Trust and Performance* Armin Falk† University of Bonn and IZA Michael Kosfeld‡ University of Zurich March 2004 (PRELIMINARY AND INCOMPLETE)
Abstract We show that a principal’s trust in the voluntary performance of the agent has a positive impact on the agent’s motivation to perform well. Before the agent decides on his performance, the principal in our experiment chooses (not) to implement a minimum performance level for the agent, thereby signaling his distrust (trust) that the agent will voluntarily perform above minimum. Under the assumption that the agent is selfish or inequity averse, trust is never optimal and the principal should always demand minimum performance. Almost 71 percent of the principals in our experiment, however, do not demand minimum performance but trust that the agent will perform voluntarily. Moreover, principals who trust induce a higher average performance by the agent and earn higher payoffs than principals who distrust. The positive impact of trust is driven by the fact that the majority of agents reveal so-called distrust averse behavior, i.e., reduce their performance if the principal signals his distrust. Our results shed new light on the possible dysfunctional effects of explicit incentives as well as the puzzling incompleteness of many economic contracts. JEL: C7, C9, M5 Keywords: Trust, Motivation, Principal-Agent, Incentives, Incomplete Contracts
*
Financial support by the Swiss National Science Foundation (Project No. ???), the Ludwig Boltzmann Institute for the Analysis of Economic Growth, and the MacArthur Foundation (Network on the Evolution of Individual Preferences and Social Norms) is gratefully acknowledged. † Institute for the Study of Labor, Schaumburg-Lippe Str. 7/9 D-53113 Bonn, Germany, E-mail:
[email protected]. ‡ Institute for Empirical Research in Economics, University of Zurich, Blümlisalpstrasse 10, CH-8006 Zürich, Switzerland, E-mail:
[email protected].
1
“In the late 1930s, when I was working for General Electric …., the company was making a big thing of plant security. … GE was especially zealous about guarding its tool and parts bins to make sure employees didn’t steal anything. Faced with this obvious display of distrust, many employees set out to prove it justified, walking off with tools and parts whenever they could. …When HP got under way, the GE memories were still strong and I determined that our parts bins and storerooms should always be open. … Keeping storerooms and parts bins open was advantageous to HP in two important ways. From a practical standpoint, the easy access to parts and tools helped product designers and others who wanted to work out new ideas at home or on weekends. A second reason, less tangible but important, is that the open bins and storerooms were a symbol of trust, a trust that is central to the way HP does business.” David Packard (1995), The HP Way, pp 135-6.
1. Introduction Among all management objectives, motivating employees is certainly one of the most important ones. Traditional principal-agent theory offers a clear advice. Principals should motivate their agents with the help of explicit incentive contracts, promotions and other forms of material rewards or sanctions. This advice rests on the assumption that agents are motivated solely by the desire to achieve income through minimal effort and to avoid risk. In this paper we empirically show that this narrow view of human motivation is questionable. Moreover, we show that introducing explicit incentives may backfire. The reason is that they are perceived as signals of distrust, which actually motivates many agents to reduce their performance. Negative consequences of explicit incentives have long been discussed in the psychological literature on “intrinsic and extrinsic motivation” (Sansone and Harackiewicz 2000, Frey and Jegen 2001). In economics, dysfunctional effects have also been shown by Gneezy and Rustichini (2000a,b), Fehr and Gächter (2002), and Fehr and Rockenbach (2003). While certainly related, these studies are different because they do not consider trust as a driving force for the motivation of agents.1 We study potentially dysfunctional effects of explicit incentives in a very simple and parsimonious set-up. In the game under study an agent chooses a productive activity x which is costly to him but increases his principal’s payoff. Before choosing x, the principal determines the agent’s choice set. He can either leave the decision about x completely at the discretion of his agent, in which case the lowest possible choice for the agent is zero, or he can force him to choose at least a minimum level x > 0. The restriction of the agent’s choice set can be interpreted, for example, as a high-control working environment, like the locked tools and parts bins David Packard remembers from is employment at General Electric (see quotation above). Alternatively, it captures also the use of high powered incentives such that 1
An exception is Frey (1993), who discusses evidence from social psychology for possible negative consequences of monitoring, which is interpreted as distrust, on workers’ effort.
