Hidden Cost of Control. A Priming Approach

Table of Content

1 Introduction

2 Experimental Design
2.1 The Principal-Agent Game
2.2 Priming
2.3 Procedural Details

3 Behavioral Predictions

4 Results
4.1 Agents' Behavior
4.2 Principals Behavior
4.3 Agents' motives

5 Data Quality

6 Conclusion

References

Appendix A

Appendix B

1 Introduction

Principal-agent relations are an important instrument to investigate conflicts of interest. In such a relation the agent has to decide about his effort level while the principal has only the opportunity to control the agent. The agent has an interest to behave self-interested since he benefits from it. At the same time the principal has an incentive to control the agent to protect him- self from opportunistic behavior of the agent. As a result of the principal’s decision to control, the agent reacts with a reduction of his performance. The reduction in the output for the principal as a result of the agents reaction is called in the literature hidden costs of control. Deci (1971) argued that there are two countervailing effects of control on the agents decision. A disciplin- ing effect (limitation of shirking) and a crowding-out effect (undermining of intrinsic motivation). Which effect dominates, is dependent of the concrete relationship. Frey (1993) argued that the disciplining effect dominates in an abstract relationship, while the crowding-out effect dominates in a more personalized relationship.¹ Falk and Kosfeld (2006), however, argue that the crowding-out effects are even in abstract relationships substantial enough to outweigh the positive effects of control.

We see the analysis of these hidden costs of control to be an interesting field of study since unintended wrong priming could reduce work motivation, respectively correct priming could increase work motivation in situations with monitoring. While in some work arrangements trusting may not always be the best decision for the principal and controlling his agents is likely to be costly, principals may be interested in avoiding hidden costs of control before they arise. Like Cohn and Maréchal (2016) say, ”priming can be a low-cost and effective tool for policy makers and organizations to promote desirable behavior.” Therefore it has high relevance for real economic issues.

In this paper we analyse how semantic priming affects the hidden costs of control of a principal-agent relationship. For this we conducted a trial in the internet based on the two player one shot two-stage principal-agent game model of Falk and Kosfeld (2006). As Schmelz and Ziegelmeyer (2015) con- clude from their study, the internet is a viable alternative to the laboratory for the experimental investigation of control aversion. For the priming three different texts were created, which are read before the main experiment, fol- lowed by an exit survey to control for the understanding of our priming and to check for several other variables like beliefs of our participants.

We argue in this paper that priming has an effect on hidden costs of control in an principal-agent relation by influencing the crowding out or crowding in of the agent’s non-monetary motivation. This means we expect that the higher or lower intrinsic motivation through the priming has an effect on agents effort compared to the control group. Moreover, we expect higher hidden cost of control in the negative priming treatment, respectively lower in the positive priming treatment since we expect our priming to affect the control aversion of our agents. If we assume that that people already have a control aversion without priming as shown by Falk and Kosfeld (2006) or by Schmelz and Ziegelmeyer (2015), we expect the increase in the hidden costs of control in the negative priming treatment to be larger than the decrease in the positive priming treatment, because we expect higher consistency of our priming text with their initial beliefs.

Our paper offers a contribution to the literature on incomplete contracts and builts on the results of Falk and Kosfeld (2006), Ziegelmeyer et al. (2012) and Schmelz and Ziegelmeyer (2015) to investigate the robusness of hidden cost of control to design variation. We extend their research by analysing how intrinsic motivation and control aversion can be influenced by the use of priming.

Furthermore, we extend the literature which analyses whether priming alters individual’s decisions when there are monetary incentives. While research is done about priming in economic experiments in a public good game by Drouvelis et al. (2015), in a trust game by Posten et al. (2014) or Al-Ubaydli et al. (2013) or in a bargaining game by Burnham et al. (2000), to give some examples, we extend this field of research by a priming experiment in a principal-agent game.

Summarized our study is able to replicate the results of Falk and Kosfeld (2006) in a small sized internet experiment and are additionally in line with the results of Schmelz and Ziegelmeyer (2015). Our results show hidden costs of control in an abstract internet setting. We find that agents can be intrinsically motivated and that control is crowding-out intrinsic motivation. Moreover, agents show aversion to control. Since agents show heterogeneity, these results only account for some types of agents. Furthermore, we show that principals anticipated the agents behavior and acted rationally given their beliefs. We find also different types of principals acting in line with the ”self-fulfilling prophecy of distrust”. We find ambiguous effects on priming on all these results. Therefore, we see room for improvement for our priming approach.

