Case Studies and Theory Development in the Social Sciences - Alexander L. George [74]
Truly important and significant hypotheses will be found to have “assumptions” that are wildly inaccurate representations of reality … the relevant question to ask about the “assumptions” of a theory is not whether they are descriptively “realistic,” for they never are, but whether they are sufficiently good approximations for the purpose in hand. And this question can be answered only by seeing whether the theory works, which means whether it yields sufficiently accurate predictions.274
Firms operating in a market, for example, behave as if they know the underlying cost and demand functions posited by economic theory, even though they do not go through the actual complex mathematical calculations posited by economic theory.275
Friedman is right in the sense that all theories are simplifications of reality. His argument is also consistent with the D-N model, in that D-N explanations are satisfied by statements of regularity that invoke as if assumptions regardless of whether the posited causal mechanisms are in fact operative. But for this reason, Friedman’s analysis confronts the same “barometer problem” that afflicts the D-N model: he cannot distinguish a good predictive relationship from a good causal explanation. In contrast, researchers seeking to explain phenomena via causal mechanisms must acknowledge that their theories are in trouble if the mechanisms their theories posit are not consistent with the observed processes at a more detailed or micro level of analysis. For example, economists modify their theories when individuals act out of altruism or other social motives that deviate from the assumptions of rational choice theory. Notably, the 2002 Nobel Prize in economics was awarded to Daniel Kahneman for working toward more accurate microlevel mechanisms that identify common cognitive biases that depart from the assumptions of rational decisionmaking.
In contrast to approaches that emphasize causal effects or predictive capacity, which draw on regularity of association and congruity of magnitude as sources of causal inference, explanation via causal mechanisms also draws on spatial contiguity and temporal succession, two additional sources of causal inference discussed by Hume. In particular, explanation via causal mechanisms involves a commitment in principle to making our explanations and models consistent with the most continuous spatial-temporal sequences we can describe at the finest level of detail that we can observe. For example, the barometer cannot be characterized as having “explained” the weather, since we know from our observations at levels of greater detail that processes involving air pressure, temperature, and so on continually interact, accounting for both the barometer readings and the weather.
More generally, in this view an adequate explanation requires also the specification of hypotheses about a causal process that brought about the observed correlation.276 Thus, while covering law explanations of the D-N type bear a superficial resemblance to mechanism-based explanations (in that a covering law explanation can simply be restated in more detailed and contingent terms to mimic a mechanism-based explanation), the two forms are profoundly different. Mechanism-based explanations are committed to realism and to continuousness and contiguity in causal processes.277 While we can posit macrolevel social mechanisms and test them against macrolevel phenomena, macrolevel theories must be consistent with what we know about individual-level behavior. In principle, a mechanism-based approach to explanation even requires that social theories be consistent with what we know about the chemical, electrical,