At one time I mistakenly thought it would be a good time to produce a second edition of my 1989 book on the methodology of economic model building. As I will explain, I changed my mind after I conducted a simple survey of my colleagues and learned that most economists today see model building very differently than I did twenty-five years ago. It became apparent to me that the planned second edition of my 1989 book would be a big mistake. So, instead, I decided to write a different book – one more appropriate for today’s economists and students of economics. It might also serve as a lesson for methodologists.

The 1989 book (The Methodology of Economic Model Building: Methodology after Samuelson) was directed at methodologists and economic model builders who think they know something about methodology. Actually, the only thing the latter seemed to know at that time was that for one to be taken seriously, one’s model must be testable. So in the 1989 book – using some examples of very simple Keynesian models – I set about demonstrating that for all practical purposes an empirical test would require far too many more observations than are possible. For example, for any non-stochastic model that includes a Cobb-Douglas production function, a non-stochastic test using a logical conjunction of exact observations of the variables involved might require a quarter-million observations! That is, the conjunction of observation reports would form a compound statement that constitutes just one possible counter-example – one that would logically refute that model. Of course, a stochastic model would take even more observations.