The Effects of Relative Performance Information and Transparency on Knowledge Sharing and Productive Effort

Abstract

Using a real-effort experiment, we investigate individuals’ simultaneous decisions regarding how much effort to exert and how to allocate this effort when working on a productive task (productive effort) and sharing knowledge (knowledge-sharing effort). Specifically, we examine the effects of relative performance information (RPI) and the transparency of individual knowledge-sharing decisions (transparency). We find that RPI increases productive effort but reduces knowledge-sharing effort over time. In contrast, transparency does not influence productive effort but increases knowledge-sharing effort. Furthermore, the results indicate that transparency can counteract the adverse effect of RPI on knowledge-sharing effort over time without impeding productive effort. We link our results to the psychological concepts of social comparison, impression management, and reciprocity. Implications for research and practice are discussed.

Keywords:

• Relative performance information
• Performance management
• Knowledge sharing
• Cooperation
• Transparency

JEL codes:

• M410

Acknowledgements

We thank Victor Maas (associate editor), two anonymous reviewers, David Derichs, Jan Diebecker, Robin Dresenkamp, Sebastian Exler, Christoph Kirchner, Weiming Liu, Teemu Malmi, Christian Rose, Ivo Tafkov, and participants at the European Network for Experimental Accounting Research (ENEAR) Summer School 2015, the University of Münster Experimental Research in Management Accounting Workshop 2016, the EAA Annual Congress 2018, CAAA Annual Conference 2018, AAA Annual Meeting 2018 and the 3rd Conference on Behavioral Research in Finance, Governance, and Accounting (BFGA) for their helpful comments and suggestions. We appreciate the financial support provided by The Chartered Institute of Management Accountants (CIMA).

Disclosure statement

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

Notes

1 In such settings, employees may nevertheless build a reputation as people willing to share knowledge by word-of-mouth. However, the likelihood of widespread transparency is lower. In our experiment, we manipulate transparency by informing subjects about the exact number of pieces of knowledge that were shared in the Transparency present condition and not providing subjects any such information in the Transparency absent condition. Both manipulations do not perfectly represent the real world, which is characterized by shades of gray rather than clear black and white. We choose this manipulation to ensure internal validity.

2 Similar to prior research (e.g., Choi et al., 2016; Tafkov, 2013), we employ this task because it allows a strong test of the underlying theory, although it is not similar to employees’ real-world tasks.

3 In contrast to Hannan et al. (2013), we do not vary the difficulty of the anagrams to keep the informational value of shared knowledge constant.

4 The instructions explicitly state that group members can send information on different anagrams (i.e., they possess private knowledge). Additionally, they are told that the differing order of anagrams prevents them from deducing which group member has sent them information, eliminating the possibility of bilateral reciprocity within the round and of nontransparency conditions.

5 Note that for the first set the countdown starts with the beginning of the round because the first set is automatically displayed at the beginning of the round.

6 Participants cannot use knowledge sharing to mislead and thus sabotage their peers because the software allows sharing only the first letter of the solution and no other information.

7 This design choice could be considered prone to ceiling effects, as many participants would like to share more knowledge. However, only in 85 out of 960 (160 participants á 6 rounds) cases (8.85%) participants share the full amount of 20 letters per round with their peers at all, which makes a ceiling effect highly unlikely.

8 As outlined in Footnote 1, we choose this operationalization of transparency to ensure internal validity. In our experiment, transparency gives participants a general impression of whether their peers share more or less knowledge without any uncertainty. Obviously, an absolute amount of knowledge sharing is difficult or extremely costly to measure in reality – otherwise, knowledge sharing could easily be contracted on, which could resolve the issue of insufficient knowledge distribution within organizations. However, a plethora of problems arise when attempting to make knowledge sharing measurable and contractible. For example, the shared knowledge needs to be novel to the recipient and of adequate value; consequently, the recipient’s assessment is necessary for a meaningful measure. This leads to two major behavioral issues: Senders could just forward anything for their ‘sharing count’ (if quality were irrelevant in the assessment of the sent items), while recipients could follow their own agenda in evaluating the shared knowledge (if quality were relevant in the assessment of the sent items). However, in our experiment, none of these issues leads to problems. The shared letters are novel and valuable by design, irrespective of participants’ prior knowledge and experience. Thus, only useful knowledge is disseminated that does not have to be evaluated by the recipient. Hence, adverse recipient behavior cannot occur. While our manipulation ensures internal validity, the attempt to implement our manipulation directly as a management control in companies would most likely lead to adverse consequences.

9 All p-values in this paper are reported on a two-tailed basis.

10 We also rerun our analysis using bootstrapping with 1,000 replications and clustering according to group as a robustness check (Wooldridge, 2013). The main results remain inferentially identical.

11 In addition, we use the number of shared letters (i.e., a maximum of 20 letters shared per round) as an alternative dependent variable as a robustness test. The results remain inferentially identical.

12 The ANOVA contains RPI, Transparency and the interaction as independent variables. We do not find an effect for Transparency (p = 0.87) or the interaction (p = 0.68).

13 Notably, we asked participants about their expectations in the PEQ at the end of the experiment. However, asking participants before the main part of the experiment could have revealed our research question and thus influenced our measurements. Hence, we deliberately chose this approach, thereby relying on participants’ ability to recall their unbiased expectations after the experiment.

Additional information

Funding

This work was supported by Chartered Institute of Management Accountants [grant number n/a].