Money, Banks and Finance in Economic Thought

To Randomize or to not Randomize? Princetonian Solutions for the 1970s Passive Observation Bias

Brackmann Netto Arthur, USP

What is now closely related to – if not identical – to endogeneity, was in the first half of the 20th century more broadly defined as the problem of passive observation. Up to the 1980s, endogeneity was not yet well defined. The several sources of the bias, simultaneity, measurement error and omitted variables, were gradually rather than simultaneously faced during the 20th century. Measurement error was the first to be tackled through the errors-in-variables models in the 1930s. Errors-in-equations models solved the simultaneity bias during the 1940s and 1950s. However, identifiability was far from being the solution for one of economics everlasting conundrums: how to make causal claims about social phenomena. Passive observations came with the impossibility of intervening in the data. This lack of intervention that before manifested as simultaneity and measurement errors, during the 1960s, emerged as observable and unobservable omitted variables. In contrast with the previous problems, the solution for omitted variables divided opinions. Two Princeton young scholars solved the new manifestation of the passive observation bias of the 1960s and 1970s: James Heckman and Orley Ashenfelter. Although both studied and published together in the first half of the 1970s, their solutions (written during the second half) are now on opposite sides of the debate. Heckman (1979) proposed his self-selection model, which dealt with the bias modeling it explicitly. Ashenfelter (1978) presented the method of Difference in Difference, dealing with the bias through artificial randomization. Nowadays, their solutions resonate on the ‘randomistas’ and structural modelers debate. What happened during the 70s that led two concurrent research agendas to develop from the same problem and from previous colleagues who even published together years before their seminal papers? This paper intends to shed some light in the answer to these questions through observing more closely the development of the endogeneity bias and its solutions during the 1970s using bibliometric and primary sources of information.


Keywords: Heckman, Ashenfelter, Endogeneity, Passive Observation, Ranzomization