Charity, incentives, and performance Oege Dijka and Martin Holménb

February 2015

Abstract We propose that donating profits to charity may improve firm performance through reduced moral hazard and increased effort in incomplete contract environments. This proposition is tested and confirmed in an incomplete contract principal-agent laboratory experiment where principals’ profits are donated to charity. The results show that both principals and agents have higher earnings in treatments where principals are working on behalf of a charity. Only in the charity treatments do agents respond positively to the effort levels suggested by the principals, and do higher requested levels of effort result in higher principal earnings. JEL: D01; L33; M52 Keywords: Charity, Incomplete contracts, Experiments ___________________________________________

We are grateful for helpful comments and suggestions by Martin Dufwenberg, Michael Kirchler, Gregory C. Leo, Steen Thomsen, and conference participants at Western Economic Association Meeting 2012, the summer workshop at Copenhagen Business School 2012, the Nordic Behavioral Economics conference 2012, the Nordic Corporate Governance Network meeting 2012, and seminar participants at University of Rome "Tor Vergata” and University of Stavanger. Financial support by VINNOVA (grant 2010-02449) and Jan Wallander and Tom Hedelius Foundation is gratefully acknowledged. a University of Gothenburg, Centre for Finance and Department of Economics, Box 640, 405 30 Gothenburg, Sweden and Radboud University Nijmegen, P.O. Box 9108, 6500 HK Nijmegen, E-mail: [email protected] b Corresponding author. University of Gothenburg, Centre for Finance and Department of Economics, Box 640, 405 30 Gothenburg, Sweden. Phone: +46 31 786 6442, E-mail: [email protected]

Charity, incentives, and performance

February 2015

Abstract We propose that donating profits to charity may improve firm performance through reduced moral hazard and increased effort in incomplete contract environments. This proposition is tested and confirmed in an incomplete contract principal-agent laboratory experiment where principals’ profits are donated to charity. The results show that both principals and agents have higher earnings in treatments where principals are working on behalf of a charity. Only in the charity treatments do agents respond positively to the effort levels suggested by the principals, and do higher requested levels of effort result in higher principal earnings. JEL: D01; L33; M52 Keywords: Charity, Incomplete contracts, Experiments

1    

1. INTRODUCTION Advocates of Corporate Social Responsibility (CSR) argue that being a good corporate citizen, e.g. by donations to charity, can also make financial sense, and indeed there is some evidence that CSR is correlated with better financial performance (see e.g. Smith, 1994; Margolis et al., 2007; Edmans, 2012). However, the exact mechanism through which CSR and donations to charity may improve financial performance is not well understood. Kitzmueller and Shimshak (2012) point out that donations to charity, and CSR in general, could increase performance due to interactions with both the product market and the labour market. In the labour market having a reputation for being a good corporate citizen could 1) aid in recruitment and screening of new employees (e.g. Greening and Turban, 2000) and 2) reduce moral hazard in incomplete contract environments (Bowles et al, 2001). On the latter point, Benabou and Tirole (2010) point out that in environments where firms can renege on implicit contracts and promises with their labor force, and where employees can shirk at noncontractible parts of their jobs, having a reputation for being socially responsible may increase trust, lessen moral hazard, and increase performance. Krueger and Mas’ (2004) results corroborate these arguments. They analyze the reduced quality of produced tires at Bridgestone/ Firestone following labour strife. In July 1994 Bridgestone/ Firestone announced that they would deviate from the industrywide tradition by moving from an eightto a 12-hour shift. They would also cut pay for new hires by 30 percent. The company was breaking implicit contracts with the employees and the conflict was not resolved until December 1996. Krueger and Mas (2004) document significantly higher defect rates for tires manufactured during the dispute. This paper uses experiments to test the hypothesis that donations to charity reduce moral hazard and improves performance in incomplete contracts environments. Our experimental design will address the following research questions: i) Do agents respond differently to the contracts offered by charity principals? ii) Do charity principals make different contract offers? iii) Is the overall efficiency in an incomplete contract framework different when the principal donates to charity? iv) How are any gains associated with a charity principal distributed among the agent and the principal? Our experimental setup closely follows Fehr et al. (2007). The principal makes a take it or leave it offer to the agent. The principal decides whether the contract either consists of i) a fixed wage 2    

independent of effort, ii) a fixed wage, a bonus, and an expected effort level, or iii) a fixed wage, a required effort level and a fine that is levied with a certain probability if the agent shirks, i.e. does not put in sufficient effort. In some of our experimental treatments we let the principals’ earnings go to charity, in our case the Swedish Red Cross. However, this introduced both a charity component and the fact that the subject making decisions as principal no longer had a personal profit motive. To distinguish these two effects, four treatments were run: i) a pure charity treatment where the principal’s earnings went to the Red Cross, ii) an added charity treatment where the principal’s profits were both paid out to the subject and donated to the Red Cross, iii) a non-charity non-self interested principal treatment where the principal’s earnings were paid out to another subject participating in the experiment, and iv) a baseline purely self-interested principal treatment where the principal’s earnings were paid out directly to the subject acting as principal. Our results show that efficiency is highest in the pure charity treatment. Furthermore, both the principals and the agents are better off when the principals are working on behalf of a charity. The two-by-two design lets us distinguish between the effect of charity per se (the two charity treatments) and the effect of having a non-self interested principal (the pure charity treatment and the treatment where the principal’s earnings were paid out to another subject participating in the experiment). The increase in earnings for the principal is mainly due to the charity aspect, whereas the increase in earnings for the agent is mainly due to a non-self-interested principal. One mechanism through which donations to charity could increase performance is through higher trust levels. With incomplete contracts employers need to have some trust that their employees actually put in effort, and employees need to have some trust that their employers will actually reciprocate and reward their effort (see e.g. La Porta et al., 1997). An employer that shows a lack of trust could demotivate employees, who respond to this lack of trust by shirking. However, our results do not suggest that that the efficiency improvements associated with charity are driven by higher trust (-worthiness). The results of two Trust Games we ran at the beginning and end of the experiment do not reveal any higher trustworthiness of agents in the Charity treatments. Furthermore, if trust per se was the main mechanism we would expect to see higher fixed salaries and more non-incentivized 3    

