19
In the case of prescriptive models, probabilistic estimates of the effects of the prescribed
behavior in the past are impossible when the situation being modeled has not happened. An
obvious example of this sort of situation is nuclear deterrence. There has never been a case of a
nuclear war, therefore quantitative tests of nuclear deterrence theory are impossible. We can test
non-nuclear deterrence, but then the link between the test and the model would remain unclear.
We can create experimental tests, but these are a very different thing – they predict behavior in
controlled circumstances, not in real-world conditions (and experimental tests are themselves
limited in the extent to which we can recreate the incentives and disincentives of nuclear war).
45
In the case of descriptive heuristic models, probabilistic information does not necessarily
help us to determine whether the model describes political processes. Recall that the logic of the
predictive use of rational choice theory is that it does not matter whether or not people behave
rationally, as long as we can predict outcomes modeling their behavior “as if” they do. In other
words, the predictive use of formal models does not claim to describe political processes, only to
predict outcomes. If we are interested in process tracing, formal models used as heuristics can
help to identify where and how rational action did (and did not) play a role whether or not it is
shown to be a statistically significant predictor of outcomes. In other words, such models can
help us describe specific cases whether or not they describe the general case.
Furthermore, sometimes in using a descriptive heuristic model we are less interested in
the question of how often something happens, and more interested in whether or not it can
happen.
46
In this instance, a small number of processes-tracing case studies that illustrate the
model can be of more use than a large-n quantitative analysis. The case studies tell us nothing
about the predictive power of the model in general, because the cases have been chosen on the
45
For a review of this literature, see Jervis 1989.
46
This point is made by McKeown 1999, 172-3.
Convention