I was not aware of the paper/polemic on the state of macroeconomics by Paul Romer or the controversy about “science” thereof that has been going on for last year plus, other than vaguely that there was such a controversy without knowing the details. To be completely honest, the battle lines are not necessarily new–most of it has been known before. The particulars of the debate, the centrality of the identification problems and that it is often difficult, if at all possible, to evaluate the moving parts of the theory with the live data, are hardly new either. The significant development, I suppose, is that the parties involved have finally decided that there is something seriously rotten where the parties involved in the debate are simply trolling each other without changing their mind and that this is no science, but “politics.”
Romer rightly points out that the problem is that the language and tone of the debate changes from that of science to that of “politics.” The goal of politics is to score points and to win, to advance your agenda and to shoot down the other side’s agenda. This is the trouble with what I’d been calling “wonkism” and “data science mentality.” (It is amusing to note that Romer is startled to discover that a lot of “data science” runs on the kind of models that caused such challenges in macroeconomics, or, for that matter, for heliocentric models of the solar system. I happen to think he is right to be startled and disturbed: data science is NOT science. It is the opposite of science, in that it operates on recognizing patterns in large scale data without theorizing or rigorous hypothesis testing–exactly like “calibration” in macroeconomics. As such, it is liable to fall into the same sort of cargo cultish traps that Feynman talked out.)
The trouble with political science echoes that of macroeconomics. In the end, what prompted macroeconomists to take the stances they did was that their decisions did not exist in a universe sealed off from policymaking. They knew that ideas shaped policy: what prompted Robert Solow to react as he did, as Romer sees it, was the implication on monetary policy from the argument being put forward by Lucas and Sargent. Lucas and Sargent certainly couched their argument in terms of potential policy implications. And the surest way to draw attention from a large audience is to claim that a new, half-baked theory has serious implications on the “real life,” not the abstracted, theoretical life of academics. In the end, political scientists have even worse problem than macroeconomics. Macroeconomics and politics are simply difficult to disentangle so that it is just difficult to think about the former without the latter. Political science IS about politics. It is IMPOSSIBLE to think about political science without linking it to politics. And most people, political scientists included, are too interested in politics to regard their studies as a purely theoretical exercise independent of real life. Even if they might be, and they might try to maintain it as such, their audiences, university administrators, and their colleagues who do like politics insist that you do. Personally, I found it difficult to maintain what Romer calls “Feynman integrity” if the topic of discussion in classroom comes too close to home: the “Arab spring” made it difficult to talk “seriously” without prejudice about elections and democracy in comparative context and every election year makes it difficult to talk about elections in U.S. My personal recourse had always been to pretend the election is not taking place and draw examples from far enough in the past that students have no preconception about what I’m talking about. This is one of the factors that led me out of academia–the powers-that-be did not like it. In fact, they were fairly insistent that I bring in current politics, with all the associated with problems–namely, “politicizing” the classroom and research.
Why should anyone outside academia care about purely theoretical endeavors anyways? Macroeconomists get the reputation that they do with the mass public because they are public intellectuals. Nobody not in economics business know the technical lingo. They do have political agendas, however, and there is a value to having famous academics on their side as props. Academics as policy advisors are there because they have only one hand–the hand that supports the agenda of the political actors whom they happen to be allied with. Truman had no use for multiarmed economists, but scientific integrity requires multiple arms. Political science has the same problem: even in 2016, they are brought in to say something about whether Trump’s voters are racists, a bunch of dumb hicks, or are genuinely driven by economic agenda. Data is ambivalent enough (and possibly, all of them are true enough) that evidence of support can be found for all of them. But in the end, it is not the academics who are saying these, but various political actors. Academics are called in simply to add flavor to arguments not of their making, just like macroeconomic theories of various schools are invoked and the credentials of their advocates recited to feed political arguments behind various policies. And this is what gets academics invited to fancy parties, important meetings, and TV shows. Nobody wants to be the theorist who discovers something that could lead to cure for cancer–because nobody will know about them, at least in the present day environment. Everyone wants to be the clinician who does cure cancer and bask in the accolades, even if they know nothing about actual practical medicine.
In some sense, it is remarkable that theoretical physicists got the fame that they did: Albert Einstein did not invent the atomic bomb. He merely made theoretical contributions that made it possible. How did he get to be famous, not, say, Edward Teller? In this sense, Lee Smolin’s issue with theoretical physics is exactly the opposite that of macroeconomics or political science. There is no “politics” there. People are not interested in the string theory because they need props to support their political agendas. They simply think string theorists are extremely smart people who have fascinating things to say about the universe that are beyond comprehension. And they are exactly right–even if it may not be really”science,” consisting of fragile theories resting on empirical bases. This is an example of a science that enjoys such credibility that it can literally spin half-baked fantastic yarns that may or may not be true–but sound cool and amazing–and still be trusted. But, for the most part, quantum mechanics, relativity, and the nature of universe have no obvious implications on policy or everyday lives of normal people. They are simply a wonderful intellectual diversion, like some sort of new age religion (and enough people are making money off of exploiting that, including, apparently, some people with legitimate physics degrees). Economics and political science, social sciences in general, indeed, do not enjoy that kind of luxury: they deal with subject matter too close to home, if you will, for policymakers and normal people like. They are welcome as long as they provide support for the policy and other political agenda. They are useless eggheads if they do not. They need to become wonks, not scientists, in order to be relevant at all.