‘’trust contracts’’ instead of incentivized bonus contracts in the charity treatment. We do not observe such differences for the contracts in the charity treatments. Another potential mechanism through which donations to charity could increase performance is through increased reciprocity and (indirect) guilt-aversion (see e.g. Rabin, 1993; Fehr et al., 1997; Charness and Dufwenberg, 2006). Miettinen (2011) shows in a theoretical model that guilt may affect the optimal contract when the agent feels bad when not reaching the target level set in the contract. Better performance can then be implemented with lower risk and the solution is closer to first-best. Furthermore, some of the higher payoff to the principal is shared with the agent, i.e. both the principal and the agent are better off. And we indeed find that it is only in the charity treatments that the communication of higher effort expectations actually leads to higher effort and higher earnings. One interpretation of this result is that when an agent is indirectly working for a charity, there is a bigger psychological cost, i.e. more guilt, to "letting down" the principal. Another interpretation is that the agents simply have a pure preference for donating to charity (Gregg et al., 2011). For charity-motivated agents the principal’s expectations work as a fundraising solicitation (Yoruk, 2009; Edwards and List, 2013). We think our results have two main implications. First, they provide additional evidence that donations to charity may improve firms’ financial performance. Second, the results suggest that one mechanism by which donations to charity may improve performance is through more efficient contracting in incomplete contract environments typical within firms. The rest of the paper is organized as follows. The next section presents our experimental setup. The results are presented in section 3 and section 4 concludes.

2. THE EXPERIMENT 2.1. Setup of the Experiment We closely follow the setup of Fehr et al. (2007). Where we differ from their paper is first that, in order to increase the number of observations, we let subjects play both the role of principal and agent, and randomly pay out only one of these roles. Second we allow principals to choose from offering either trust contracts, bonus contracts or fine contracts, whereas in the original paper principals could 4    

typically only choose among two of these contract types in any given treatment. Finally we computerized the experiment with z-Tree (Fishbacher, 2007) in order to speed up the experiment. The experiment lasts ten rounds. Each round a principal is matched with a random, anonymous and unique agent. The principal makes a wage offer to the agent, who can then either accept or reject the offer. If the agent rejects both the principal and the agent earn zero. The offer can be of three types: i) A Trust Offer consisting of a fixed wage and a suggested level of effort. The principal pays the fixed wage independent of effort level chosen by the agent. ii) A Bonus Offer consisting of a fixed wage, a suggested level of effort and a promised bonus. The agent selects her level of effort and this is observed by the principal. Only after the principal observes the actual effort exerted by the agent, does the principal decide on the actual size of the bonus to be paid. iii) A Fine Offer, consisting of a fixed wage, an expected level of effort and a fine that is levied with 33% probability in case the agent shirks, and provides less effort than stipulated in the contract. After observing the details of the offer, the agent decides on effort level (e). The earnings of the agent consist of the tokens paid by the principal (thus fixed wage plus perhaps a bonus minus perhaps a fine) minus the cost of effort given by table 1. Table 1. Cost of effort in tokens to the agent.

e c(e)

1 0

2 1

3 2

4 4

5 6

6 8

7 10

8 13

9 16

10 20

The profit for the principal (p(e)) consists of the production from the effort by the agent (see table 2), minus the wages and bonuses paid to the agent plus perhaps fines paid by the agent. Table 2. Gross profit of effort for the principal.

e p(e)

1 10

2 20

3 30

4 40

5 50

6 60

7 70

8 80

9 90

10 100

Note that the equilibrium level of effort is zero for both trust and bonus offers. Positive incentivecompatible effort can only be maintained with the fine offer. For details we refer to the discussion in Fehr et al. (2007).

5    

Each round, subjects played both the principal role and the agent role, but in both roles were matched with a different principal and a different agent every round. Earnings thus accrued in a principal earnings account and an agent earnings account. At the start of the experiment, each account contained 125 tokens. At the end of the experiment, tokens were exchanged into Swedish Kronor (SEK) at a one to one rate and were paid out to the subject. Donations to the Red Cross were added up and transferred by bank wire to the organization. Proof of the donation was displayed on an announcement board outside the experimental lab. The SEK/Euro exchange rate was about 0.11 at the time of the experiment, such that each token was worth approximately €0.11.

2.2. Treatments Four treatments were run in order to disentangle the charity effect from the effect of having a non-self-interested principal. T1: The Pure Charity Treatment: In this treatment the earnings of the principal account were donated to the Swedish Red Cross (Röda Korset), whereas a subject would be paid the earnings in his or her agent account. T2: The Added Charity Treatment: In this treatment the subject either earns her principal or agent account both with 50% probability. In case the principal earnings are selected an equivalent amount was donated to charity.1 T3: The Other Treatment: In this treatment the subject would either earn his or her agent earnings or the principal earnings of another subject in the room. Thus as in the Pure Charity Treatment the decisions of the principals in this treatment would not directly affect their own payoffs. T4: The Self Treatment: This treatment is the baseline treatment, equivalent to Fehr et al (2007), where the subject either earns her principal or agent earnings, both with 50% probability. Thus, except in the Pure Charity Treatment one of the two accounts, agent or principal, would be randomly selected for payout to the subject. In the Pure Charity Treatment, the agent account is always selected.                                                                                                                         1

 The  principal  earnings  reported  in  the  results  section  will  refer  to  the  earnings  of  the  principal  in  the  game,  so   not  the  total  earnings  of  the  subject  and  the  charity  taken  together.    