2
choosing x < x is not in the interest of the agent, or other formal elements determining the explicit contract of the agent. For example, if x represents the amount of working hours the agents spends at the firm, x can be interpreted as a minimum presence requirement. Or, if x represents the number or the quality of items an agent has to produce, x is the minimum number or quality the agent is forced to deliver. Since x is costly to the agent, according to traditional economic theory the agent will choose the lowest possible x, which is zero if the principal does not restrict the agent’s choice set and x > 0 if he does. Since the principal’s payoff is increasing in x, he will, hence, always be better off restricting his agent compared to leaving his agent’s choice set unchanged. However, it is conceivable that there are some agents with a high work morale, who in principal are willing to perform above the minimum requirement, i.e., they choose x > x. If such an agent learns that the principal forces him to choose x ≥ x, what does this signal to him? He may believe that the principal does not trust him to perform above x because otherwise it would not have made sense to restrict his choice set in the first place. In other words, the decision to restrict the agent may signal that the principal does not think the agent has a high work morale and is instead likely to shirk. Thus, restricting the agent can be perceived as a signal of distrust. Moreover, it is likely that this signal reduces an agent’s motivation to perform. After all who likes being thought of as not being trustworthy – in particular if one is? In this sense explicit incentives may backfire, resulting in a lower, rather than a higher, performance. We show that a majority of the agents in our experiment (57 percent) exhibit the above described behavior of so-called “distrust aversion”, i.e., they reduce their performance when restricted by the principal. Only a minority (18 percent) choose the lowest possible activity, i.e., act in line with the traditional economic model. In consequence, principals who do not restrict their agents’ choice set receive a significantly higher performance (31 percent more on average) and hence earn a higher income than principals who do restrict their agents. The motivational impact of trust is correctly anticipated by the majority of the principals: almost 71 percent chose not to restrict their agent but to trust instead. Our results provide a behavioral foundation for the observed incompleteness of many economic contracts. The paper is organized as follows. Details of the experimental design are explained in the following section. Section 3 analyzes the behavioral predictions based on standard selfish preferences, inequity aversion, and distrust aversion. Results of the experiment are presented in Section 4 and 5. Finally, Section 6 concludes.
3
2. Experimental design Our design philosophy was to set up an experimental game that allows studying the potential effects of trust on motivation in the most parsimonious way. We therefore implemented the following two stage principal agent game. The agent chooses a productive activity x, which is costly to him but beneficial for the principal. The cost for the agent is c(x) = x, while the benefit for the principal is 2x, i.e., the marginal cost of providing the productive activity is always smaller than the marginal benefit. The agent has an initial endowment of 120 while the endowment of the principal is 0. The payoff functions are therefore given by
πP = 2x for the principal and
πA = 120 – x for the agent. Before the agent decides on x, the principal determines the choice set of the agent. He can either restrict (restrain, constrain?) the agent, in which case the latter can choose any integer value x∈{10, …, 120}, or he can leave the choice set unrestricted to x∈{0, …, 120}. Thus, the principal can force the agent to choose an activity of at least 10, but he can also leave the decision completely up to the agent. Figure 1 illustrates the two stages of the principal agent game. Figure 1: The principal-agent game Principal
don’t restrict
restrict
Agent
Agent
choose x
0
choose x
x
120
10
x
120
In the experiment we made use of the strategy method to elicit the agents’ decisions. Before knowing the actual choice of their principal the agents had to decide on x for both
4
possible cases. On their computer screen they were asked to choose x∈{10, …, 120} in case their principal forces them to choose at least 10, and to choose x∈{0, …, 120} in case their principal does not restrict their choice set. We used the strategy method to gain direct information of individual types as will be discussed in detail below. After principals and agents had made their choices the experiment was over. They were then asked to answer a short questionnaire including motivational questions as well as some questions about how they would behave in some hypothetical workplace scenarios (see below). The experiment was computerized and we used the software “z-tree” (Fischbacher (1999) to run the experiment. Upon arrival to the lab subjects were randomly allocated a role as principal or as agent. A total of 144 subjects participated in our experiment (72 principals and 72 agents). A session lasted, on average, 45 minutes. Subjects were students from the University of Zurich and the Federal Institute of Technology in Zurich. On average, a subject earned CHF 24. 3. Behavioral predictions In discussing possible outcomes of our game we concentrate on three different types of possible motivations: selfishness, inequity aversion and distrust aversion. Let us consider the “standard” case first. If agents are selfish, i.e., only interested in maximizing their incomes, they choose the minimum x. This implies that they choose x = 10 if they are restricted and x = 0 otherwise. As a consequence, principals who want to maximize their payoffs should always restrict their agents’ choice sets. According to this standard economic benchmark solution agents are opportunistic and therefore it inevitably pays to rule out the most opportunistic choices. Since there is no trust or voluntary cooperation in the first place there is also no potential for a crowding out of trust via restricting the agent. There is ample evidence, however, that many people are not acting in a purely selfish manner but are endowed with social preferences (for an overview of the experimental literature see Fehr and Gächter 2000 and Camerer 2003). In the dictator game, e.g., proposers often give positive amounts of money (Forsythe et al 1994 xxx). This finding has recently been explained in terms of inequity aversion (Fehr and Schmidt 2000; Bolton and Ockenfels 2000). In addition to their preference for material payoffs, inequity averse subjects have a dislike for inequitable outcomes, i.e., they are willing to sacrifice money in order to reach more equitable outcomes. In light of the experimental evidence and according to the theory of inequity aversion, a substantial fraction of agents are hence expected to choose an activity x
5
that is strictly larger than the respective minimum. However, if an agent’s choice of x is larger than 10 (i.e., the agent has a relatively strong preference for equitable outcomes), inequity aversion implies that the agent’s choice is independent on whether he is restricted by the principal or not. The reason is that inequity aversion preferences are based on payoff distributions only and that the minimum x = 10 is not binding. Only weakly inequity averse agents who choose 0 < x < 10 if not restricted, will choose a higher activity (namely, x = 10) if restricted. Taken together, as it holds for the standard economic case, a payoff maximizing principal cannot lose anything (but possibly gain something) from restricting inequity averse agents and hence should always restrict the agent’s choice set. 2 We now turn to a different type of motivation, which we call distrust aversion. An agent is called distrust averse if he dislikes not being trusted and if he therefore responds negatively to a signal of distrust. What does distrust aversion imply for our stylized principal agent game? In this game a principal decides to restrict his agent if he believes that his agent will perform poorly, i.e., choose x < 10. Thus, if the principal restricts his agent, the message conveyed to the agent is that he is not trusted to perform well. A distrust averse subject reacts to this signal by actually choosing a lower performance than he would have chosen in case the principal had trusted him. In other words a distrust averse agent chooses a lower x when he is restricted than we he is not restricted. If faced with distrust averse agents, the optimal strategy for the principal is different from the optimal strategy before: a principal is better off trusting and not restricting the agent. The quantitative importance of these three motivational types is an empirical question. If the fraction of selfish or inequity averse agents is sufficiently high it pays to restrict, thereby ruling out the most opportunistic choices. If on the other hand there are relatively many distrust averse agents, a principal is better off trusting his agent. In the next section we will answer this question not only on an aggregate level. Since we employ the strategy method to elicit the agents’ decisions, we can directly identify all three motivational types and determine their relative frequencies.