This paper is structured as follows. In Section 2 we explain the design of our experiment. Particularly we show the principal-agent game more in detail, illustrate our priming approach and discuss further our procedure for the experiment. Section 3 discusses the behavioral prediction of the experiment. In Section 4 we present our results and section 5 concludes.

2 Experimental Design

2.1 The Principal-Agent Game

Our design of the experimental game sticks to the design chosen by Falk and Kosfeld (2006). The reason is that on the one hand their experiment is easy reproducible with our limited resources and on the other hand we want to reproduce their survey in a neutral manner since they used in the instructions for their experiment words like ”control” which we suspect to be already framing. For this reason we implemented in our set up two different priming groups and one control group.

In the two-stage principal-agent game from Falk and Kosfeld (2006) we have an agent (participant A) who chooses a productive activity x. This activity is costly to him but beneficial for the principal (participant B). Therefore, the cost function for the agent is c(x) = x. The benefit for the principal is equal to 2x. This means the marginal benefit for the principal is always higher than the marginal costs for the agent. The agent has an initial endowment of 120 points while the principal has an initial endowment of 0 points. Thus, the payoff functions are given for the principal by [Abbildung in dieser Leseprobe nicht enthalten] and for the agent by [Abbildung in dieser Leseprobe nicht enthalten]. In the first stage the principal is able to determine the agent’s choice set. Therefore the principal can choose whether he wants to restrict the agent’s choice set or not.² If he restricts, the agent’s choice set for his productive activity can be any integer value of x ∈{x, x + 1, ..., 120}. If he chooses not to restrict, the choice set of the agent is unchanged and can be any integer value of x ∈{0, 1, ..., 120}. For the level of the control treatment we have chosen a fixed level (x = 10) as chosen in Falk and Kosfeld (2006) as main treatment. In the second stage the agent has to decide on his productive activity x without knowing the principals’ decision. Since we made use of the strategy method, the agents had to make their decision for both cases, if their choice set is restricted and if their choice set is not restricted by the principal.³ With this method we are able to collect additional information about the individual types of the agents. Finally, we ask about the participants’ beliefs to make conclusions about the motives of their choices.

2.2 Priming

The concept of priming is described by Hertel and Fiedler (1998) as the procedural feature that some previously activated information impacts on the processing of subsequent information. It describes the fact that a mental representation of a phenomenon can have outside of its context an effect on behavior. This arises from the limited cognitive abilities of people, such that these mental representations have behavioral effects even if they are irrelevant in the given situation. This can be seen as a spillover effect. The mental representations of a person are affected by their experiences, while the intensity of the spillover effects of the mental representations on the behavior is determined by the individual accessibility (Al-Ubaydli et al., 2013).

In our case the mental representation of a work environment with a given business culture is based on the individuals experience as example made as a principal or as an agent made in the past while working. The accessibility is only present if an individual has an intrinsic motivation, because without intrinsic motivation the agents’ choice of their productive activity will be always zero.

From this we infer, that if there is an intrinsic motivation, then it should be possible to prime the mental state with a control or a trust work environment such that we are able to predict the effect of the prime on the decision making process of our participants.

To do so we have created a semantic prime in form of three short texts which have to be read by our participants before performing the main experiment. The texts are about a work relationship in which an agent is controlled by a principal. Depending in which treatment group participants are, they receive a positive or a negative priming text or a control text. The three texts are mainly the same and differ only by their priming sentence (in bold formatting). Following our three treatment texts will be shown.

Positive priming: ”Simon is 27 and moved to London from Essen, where he studied economics, to start his new job. In his job Simon is working with sensitive data. In his break, he wants to check his emails on his work computer. However, when he wants to access his private email server a message pop’s up. The message reads: Private email servers and social media are blocked. We control our internet activity because we want to protect our data from malware, (since we had bad experiences in the past.)

His colleague Tim shows up and they go to the cafeteria and grab a drink.”