6    

Oege Dijk 24/2/2015 13:47 Formatted: English (US)

The two-by-two design distinguishes between the effect of charity and the effect of having a non-self interested principal, respectively, and is outlined in table 3.

Table 3. Treatments

Principal does not earn own payoff Principal earns own payoff

Principal’s payoff donated to charity Charity

Principal’s payoff not donated to charity Other

Added Charity

Self

All subjects received the same initial instructions, corresponding to the Self Treatment, and only after reading the instructions were they informed about the particular setup of that treatment through an overhead slide shown in the laboratory (see Appendix A2). The treatment conditions were the same for all 10 periods of the experiment. All subjects only participated in one of the four treatments.

2.3. Trust Game In order to further investigate the mechanism behind any effect of charity donations on contracts, effort and earnings, we also run two Trust Games (Berg et al., 1995), both at the beginning and the end of the experiment. This way we can measure abstract trust and trustworthiness levels in the different treatments in a way that is more straight-forward to analyse and separate from the principal-agent game itself. Repeating the trust games once before the experiment and once at the end of the experiment allows us to track whether trust levels change over the course of the experiment as subjects learn about the behaviour of others. For these Trust Games, principals received an additional 10 tokens at the beginning of round 1 and round 10 of which they could send 0, 2, 4, 6, 8 or 10 tokens to their agent. The amount of tokens sent were then tripled, and the agent was then asked using the strategy method to decide how many tokens to send back to the principals in case they had received 6, 12, 18, 24 or 30 tokens. The results of the Trust Games were only announced after period 10 had ended, and the earnings were then added to the respective principal and agent accounts. 7    

2.4. Experimental Implementation The experiment was programmed with the help of the z-Tree software package (Fischbacher, 2007), and subjects were recruited through the online recruitment system ORSEE (Greiner, 2004). The experimental sessions were run in December 2011 and April 2012 at the University of Gothenburg with bachelor and master students from various disciplines. Three sessions were run for each treatment. Overall 140 subjects showed up for the experiments. Due to different show-up rates the total number of subjects varies between treatments. The average earnings were SEK 270 (the equivalent of roughly Euro 30) and in total SEK 16510 were donated to Charity. Including time for instructions and final payment, the experiment lasted about two hours.

3. RESULTS 3.1. Earnings We first look at the earnings in the principal- and agent-accounts across different treatments. Rejections of offers were extremely rare (less than 1% of the offers) in all four treatments. Earnings for both principals and agents are significantly higher in the pure charity treatment than in the baseline self-interested principal treatment (see figure 1 and table 4). Thus it is not the case that the principals used their charity status to increase just their performance: agents are better off as well. Principal earnings are on average 5 tokens higher in the pure charity treatment (27.8 vs 22.9). Agent earnings were 6 tokens higher on average as well (26.4 vs 20.4 tokens). Principal earnings in the Added Charity treatments were also higher than the baseline, but not significantly so, and principal earnings in the Other was even slightly lower than the baseline, although again not significantly. Agent earnings in the Other treatment however were significantly higher than in the baseline Self Treatment. Figure 1 here

8    

Table 4. Earnings Overall In this table we report Wilcoxon signed rank tests on subject’s average earnings over 10 periods, where T1, T2, and T3 are compared to the baseline Self treatment (T4). Standard deviations are given in parentheses. N signifies the number of subjects in each treatment. *** Significant at the 1% level; ** Significant at the 5% level; * Significant at the 10% level.

Principal Agent N

T1: Pure Charity

T2: Charity

27.8 (12.3)* 26.4 (4.9) *** 36

24.3 (15.0) 22.4 (5.1) 40

Added T3: Other 20.8 (13.9) 24.5 (4.8) ** 30

T4: Self 22.9 (12.3) 20.4 (7.5) 34

This difference in earnings is not caused by a difference in the type of offers made. Indeed, figure 2 shows that the distribution of offer types is very similar across treatments with a clear preference for the bonus offer similar to Fehr et al. (2007). A Kolmogorov test finds no significant differences between treatments (p>0.7). Thus it seems that the charity manipulation does not result in a preference for more trusting contract types (Trust and Bonus). Clearly the increased earnings do not stem from different types of offers, but rather the details of the offers themselves. Figure 2 here

3.2. Bonus offers As most offers were of the bonus type (see figure 2), we will mainly analyze these to understand why earnings were higher in the charity treatment. Almost all observations of trust offers and fine offers were in the first few rounds. As subjects gained more experience they settled on the bonus offer type. Fehr et al. (2007) also find that the fraction of bonus (trust) offers increases (decreases) over time. The bonus offer consisted of a fixed wage, an expected level of effort and a promise of a bonus. Only after the agent had selected her level of effort, the principal could determine the actual size of the bonus to be paid.

9    

Table 5. Earnings in Bonus contracts In this table we report Wilcoxon signed rank tests over the average earnings in periods in which the subjects selected a bonus contract (principal) or received a bonus contract (agent), where T1, T2, and T3 are compared to the baseline Self treatment (T4). Standard deviations are given in parentheses. N signifies the number of subjects in each treatment. *** Significant at the 1% level; ** Significant at the 5% level; * Significant at the 10% level.

Principal Agent N

T1: Pure Charity

T2: Charity

31.6 (10.9)** 26.8 (5.6)*** 36

27.5 (15.8) 20.8 (5.4) 40

Added T3: Other 23.4 (14.3) 23.4 (4.5)** 30

T4: Self 24.3 (14.1) 16.9 (8.0) 34

When focusing on the Bonus contracts the pattern is the same as for the total sample: Principal earnings are highest in the Pure Charity Treatment, lowest in the Other treatment, with the Added Charity and the Self Treatments falling somewhere in between (see table 5). When we assign a Charity dummy to all bonus contract observations in the Pure Charity and Added Charity treatments, and a Not-Self dummy to all the observations in the Pure Charity and the Other treatments, and regress the earnings on these dummies we see that it is mainly Charity aspect and not the Not-Self aspect that contributes to the increase in earnings for the principal (see table 6, model 1). The increase in earnings for the agent is on the other hand more driven by Not-Self than by Charity (table 6, model 2). Table 6. Determinants of earnings with bonus contracts In this table we report OLS-regressions with Principal Earnings (model 1) and Agent Earnings (model 2) in the bonus contracts as dependent variables. Charity is equal to one for T1 and T2 and zero otherwise. Not-Self is equal to one for T1 and T3, and zero otherwise. Coefficients are reported with standard errors in parentheses. Standard errors are clustered on subject. *** Significant at the 1% level; ** Significant at the 5% level; * Significant at the 10% level.