4. Results In the results section we first present the agents’ behavior on the aggregate and on the individual level. We then turn to the behavior of the principals. 2
Formally, according to the model of Fehr and Schmidt (2000), inequity averse agents choose either x = 0 if β < 1/3 or x = 40 if β > 1/3. In the first case it pays to restrict, in the second case the agent’s decision is unaffected.
6
4.1 Agents Our first result shows that, on average, restricting the agent has an adverse effect on the motivation to choose a high x. Result 1: Agents’ choices of x are significantly higher if principals do not restrict. As a consequence, it does not pay to restrict. Support for Result 1 comes from Table 1 and Figure 1. Table 1 shows the average and median choices as well as the standard deviations of x if the principal restricts and otherwise. This table conveys a clear message. Both average and median values of x are higher in the no restriction case compared to the restriction case. On average, agents choose 31 percent higher values of x if they are not restricted by their principal and the median is twice as high. These differences are statistically significant at any conventional level (Wilcoxon signed rank test, p < .001). Thus, both in contrast to the standard economic model and to the model of inequity aversion, agents not only choose above the minimum, they also choose more if the required minimum is lower. Table 1: Agents’ offers dependent on the principals’ restriction decision Principal does restrict
Principal does not restrict
Average offer
17.5
23.0
Median offer
10
20
13.57
17.97
72
72
Standard deviation Number of observations Wilcoxon signed rank test
p < .001
Figure 2 shows how the x-choices are cumulatively distributed dependent on the decision of the principal.3 Obviously, the distribution of x is very different under the two conditions. Most notably, if the principals restrict their agents, the by far most frequently chosen x is 10 (dark bars ???). If restricted, 53 percent of the agents choose exactly the minimum x, and only about 20 percent choose x > 20. In sharp contrast only 32 percent choose x ≤ 10 in the no restriction condition and more than 47 percent of the agents choose x > 20. While only 7 percent of the agents choose the payoff equalizing level of x = 40 if restricted, 24 percent do 3
Figure 2 shows all observations for x ≤ 50. In both conditions there was one x above 50. In the restriction case this was x = 95, in the no restriction case it was x = 105. These two cases are summarized in Figure 2 as x > 50.
