Performance Feedback with Team Incentive: A Field Experiment in Chinese Factories

ABSTRACT

This paper employs a field experiment to investigate in which information environment team-based incentives work better. The experiment was conducted in two spinning factories in Henan, China. We focus on workers who were doing the same individualistic task but still were paid according to team performances. For about three months, we have given three different types of performance feedback, baseline, intra-team, and inter-team feedbacks. We find that workers’ productivity was highest with the intra-team feedback and lowest with the baseline feedback, which suggests that peer pressure and group status concern are of importance in making team incentives work.

KEYWORDS:

• Relative performance feedback
• peer pressure
• group identity
• field experiment
• chinese factory

JEL:

• Codes: C93
• M50
• M54

Acknowledgements

We are grateful to Syngjoo Choi, Subhasish M. Chowdhury, Chulyoung Kim, Duk Gyoo Kim, SeEun Jung, Euncheol Shin, and the participants of the Experimental Economics workshop at Seoul National University 2019 for their valuable comments.

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Ethical Approval

Yonsei University Ethics approval was obtained for this project, with ethics approval number 7001988-201805-HR-367-04.

Notes

1 Ledford Jr (1995) report that 70% of large firms used some form of team incentives. Moreover, ‘the use of various forms of team production (including methods such as quality circles) have increased in recent decades.’ (Lazear and Gibbs 2009, 204) See Lazear and Shaw (2007), Babcock et al. (2015), and Bandiera, Barankay, and Rasul (2013) for more discussions on the use of team in modern organizations.

2 According to Tajfel and Turner (1979), social identity has three components: categorization, identification, and comparison. Categorization is the process of putting people into categories. Identification is the process by which we associate ourselves with certain groups. Comparison is the process by which we compare our group with other groups.

3 We focus on these workers because (i) there were a number of them in relatively small factories, (ii) individual performances were easy to measure, and (iii) they were paid according to team performances. About the reason why team incentives were given, see footnote 13.

4 We could have considered no feedback treatment as well, but decided not to, because given limited time, we wanted to focus on the inter-team and intra-team feedbacks, and the baseline feedback is more comparable to the other feedbacks than no feedback.

5 See for example Azmat and Iriberri (2016) and Sausgruber (2009).

6 See the discussion at the end of Section 2.

7 Evidence on the effects of performance feedback is mixed. For example, Blanes i Vidal and Nossol (2011), Kuhnen and Tymula (2012) and Charness et al. (2014) find significant positive impacts of relative performance feedback, while Guryan, Kroft, and Notowidigdo (2009), Eriksson, Poulsen, and Villeval (2009) and Bellemare, Lepage, and Shearer (2010) find no significant effects. Barankay (2012) find that removing relative performance feedback improve productivity.

8 It may be attributable to the cultural difference between the west and the east. We thank an anonymous referee for drawing our attention to this point.

9 Blazovich (2013) shows that teammates’ familiarity with one another influences the effectiveness of the compensation scheme.

10 With the noise being a normal random variable, $q_{ij}$ can be negative, which is unrealistic. However, the purpose of this assumption is to derive a simple closed form solution from which we can obtain a clear intuition. We strongly believe that the intuition will remain the same even if we relax this assumption.

11 It may be the case in reality that people care more about the ranks than the absolute values of $q$’s. However, even if so, the intuition would be the same. Any increasing function of $q$ would lead to qualitatively similar results.

12 Because in the experiment workers knew their own individual performance on top of the average performance of their team, the actual inference problem by workers was slightly more complicated than the one presented here. However, even if we take into account this, the qualitative results derived here do not change.

13 In sequential equilibrium, the beliefs are consistent with the strategies, and the strategies are optimal given the beliefs (see e.g., Osborne and Rubinstein 1994). Here, the belief does not have to be consistent with the strategy. All we need is the assumption that the other workers believe that $x_{ij}=x^{\ast }$ for some known value $x^{\ast }$.

14 The derivation is presented in Appendix A.

15 We could not incorporate this idea to our experimental design for logistic reasons.

16 It is conceivable that intra-team dynamics may be complicated by inter-team competition, which our model does not take into consideration. Many issues in team production including this hypothesis remain to be investigated.

17 We focus on these workers because (i) there were a number of them in relatively small factories, (ii) their individual performances were easy to measure, and (iii) they were paid according to team performances.

18 The reason is not entirely clear, but we think that the team-based incentive scheme is used because it is more convenient (or less cumbersome) to implement than an individualized incentive scheme. Also, the managers and the workers seemed to believe that each individual worker’s contribution to the team was more or less homogeneous, so they might believe that there was no reason to differentiate the compensation.

19 The machines and time that a worker was responsible for were shifted regularly. In factory A, every four days evening team became afternoon team, and afternoon team became morning team. When changing the working time, they also changed the machines they were responsible for. Similarly, in factory B, the workers changed the machines and time every seven days.

20 Thus, the ‘unproductive’ time can be divided into two: the time when the spindles are spinning and the time when they stop. The workers in factory A had the incentives to minimize the latter (i.e., the machines’ idling time), instead of the total unproductive time.

21 See Appendix B for a sample feedback.

22 We are aware of the weakness of our implementation, namely that the sequence of the treatments might matter. However, because the same type of feedback has been given daily for a relatively long time (about a month), we believe that the sequence effect would be minor. And, the results from the factories with different sequences of the treatments (see Figure 1 for instance) seem to suggest that the sequence effect was indeed minor.

23 Recall that the sequence of the treatments was C, T1 and then T2 in factory A, whereas C, T2, and T1 in factory B.

24 In factory A, a worker was responsible for 1920 spindles, whereas in factory B, a worker was in charge of about 2600 spindles. And, a worker worked for 8 hours a day.

25 The number of workers in a team was constant for most days, but occasionally there were absent workers due to illness or personal businesses. In such occasions, the number of spindles that a worker was in charge of was larger.

26 Here the belief does not have to be consistent with the strategy. All we need is the assumption that the other workers believe that $x_{ij}=x^{\ast }$ for some known value $x^{\ast }$.

Additional information

Funding

This research was (in part) supported by the Yonsei Signature Research Cluster Program of 2021 (2021-22-0011).