Constant Charity Not-Self

Principal Earnings (1) 25.4*** (1.8) 4.5** (2.2) 1.5 (2.2) 10  

 

Agent Earnings (2) 21.3*** (1.1) 1.5 (1.0) 3.7*** (1.0)

R2 N

0.01 1023

0.02 1023

As can be seen in table 7, there are some interesting differences between the bonus offers in the four treatments. The most salient of these is that principals in both the Charity treatments and the Other treatment have significantly higher effort expectations than the principals in the Self treatment, and also promise significantly higher bonuses. The principals’ Expected effort in the Charity treatment is on average 8.6 (out of a maximum of 10). In the Self treatment the principals are less demanding and only request 7.3 units of effort on average, and the Added Charity and Other treatments fall in between. Table 7. Average bonus offer characteristics In this table we report the average offer characteristics, as well as effort and actual bonus choices, across the different treatments (with standard deviations in parentheses). We used a clustered OLS regression with treatment dummies to test for the differences between T1, T2, and T3 and the baseline treatment T4. *** Significant at the 1% level; ** Significant at the 5% level; * Significant at the 10% level.

Treatment

Wage Expected effort Promised bonus Actual effort Actual bonus N

T1: Pure Charity T2: Added N=285 Charity N=297 21.4 (7.7) 20.8 (8.7) 8.6 (1.5)*** 8.2 (1.9)** 31.4 (13.6)*** 28.4 (12.8)*** 7.2 (3.0)*** 5.9 (3.3) 17.7 (15.4)*** 9.9 (12.5) 36 40

T3: Other N=206

T4: Self N=235

21.6 (10.0) 8.1 (1.7)** 33.2(15.4) *** 5.8 (3.3) 11.1 (12.8) 30

20.1 (12.2) 7.3 (1.8) 19.9 (12.3) 5.4 (3.2) 7.7 (11.3) 34

As can be seen in the lower part of table 7 it is only in the Pure Charity treatment that the promises of higher bonuses are actually borne out, and it is only in the Pure Charity treatment that average effort by the agents is significantly higher than in the baseline. The average Actual effort level is 7.2 in the Pure Charity treatment vs 5.4 in the Self treatment. This difference of 1.8 translates into an increase of production of 18 tokens. In return for this higher production principals pay out higher Actual bonus (17.7 vs 7.7 tokens), ensuring that both principal and agent have higher earnings in the Pure Charity Treatment. 11    

When we regress the effort decision by the agent on the attributes of the offer (i.e. Wage, Expected effort and Promised bonus), we see that in the Self and Other treatments the actual level of effort is not significantly related to either the Expected effort or the size of the Promised bonus (see table 8). Only in the two Charity treatments does the Expected effort level have a significant influence on the actual effort decision.2 The coefficient is 0.41 in both Charity treatments. Furthermore, it is only in the Pure Charity treatment that the promise of a higher bonus actually elicits higher effort among agents. Table 8. Regressions with actual effort as dependent variable In this table we report OLS-regressions with agent effort as dependent variable. Coefficients are reported with standard errors in parentheses. Standard errors are clustered on subject. *** Significant at the 1% level; ** Significant at the 5% level; * Significant at the 10% level.

Treatment Constant Wage Expected effort Promised bonus R2 N

Pure Charity (1) 1.90** (0.8) 0.03** (0.02) 0.41*** (0.09) 0.03*** (0.01) 0.09 285

Added Charity (2) 2.1*** (0.7) 0.02 (0.02) 0.41*** (0.10) 0.00 (0.02) 0.06 297

Other (3) 3.8*** (0.9) 0.06*** (0.02) 0.10 (0.10) -0.01 (0.01) 0.03 206

Self (4) 2.7*** (0.9) 0.08*** (0.01) 0.06 (0.14) 0.03* (0.02) 0.12 235

One interpretation of these results is that in the two Charity treatments the promised bonus is more credible: the agent trusts more that the bonus will actually be paid out. However in that case the size of the promised bonus should increase effort in the Added Charity treatment as well, yet it does not. Another interpretation is that the psychological cost (i.e. guilt aversion) of shirking goes up in the Charity treatments: The agent does not want to disappoint the principal. This would increase effort in the Added Charity treatment as well, even though the size of the promised bonus does not seem to induce additional effort.                                                                                                                         2

 We  also  ran  a  probit  regression  on  the  likelihood  of  the  agent  putting  in  sufficient  effort  (i.e.  equal  or  more   than  the  expected  effort)  on  treatment  dummies.  Controlling  for  offer  characteristics,  agents  in  the  charity   treatment  were  significantly  (p<0.01)  more  likely  to  put  in  sufficient  effort  compared  to  the  self  treatment.      

