Case Studies and Causal Inference
| Nicolai Foss |
Can case studies — in the extreme: a study of a single case — play any systematic role in causal inference? If so, how? These are the questions posed in a paper by brilliant LSE mathematical sociologist, Peter Abell, forthcoming in the European Sociological Review. The paper is essentially a summary of Abell’s work over more than two decades with building stronger foundations for “qualitative” or “case study” research (a more comprehensive statement can be found in the “A Case for Cases” paper on Abell’s site).
Of course, in the standard statistical interpretation of causal inference, N should be large, and certainly not equal to 1. And most social scientists believe there is no explanation without generalization (an issue discussed at length by Popper, Dray, Collingwood, and others in the philosophy of history as well as by more recent social scientists such as Ragin and Goldthorpe — and James March (here)), so causal inference is predicated on generalization and comparative method.
Many economists seem to think that economics offers a universal action theory which represents generalizations that can be applied in particular, but Abell dismisses such an argument: “it is difficult to give credence to the view that in the analysis of rarely repeatable case studies, we can rely upon the good services of a universal theory of action, which for all described [causes], surrendes the appropriate conditional action for all actors” (p. 5).
Abell argues that there is a way to make causal inferences without necessarily resorting to generalization and comparative method, namely what he calls “narratives,” that is, chronological statements detailing actors, actions, conditons, states, and consequences (mathematically, he conceptualizes these as graphs). Abell adds a Bayesian element by arguing that participants’ self-reports, reports by witnesses, assessments by expert judges, etc., represent items of evidence that probabilistically are relevant for assessing hypotheses of the presence (or absence) of causal links. Causal explanation can then form the basis of comparison and generalization (rather than the other way around, as in the standard understanding of causal inference).
The basic idea seems to tie in closely with established case study practice and implicit understandings of what is going in “qualitative” research. However, Abell shows that it is entirely possible to be formal and rigorous, in the sense of applying fairly standard (Bayesian) statistics as well as graph theory, to particular cases, giving “qualitative” method a foundation which is has hitherto been lacking.