7
so if they are not restricted. Taken together, Figure 2 shows that while we see rather high xchoices from unrestricted agents, the mass of x-choices is centered at 10 if the principals do restrict. Figure 2: Cumulative distribution of x dependent on the restriction decision of the principal 1 cumulative frequency
0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0
5
10
15
20
25
30
35
40
45
50
x not restricted
restricted
The aggregated results shown in Table 1 and Figure 2 indicate that there are agents who reveal distrust aversion, i.e., who choose lower performance levels if restricted than otherwise. However, there must also be other types of agents given that 32 percent choose x ≤ 10 in the no restriction case. Since agents had to state their x-choices for both possible scenarios we are in the position to exactly determine the relative frequency and importance of the different types of behaviors. Result 2: There is a strong heterogeneity in agents’ motivation. We see selfish, inequity averse and distrust averse agents, the last group being the majority. Support for Result 2 comes from Figure 3 and Table 2. Figure 3 plots for each of the 72 agents his x-choice if restricted minus his x-choice if unrestricted. If all agents were selfishly motivated we should see that this difference is positive and close to ten for all agents. Figure 3
8
instead shows that there are three main types of behavioral patterns. For the first group of agents to the very left the effect of restricting is in fact positive as hypothesized by the standard principal-agent model. 25 percent of the agents belong to this group. For a second group of agents, the restriction decision seems to be irrelevant as they choose the same x under both scenarios. This group consists of 18 percent of all agents. Finally, for 57 percent of the agents restricting has a negative effect on their performance. They choose a lower x if restricted than otherwise. Figure 3 nicely summarizes the costs and benefits of the principal’s decision to restrict the agent. If agents were only of the first two behavioral types it would always pay to restrict. Either the agent is opportunistic and simply chooses the minimum performance level or he is inequity averse, in which case he does not care about the possible signal of distrust. Given the large fraction of agents who show a negative behavioral reaction to being restricted, however, it is obvious that the cost of restricting are nonzero and, what is even more important, outweigh its benefits. Figure 3: Costs and benefits of restricting: individual differences
x-choice if restricted minus x-choice if not restricted
15 10 5 0 -5 -10 -15
Effects of restricting positive: 25 percent neutral: 18 percent negative: 57 percent
-20 -25 -30 -35 1
4
7
10
13
16
19
22
25
28
31
34
37
40
agents
9
43
46
49
52
55
58
61
64
67
70
Figure 3 shows for each agent whether the performance response to being restricted is positive, neutral or negative. A closer inspection of the data allows an even more detailed classification (Table 2). In light of our discussion in Section 3 we classify an agent as selfish if he chooses 0 or 10 dependent on being restricted or not. 18 percent of the agents are of this type. A subject is called inequity averse if he chooses more than zero if not restricted, and not less if restricted. Among these agents, who represent 21 percent of the cases, the most agents choose in fact the same x in both scenarios, most often either 10 or 40. The largest group of agents, however, consists of distrust averse subjects who show a negative response to being restricted (see Figure 3). 57 percent reveal this behavioral pattern. They are responsible for why it does not pay to restrict on average. Their mean (median) choice of x if restricted is 18.7 (12) while it is 32.3 (30) if they can decide freely. Finally, there is a small class of three subjects, who do not fall in any of the three categories. These agents are not selfish according to our definition but choose more if they are restricted than otherwise.4
Table 2: Type distribution Selfish5
Inequity averse
Distrust averse
Other
Number
13
15
41
3
Percent
18
21
57
4
Mean x if restricted
10.1
21.0
18.7
16.3
Mean x if not restricted
0.0
20.3
32.3
8.3
So far we have assumed that agents who choose a lower x if restricted than otherwise do so for reasons of distrust aversion.6 Note that distrust aversion rests on two premises. First, agents must perceive the restriction of their choice set as a signal of distrust. Second, they have to negatively respond to this signal. To check whether these two considerations and thus the argument of distrust aversion actually plays a role, we asked all participants two questions after they had made their decisions in the experiment. The first question was, whether the restriction by the principal is perceived as a signal of distrust in the sense that the agent is not expected to choose x > 0. In the second question we asked whether they would choose a lower x if the principal considers them not to be trustworthy compared to if the principal 4
In Appendix 1 we report all choices of the agents together with how they are classified. One of the subjects classified as selfish actually chose x = 11 when being restricted and x = 0 if not. This is why the mean x if restricted is 10.1 and not 10.0. 6 Alternative explanations might be that the restriction decision of the principal either creates an anchor, makes is more difficult for the agent to look fair by choosing a high x, or signals a selfish principal. 5
10
considers them to be trustworthy.7 Figure 4 shows the distribution of answers to the two questions for all subjects who participated in the role of an agent in the experiment. 86 percent of the agents agree that the principal’s decision to restrict does signal distrust, only 14 percent do not agree. Likewise, 72 percent state that their performance is affected by how the principal perceives them. If they think the principal considers them as not being trustworthy they perform less than otherwise. 24 percent of the agents do not agree to this statement. Of course, the agreement to these two questions should be higher for those agents, whose behavior has been classified as distrust averse, than for the other types. This is in fact what we find. We constructed a dummy variable for an agent’s type (1 if distrust averse, 0 otherwise). Likewise we constructed dummy variables, which take the value 1 if the agent either ‘agrees’ or ‘fully agrees’ to the respective question and 0 otherwise. The Spearman rank correlation between distrust averse subjects and their support for the questions is positive and significant (p = 0.0651 for the first and p = 0.0000 for the second question, respectively; n = 72). Taken together the data support the relevance of trust as an important motivator to perform well. Figure 4: The important motive of distrust aversion (all subjects in the role of agents, n = 72)
relative frequency
0.6 0.5 0.4 0.3 0.2 0.1 0 Do not agree at Do not agree all
Don't know
Agree
Fully agree
Does restricting signal distrust? I transfer less if principal thinks I am not trustworthy compared to if he thinks I am trustworthy.
7
The exact reading of the two questions is: Question 1: Suppose you are participant of type A [agent]. If B [the principal] forces me to transfer at least 10, this obviously signals that he does not trust me to transfer a positive amount. Question 2: Suppose you are participant of type A [agent]. I transfer less if B [the principal] thinks I am not trustworthy compared to if he thinks I am trustworthy.