12    

Another interpretation is that the results are driven by the fact that the agents simply have a pure preference for donating to charity (Gregg et al., 2011). Agents are better off in the Other and Pure Charity Treatment (table 5), consistent with the idea that there is no reason for the principal not to provide a bonus. Principals are only better off in the Pure Charity treatment, consistent with agents being charity-motivated and expected effort being similar to a fundraising solicitation (Yoruk, 2009; Edwards and List, 2013) One concern is that efficiency is higher in the Added Charity treatment since the principal earnings are paid out both to the subject and donated to charity. Thus, agents focusing on efficiency might actually put in more effort in the Added Charity treatment for other reasons than charity per se. But this would also imply a higher general effort level, not a stronger sensitivity to expected effort. Furthermore, the Expected Effort effect is the same in the Pure Charity and Added Charity Treatments. In table 9, we regress the actual bonus on Wage, Promised bonus and Actual effort. In all four treatments the size of the actual bonus does respond to the observed effort level. However the effect is more pronounced in the Pure Charity than in the other treatments (coefficients of 3.10 vs 2.00, 2.12 and 2.20 respectively, a difference of 3-4 standard errors). Thus it seems that agents are right to take the principals’ Expected effort level more seriously in the Pure Charity treatment. However in the Added Charity treatment the principals do not respond as generously to the Actual effort level and still the agents respond strongly to the Expected effort level (see table 8). This suggests that the significance of the Expected effort level for the two Charity treatments in table 8 stems from the agents’ behavior due to the Charity manipulation and not from higher reciprocity from Charity principals.

13    

Table 9. Regressions with actual bonus as dependent variable In this table we report OLS regressions with actual Bonus paid as dependent variable. Coefficients are reported with standard errors in parentheses. Standard errors are clustered on subject. *** Significant at the 1% level; ** Significant at the 5% level; * Significant at the 10% level.

Treatment Constant Wage Promised bonus Actual effort R2 N

Pure Charity (1) -6.9 (8.3) -0.23** (0.11) -0.23 (0.22) 3.10*** (0.23) 0.45 285

Added Charity (2) 4.4 (4.6) -0.20 (0.15) -0.08 (0.05) 2.00*** (0.29) 0.23 297

Other (3) 9.4 (3.8) -0.30** (0.14) -0.12 (0.08) 2.12*** (0.44) 0.41 206

Self (4) -6.0*** (1.8) -0.13 (0.07) 0.22*** (0.07) 2.20*** (0.32) 0.52 235

As we have seen, in the two Charity Treatments the agent reacts very strongly to the Expected effort level. This would imply that principals can easily increase their profits by increasing the Expected effort level, and this is indeed borne out by the results reported in table 10: in the Pure Charity and Added Charity treatments the principal’s earnings increase dramatically, 3.97 and 4.38 tokens, respectively, per extra level of Expected effort, whereas in the Self and Other treatments increasing the Expected effort level has no significant effect on earnings. Table 10. Regressions with principal earnings as dependent variable In this table we report OLS regressions with actual principal earnings as dependent variable. Coefficients are reported with standard errors in parentheses. Standard errors are clustered on subject. *** Significant at the 1% level; ** Significant at the 5% level; * Significant at the 10% level.

Constant Wage Promised bonus Expected effort R2 N

Pure Charity (1) 14.6* (8.8) -0.64*** (0.17) -0.07 (0.21) 3.97*** (0.76) 0.05 285

Added Charity (2) 8.5 (5.3) -0.71*** (0.19) -0.02 (0.19) 4.38*** (1.19) 0.07 297 14  

 

Other (3) 13.5 (9.6) -0.24 (0.26) 0.06 (0.17) 1.87 (1.25) 0.04 206

Self (4) 23.5*** (6.9) -0.30* (0.16) 0.03 (0.17) 1.17 (1.21) 0.05 235

3.3. Trust game In the two Trust Games we ran (one at the beginning of the first and one at the beginning of the final period), we do not find any evidence that agents act more trustworthy with a charity principal (see table 11, models 3 and 4). The fraction of tokens returned is not significantly higher in the Charity Treatments. Our interpretation is that in the absence of a clearly stated expected level of performance, the charity treatment does not induce higher performance. In the Trust Game using the strategy method, subjects acting as agents are asked for each potential amount of tokens that the principal could have sent how much they would return. Thus subjects do not know how much the principal has actually sent and do not know how much the principal expects in return. This is in contrast to the contracting part of the experiment where principals clearly communicate an expected level of effort with every contract. In the first period we do not find significant evidence of higher levels of trust among Charity principals (see model 1). However after ten periods the level of trust seems to have increased among Charity principals while it has slightly decreased among self-interested principals (model 2), resulting in a significant difference of tokens sent from the principal to the agent. Since agents in the Charity treatment do not return more Tokens, Charity may principals falsely believe that trustworthiness has increased among the agents after ten periods. Table 11. Regression on Trust Game outcomes In this table we report OLS regression with amount of tokens sent by the principal (models 1 and 2), and the average fraction of tokens returned by the agent (averaged over all five possible transfer levels, 6, 12, 18, 24 and 30), as dependent variable (models 3 and 4). The baseline number of tokens sent and returned in the Self treatment is captured by the constant. Three dummy variables account for the effect of the various treatments. Coefficients are reported with standard errors in parentheses. *** Significant at the 1% level; ** Significant at the 5% level; * Significant at the 10% level.

Constant Charity Added Charity Other

Trust sent, Period 1 (1) 5.93*** (0.44) 0.78 (0.61) -0.50 (0.60) 0.53 (0.64)

Trust sent, Period 10 (2) 5.50*** (0.52) 1.65** (0.75) 0.37 (0.73) 0.96 (0.78) 15  

 

Trust returned, Period 1 (3) 0.45*** (0.03) -0.07 (0.04) -0.05 (0.04) -0.08* (0.05)

Trust returned, Period 10 (4) 0.36*** (0.04) -0.04 (0.05) -0.08 (0.05) 0.01 (0.05)

