Towards the end of this year, The American Political Science Review will publish its 100th anniversary issue. In researching for a submission to this centennial issue, I examined what political scientists have been saying for the past 100 years, and in doing do something very odd struck me: that the arguments that I have been having for a decade with my colleagues about the idea of a science of politics being at all possible are the same arguments that have been going on in the pages of The American Political Science Review since its inception.
Then and now, political scientists tend to fall into two camps. In the first camp are those who wear the badge of ‘scientist’ and see their field as a predictive enterprise whose job it is to uncover those general laws of politics that ‘must’ be out there. The second camp contains those who think the former project logically untenable. For years now I have tried (largely in vain) to convince my colleagues in the first camp that the idea of a political ‘science’ is inherently problematic. I have marshaled various arguments to make this case, and each of these has been met by a some variant of; ‘political science is a young science’; ‘what we face are problems of method’; and that ‘more ‘basic research is required’. Then, with ‘more and better methods’ we will make ‘sufficient’ progress and ‘become’ a science. I remain unconvinced by this line of argument, but it was enlightening to see it played out again and again over a century.
Discovering that these same arguments have been going on for 100 years was both heartening (I was in good company) and depressing (‘round and round we go’). But in doing so I discovered something else. If political science is a ‘science’ by virtue of its ability to predict, as many of its ‘scientific’ brethren maintain, then it really should have been abandoned years ago since the prediction rate of my field over the past 100 years is less than what would be achieved by throwing darts at dartboard while wearing a blindfold. To see why this is the case consider the following potted history of political science.
From its inception in 1906 until World War One American political scientists took ‘public administration’ as its object and the Prussian state as the model of good governance. Sampling on this particular datum proved costly to the subfield however when the model (Germany) became the enemy during World War One and the guiding models of the field collapsed. Following this debacle, political science retreated inwards during the 1920s and 1930s. One can scan the American Political Science Review throughout these tumultuous decades for any sustained examination of the great events of the day and come up empty. What I did find however were reports on constitutional change in Estonia, committee reform in Nebraska, and predictions that the German administrative structure will not allow Hitler to become a dictator.
After World War Two this lack of ‘relevance’ haunted the discipline and its post-war re-founders sought to build a predictive science built upon the process notions of functionalism, pluralism, and modernization. These new theories saw societies as homeostatic systems arrayed along a developmental telos with the United States as everyone’s historical end. Paradoxically however, just as the field was united under these common theories, they were suddenly, and completely, invalidated by the facts of the day. At the height of these theories’ popularity, the United States was, contrary to theory, tearing itself apart over civil rights, Vietnam, and sexual politics while ‘developing’ countries were ‘sliding back’ along the ‘developmental telos’ into dictatorships. Despite these events being the world’s first televised falsification of theory, once again political science turned inward and ignored the lesson waiting to be learned – that prediction in the social world is far more difficult than we imagine, and the call for more ‘rigor’ and ‘more and better methods’ will never solve that problem. Our continuing prediction failures continue to bear this out. Since its ‘third re-founding’ in the 1980s till today, political science has predicted the decline of the US (just as it achieved ‘hyper-power’ status); completely missed the decade long economic stagnation of Japan (just as it was supposed to eclipse the US); missed the end of the Cold War, the growth of international terrorism, and the rebirth of religion in politics.
After reviewing this catalog of consistently wrong calls, a very simple question occurred to me. If political science is a science by virtue of its ability to predict, and its prediction rate is so awful, can it be a science even in its own terms? I would say that it cannot. But this answer itself begged another, and I think more interesting, question; why is my field’s ability to predict so bad? The answer to this question is not found in the pages of the American Political Science Review. Rather, it is found in how political science as a discipline, through its training, thinks about probability in the social world. To see why this is the case I ask the reader to follow me through three ‘possible worlds’ that have three different probability distributions, and then decide which world it is that political science studies – and which one it thinks it studies.
Our first (type-one) world is the world of the dice roll where the generator of outcomes is directly observable. Here we live in a world of risk. We know when throwing a die (the generator) that there are six possible outcomes. Given the ability to directly observe the generator and a few dozen throws of the die, the expected and actual means converge rapidly via sampling, and this is sufficient to derive the higher moments of the distribution. This distribution, given the known values of its generator, is reliably ‘normal’ and sampling the past is a good guide to the future. One is not going to throw a ‘300’ – there are only six sides on the die – and skew the distribution. This type one world is reliably Gaussian, and is, within a few standard deviations, predictable. Political science thinks it operates in this world. This is the familiar world of the bell-curve.
