Self-selection and the power of incentive schemes: an experimental study

Abstract

We examine how self-selection of workers depends on the power of incentive schemes and how it affects team performance if the power of the incentive schemes is increased. In a laboratory experiment, we let subjects choose between (low-powered) team incentives and (high-powered) individual incentives. We observe that subjects exhibiting high trust or reciprocity in the trust game are more likely to choose team incentives. When exposed to individual incentives, subjects who chose team incentives perform worse if both the unobservable interdependency between their efforts and their incentive to cooperate under team incentives are high.

Keywords:

• incentive scheme
• self-selection
• laboratory experiment
• trust
• reciprocity

JEL Classification:

• C91, J33, M52

Acknowledgements

Financial support from CPB Netherlands Bureau for Economic Policy and the Dutch National Science Foundation (No's. NWO-VENI 051.14.023 and NWO-VICI 453.03.606) is gratefully acknowledged. For valuable discussions and comments, we would like to thank Marcel Canoy, Robert Dur, Jan Potters, Arthur Schram, Jeroen van de Ven, Frans van Winden, participants a seminar at the University of Amsterdam, the Conference on ‘Incentives and Work Motivation in the Public Sector’ 2005 at Erasmus University Rotterdam, and the 2006 EALE conference in Prague. The usual disclaimer applies.

Notes

¹ See, e.g., Drago (1991) and Lazear (2000). Gibbons (1997) and Prendergast (1999) provide overviews. Sunde (2009) and Franke (2011) observe similar effects in sports tournaments.

² See, e.g., Kandel and Lazear (1992), Gneezy and Rustichini (2000), Minkler (2004), Canton (2005), and Sliwka (2007).

³ Experiment instructions are available at http://www.sanderonderstal.com/Instructions/SelfselectionP2.doc.

⁴ Subjects only received instructions for the stage that they were on, and were not informed about the stages to follow.

⁵ Vyrastekova and Onderstal (2010) discuss this ‘behind the veil of ignorance’ design and observe that subjects behave similarly as in the standard design. Moreover, Vyrastekova and Garikipati (2005) provide evidence that transfers are a reasonable measure of beliefs in positive reciprocity.

⁶ Subjects learned the outcomes for the trust game only at the end of the experiment so that these would not affect their behaviour in the rest of the experiment.

⁷ Subjects did not know how many others chose the same scheme. Therefore, they could not condition their behaviour on the self-selection procedure outcome. In each session, more than four subjects self-selected into each scheme so that we did not observe repeated interaction of a fixed matched pair.

⁸ In the experiment, we used neutral labels: ‘PULL’ instead of e = H, ‘PUSH’ instead of e = L, ‘GIVE’ instead of r = R, and ‘KEEP’ instead of r = 0.

⁹ The parameters are consistent with the following production game. The costs of effort are c(e_i ), with c(L) = 0 < c(H) < H − L. The costs of reward are k(r_i ) where k(0) = 0 and 0 < k(R) < R. An agent's effort raises her own output, while her reward raises the output of her team member (which could be interpreted as helping the other person or being cooperative with her). More precisely, the relationship between the efforts and the rewards of players i and j forming one team and the output o_i of player i is o_i = e_i + r_j , {i, j} = {1, 2}. Player i's payment equals p, = αo_i  + (1 − α) o_j , {i, j} = {1, 2}, where α is a measure of the extent to which an agent's own output determines her payment. In the extreme case where α = 1, only her own output determines what she gets, while the other extreme α =  indicates that the payments are only based on total team performance. The parameters used for the parametrisation P1 (P2) are: H = 18(14), L = 2(0), c(H) = 12(8), R = 16(14), k(R) = 8(8), α_INDI = , and α_TEAM = .

¹⁰ Note that we used a cautious design in that subjects did not have the opportunity to sort themselves into teams endogenously or to build reputation. Free-rider problems could be alleviated if the subjects are able to form groups endogenously on the basis of historical information or repeated interaction. These options remain open for future research. Keser and Montmarquette (2009) is a promising first step.

¹¹ We chose this approach so as to let subjects earn approximately the same amount of income as in the other sessions. In our analysis, we only compare data on the TEAM payment scheme without self-selection in rounds 1 to 10 to data on TEAM payment scheme with self-selection (rounds 1 to 10 as well).

¹² We tested for the possibility of multi-collinearity between trust and reciprocity, and found this not to be of importance. We also estimated the model with a gender dummy as an additional control, but this variable was insignificant and did not change our trust and reciprocity coefficient values substantially.

¹³ In the long run, employees who prefer low powered incentives may leave the firm and find jobs that better suit their preferences (see, e.g., Delfgaauw (2007)).