R2 N

0.02 140

0.02 140

0.01 140

0.01 140

4. CONCLUSION AND DISCUSSION This paper tested the hypothesis that donations to charity influence how agents respond to proposed contracts in incomplete contract environments. The answers to our research questions are as follows: i) When a principal’s profits are donated to charity, agents respond more strongly to the suggested level of effort and pure charity principals show higher commitment to the offered contract and respond more generously to the agents’ effort. ii) Charity principals do, however, not make different contract offer. iii) The overall efficiency of the firm is higher when the principal makes donation to charity, iv) The gains associated with charity are split between the principal and the agent. We conclude that firms associated with charity could potentially outperform other firms through more efficient contracting, and seemingly wasteful donations to charity could actually pay off in the long run. Our results contribute to the literature along two dimensions. First, the link between donations to charity, Corporate Social Responsibility and firm performance is much researched but not well understood. Our experimental analysis documents that performance improves when the principal is associated with donations to charity. Our experimental approach also lets us identify possible mechanisms for the positive relation between donations to charity and performance. The mechanism is more efficient contracting in incomplete contract environments. However, more efficient contracting does not seem to be driven by increased trust per se. We propose that the positive affect for the firm garnered through an association with a charitable purpose makes it psychologically harder for employees to shirk. Alternatively, agents are charity-motivated and performance is improved in the Charity treatments as a result of expected effort operating as a kind of fundraising solicitation. We think this paper is related to three strands of literature. First, our results are related to empirical studies investigating the relation between corporate profitability and socially responsible behavior in general. Meta-analyses suggest that there is no negative or even a slightly positive relation

16    

between socially responsible behavior and corporate performance (see e.g. Margolis et al., 2007). Especially charitable contributions and environmental performance appear to be positively related to corporate profitability. Second, our results might explain the performance of firms with non-profit charitable foundation as controlling shareholders, i.e. firms like Bertelsman, Heineken, Robert Bosch, and Carlsberg (Thomsen and Conyon, 2012). These foundations have no owners and the foundation charters typically stipulate that profits must be distributed to charity. Incentive and agency theory suggests that these firms should not be competitive except in certain industries such as education and health care (Hansmann, 1980). In other industries, theory suggests that the effect of foundations not having a personal profit motive to monitor managers and the foundation being entrenched from the market for corporate control through majority shareholdings should lead to worse profitability (see e.g. Jensen and Meckling, 1976; Stulz, 1988). However, empirical studies have found that the performance of foundation controlled companies is on par with or even slightly better than firms where the controlling shareholders have profit motives (Herrmann and Franke, 2002; Thomsen and Rose, 2004). Third, our laboratory experiments with donations to charity are related to Gneezy et al. (2010). In their field experiment, revenues from fixed price souvenir photos only slightly increased when half of the revenue went to charity. However, revenues increased substantially when participants could pay what they wanted with the same charitable component. Gneezy et al. (2010) label this effect shared social responsibility, and argue that it works because customized contributions allow customers to directly express social welfare concerns.

17    

AUTHOR BIOGRAPHY Dr. Oege Dijk received his Ph.D. from the European University Institute in Florence, Italy, and after a two year sojourn at the University of Gothenburg, he recently returned to his native Netherlands to start a position as Assistant Professor in Behavioural Finance at the Radboud University Nijmegen. His main research interests are the role of social status and social comparison in economics and the role of emotions on decision-making. Professor Martin Holmen received his Ph.D. from University of Gothenburg in 1998 and after ten years at Stockholm University and Uppsala University he returned to University of Gothenburg in 2009. He is professor of corporate finance and since 2011 he is the director of the interdisciplinary Centre for Finance. His main research areas are corporate governance and experimental finance.

18    

REFERENCES

Benabou, R. and J. Tirole, 2010. “Individual and corporate responsibility.” Economica 77, 1-19. Berg, J., J. Dickhaut, and K. Mccabe, 1995. “Trust, reciprocity, and social history.” Games and Economic Behavior 10, 122-142. Bowles, S., H. Gintis, and M. Osborne, 2001. “Incentive-Enhancing Preferences: Personality, Behavior, and Earnings.” American Economic Review 91, 155–58. Charnness, G. and M. Dufwenberg, 2006. “Promises and partnership.” Econometrica 74, 1579-1601. Edmans, A. 2012. The link between job satisfaction and firm value, with implications for corporate social responsibility. Academy of Management Perspective 26, 1-19. Edwards, J. and J. List, 2013. “Towards an understanding of why suggestions work in charitable fundraising: Theory and evidence from a natural experiment.” NBER working paper 19665. Fehr, E., S. Gächter, and G. Kirchsteiger, 1997. “Reciprocity as a contract enforcement device: Experimental evidence.” Econometrica 65, 833-860. Fehr, E., A. Klein, and K. Schmidt, 2007. “Fairness and contract design.” Econometrica 75, 121-154. Fischbacher, U., 2007.”Z-tree: Zurich toolbox for ready-made economic experiments.” Experimental Economics 10, 171-178. Gneezy, A., U. Gneezy, L. Nelson, and A. Brown, 2010. “Shared social responsibility: A field experiment in pay-what-you-want pricing and charitable giving.” Science 329, 325-327. Greening, D. and D. Turban, 2000. “Corporate social performance as a competitive advantage in attracting a quality workforce.” Business and Society 19, 254-280. Gregg, P., P. Grout, A. Ratcliffe, S. Smith, and F. Windmeijer, 2011. “A field experiment in charitable contribution: The impact of social information on voluntary provision of public goods.” The Economic Journal 119, 1422-1439. Greiner, B., 2004. “Forschung und wissenschaftlishes Rechnen 2003, An On-line Recruitment System for Economic Experiments.” GWDG Bericht 63. Gesellschaft fuer Wissenschafliche Datenverarbeitung, Goettingen, 79-93. Hansmann.H., 1980. “The role of nonprofit enterprise.” The Yale Law Journal 89, 835-901. Herrmann, M. and G. Franke, 2002. “Performance and policy of foundation-owned firms in Germany.“ European Financial Management 8, 261-279. Jensen, M. and W. Meckling, 1976. ”Managerial behavior, agency costs an ownership structure.” Journal of Financial Economics 3, 305-360. Kitzmueller, M. and J. Shimshack, 2012.” Economic perspectives on corporate social responsibility.” Journal of Economic Literature 50, 51-84. Krueger, A., A. Mas, 2004. “Strikes, scabs, and tread separation: Labor strife and the production of defective Bridgestone/ Firestone tires.” Journal of Political Economy 112, 253-289. 19    