In the positive priming text the blocking of private email servers and social media is explained by a protective action and is legitimated by experience in the past. These two points should emphasize the trust environment and should reflect the social norms of trust in our enterprise, respectively the trustful business culture.⁴

Negative priming: ”Simon is 27 and moved to London from Es- sen, where he studied economics, to start his new job. In his job Simon is working with sensitive data. In his break Simon wants to check his emails on his work computer. However, when he wants to access his private email server a message pop’s up. The message reads:

Private email servers and social media are blocked. We control our internet activity because we want you to work efficiently and not waste time.

His colleague Tim shows up and they go to the cafeteria and grab a drink.”

In the negative priming text the blocking of private email servers and social media is explained by negative expectations of the principal towards the work attitude of the agent. This signal of distrust should underline the restrictive business culture, which represents the control environment.

Control treatment: ”Simon is 27 and moved to London from Essen, where he studied economics, to start his new job. In his job Simon is working with sensitive data. In his break Simon wants to check his emails on his work computer. When he wants to access his private email server his colleague Tim shows up and they go to the cafeteria and grab a drink.”

In the control treatment there is no priming text. The rest of the text is still the same and is only minimally different from the other texts. The control text is necessary to keep the difference between the positive and negative priming treatment and the control treatment as small as possible. On the same time it enables us to ask the participants from all treatments the same control questions.

2.3 Procedural Details

To conduct the experiment we decided to use an internet survey tool called ”qualtrics” for which an account was provided for us by the university of Zurich. The advantage of an internet survey compared to the laboratory is that we have access to a different pool of participants, we are more flexible with the time to conduct the experiment and because we do not have a show- up fee as it is mandatory in the laboratory, we are able to payout more subject with the same budget. Since our budget was limited to 400 CHF this was a crucial point for our decision. The disadvantages as Schmelz and Ziegelmeyer (2015) report are that the social bonds are likely to be weaker, what results in lower but still significant effects than in the laboratory. Furthermore, they say data quality is an important issue since the internet setting reduces the level of external control. The survey consists of the instruction for the experiment, the control questions for the experiment, the priming or control text, the control questions for the text, the actual experiment and an exit survey asking about demographic and other information including beliefs. We used exactly the same instructions as Falk and Kosfeld (2006) but modified them slightly by changing some framing words and to account for the purpose of an online survey.⁵

As subject pool we used assessment classes in economics and business ad- ministration from the university of Halle in Germany. They are favorable, because we do not expect them to know our experiment already but expect them to be encouraged to participate in economic experiments and to have low opportunity costs. We had the opportunity to make an announcement about the experiment in class to advertise our survey. Since we had access to the mailing list of the university of the course, we next have sent the link to the survey to all the students of the course.⁶ After two weeks at the latest the survey was closed and the participants were paid in cash at the univer- sity after the course.⁷ There was no show up fee. The points earned in the experiment were converted 1 point = 4 Cents.⁸ For the payoffs the partici- pants (agents and principals) were matched randomly within their treatment groups.

Because we had to guarantee the financing of our participants, we had to make a worst case scenario calculation. The highest total sum of payoffs possible was when every agent would have transferred his whole initial en- dowment to the principal. This would have been 120 points times two equals 240 points. Since decided that one point corresponds to four cents, as a result the highest possible payoff for a couple of one agent and one principal is 9.60 Euros. With 39 couples (13 for every treatment group) this gives us 374.40 Euros (approx. 411.84 CHF). As comparison the expected average value of transfer is 40 points, such that principal and agent both have 80 points, we receive 6.40 Euros per couple. This is an expected total sum of 249.60 Eu- ros (approx. 274.56 CHF). As an element of uncertainty we had that the program only counts when a survey was finished such that in the meantime more surveys could have been opened than defined by our upper limit such that overshooting was possible. But on the same time the program counts for incomplete surveys. On the other hand we do not know the rate of subject who pick up their payoff after completing the survey. To receive the highest power but avoid overshooting we decided to set a limit at 80 participants. The participants in our experiment were randomly assigned to the two roles as agent or as principal and to the three treatment groups, in concrete the positive priming, the negative priming and the control treatment. As a re- sult we have six categories: principal control treatment (PC), principal posi- tive priming (PP), principal negative priming (PN), agent control treatment (AC), agent positive priming (AP) and agent negative priming (AN). In to- tal 73 subjects participated in our experiment. Every participant had to identify himself/herself by his student-ID to eliminate the possibility that subject participated in more than one treatment. The distribution to the categories was 12 (PC), 11 (PP), 15 (PN), 11 (AC), 12 (AP), 12 (AN).⁹ The time for completing the survey was between 8 and 14 minutes (interquartile range) and our participants earned between 2.40 and 4 euros (interquartile range). The instructions of the experiment is attached in Appendix B. Our participants are on average 23 years old, have a disposable income (without renting costs) slightly below 500 euros and check their mail about 4 times a day. 52 percentage are male, 30 percentage already had experience being a principal while 82 percentage already had experience in being employed. These divide in 40 with a short, 24 with a medium and 18 percentage with a long employment time for students in our opinion.