11
4.2. Principals We now turn to the principals. Given the average responses of the agents it is clearly not optimal for the principals to restrict their agents. However, if they do not restrict they risk getting nothing while they can guarantee at least an income of 20 if they do. The principals should therefore refrain from restricting only if they correctly anticipate the adverse effects of doing so. Our next result shows that this is in fact what most principals do. Result 3: The large majority (>70 percent) of the principals choose not to restrict. Result 3 is supported by the fact that 70.8 percent of the principals decide not to restrict while 29.2 percent do restrict. This difference is statistically significant at any conventional level (Binominal test). The data reveal that principals must have anticipated possible adverse effects of restricting their agents because they voluntarily give up the option to guarantee 20 points for themselves. In order to better understand the principals’ decisions we elicited their beliefs about the agent’s choice of x. Right after principals had made their decision in the experiment they were asked to state their expectation about the agent’s choice as well as how certain they are about this expectation.8 As it turns out principals who restrict expect lower offers compared to principals who do not restrict. The mean (median) expectation about x for those who restrict is 19.4 (10) while 25.7 (25) for those who don’t. The differences in beliefs are statistically significant (Mann-Whitney test, p = 0.0046).9 Figure 5 shows the cumulative distribution of principals’ beliefs with regard to x.10 Comparing the agents’ actual choices displayed in Figure 2 with the principals’ expectations in Figure 5 reveals that expectations were indeed quite accurate. Note in particular that 57 percent of the principals who restrict expect to get x = 10, and the actual number, 53 percent, is very close to that. In the no-restriction case principals are somewhat too optimistic. For example, they expect the equal split in 39 percent of the cases but agents choose x = 40 only in 24 percent of the cases. However, beliefs and actions are not statistically different from each other. A Mann-Whitney test does not reject that beliefs and actions are the same both if principals do not restrict (p = 0.2025) and if principals do restrict (p = 0.7603). 8
Certainty ranged from 1 (very uncertain) to 7 (very certain). In addition, principals who restrict are more certain about the expected performance of the agent than principals who do not restrict. The mean (median) value is 4.8 (5) if the principal restricts and 3.6 (3) otherwise. 10 Figure 4 shows all observations for x ≤ 50. In the restriction condition one principal stated an expected x above 50, x = 120. This is displayed as x > 50. 9
12
cumulative frequency
Figure 5: Cumulative distribution of principals’ belief about x, dependent on their decision to restrict or not to restrict 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0
5
10
15
20
25
30
35
40
45
50
x not restricted
restricted
The beliefs of the principals and the actual performance of the agents in our experiment nicely illustrate, what may be called, the “self-fulfilling property” of distrust.11 Principals who have rather pessimistic beliefs and choose to restrict, experience that their beliefs are indeed confirmed by the relatively low average performance of their agents. On the other hand, principals who have more optimistic beliefs and trust that the agent will perform above the minimum level, learn that on average agents choose a relatively high performance. In this sense, different and reinforcing “firm cultures” may emerge. In low trust firms, managers have little trust in their employees and predominantly rely on control and explicit incentives. These managers will not be surprised to see performance at the minimum, confirming their pessimistic beliefs. Following David Packard’s quotation at the beginning of the paper, the locked storeroom policy of General Electric in the 1930’s is a perfect example for such firm culture. High trust firms on the other hand are governed by the expectation of mutual trust. In these firms employees are trusted and given responsibility. This “empowerment” of agents actually produces non-minimal performance, reassuring the initial beliefs held by the managers.12 A prototypical example for this firm culture is Hewlett Packard having built the so-called “HP way”.
11
Cf. Luhmann (1968). The stability of these prototypical firm cultures is probably different. Low trust firms will encounter deviations between expectations and actual behavior to a lesser extent than high trust firms. Remember that in case the principals did not restrict they were somewhat too optimistic (Figure 5). This implies that some principals who
12
13
The coexistence of these two different belief cultures is confirmed by the questionnaire data of the principals in our experiment, as well.13 Note that according to the argument above, principals who restrict their agents should not consider distrust aversion as a relevant phenomenon. They should neither think that the decision to restrict signals distrust nor should they believe that such signals would alter agents’ behavior in an undesired way. As before we constructed a dummy variable for a principal’s type (1 if principal does restrict, 0 otherwise). Likewise we constructed dummy variables which take the value 1 if the respective principal ‘agrees’ or ‘fully agrees’ to the respective question and 0 otherwise. As hypothesized, the Spearman rank correlation between the principals’ types and the support for the questions is in fact negative and significant for both questions (p = 0.0021 for the first and p = 0.0006 for the second question, respectively; n = 72). 5. Distrust aversion at the workplace: Questionnaire results In principal-agent relationships the possibility for trust or distrust shows up in many forms. The principal can either trust his agent or he can double-check the agent’s word, can control his performance, write explicit performance contracts, or restrict his choice set by locking store- and equipment-rooms, for example. In our stylized experimental environment we have shown that signals of distrust may adversely affect an agent’s motivation and consequently his performance. This view is also supported by a set of questions we asked all subjects at the very end of the experiment. They were asked to consider being an agent in three different work situations. For each of the situations they were presented two conditions, one where the principal trusts the agent, one where he doesn’t. For both conditions we asked subjects to indicate their work motivation.14 In the first hypothetical situation the agent works in a supermarket and is responsible for checking the balances in the cash registers. In principal he could easily take out money for himself but he is assumed not to do so. In the first condition (no trust) the principal checks what the agent did; in the second condition the principal does not check behavior (trust). In the second situation the agent just started a new job and is explained what he is expected to do. In the first condition the agent has to sign a formal agreement about his working time (no have been disappointed may adjust their beliefs and choose to restrict in the future. It would be interesting to study the stability of firm cultures in a repeated interaction framework. 13 Recall, that all participants in the experiment had to answer the two questions discussed in Figure 4. Those who participated in the role of principals were asked to imagine being in the role of an agent when answering the two questions. 14 The full description of the situations and scenarios is given in the Appendix 2.
