14 Nov 2008Working Paper Summaries

Parallel Search, Incentives and Problem Type: Revisiting the Competition and Innovation Link

by Kevin J. Boudreau, Nicola Lacetera and Karim R. Lakhani

The innovation process is fraught with uncertainty. Managers often do not know ahead of time the ideal mix of individuals and skills needed to solve innovation-related problems. One way around this uncertainty is to have multiple paths, approaches, or designs explored at once. The "parallel search" principle can be used inside the firm just as it may be used more generally by pursuing "open innovation". However, having too many searchers attempting to solve the same problem can undercut the benefits if it leads to less effort and investment. The authors study the outcomes of 645 software development contests, conducted by a software outsourcing vendor, involving over 9,000 coders, to understand the relationship between parallel search and increasing competition and innovation. Key concepts include:

The key factor favoring parallel search, i.e. increasing the number of independent solvers, is the complexity of the problem at hand.
The benefits of increased searchers were curtailed when the problems were simple, indicating that the negative consequences of competition matter most for simpler problems.

Author Abstract

This paper presents econometric evidence of two independent effects of adding more competitors on innovation: 1) a competition effect whereby increasing rivalry shapes, and often decreases, incentives to expend effort and invest in innovation; and 2) a parallel search effect whereby adding greater numbers of "searchers" benefits innovation by broadening the search for solutions. We further show the importance of these effects depends on the nature of the innovation problem being solved. The analysis uses data from TopCoder's software contest platform, on which elite software developers were assigned different problems to solve within assigned groups of direct competitors. Econometric relationships are identified by exploiting random assignment and a separate instrumental variables procedure.