La Porta, R., F. Lopez-de-Silanes, A. Shleifer, and R. Vishny, 1997. ”Trust in large organizations.” AEA Papers and Proceedings 87, 310-321. Margolis, J., H. Elfenbein, and J. Walsh, 2007. “Does it pay to be good? A meta-analysis and redirection of research on the relationship between corporate social and financial performance.” Working Paper, Ross School of Business, University of Michigan. Miettinen, T., 2011. “Moral hazard and clear conscience.” Journal of Institutional and Theoretical Economics 167, 224-235. Rabin, M., 1993. “Incooperating fairness into game theory and economics.” American Economic Review 83, 1281-1302. Smith, C. 1994. The new corporate philanthropy. Harvard Business Review 72, 105-116. Stulz, R., 1990. “Managerial discretion and optimal financing policies.” Journal of Financial Economics 26, 3-27. Thomsen, S. and M. Conyon, 2012. Corporate Governance: Mechanisms and Systems. McGraw-Hill. Thomsen, S. and C. Rose, 2004. “Foundation Ownership and Financial Performance.” The European Journal of Law and Economics 18, 343–364. Yoruk, B., 2009. “How responsive are charitable donors to requests to give?” Journal of Public Economics 93, 1111-1117.

20    

Figure 1. Earnings Overall Average earnings per round (in tokens), by both Principal and Agent, separated by treatment.

Average  earnings  per  round  

29   27   25   T1:  Pure  Charity  

23  

T2:  Added  Charity  

21  

T3:  Other  

19  

T4:  Self  

17   15   Principal  

Agent  

21    

Figure 2. Offer types Offer types for the four different treatments. In all four treatments the overwhelming number of treatments were of the Bonus type.

Offer  Types   80%  

75%  

75%  

74%  

71%  

70%   60%   50%  

Trust  

40%  

Bonus  

30%   20%   10%  

15%  

20%   10%  

8%  

5%  

Fine  

20%  

18%  

8%  

0%   Pure  Charity   Added  Charity  

Other  

22    

Self  

Appendix A: Instructions A1: Baseline Experimental Instructions (Self Treatment) We welcome you to this experimental session and kindly ask you to refrain from talking to each other for the duration of the experiment, and switch off your cell phones. If you face any difficulties, contact one of the supervisors. First of all you should now carefully read these instructions. After having done so, please answer the control questions, which you find at the end of these instructions Tokens The unit of account of this experiment are called Tokens. Tokens convert into SEK at the rate of: 1 Token = 1 SEK The experiment consists of two parts. In both parts you will be initially endowed with 125 tokens, and have the possibility to earn more during the experiment. However there is also the possibility of potential losses during the experiment. If your losses exceed 115 tokens, you will not be able to proceed with that part of the experiment. By appropriately taking your own decisions you can avoid losses with certainty! General Information During this part of the experiment you will play two roles: both the role of employer and employee. You will start with 125 tokens in both roles. The experiment lasts for 10 periods. Each round your earnings as employer will be added to your employer account, and your earnings as employee will be added to your employee account. At the end of the experiment only one of the two roles will be randomly selected for payment. In each period each employer is matched with a different employee. This employee in turn will also act as an employer for another employee, who in turn acts as an employer for a third subject. Thus in every period the subject that acts as your employee will not be acting as your employer. In every period a new employee is matched to every employer in order to ensure that the same employer and employee are matched together only once in the experiment. No employer gets to know the identity of the matched employees, nor do the employees know with whom they are dealing. Your decisions as employer in one period are transmitted only to the employee matched with you in that period. Likewise your decisions as an employee are only transmitted to the employer matched with you in that period. No one else is informed about your decisions. The experiment lasts 10 periods. The first task of the experiment will only occur at the beginning of period 1 and period 10. In this task first every employer will be endowed with 10 tokens, and the employees will not be endowed with any tokens. The employer can then send a part (either 0, 2, 4, 6, 8 or all 10) of his tokens to the employee. These token are then tripled. So if the employee sends 2 tokens, the employee will receive 6, if the employer sends 4 tokens, the employee will receive 12, etc. The employee then makes the decision how much of the tokens to return to the employer. As the employee you will be asked to submit how 23    

much you will return in case you receive 6 tokens, how much you will return when you receive 12 tokens, …18 tokens, …24 tokens, …30 tokens. You earnings in this task will only be announced at the end of period 10 and only then be added to your total earnings. The main part of the experiment works as follows: As the employer you have to make an offer to the employee, who can accept or reject this offer. If the employee rejects the offer, both the employer and the employee earn 0 that round. If the employee accepts the offer, the employee has to decide how much effort to make. This effort (e) is a number between 1 and 10. For every unit of effort the employee makes, the employer makes a profit of 10 tokens. However making effort is costly to the employee: in order to make a given amount of effort the employee has to be pay an amount of tokens given in the table below: Table 1. Cost of effort to employee

e c(e)

1 0

2 1

3 2

4 4

5 6

6 8

7 10

8 13

9 16

10 20

4 40

5 50

6 60

7 70

8 80

9 90

10 100

Table 2: Profit to employer

e P(e)

