An Experimental Study of Endogenous Discretionary Controls on Employee Effort

Abstract

Discretionary controls are controls that can be applied at the ex post discretion of the superior rather than being supported by pre-committed enforceable contracts. This study examines employees’ effort levels when a discretionary control is endogenously present (chosen by the superior) relative to when it is exogenously present (assigned by the experimenters). Results from our gift-exchange experiment regarding superiors’ choice of a discretionary control run counter to prior research on superiors’ choice to use pre-committed controls in two important ways. First, we find employee effort is unaffected by the superior’s choice of a discretionary control relative to when it is exogenous. That is, the intentional choice of a discretionary control does not seem to be resented by the employee. Second, we find employee effort is lower when superiors choose not to have a discretionary control present relative to when it is exogenously unavailable. That is, we not only find no reward to forgoing a discretionary control, but rather a cost to not choosing a discretionary control. In sum, our results support the prevalence of discretionary controls in practice.

Keywords:

• Discretionary Controls
• Endogenous Controls
• Gift-Exchange

Acknowledgements

The authors would like to thank the Jensen-Wallin-Young fund for their financial assistance. We gratefully acknowledge the helpful comments of Jeremy Bentley, Jason Kuang, Victor Maas (editor), Timothy Shields, Victor van Pelt, Di Yang, two anonymous reviewers, and workshop participants at the Southern California Accounting Research Forum, Chapman University, Miami University, the Ohio State University, Ball State University, the 2018 AAA National Meeting, the 2018 AAA ABO Section Meeting, 2019 AAA MAS Section Meeting and the 2019 Western Regional Meeting of the AAA.

Disclosure statement

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

Supplemental Data and Research Materials

Supplemental data for this article can be accessed on the Taylor & Francis website, https://doi.org/10.1080/09638180.2022.2063151.

• Appendix A. Screenshots of z-Tree for EN-DC

• Appendix B. Instructions for Exogenous No Control (EX-NC)

• Appendix C. Instructions for Exogenous Discretionary Control (EX-DC)

• Appendix D. Instructions for Endogenous Control Choice (EN-NC and EN-DC)

Data Availability Statement

Data are available from authors upon request.

Notes

1 We focus on discretionary penalties to parallel pre-committed controls research (e.g., Fehr & List, 2004) and since penalties likely lead to resentment (Luft, 1994; Fehr & Gächter, 2002), which causes a response to intentions.

2 This is inconsistent with information leakage about social norms with pre-committed controls (Cardinaels & Yin, 2015).

3 There are other advantages to using discretionary controls, as noted in Chenhall and Moers (2015).

4 Trust and reciprocity of trust are a significant driver of behavior in the gift-exchange game and the similar trust game. However, other social preferences, such as distributional fairness, inequality-aversion, surplus-maximization, and altruism all may play an incremental role in determining behavior in these settings (Charness, 2004; Cox, 2004; Fehr, 2009; Ashraf et al., 2006; Brülhart & Usunier, 2012).

5 The study of discretionary controls is related to subjective performance evaluation. Subjective performance evaluation is typically used in settings where objective measures would not adequately capture the employee’s performance with respect to the expectations of the superior. The focus of research on subjective performance evaluation is on the qualitative nature of the performance evaluation (Bol, 2008; Bol & Smith, 2011). Subjective performance evaluations must entail ex post discretion by the superior because subjective evaluation can only occur post employee performance. However, discretionary controls can exist even in settings where performance is quantitative, easily measured and timely, such as superior’s discretion in sanctioning cash posters in Akerlof (1982). Thus, while our study of discretionary controls may be relevant to the study of subjective performance evaluation, employees’ performance herein can be objectively and easily measured, unlike subjective performance evaluation.