14
trust) in the second he is asked to exactly meet his working time obligation (trust). Finally, in the third situation the agent gives a job interview and reports on his qualifications and work experience. The agent offers his previous employer as a reference, who can verify the agent’s statements. In the first condition the new employer believes what the agent said and hires him (trust); in the second he hires him only after having consulted the previous employer (no trust). In Table 3 we display how subjects indicated their work motivation in all six conditions. Table 3: “How high is your work motivation?” (n = 144)
Work motivation (relative frequencies) very low Low Middle High very high
Situation 1 checking cash registers at supermarket Manager Manager controls does not (no trust) control (trust) 0.00 0.00 0.13 0.58 0.29
0.10 0.37 0.33 0.14 0.06
Situation 2 working times at new job
Situation 3 job interview
Asked to meet obligations (trust)
Formal agreement (no trust)
Employer believes statements (trust)
Employer consults reference (no trust)
0.01 0.09 0.35 0.47 0.08
0.01 0.24 0.44 0.27 0.03
0.00 0.00 0.03 0.47 0.50
0.01 0.08 0.36 0.40 0.15
The results in Table 3 resemble our findings reported in Table 1 and Figure 2. The agents’ motivation is negatively affected by signals of distrust. For all three situations the relative frequencies of agents indicating to have a high or very high work motivation are always lower in the no trust than in the trust scenario. For all three situations the differences in work motivation between the trust and the no trust scenarios are significant at any conventional level (Wilcoxon signed rank tests, p = 0.0000 for all three comparisons). Unlike our results from the experiment, however, we cannot conclude from Table 3 that it pays for the principal to trust his agent. Maybe he is better off controlling the agent, fixing requirements in an enforceable way and consulting the agent’s former employer. What the results do show, however, is that there is a cost in doing so: the agent’s motivation to perform well is lower. 6. Discussion In this paper we analyze the interaction of trust and motivation in a principal-agent relationship. We show that a majority of agents reveals so-called distrust averse behavior, i.e.,
15
they are less motivated to perform well if the principal forces them to deliver a minimum level of performance. The main mechanism behind this behavior is the principal’s signal of distrust in the voluntary performance of the agent that is generated by the decision to restrict the agent’s choice set. Agents who are distrust averse consider this signal as negative and respond reciprocally by reducing their performance. The presence of distrust averse agents generates substantial costs for the restriction decision of the principal. On average, a principal in our experiment earns 31 percent less if he restricts the agent than if he does not restrict. These costs are absent in any traditional economic model that assumes either selfish or inequity averse behavior on the part of the agent. Not surprisingly therefore, the subgame perfect equilibrium prediction of the traditional model yields the complete opposite of what the large majority of the principals in our experiment do. While the traditional model predicts that all principals choose to restrict, almost 71 percent of the principals choose not to restrict thereby earning significantly higher payoffs. Our results provide another example of the possible dysfunctional effects of explicit incentives. Previous studies have demonstrated such effects, showing how the introduction of explicit incentives in the form of rewards or punishments may reduce rather than increase the desired behavior (Gneezy and Rustichini 2000a,b, Fehr and Gächter 2002, Fehr and Rockenbach 2003). These studies are also related to the somewhat older psychological literature on so-called “extrinsic and intrinsic motivation” (Deci 1971, Lepper and Green 1978, Sansone and Harackiewicz 2000). See Frey and Jegen (2001) for a comprehensive overview of the economic aspects of this literature. While our results are consistent with the findings in these studies, the main approach of our analysis is different. First, the principals in our experiment have a new and so far unexplored decision possibility. Rather than specifying punishments or rewards inducing incentives ex-post, in our set-up a principal can determine ex-ante the choice set of the agent by fixing a positive minimum performance level. Such possibilities are common in the design of real-life principal-agent relationships (regulated working times, high-control working environments, minimum output/quality, etc.) but, to the best of our knowledge, have not yet been explored in an economic analysis. Second, our experimental design allows us to obtain detailed individual-level information on the behavior of the agents rather than only aggregate results. In particular, we can distinguish between three main different behavioral types (selfish, inequity averse, and distrust averse agents) and we can determine the relative frequency, and hence behavioral importance, of each type. Finally and perhaps most importantly, our study emphasizes a new mechanism for possible
16
