Think first about the following question: Why do we take seriously the claims to approximate truth of many natural scientific theories? We do so, in my view, mostly when such theories make predictions about observable phenomena that exhibit accuracy, precision, reliability, etc. These qualities do not exhaust the virtues of good natural scientific theories, but when a theory enjoys them to a great extent, its verisimilitude is hard to deny. The striking fact of a theory’s extraordinary predictive success cries out for explanation. The only non-miraculous explanation for this success, really, is that it is approximately true. A theory may have other features to recommend it, even if its predictions come up short, but the less compelling a theory’s predictive track record, the easier it is to explain its good qualities without invoking its approximate truth.
What, then, should be our view of social scientific models, the overwhelming majority of which fail to meet ordinary natural scientific standards of predictive success? If a model is internally consistent, then it correctly describes some possible world. One view of good social scientific models is that, even though they do not correctly describe our world (perhaps due to intrinsic complexities in our world), they correctly describe very similar (but much simpler) worlds. The natural sciences take this approach from time to time. Consider the ideal gas law—no one seriously believes that any gas in our world is ideal in the relevant sense, but most chemists believe that the world of ideal gases is very similar to our world. Social scientists may defend a model, knowing that it is not strictly correct in every particular, by insisting that the world of the model is so similar to reality that the model is still of great scientific value.
One problem with this defense is that the ideal gas law, like other successful natural scientific theories, makes very, very good predictions. Such successful predictions are very few and far between in the social sciences. If social scientists want to persuade others of the similarities between the worlds of their models and the real world, predictive success has to be the centerpiece of their argument.
A second, perhaps more serious problem with this defense is that social scientific models frequently incorporate ideas that we not only know to be false, strictly speaking, but also know to be not even approximately true. In rational choice models of human decision-making, for example, it is supposed that humans have complete, transitive preferences over their options. We know this is not true—indeed, we know this is not even approximately true. Yet it continues to maintain a remarkable presence in many prominent social scientific models. Examples of this kind severely undermine the case outlined above for garden variety models in the social sciences.
If social scientific models do not qualify as successful scientific descriptions of our world—indeed, not even approximately true abstractions—then what ought we to make of them? The key to understanding social scientific models, in my view, is to recognize that many phenomena of interest need not occur only in our world, or only in very similar worlds. It is conceivable, for example, that recessions much like the ones that take place from time to time in our world take place from time to time in other, very different worlds. Some of these worlds may be much simpler than the real world, in that they obey only a handful of basic laws. Studying the properties and behavior of recessions in these worlds is more feasible than doing so in the real world, in which they co-occur with countless other complicating factors, even if these worlds have little else in common with the real world.
The challenge with investigating recessions in very different worlds is that features of recessions in those worlds may not carry over to the real world. These features may be intricately linked with the world in which they’re embedded, a world very different from our own. This is why robustness is important in such investigations. If we study a range of worlds that differ from each other along many of the same dimensions along which they each differ from the real world, then we may conclude that robust features of the phenomena being studied can conceivably carry over to our world, even if every world we study happens to be very much unlike our world.
As social scientists produce many such characterizations of phenomena of interest, each of which has properties and behaviors that do not necessarily depend upon the idiosyncrasies of the worlds in which they’re being studied, these characterizations may be compared with respect to their likeness to real world phenomena. The closer the match, the more compelling the characterization—though short of meeting ordinary scientific standards of predictive success, these characterizations need not enjoy the epistemic credentials of established physics, chemistry, etc.
On this understanding of social scientific models, it makes sense to separate two distinct kinds of debate. The first concerns the extent to which the account of phenomena embedded in the model corresponds to reality. The second concerns the extent to which the world of the model is a suitable environment in which to study the phenomena of interest. Note that this does not include debating the extent to which the world of the model corresponds to reality. The greater the extent to which this is the case, the better (of course), but the cost of this is typically greater analytical complexity. In light of this, it seems appropriate for some researchers to err on one side of this tradeoff, while the rest err on the other side.
This may not be the only way to do good social scientific research. This may not even be the best way to do social scientific research. It is, however, one way—one that is practiced much more often than other ways in some social sciences (e.g., economics), while playing a more minor role in others (e.g., cognitive psychology). In the end, I’m a pluralist. No one way of conducting social scientific inquiry has proven its worth to such a degree that other ways must be swept into the dustbin of history. The ‘model phenomena realistically, albeit in unrealistic settings’ method, if you will, has its limitations, to be sure, for which it ought to be criticized, if only to keep researchers’ eye on the prize, but it has something intellectually valuable to offer, too. Proponents of this approach do not consistently practice what they preach—this, too, ought to be extensively criticized—but when they do, they frequently yield novel perspectives on challenging problems, insights into which are always welcome, however they may be found.