Self-Control at Work

Self-control problems change the logic of agency theory: workers not only fail to work as hard as employers would like, they fail to work as hard as they themselves would like. In response, firms can use incentive pay to affect the self-control problem, not just moral hazard. We describe the results of a year-long field experiment on data entry workers designed to test the empirical importance of these ideas. First, we find that workers will choose dominated contracts—which pay less for every output level but have a steeper slope—in order to motivate themselves. Second, their effort increases significantly as the (randomly assigned) payday gets closer. Third, these two effects are linked: the demand for dominated contracts (and their benefits) is concentrated amongst those with the highest payday effects. Finally, as workers gain experience, they appear to learn about their self control problems: the correlation between the payday effect and the demand for the dominated contract grows with experience. Both payday and contract effects are quantitatively large in magnitude when benchmarked against the impact of a change in the slope of incentives or of a year of education. These results together suggest that selfcontrol, in this context at least, meaningfully alters the firm’s contracting problem. * Supreet Kaur: Department of Economics, Harvard University ([email protected]); Michael Kremer: Department of Economics, Harvard University and National Bureau of Economic Research ([email protected]); Sendhil Mullainathan: Department of Economics, Harvard University ([email protected]). We thank Lawrence Katz, David Laibson, John Matsusaka, Ted O’Donoghue and participants at numerous seminars for helpful comments, and Raluca Dragusanu for outstanding research assistance. We thank Dominik Bulla, Karthik R, and the Centre for Innovative Financial Design (IFMR) for operational support and Dan Bjorkegren, Matthew Darling, and Nicolas Papamichael for expert research assistance in specific parts of the project. We gratefully acknowledge funding from Harvard University, the International Finance Corporation, and the National Science Foundation. All errors are our own.