6 Christ (2013) uses a setting where the superior chooses the employee’s wage prior to the effort decision, such that the superior earns 2 × effort – wage. Fehr et al. (1998) refer to it as a modified gift-exchange and we adopt that label. The payoff structure used in Christ (2013) diverges from that of standard gift-exchange payoff structure where the superior’s return to employee effort decreases as wages increase (e.g., Fehr et al., 1993; Fehr et al., 1998; Hannan et al., 2002; Hannan, 2005; Kuang & Moser, 2009, 2011; Brown et al., 2015). Research, most specifically Fehr et al. (1998), who directly compare the two games, suggests that participant behavior is similar across the two payoff structures.

7 Further, in all these games the social surplus has a non-negative relation to effort and each has a similar equilibria.

8 The strength of the control ranges from being able to motivate less than 10% of effort or unable to stop any misreporting (e.g., Falk & Kosfeld, 2006; Cardinaels & Yin, 2015) to full compliance (Kuang & Moser, 2009).

9 Fehr and Rockenbach (2003) and Fehr and List (2004), using a trust game, do not compare exogenous controls to endogenous controls. However, they do find that endogenous no controls elicit more pro-social behavior than exogenous no controls. Therefore, in the trust game setting, while we do not have evidence that the intentional choice of controls is resented, we do have evidence that the intentional choice to not use controls is appreciated.

10 We are conservative in relying on Douthit et al. as many aspects differ. Employees’ action (effort vs. reporting), relevant norms (reciprocity vs. honesty), control strength (moderate vs. strong), nature of norms (social vs. moral), superior activity (chosen wage vs. none), cost to employees as effort/honesty increases (exponential vs. linear) differ between studies. Given the myriad distinctions, we do not predict we will replicate it.

11 The study was reviewed and approved by the relevant Institutional Review Board.

12 In the experiment, we refer to the superior as ‘employers.’ For clarity, we persist with the term ‘superior.’

13 Earnings in the experiment were expressed in Lira where 2 Lira = $1 USD.

14 We use this payoff structure for the same reason noted in Fehr et al. (1993) in that it makes superior outcomes strictly non-negative. Since participants’ losing money is unenforceable due to lab and IRB rules, this is important. Kuang and Moser (2011, fn. 6) and Brown et al. (2015, fn. 10) reiterate this and note that research generally finds that reciprocity is a key driver even in payoff structures that divorce the return of effort from wages (e.g., Fehr et al., 1997; 1998; Choi, 2014). Also, this is the standard gift-exchange payoff function, which aids in comparability to the closely related study of Kuang and Moser (2009) and broader gift-exchange research in accounting (e.g., Hannan, 2005; Kuang & Moser, 2011; Brown et al., 2015).

15 Extremely high wages are hard to explain, as wages above 90 ensure the social surplus decreases with effort and wages at 120 mean that effort does not increase superior pay (Charness, 2004). Such high wages are rare, with wages above 90 (of 120) occurring in 3% (0.5%) of our observations. Our results are robust to excluding these high wages. As the reduction in effort from the choice of a pre-committed control occurs in a similar gift-exchange (Kuang & Moser, 2009), trust game (Fehr & List, 2004), and with no secondary trust acts (e.g., Burdin et al., 2018), it does not appear relative trust inherent in gift-exchange payoffs reduces the effect of choosing a pre-committed control.

16 Certain discretionary actions by the superior (forgone bonuses, or raises, forced unpaid leave, undesirable/extra work) decrease employee utility. We operationalize discretionary controls using pecuniary penalties. Consistent with induced value theory in abstract experiments, we do not intend for our operationalization to map into any specific control in practice, but rather reflect any control whereby discretion from the superior can decrease employee’s utility.

17 In dictator settings (Falk & Kosfeld, 2006; Christ, 2013), the control can enforce 25% of what the dictator must transfer to achieve equity. In the ‘gift-exchange’ version of these studies, if one assumed superiors transferred their entire endowment to subordinates then the control could enforce 18% of what need be returned to achieve equity. In the trust game of Houser et al. (2008), with a maximum superior investment, the control can enforce 27% (4/15) of what is needed for equity. On the other hand, the control in Kuang and Moser’s (2009) gift-exchange game can motivate maximum effort. Finally, Brunner and Ostermaier (2019) have a hurdle contract in a budgeting setting. This is quite different than the other controls but is a strong control as with the control the superior expects 67% of earnings whereas with no control the superior expects 0% of earnings. Thus, our moderate control is within the control strengths considered by prior research on pre-committed controls, and should not overwhelm the effect of intentions.

