In this Epilogue, I will be talking about methodology but also about ideology. By methodology I am not talking about the kind of issues philosophers consider methodology to involve, such as how to (or the need to) justify one’s knowledge claims. Instead, I always mean the low level questions about why economic model builders assume what they assume. As I noted in Chapter 9, my PhD thesis was about the most popular methodological concern among economic model builders of the 1960s – specifically, the view that, to be taken seriously, mathematical models must be testable. A lot of people then thought that this concern was due to the writings of the philosopher of science Karl Popper. This was a common mistake. The concern was not due to Popper; instead, it was due to the work of the economic theorist Paul Samuelson. And again, Samuelson’s concern for testability was a response to the late 1930s critics of the use of mathematics in the development of economic theory. As I have also noted, with his famous 1941 PhD thesis [1947/65], The Foundations of Economic Analysis, Samuelson set about demonstrating how one can construct economic models that are testable. As he stated in this thesis, he would always have the methodological question of testability in mind when building his economic models.

Unlike Samuelson, I, of course, did not receive a Nobel Prize for my PhD thesis, but mine, nevertheless, was also concerned with testability of economic models. At the time, I naively thought economic model builders were not just talking about the logical problem of testability, but instead I thought they were actually talking about the process of collecting sufficient empirical data to perform a test of a model. However, as I explained in Chapter 9, even with simple Keynesian models, the number of observations necessary to construct just one single observation-based test of such a model would far exceed what is realistically possible in the real world. Interestingly, as I noted in the Preface, one particular example stood out: I showed that any model that assumes its production function is Cobb-Douglas in form can require a quarter million observations!