1. ^aDepartment of Economics, Northwestern University, Evanston, IL 60208, USA. E-mail: m-zelder@northwestern.edu
2. ^bCenter for Health and the Social Sciences, University of Chicago, Chicago, IL 60637, USA

Abstract

Economists often decry the perceived tendency towards selective reporting of empirical results ("specification search") in scholarly work. Yet, economists have largely neglected to analyze the incentive structures underlying this phenomenon of econometric "cons". This paper endeavors to provide this analysis, posing a game-theoretic model of specification search. In this three-player game (author, journal, and profession), academic authors choose whether to report the "true" t-statistic associated with an empirical result, or whether to "con" by reporting a distorted t-statistic. Subsequently, both journal and profession must choose whether to bear the cost of "scrutinizing" the author's work (e.g., by reanalyzing his data). Multiple perfect Bayesian equilibria are found, including one where authors "con" and arc not detected, an equilibrium which may be Kaldor–Hicks efficient. Moreover, public and private mechanisms (existing or proposed) to curtail "conning" seem ineffectual.