After explaining why I thought the two chief characteristics of "real" science are minimization of Type 1 error and pursuit of causal mechanisms, I point out
Economists have been seduced by a different vision of science, with its roots in those same ancient geometric proofs: that the foundation of science rests on a deductive theory, where “explanation” means “producing a story that can itself be expressed as a deduction from top-level theory”. This is a dream of mathematics and much of physics. It is not characteristic of scientific work in general, however, and its usefulness is limited in fields of research that are extremely complex and dependent on a vast array of contingent factors. Geology would be such a case, where certain general concepts (like plate tectonics) are broadly accepted, but what matters in practice is the knowledge of concrete mechanisms, such as what forces are at work in the subduction of plates or, on a more mundane level, the movement of subsurface water through various soil and rock strata. Geologists cannot give you a general theory of earthquakes, but they can describe rather accurately the process by which the force generated by plate collisions is stored and transmitted in specific formations, and the same can be said for the skills of the hydrologist you might consult before deciding whether you ought to build your house on a particular slope.Any my conclusion:
Causal mechanisms are the preferred subject matter of science for several reasons. They appease our curiosity. They are more likely to be testable in ways that minimize Type I error than process-independent claims about prior and subsequent states. Above all, they provide knowledge that is frequently useful even when it is incomplete—as knowledge invariably is in complex domains. If you know some mechanisms but not all of them, you still know something about how a system works.
So what would a scientific economics look like? I have mostly answered this already: it would look like other sciences whose objects of study are complex, heterogeneous and context-dependent. It would study mechanisms primarily and end states only for heuristic purposes. It would be predominantly empirical, where this encompasses both statistical work and direct observations on economic behavior (which may also entail statistical analysis). It would ruthlessly identify potential sources of Type I error and strive to eliminate them in hypothesis testing. Experimentation, in the lab and in the field, would become more common, but even more important, primary data collection of all sorts would be accorded a very high value, as is the case in all true sciences. Its macro models would come to look like macro models in hydrology or biogeochemistry: simultaneous differential equations representing mechanisms rather than static end states embodying (a single) equilibrium. Economists would increasingly find it useful to collaborate with researchers from other fields, as their methodological eccentricities are abandoned. Finally, there would be a much clearer distinction between the criteria governing scientific and policy work, insulating the former from some of the influence exerted by powerful economic interests and freeing the latter to adopt an ecumenical and risk-taking approach to tackling the world’s problems.