I used to be much more naive and optimistic about the prospect of a “social science,” where society can be studied in a manner that is actually “scientific,” completely detached from agenda, values, and anything normative and reliant only on logic, data, and, most importantly, finding where the data contradicts the logic and forces new thinking. I am increasingly convinced that this is not possible. Real world needs social science as trophies and props for politics, not for understanding, for this role, it offers too enticing a set of rewards. Feynman integrity is not an easy commodity to come by, especially when people are interested in the outcomes too much.
PS. Comment on another blog offers this definition of Feynman’s caution about “cargo cult science:
“Try hard to understand, what you are doing. Do not rely on formalities.”
I think this is fundamentally right: you “believe” in a cult. You trust that the ceremonies, the people, the formalities, or the institutions somehow “work” even if you don’t know how. You set up a model with 100 different moving parts that literally cannot be wrong because 50-60 of the moving parts will be doing something right at any one time. So Google Translate is always right, and it is designed and tweaked with, to be “always” right. After all, millions of people depend on it to translate their documents. Asking how and where Google Translate gets things wrong is likely to be met with the response, “don’t worry, the Google Team will make sure that it will be right even in those contexts too.” I suspect that they will, some day, but that’s not what the question is asking. The question is asking really about how Google Translate approaches language may not fit how people write poetry or love letters. Maybe computers can be made smart enough that they can recognize patterns in poetry or love letters even better than humans…or not. But we want to know about the patterns and the logic behind it, perhaps even more than translate Evgenii Onegin. Who cares if the computer can recognize the patterns?
PPS. This is a great article on Romer and his argument. The problem with human foresight goes back to the fundamental problem shaping all of social sciences, though: the problem of omniscience goes back to the logical problems posed by Divine omniscience, omnipotence, and beneficence. Science fiction writers took on this too–this is the central problem in both Foundations series by Asimov and Dune series by Herbert. The answer, offered by Nietzche and Kierkegaard, and in more formal terms by people like Leonard Savage, is that truth is subjective and the full truth unknowable (and this, of course, is captured by the parable of the blind men and the elephant, which, again, is many centuries old.) I think the clinician-lab scientist analogue is good: clinicians know what works and what doesn’t, even if the theoretical underpinnings may not be fully worked out; they want to know what works better or why what doesn’t work doesn’t work–fairly limited questions, not grand questions about the universe. Focused theoretical studies can help address these problems. Still, this is no panacea when the whole edifice might be wrong, in which case an intellectual revolution is needed that can take centuries to clear out–and that’s when theorists are treating each other decorously and not making political fights out of them: Thomas Kuhn taught us this, about physics. I wonder if the reason Einstein gets reverence he does for purely theoretical work is that physics underwent the great intellectual tumult in the form of the Copernican revolution that went on for centuries that taught them both humility and patience, that physics, at least as practiced by humans, cannot grasp much of the universe at all, and that such advances may come at a very slow pace. As Amartya Sen said, people are more complex than the physical universe (paraphrased). Yet, we in social sciences arrogate to explain far more of humanity than physics does with the universe with far simpler models and propose to use them as basis for serious policymaking. So who is going to be the social science’s Copernicus? Where’s (the old institutions of) Catholic Church when you need them?
Of course, that presupposes that we have come to the crisis of epicycles, or calibration, or data mining or whatever, already. The critics of the Romer piece are not wrong to say that their approach does do some pretty impressive things, or, in other words, Google Translate does do wonderful job for translating most standard workaday prose between common languages. Romer, in a sense, is making fuss over how Google Translate has issues with Pushkin, so to speak. Google Translate Team, perhaps like Sargent, might retort, “is Pushkin important”? That’s a good question, actually. For the aficionados of Russian literature, yes, but there aren’t that many aficionados of Russian literature, at least among users of Google Translate who’d actually use the service to translate things from Russian. And one might say, in order to truly appreciate poetry, you need to learn the language anyways, precisely because poetry does not translate so well.
I don’t think this is necessarily “wrong” on its face, but the line of thinking sounds eerily like the principles of newspeak from Orwell: how to reduce the language to its “functional” parts, with the definition of “functional” based on too crude a utilitarian criterion, with the consequence that the thought is reduced to similarly crude utilitarian conceptions. This sounds like groupthink to me, exactly the kind Romer describes: when I was in grad macro class, for example, I was bothered, amused, and distressed that what they called “microfoundations” either looked nothing like microeconomics I recognized or were predicated on too narrow and restrictive set of assumptions that were driving their results. The macro people were right to point out that they were just toy models, for representing certain ideas and their implications. But I knew that people were unwilling or even unable to break out of these thought patterns even in context of describing and understanding the real world as they became too accustomed to seeing the world in terms of their toy models. Of course, this is also the problem that shows up in spatial models in conjunction with DW-Nominate scores (and related techniques). Before there were DW-Nominate scores, people were usually willing to chuckle and say that spatial models are a bit wobbly. Now that there are “hard” numbers (never mind where they came from, and the fundamental identification problems they pose–the numbers come from how legislators voted, but who knows if they voted for ideology, for constituents, for deals, or what?), people are not only taking spatial models as gospel truth, they are not even willing to think outside the unidimensional models–because 95% or whatever of the votes nowadays are “explainable” by the model, so yeah, Pushkin doesn’t matter in Russian language, while technical manuals and business documents do–because they are reliably Google Translable. Newspeak breeding newthink indeed.