18 The presence of a discretionary control is costless, regardless of the source of its presence. When the discretionary control is present, the choice to enforce the penalty is costly and at the superior’s discretion, regardless of its source.

19 The negative effect of endogenous pre-committed controls has been found across subject types, including general undergraduate, business majors, MBAs and CEOs. (e.g., Fehr & List, 2004; Christ, 2013; Kuang & Moser, 2009).

20 In zTree, superior participants who could use the discretionary control were provided with two options: ‘Use the conditional payoff cut plan’ or ‘DO NOT use the conditional payoff cut plan’ (see online appendix). The capitalization of the words ‘DO NOT’ could focus superior participants on not using the discretionary control to enact the penalty – though between 40% (EX-DC) and 53% (EN-DC) chose to do so when possible. We used this capitalization for emphasis to ensure superior participants easily knew which option was which.

21 Screenshots of the interface for the endogenous treatment and all instructions are available on the online appendix.

22 The data from EN-DC (Chosen) represents the periods where the superior in EN chose the discretionary control, ignoring periods where the superior chose to not have the control. The data from EN-NC (Chosen) represents the periods where the superior in EN chose to eschew the control. These data both contribute to our data on employee effort for EN-NC and EN-DC. Since we elicit effort responses to both possible states (DC and NC) simultaneously in the EN condition we have data on both regimes for every employee, every period. These two subsets of our data are part of the data used in a methodological paper (Davis, 2018). While these data are a subset of some of the observations in our EN condition, the comparisons considered and research question in this study are novel.

23 Robust clustered standard errors can still have biased standard errors when the number of clusters is not large enough. In untabulated analyses, we repeat all tests using the wild cluster bootstrap-t procedure (Cameron et al., 2008), which corrects for small number of clusters studies. Unless otherwise noted, all variables of interest retain their (non)significance relative to the 10% p-value cutoff if they were reported as (non)significant.

24 Untabulated repeated-measures ANOVAs find no main period effect or interaction with discretionary control for wages, requested effort, or actual effort (each p > 0.10, two-tailed), suggesting learning did not change behavior.

25 An untabulated logit regression finds that the likelihood of the choice to make a discretionary control present does not change across periods with endogenous choice (z = 0.58, p = 0.56, two-tailed). A series of Tobit regressions find that wage offers and requested effort in the endogenous choice condition do not vary significantly across periods (each p > 0.15). A series of repeated-measures ANOVAs suggests there is a main period effect on wages, requested effort, and effort in the EX-DC and EN-DC conditions collectively (each p < 0.06, two-tailed), but no interactions between condition and period (each p > 0.61, two-tailed). This suggests that the differences between conditions do not materialize over time and learning does not occur differently between conditions or drive our results.

26 A series of repeated-measures ANOVAs suggests there is no main period effect or interaction between condition and period on wages, requested effort, and effort in the EX-NC and EN-NC conditions (each p > 0.48, two-tailed). This is additional evidence that learning did not occur differentially across conditions or drive our results.

27 The main effect of EN-NC in column 2 (3) captures the effect of EN-NC vs. EX-NC when wages (requested effort) is 0, which is not possible in our setting where the minimum wage (requested effort) is 20 (0.1).

28 The wild cluster bootstrap-p procedure alters this result to t = −1.67, p = 0.100, two-tailed.

29 This point comes into sharper relief by noting that superior earnings in EX-NC (24.88) are higher than in EN-NC (18.53) (p = .03, two-tailed).

30 While an interesting result, care should be taken in drawing direct comparisons across these studies as the signal of trust, the strength of this signal, and the form of control all vary across studies.