Our second world (type-two), is a world with fat tails (Gauss plus Poisson) where uncertainty rather than risk prevails. An example of the generator here would be a stock market. Although one can sample past data exhaustively, one does not observe the generator of reality directly. Consequently, one can ‘throw a 300’ since large events not seen in the sample may skew the results and become known only after the fact. For example, stock market returns may seem normal by sampling, but a ‘Russian Default’ or a ‘Tequila Crisis’ may be just around the corner that will radically alter the distribution in ways that agents cannot calculate before the fact. This is a world of uncertainty as much as it is risk. Agents simply cannot know what may hit them, though they may be think that the probability of being hit is small.
Our third possible world (type-three) is even more unsettling. Imagine a generator such as the global economy. In this case, not only can one not see the generator directly, agents can sample the past till doomsday and actually become steadily more wrong about the future in doing so. As two probabilists, Nassim Taleb and Avatel Pilpel, put it, with such complex generators “it is not that it takes time for the experimental moments…to converge to the ‘true’ [moments]. In this case, these moments simply do not exist. This means…that no amount of observation whatsoever will give us E(Xn) [expected mean], Var(Xn) [expected variance], or higher-level moments that are close to the “true” values…since no true values exist.”
To see what this means, consider the following example. Macroeconomics, like political science, has had at least four general theories of inflation over the past fifty or so years, which suggests two things. First, that these theories cannot be general theories since they change every decade or so. Second, that such theories might be thought of as general (at the time they were constructed given the sample that they were derived from) but such theories must become redundant since the actual sources of inflation change over time.
For example, if the agreed-upon causes of inflation in one period, (monetary expansion) are dealt with by building institutions to cope with such causes (independent central banks), this does not mean that inflation becomes impossible. Rather, it means that the conditions of possibility change such that the theory itself becomes redundant. In such a world outcomes are fundamentally uncertain since the causes of phenomena in one period are not the same causes in a later period. Given this, when we assume that outcomes in the social world conform to a Gaussian distribution we assume way too much. Any sample of past events can confirm the past, but cannot be projected into the future with the confidence we typically assume. Take away that prior assumption of ‘normality’ in the distribution and standard expectations regarding prediction fall apart.
Given this, which world is the world most likely studied by political scientists? Our type-one world can be ruled out since if the world was so predictable our theories should be able to predict accurately. Given the record in this regard, it is safe to conclude that the world we occupy is not this one. Our type-two world seems suspiciously normal most of the time, but our theories ‘blow up’ much more than they should since most of the action occurs in the tails and we cannot see the generator of outcomes. This sounds more like the world where people actually live.
A type-three world is even worse however, since in a type-three world all bets are off as to what the future may bring. Humans do not however deal particularly well with such uncertainty and try to insulate themselves from it. Whether through the promulgation of social norms, the construction of institutions, or the evolution of ideologies, the result is the same. Human agents create the stability that they take for granted. In taking it for granted however they assume the world to be much more stable than it actually is. Consequently, our theories about the world we live in tend to assume much more stability, and thus predictability, than is warranted.
In short, we cannot live in a type-three world, so we build institutions, cultures, and societies to cope with uncertainty. But when we are successful at doing so we assume we live in a type-one world of predictability and develop theories to navigate such a world. Unfortunately, we actually have succeeded only in constructing our type-two world of fat tails, and this is why we are constantly surprised. We think (and model) type-one while living type-two. Meanwhile, as a discipline, we refuse to admit the possibility of a type-three world generating both the others.
The result is that the action is in the tails, and we, given our type-one assumptions and models, are blind to what is going on there. So we focus, like the proverbial drunk under the lamp-post, on the middle of the distribution since that is where the (theoretical) light is; and like the proverbial drunk, we are constantly surprised that our keys are actually to be found somewhere else entirely. Political science may have reached the ripe old age of 100, and I congratulate it for doing so. It did so however by imagining the world to be quite different from what it is, and by completely ignoring its predictive failures. If however political science wants to be around for another 100 years it may want to think a bit more about what those failures are trying to tell us.