Table 1: Socio-economic characteristics

Abbildung in dieser Leseprobe nicht enthalten

3 Behavioral Predictions

In standard economic theory we would assume the agents to be selfish and only maximizing their own payoff. Accordingly, they would always choose the minimum x = x since x is costly, which would be in the case of no restriction x = 0 and in the case of restriction x = 10. Consequently, selfish principals maximizing their own payoff would always control because of the opportunistic behavior of the agents. Therefore, a restriction of the choice set of the agent would rule out the most opportunistic behavior of the agent and would make the principal better off.

However, results from experimental research are consistent with the social- preference theories which predict that a substantial fraction of agents are endowed with social preferences and do not act in purely selfish manner.¹⁰ These agents are expected to choose an x that is larger than the minimum value. As long as these social-preferences are small and the agent chooses 0 < x < x the principal still gains from controlling. If the agent, however, has strong social-preferences and chooses x > x the payoff of the principal is independent of his choice, whether he controls or not, because the constraint is not binding anymore. As a result the principal can only gain but never loose when restricting the choice set of the agent. Thus a payoff maximizing principal should always control the agent.

If we now assume agents having non-monetary motivations like other-regarding preferences to perform in the principal’s interest or intrinsic preferences for autonomy in decision-making, the decision of the principal to control or not has an influence on his payoff. The decision of the principal to control im- plicitly signals the agent that the principal does not expect him to choose a high value of x. Agents may interpret this as an indication for the principal’s expectation of the agent’s motivation, they may take it as a signal of distrust or undermining of their sense of autonomy.¹¹

[...]

¹ Supporters of Frey’s proposition are eg. Antonakis and Atwater (2002), Stanton (2000), Dickinson and Villeval (2008) or Bartling et al. (2010).

² To avoid demand effects we did not use words like ”control” or similar as Falk and Kosfeld (2006) did.

³ The strategy method is legitim since Falk and Kosfeld (2006) have shown that the behavior of the agents is not an artifact of the strategy method since their results with the strategy method did not significantly differ from the results of a control treatment using the specific response method.

⁴ To avoid confounders the message is formulated concisely without mentioning any third-parties.

⁵ For further details see Appendix A.

⁶ As a backup we had the permission to do the same at the university of Essen in Germany if our response rate would have been to small.

⁷ We are aware of the fact that time discounting may have an influence. For this reason we have chosen the period between sending the link and paying out the participants as short as possible.

⁸ In the experiment of Falk and Kosfeld (2006) they paid 1 point = 20 centimes with a show up fee of 10 francs in their laboratory experiment, while Ziegelmeyer et al. (2012) paid 1 point = 15 cents without any show up fee in their online survey

⁹ The deviation from the upper limit in qualtrics comes from the fact that some opened the link while the last subject below the upper limit still did not finish the survey. Further- more, there were 20 unfinished surveys, from them 7 stopped before the actual experiment started but started the survey once again and completed it, 7 stopped before the ex- periment started and only 6 stopped during the actual experiment. We only considered completed surves for our analysis.

¹⁰ See eg. Fehr and Gächter (2000), Camerer (2003), Forsythe et al. (1994), Kagel and Roth (1995), Fehr and Schmidt (1999), Bolton and Ockenfels (2000), Charness and Rabin (2002)

¹¹ See Sliwka (2006), Ellingsen and Johannesson (2008) and von Siemens (2013) for further information