dysfunctional effects of incentives, i.e., distrust. Previous studies have shown that incentives may undermine motivation because they provide new information regarding the importance or the cost of the task (Gneezy and Rustichini 2000b, Benabou and Tirole 2003), because they insult the agent (Gneezy and Rustichini 2000a), or because they are in conflict with social norms of fairness and cooperation (Fehr and Gächter 2002, Fehr and Rockenbach 2003). Our study shows that distrust on the part of the principal represents an additional mechanism. By introducing explicit incentives, the principal signals that he does not trust the agent to perform otherwise. If the agent is distrust averse, he regards this signal as negative and responds reciprocally by reducing his voluntary performance. Our study does not say, of course, that explicit incentives will never work. Quite to the contrary, if incentives are strong enough, they will increase an agent’s performance also in our model. It is easy to imagine, e.g., how the costs and benefits of restricting the agent’s choice set change if the principal can enforce not only a performance level of 10 but, say, 20 or 50 (cf. Figure 3). Since only few agents may be willing to choose 50 voluntarily, a restriction to x ≥ 50 will clearly increase the principal’s payoff. Thus, average performance and the strength of incentives will form a non-monotonic relationship (cf. Gneezy 2004). However, in many situations strong incentives are not available. For example, a principal can regulate the working time of an agent to a fixed schedule from 9am to 5pm, but it is usually not possible to force agents to work, say, at least 12 hours per day. Nevertheless, many agents may be willing to work some extra time if necessary. What our results say is that in these situations it may be better not to use explicit incentives at all and to trust that the agent will work 8 hours per day (and sometimes even more than that), than to use weak incentives by forcing them to a fixed schedule from 9am to 5pm. In consequence, our paper also contributes to the literature on incomplete contracts. According to traditional economic theory, the incompleteness of many real-life economic contracts presents puzzle. Contracts often fail to include important, verifiable obligations of the contracting parties (or they do so only in vague terms), and they often make actions less sensitive to verifiable information that would be optimal. While bounded rationality on the part of the contracting parties might account for some of this incompleteness, the key question remains why many contracts appear to be left deliberately incomplete. Several explanations have been offered by now. For example, in some situations incomplete contracts may be completed by renegotiation design to achieve first best outcomes, so there is no need to write more complicated contracts (Aghion, Dewatripont, and Rey 1994, Nöldeke and Schmidt
17
1995). In other cases, complete contracts are of no value because they are not able to improve the situation (Segal 1999, Hart and Moore 1999). Finally, incomplete contracts may actually be superior to more complete contracts. Allen and Gale (1992) and Spier (1992), e.g., argue that the offer of a more complete contract may lead the other party to draw negative inferences about the first party’s type. Bernheim and Whiston (1998) show in a repeated game setting that if some obligations are non-contractible it may be better to leave other aspects unspecified, giving rise to so-called “strategic ambiguity” in the design of the contract. Our paper provides a different rationale for the observed incompleteness of many reallife contracts. In particular, we take for granted that agents are heterogeneous and that many agents have non-selfish preferences, in the sense that they are generally willing to perform above minimum even if this is costly to themselves. However, since a large majority of these agents reveal distrust averse behavior, i.e., perform less if the principal signals his distrust in their voluntary performance, the average induced performance negatively relates to the completeness of the contract offered by the principal. In consequence, an incomplete contract, where the agent has full discretion, outperforms and leads to a higher payoff to the principal than a more complete contract, where the principal eliminates the agent’s most opportunistic choices.
18
References Aghion, P., Dewatripont, M., and Rey, P. (1994) “Renegotiation design with unverifiable information,” Econometrica 62, 257-282. Allen, F. and Gale, D. (1992) “Measurement distortion and missing contingencies in optimal contracts,” Economic Theory 2, 1-26. Benabou, R. and Tirole, J. (2003) “Intrinsic and extrinsic motivation,” Review of Economic Studies 70, 489-520. Bernheim, B.D. and Whinston, M.D. (1998) “Incomplete contracts and strategic ambiguity,” American Economic Review 88, 902-932. Bolton, G.E. and Ockenfels, A. (2000) “A theory of equity, reciprocity, and competition,” American Economic Review 100, 166-193. Deci, E. (1971) “Effects of externally mediated rewards on intrinsic motivation,” Journal of Personality and Social Psychology 18, 105-115. Fehr, E. and Gächter, S. (2002) “Do incentive contracts undermine voluntary cooperation,” Institute for Empirical Research in Economics, Working Paper 34. Fehr, E. and Rockenbach, B. (2002) “Detrimental effects of sanctions on human altruism,” Nature 422, 15 March 2002, 137-140. Frey, B.S. (1993) “Does monitoring increase work effort? The rivalry with trust and loyalty,” Economic Inquiry 31, 663-670. Frey, B.S. and Jegen, R. (2001) “Motivation crowding theory,” Journal of Economic Surveys 15, 5, 589-611. Gneezy, U. and Rustichini, A. (2000a) “A fine is a price,” Journal of Legal Studies 29, 1–17. Gneezy, U. and Rustichini, A. (2000b) “Pay enough or don’t pay at all,” Quarterly Journal of Economics 115 (2), 791–810. Gneezy, U. (2004) “The W effect of incentives”, mimeo, The University of Chicago Graduate School of Business. Hart, O. and Moore, J. (1999) “Foundations of incomplete contracts,” Review of Economic Studies 66, 115-138. Lepper, M. R. and Greene, D. (1978) The hidden costs of reward: new perspectives in the psychology of human motivation, Lawrence Elbaum Associates, John Wiley and Sons, NJ. Luhmann, N. (1968) Vertrauen, 4. Auflage, 2000, Lucius & Lucius, Stuttgart. Nöldeke, G. and Schmidt, K.M. (1995) “Option contracts and renegotiation: a solution to the hold-up problem,” RAND Journal of Economics 26, 163-179. Packard, D. (1995) The HP way, HarperCollins, NY. Sansone, C. and Harackiewicz, J.M. (2000) Intrinsic and Extrinsic Motivation, Academic Press. Segal, I. (1999) “Complexity and renegotiation: a foundation for incomplete contracts,” Review of Economic Studies 66, 57-82. Spier, K. (1992) “Incomplete contracts in a model with adverse selection and exogenous costs of enforcement,” RAND Journal of Economics 23, 432-443.
