description without causation, causation without explanation

Over at SocArXiv, two University of Michigan political scientists just posted a wonderful, short comment on my stylized facts paper. In the original paper, I argue that stylized facts are empirical regularities in search of explanation, that the production of stylized facts should be understood as an important component of social scientific practice, and that stylized facts are capable of doing political work even in the absence of well-established causal explanations. In their comment, Crabtree and Fariss (C&F) offer a nice clarification in the context of experimental social scientific research programs.

When I wrote my paper, I had in mind the largely about descriptive statistics produced by sociologists and economists, often from survey or administrative data, which then feed the theoretical enterprise of modeling and the hunt for mechanisms. C&F rightly note that social scientists are also experimentalists, and that these experiments produce a particular kind of knowledge that is similar to, but distinct from, the sorts of stylized facts I had in mind. They write: “`Robust stylized facts,’… are empirical regularities in search of theoretical explanations. These facts make causal claims, are produced by experimentation, and are supported by internally valid findings.” (4)

I like this distinction, and want to offer a shorthand for it. Stylized facts offer description without causation. Robust stylized facts offer causation without explanation. That is, the kinds of findings explored by C&F under the heading of robust stylized facts claim to have strong (“robust”) predictive validity. Like the phenomena of the natural sciences discussed by Ian Hacking (one of the motivating sources for my paper), robust stylized facts can be routinely produced through manipulations – but we don’t yet know why.

That said, I’m left with a terminological quibble. If robust stylized facts are internally valid, repeatable products of intervention, why not just call them “phenomena”? What do we gain by expanding the concept of stylized fact to include them? Perhaps this expansion better maps onto contemporary discourse – I should go read the experimental political science literature more closely to see what they talk about when they talk about stylized facts. But, for me, the primary importance of writing a paper about stylized facts as a narrowly descriptive phenomena was to help make sense of the role of description in social science, and the way that social sciences grappled with their mass of non-experimental findings as parts of research programs.

Put differently: I think “robust stylized fact” makes sense on its own as a concept, and perhaps it will serve as a useful tool for helping scholars distinguish between things they are calling stylized facts that derive from repeatable interventions (“robust stylized facts”) and those that derive from aggregate observations (“stylized facts”). But I worry that treating this as an expansion to the concept, rather than a close but disjoint cousin, might detract from the analytical utility of the concept of stylized facts.

Author: Dan Hirschman

I am a sociologist interested in the use of numbers in organizations, markets, and policy. For more info, see here.

5 thoughts on “description without causation, causation without explanation”

  1. I enjoyed reading the article. The use of the specific concept of the “stylized fact” (so that it would turn up in a keyword search) seems to be one way of talking about an emphasis on generating well documented empirical regularities. That is, careful descriptive research as an essential precursor to theorizing. In my work, I’d say an example is that prior to the mid-2000s, nobody had taken seriously the empirical pattern (stylized fact?) that Black incarceration rates are higher in liberal Northern states than in the states of the Old Confederacy. But perhaps this empirical regularity (which was there all along) did not become a stylized fact until people started talking about it? Is that the idea? Once you spot it, you can theorize it and it isn’t actually difficult to explain it theoretically, once you pay attention to the fact itself.

    I’ve become more and more convinced that descriptive research is undervalued in sociology and that we need much more solid empirical research to undergird theorizing. Do you see your paper as a call for more descriptive research? Or do you see it as a call for writing more synthetic descriptive papers that identify empirical regularities across descriptive projects and give them names? Or is the idea of stylized fact entirely about discourse and how people talk about empirical regularities?

    Regarding the point in the blog, i.e. distinguishing between bivariate (or even multivariate) correlations that hold up to careful empirical scrutiny and relationships that are supported by experimental evidence that can support cause-effect assertions, I’m not sure the distinction is as important as you or your commenters think. A carefully controlled randomized experiment can isolate the independent variable as the only thing that can account for variations in the dependent variable, which is helpful, but the experiment does not, as a rule, explain WHY that happens. A simple-minded example is a careful experiment that demonstrates that flipping a light switch turns the light on or off: this does not explain how electricity work or reveal the wiring that is necessary to make the phenomenon work. Less simple-minded examples are experiments that can readily reveal implicit bias or hidden prejudices in evaluations; these are quite easy to design and quite regularly reveal prejudice. But they do not tell you why or how these prejudices come into being (what the neural processes are) nor do they explain variations between people or within people across time or place in the extent of these implicit biases.

    In the language of stylized facts, the stylized fact is that prejudiced judgments can be regularly found across a wide range of empirical measures of them. This stylized fact (supported by experiments) seems to me to be roughly comparable to stylized facts generated from repeated correlational studies.

    Like

    1. Hi OW! Thanks for the comment.

      In re: “I’ve become more and more convinced that descriptive research is undervalued in sociology and that we need much more solid empirical research to undergird theorizing. Do you see your paper as a call for more descriptive research?” Yes! Specifically, I think we should revalue description, and do a better job of understand its place in both the academic research process and in the political circulation of social scientific knowledge. Descriptions are both important and powerful, and we need to have better ways of talking about different kinds of description, different levels of description, different tools for describing, for aggregating descriptions, etc. And then we need to make sense of how those descriptions can circulate as intended – or not.

      In re: “A carefully controlled randomized experiment can isolate the independent variable as the only thing that can account for variations in the dependent variable, which is helpful, but the experiment does not, as a rule, explain WHY that happens.” This is what I think I was trying to get at with the distinction between causation and explanation, which I’m still trying to fully work out for myself. The kind of causal knowledge produced by experiments is a very thin sort (which does not make it useless by any stretch!). And there is certainly more fluidity between description and explanation than is convenient for trying to pin down and carefully define a set of distinct terms. One of the dynamics I want to capture in this whole discussion is how we aggregate up (individual data points to a stylized fact, say) and dig down (from a bivariate trend to a multivariate decomposition to some kind of unpacking of causal mechanisms, etc.). Some times, “description” vs. “causation” or “explanation” might be more directions on a continuum than stable, discrete points (Abend makes a similar argument about facts and theories, which I believe he gets from Alexander).

      Like

  2. Hey Dan: looking forward to hearing more of your thoughts on description. This reminds me: the paper I gave at the Vancouver SSHA on historical sociology and description (and at which you and I discussed whether Foucault was causal or descriptive) is published. Would love to hear your thoughts if you get around to reading it: https://www.academia.edu/2279881/Occluding_the_Global_Analytic_Bifurcation_Causal_Scientism_and_Alternatives_in_Historical_Sociology_

    Like

Leave a Reply

Please log in using one of these methods to post your comment:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s