1 10

2 20

3 30

As effort is costly to the employee and only benefits the employer, the employer has to make an offer to the employee to compensate him/her for his/her effort. There are three types of offers the employer can make to the employee: a trust offer, a bonus offer and a fine offer. A trust offer consists of two components: a fixed wage w, and a desired level of effort e*. If the employee accepts the offer, he/she will earn the fixed wage independent of how much effort he/she makes. A bonus offer consists of three components: a fixed wage w, a desired level of effort e*, and a promised bonus b*. The promised bonus is the amount of tokens the employer promises to pay the employee if he/she puts in at least e* effort. After observing the employee’s effort the employer sets the actual bonus b. This can be lower, equal or higher than the initial promised b*. A fine offer also consists of three components: a fixed wage w, a desired level of effort e* and a fine f. If the employee puts in less effort than e*, the computer will roll a dice. If the dice lands either 1 or 2, the employee has to pay the fine f to the employer. However if you choose the fine offer, the employer always has to pay a fixed fee of 10 tokens. For the trust offer and the bonus offer you don’t have to pay this fixed fee. There are a few additional rules on the size of the wage, bonus and fee: 1. The fixed wage w should not exceed 100 tokens. 2. The fixed wage w should not be below the cost of effort expected (e*). For example if the expected effort e*=6, then the fixed wage w should at least be equal to 8. (see table 1 above). 24    

3. The fixed wage and promised bonus together should not exceed 100 tokens. 4. The fixed wage and actual bonus together should not exceed 100 tokens. 5. The fine should not exceed 13 tokens. At the end of a period, several different situations are possible: 1. The employee rejected the offer. In this case both the employers profit and the employee’s income are equal to 0 this period. 2. If an offer was accepted, the calculation of profit and income depends on the kind of offer the employer made to the employee 2a) Trust contract Employers profit P = 10 × actual effort level (e) − fixed wage (w) Employee’s income I = fixed wage (w) − cost of actual effort level c(e) 2b) Bonus contract Employers profit P = 10 × actual effort level (e) − fixed wage (w) − actual bonus (b) Employee’s income I = fixed wage (w) + actual bonus (b) − cost of actual effort level c(e) 2c) Fine contract If an offer with potential fine was accepted, the employer has to incur a fixed cost of 10 tokens in any case. Three cases have to be distinguished, depending on whether the employee has to pay the fine or not. (i) The employee worked more than or exactly what was demanded. In this case the employee does not pay the fine. Employers profit P = 10 × actual effort level (e) − fixed wage (w) − fixed cost (10) Employee’s income I = fixed wage (w) − cost of actual effort c (e) (ii) The employee worked less than was demanded and the dice shows the numbers 1 or 2. In this case, the employee has to pay the fine. Employer’s profit P = 10 × actual effort level (e) − fixed wage (w)− fixed cost (10) + fine (f ) Employee’s income I = fixed wage (w) − cost of actual effort c(e) − fine (f ) (iii) The employee worked less than was demanded and the dice shows the numbers 3, 4, 5 or 6. In this case, the employee does not have to pay the fine. Employer’s profit P = 10 x actual effort level (e) − fixed wage (w) − fixed cost(10) Employee’s income I = fixed wage (w)− cost of actual effort c Every period proceeds in at most six stages: Stage One (only in period 1 and 10): You are the employer. Select how many tokens to send to the employee (either 0, 2, 4, 6, 8 or 10 tokens). Click next.

25    

Stage Two (only in period 1 and 10): You are the employee. Indicate how many tokens you would send back in case the employer sent you 2 tokens (and thus you received 6), how many if the employer sent you 4 tokens (and thus you received 12 tokens), etc. Click next. Stage Three: You are the employer. Decide what kind of offer to make to the employee: trust offer, bonus offer or fine offer. Fill in the size of the fixed wage, expected effort and promised bonus or fine if applicable. Click the red select button, and then the grey confirm button at the bottom of the screen. Stage Four: You are the employee. You will see the offer made by your employer on the screen. Choose whether you accept or reject this offer. If you reject, both you and the employer will earn 0 tokens this round. If you accept the offer, indicate how much effort you will make this round. Click next. Stage Five: You are the employer. If you selected a trust offer you will be informed of the effort made by your employee. Click next. If you selected a bonus offer you will be informed of the effort made by your employee, and are asked to determine the actual bonus. Click next. If you selected a fine offer you will be informed of the effort made by your employee, the roll of the dice, and whether your employee had to pay a fine or not. Click next. Stage Six: In stage six you will be informed of your earnings this round. On the left will be shown how much you earned this period as employer , and on the right will be shown how much you earned this period as employee. Please keep in mind that the entire experiment cannot progress to the next stage unless you click the “next” button. If you have any questions, please ask us

26    

A2: Treatment Differences in Instructions. In all treatments the subjects read the same baseline instructions as in appendix A1. After all questions about the instructions had been cleared up, subjects in the Pure Charity, Added Charity and Other treatments received the final instructions on an overhead projector slide. The text on these slides differed by treatment: Pure Charity: There is one final twist to this experiment: in case your employer-account gets selected for pay-off (with 50% probability), we will instead donate this same amount of SEK to the Swedish Red Cross (Röda Korset). We will post proof of our donation on the wall outside the lab. In case your employer account gets selected and we make this donation, we will pay you your employee-account earnings instead. In case your employee account gets selected for pay-off (50% probability), you will simply earn your employee account earnings. Thus with 50% probability your employer-account earnings will be donated to charity, and you will earn your employee account earnings in any case. Added Charity: There is one final twist to this experiment: In case your employer account gets selected for pay-off (with 50% probability), we will donate this same amount of SEK to the Swedish Red Cross (Röda Korset). We will post proof of our donation on the wall outside the lab. In case your employee account gets selected for payoff (with 50% probability), you will simply receive your earnings in this account and we will not donate to the Swedish Red Cross. Other: There is one final twist to this experiment: In case the employer account gets selected for payoff (with 50% probability), you will not earn the earnings of your own employer-account, but the earnings in the employer-account of a random other subject in this room. Conversely, your employeraccount earnings may determine the earnings for some other subject in this room. In case your employee account gets selected for payoff (with 50% probability), you will simply receive your earnings in your own employee account.

27    

Appendix B: Screenshots. B1: Offer decision by the Principal

B2: Effort Decision: Agent

28    

B3: Bonus Decision: Principal

29