19
Appendix 1: Table A1: Agents’ x-choices if choice restricted and if not restricted agent
x choice if restricted
x choice if not restricted
type classification
1 2
11 10
0 0
selfish selfish
3
10
0
selfish
4
10
0
selfish
5
10
0
selfish
6
10
0
selfish
7
10
0
selfish
8
10
0
selfish
9
10
0
selfish
10
10
0
selfish
11
10
0
selfish
12
10
0
selfish
13
10
0
selfish
14
10
5
inequity averse
15
10
5
inequity averse
16
10
10
inequity averse
17
10
10
inequity averse
18
10
10
inequity averse
19
10
10
inequity averse
20
10
10
inequity averse
21
12
12
inequity averse
22
18
18
inequity averse
23
20
20
inequity averse
24
35
35
inequity averse
25
40
40
inequity averse
26
40
40
inequity averse
27
40
40
inequity averse
28
40
40
inequity averse
29
15
17
distrust averse
30
18
20
distrust averse
31
18
20
distrust averse
32
29
31
distrust averse
33
14
18
distrust averse
34
10
15
distrust averse
35
35
40
distrust averse
36
35
40
distrust averse
37
35
40
distrust averse
38
12
20
distrust averse
39
10
20
distrust averse
40
10
20
distrust averse
41
10
20
distrust averse
42
10
20
distrust averse
43
10
20
distrust averse
44
10
20
distrust averse
45
20
30
distrust averse
46
30
40
distrust averse
47
30
40
distrust averse
48
30
40
distrust averse
49
35
45
distrust averse
20
Table A1 contd. agent
x choice if restricted
x choice if not restricted
type classification
50 51
40 95
50 105
distrust averse distrust averse
52
13
24
distrust averse
53
10
25
distrust averse
54
10
25
distrust averse
55
10
25
distrust averse
56
15
30
distrust averse
57
12
30
distrust averse
58
10
30
distrust averse
59
10
30
distrust averse
60
10
30
distrust averse
61
10
30
distrust averse
62
15
35
distrust averse
63
20
40
distrust averse
64
11
40
distrust averse
65
10
40
distrust averse
66
10
40
distrust averse
67
10
40
distrust averse
68
10
40
distrust averse
69
10
40
distrust averse
70
20
10
other
71
14
7
other
72
15
8
other
21
Appendix 2: A full description of the situations and conditions. Situation 1 You began a new vacation job in a supermarket. Your task is to check the balances in the cash registers in the evening, meaning that you examine whether the amounts of money in the cash registers correspond with the entries. In theory, you could easily swindle the supermarket by simply removing money from the cash register. Condition 1: You examined the cash registers conscientiously and without cheating, and reported the results honestly. You realize on the way home that you forgot your umbrella. When you enter the supermarket, you see that the manager is again examining the amounts in the cash registers, as if he undoubtedly mistrusts you. Condition 2: You examined the cash registers conscientiously and without cheating, and reported the results honestly. You realize on the way home that you forgot your umbrella. When you enter the supermarket, you meet the manager, who is leaving for the day. He wishes you a pleasant evening. Situation 2 You have a new job. Your boss explains your tasks to you as well as the amount of work expected of you. Condition 1: Before starting your work, you have to sign a binding agreement. This defines your working times exactly. Condition 2: Your boss asks you to follow the work times exactly. Situation 3 During a job interview, you presented your knowledge, experience, and qualifications truthfully. You provide your previous employer as a reference who could confirm your information. Condition 1: The new employer believes your information and hires you. Condition 2: Your new employer does not hire you until he has gathered information about you from your previous employer and confirmed the accuracy of your information.
22
