In general, I shy away from using the term “social science,” because I do not think that economists can aspire to the same level of falsifiability as physicists. I believe that the difference between social science and natural science boils down to this:
In natural science, there are relatively many falsifiable propositions and relatively few attractive interpretive frameworks. In the social sciences, there are relatively many attractive interpretive frameworks and relatively few falsifiable propositions.The reason that there are relatively few falsifiable propositions in the context of social phenomena is that there are many causal factors, and decisive experiments are rarely possible. Social phenomena are characterized by high causal density, to borrow a term from James Manzi.
As a result, economics is closer to history than to physics. If a historian wants to examine the causes of the decline of Rome, or the decline of empires in general, he or she will provide an interpretive framework. That framework cannot be falsified, but readers can compare it to other frameworks and make judgments about its plausibility.
. . .Economists who employ models think of themselves as “doing science,” meaning that they are generating falsifiable propositions. However, in practice, they rarely reject their preferred models. Instead, they explain away anomalous observations. In that sense, they are really using their preferred models as interpretive frameworks.
I recommend reading the whole post as he throws in a couple of examples.
Of course, Hayek's famous essay --- 'The Theory of Complex Phenomena' is worth mentioning, deals with exactly the same topic. Excellent insights.
ReplyDeletehttp://www.libertarianismo.org/livros/fahtcp.pdf
Thanks for sharing this, Mahesh!
DeleteThis comment has been removed by the author.
ReplyDeleteYou make some good points, Dallas, but I think you're making his claim stronger than it is.
DeleteFor instance, he says "I do not think that economists can aspire to the same level of falsifiability as physicists." So he's talking specifically about physicists, and presumably about those who can do controlled experiments.
He goes on
"In natural science, there are relatively many falsifiable propositions and relatively few attractive interpretive frameworks. In the social sciences, there are relatively many attractive interpretive frameworks and relatively few falsifiable propositions."
So here he's making the point that it's a relative thing. Astrophysics might be confined to "interpretive frameworks" rather than falsifiable propositions, but taken as a whole, the natural sciences can rely more on the latter than the former. We can still be ignorant of modes of action in pharmacology, but we can formulate and test falsifiable hypotheses in pharmacology to a much greater degree than we can in economics.
I think to clarify this, we'd have to say something like the following:
Economics is fundamentally empirical, but our theories start with more general observations about reality (e.g. people prefer more to less) and deduce more specific results (e.g. beef and pork are substitutes). We can use introspection and deduction to come to conclusion about causal links in economics (which is something we can't do in astrophysics), but "testing" them in an inductive framework is always more problematic in economics because we can't control for all (or even most) of the relevant factors. The astrophysicist's data is better than the economist's because planets don't have wills and don't make decisions.
In (many of) the natural sciences, we can apply a treatment to A and leave B untreated to see what happens. If something happens to A and not to B,we can attribute that "something" to the treatment we applied. We just can't do the same thing in economics.
This isn't to say that we shouldn't look at the outside world or apply statistical tools to social phenomena, but we should be more careful about the conclusions we draw. We should be mindful that the Duhem-Quine thesis weighs more heavily on the study of our subject matter than it does in the natural sciences, even astrophysics.
I don't think that makes economics "not a science" but it makes it less of what most people think a science is.
Sorry for rambling. I look forward to reading your thoughts.
Thanks for clarifying! I think you are right that I was probably thinking Arnold was making a stronger claim than he actually was. I think you are right that he is saying that it is a relative thing. I could probably even agree with the notion that we could arrange different fields along a continuum of falsifiability. But I still don't buy that economics would sit closer to history on this continuum than physics or many other "natural" sciences.
DeleteHe wants to say that decisive experiments are rarely possible in economics and that's why we are less falsifiable. Okay. But how often are they possible in cosmology? fire ecology? climatology? meterology?
I also think Arnold puts too much weight on simple falsification. If you take the Duhem-Quine thesis seriously, even the most experimental sciences don't truly produce falsifiable propositions. I don't think this means we can never discover objective truths about the universe. But I do think it means that discovering those truths is more complicated than the simplified version of the scientific method we're taught in high school.
As an aside...
"In (many of) the natural sciences, we can apply a treatment to A and leave B untreated to see what happens. If something happens to A and not to B,we can attribute that "something" to the treatment we applied. We just can't do the same thing in economics."
I think Josh Angrist would disagree.
So correct me if I'm wrong but it seems like your'e saying something like this:
DeleteBoth social and natural scientists use observational data in their work, so either both are "scientists" or neither are.
If that's a good approximation of what you're saying, then I suppose my response would be that there's something crucially different about the data we use in the social sciences that makes it necessary for us to lean on our interpretive framework more than on falsifiable hypotheses.
Data in the social sciences will always be less complete in terms of encompassing all the relevant information needed to generate falsifiable predictions than data in the natural sciences.
Additionally, we can know more about the causal mechanisms in social science due to our ability to use introspection and deduction. To learn about causal chains in the natural sciences we have to find data (which is generally better than social science data) and use induction to understand what's going on.
This implies that what looks like hypothesis testing in social science is really more like illustrating what we determine via introspective and/or deductive theory. Getting your interpretive framework right is crucial in social science because the data you have is relatively bad.
Anyway, I'm rambling again, but I think there's some nuance to Kling's overall point that he left out. I think once that missing part is stated properly, it makes his conclusion stronger.
But that's just me.
I definately don't agree with the way he contrasts the natural and social sciences. It sounds like he thinks of all of the "natural sciences" like a high school physics lab, where experiments can be set up to study a limited number of causal factors. But this isn't really true of lot of fields that we consider "natural" science.
ReplyDeleteWhat about the study of drugs (pharmacology)? You'd be surprised how little we actually know about how most drugs work and that's because the human body is so complex.This is true of even very common drugs. For example, check out the physicians note that comes with Tylenol. It clearly states that even though the drug has been proven effective "the site and mode of action have not been elucidated." There are theories obviously, but it is hard to distinguish between them. Are pharmacologists not doing science?
Or what about astrophysics? There is very limited scope for experiments in this field for obvious reasons. So these guys have to typically rely on observational data to make inferences about the complex universe that surrounds them. That sounds a lot like economics to me. So i would say that if economists are not scientists, then neither is Stephen Hawking. :P
I replied to your other comment. Sorry!
DeleteNo worries! it was essentially the same comment.
DeleteDallas, interesting thoughts related to pharmacology. I think that's obviously why Imbens and Rubin address biomedical sciences in their new causal inference book. So much of what I have learned in apied econometrics I've learned actual reading Peter Astin and others in medical stats journals. To Levi's point, I agree in economics (and other fields as you mention) you can't get these perfect experimental comparisons between A and B, but high causal density impacts lots of fields. Maybe the hard sciences aren't as hard as most would like to think. The public in general I think believes anything published in a scientific journal was carried out with the precision you are talking about in a physics lab. I simply cringe whenever I hear someone tout a 'study' that finds this or that. The devil is definitely in the details. I guess I agree with you both, in that falsification is empirically challenging for economics, but also for many other fields. And I think the danger is that the public and media and politicians and voters unfortunately just don't get the credibility revolution and the implications of doing 'science' with observational data in the context of high causal density. Economists for some reason get a bad rap in this regard more often than other researchers.
ReplyDeleteI don't disagree that "falsification is empirically challenging ... for many other fields" but I think there's something unique about human beings that makes social sciences qualitatively different from any of the physical sciences. I suppose I'm subject to a reductionist critique, but I think reductionists are their own worst enemies.
Delete