrew even blogged about Goodman vs. PRL, with Andrew taking the “I’m a physicist by training, so I can make fun of physicists” out (hey, if undergrad degrees count, that makes me a mathematician).

        • I’m not a Physicist but I’ve collaborated with some, and my impression is they treat PRL as very respectable indeed. My sample was mostly from the condensed matter physics community.

          PRL is good so long as it sticks to Physics, I think.

  3. One aspect of this tendency hasn’t been mentioned: grants. If you can show that your favorite statistical physics system or analysis technique has real-world applications, this may well improve your chances of getting funded. Even if you don’t believe that your spherical cow is a good model of bovine physiology, creating and selling a plausible story about its potential agricultural utility can be an effective, if unethical, means of arguing the importance of your research to the NIH/NSF.

  4. The big difference between physics and the social sciences are:

    A. Things change in the social sciences. Human beings react and strive. They aren’t planets.

    B. When things don’t change in the social sciences, we don’t want to hear about it. For example, Jason Richwine got fired in 2013 for having discussed in his Harvard treatise the relative stability of Hispanic IQs over the generations and how that key fact is important for grounding the immigration debate empirically rather than just letting pundits free associate about the Statue of Liberty.

  5. the guy who bragged that when he took up biology as a hobby he could do better than the biologists, who bragged that when he took up painting as a hobby he was an awesome painter

    Obviously, if he bragged about it, it must be true!

    the idea that some clever mathematician or physicist can derive universal laws of social behavior

    Not exactly but close enough: Just wait – any minute from now, Steve Hsu and his amazing Chinese sequencers will uncover the molecular basis of intelligence. He says so. For the past several years. It must be true then.

    • If you’re the same DK that comments on my blog, then I have to question your reading comprehension capability.

      I have specific estimates for what sample size will be necessary to get a handle on genomic prediction of cognitive ability. We are nowhere near that sample size at the moment, as I have said many times on the blog.

      http://arxiv.org/abs/1310.2264

      You might be confusing the goal of solving the *whole problem* with the much easier problem of finding a few genome-wide significant hits on some SNPs. That has been accomplished by the SSGAC collaboration recently, and published Science. The statistical power required is similar to what I had anticipated a few years ago.

      http://infoproc.blogspot.com/2013/05/first-gwas-hits-for-cognitive-ability.html

  6. iirc, in the 80s, didn’t the Chaos theory people claim they had a theory of everything ?

    PS: given that Feynman was smart, he also seemed kinda mean – or maybe just lacking in social skills and allowed to get away with it cause he was so smart.
    As the story goes, Danny Hillis, an MIT grad student, got a lot of VC money for a new kind of computer; and somehow roped Feynman in as an advisor.
    So Hillis and the kids are sitting on the floor of the space for their new company – they don’t even have chairs – and there is a knock on the door.
    Seargeant Feynman reporting for duty , sir.
    So the kids don’t know what to do, so they send Feynman out to get office supplies….

    Later on, someone is hanging around, and sees Feynman point a bowl at an MIT grad student in physics (ie, not stupid) who is female, and feynman says food
    she goes an gets it
    so the observer says to the student, wtf ?, you wouldn’t put up with that from anyone ?
    and the student say, he is the only person who has ever been able to explain thermodynamics to me…….

  7. imo, having been around MIT and harvard, is that a lot of profs get there by being highly competitive (and smart, er then me)
    I mean, like highly
    So Einstein is having dinner at someone’s house in princeton, and there is a rubik cube like puzzle on the table
    Albert says to his host, can I borrow this ?
    Albert returns it the next day with a big smile: Oppie (oppenheimer) couldn’t do it…

Leave a Reply to Nick Cancel reply

Your email address will not be published. Required fields are marked *