Institutional, idiosyncratic and physiological aspectes of corruption

Transcription

1 Institutional, idiosyncratic and physiological aspectes of corruption Tarek Jaber-López LEE and Economics Department, Jaume I University, Castellón, Spain Aurora García-Gallego LEE and Economics Department, Jaume I University, Castellón, Spain Nikolaos Georgantzís GLOBE and Economics Department, University of Granada, and LEE, University Jaume I, Castellón, Spain Georgia Michailidou LEE and Economics Department, Jaume I University, Castellón, Spain Pandelis Perakakis LEE and Economics Department, Jaume I University, Castellón, Spain Abstract We gratefully acknowledge financial support by the Spanish Ministry of Science and Innovation (project ECO /ECON), Bancaixa (project P1-1B ), Junta de Andalucía (project P07- SEJ-03155) and University Jaume I for financial support on grant (PREDOC/2010/25). Special thanks go to all of the members of the LEE. Presenting author: Economics Department, Universidad de Granada, Campus de la Cartuja, Granada (Spain). ngeorgantzis@ugr.es. Phone number: Fax

2 In this paper we report results from an experimental study on corruption. We study the role of a suing option available to the loser of an auction in which two bidding firms post combinations of quality and bribe to a social planner who then decides on the winning proposal. Quality generates an externality which affects the payoffs of all players, including a group of passive receivers ( citizens ) introduced as a control factor. In a pre-play session, we elicit subjects risk attitudes and other-regarding preferences. We designed a four treatment experiment, where 22 independent groups played a repeated version of the game. We find that firms have a preference for quality and aversion to bribe. We also find a positive correlation between a subject s risk aversion and his/her posted bribery when suing option exists, and a negative correlation between a subject s risk aversion and his/her frequency of denouncing. Finally a negative correlation exists between a pro-social planner and risk aversion in the baseline game, and a positive correlation in the suing-option treatment. Overall the suing option has a significant quality-boosting and bribery-reducing effect, although it is adopted less and less towards the end of the session. The existence of a passive receiver of the quality externality significantly increases investments on quality (and reduces bribes) when citizens are receivers of the externality. Emotional arousal measured by skin conductance reflects strong emotions when deciding not to bribe in the baseline and to bribe on the suing option treatment. Both findings are compatible with arousal due to dissonance with own monetary reward maximization or else due to actions against the majority of the cohort. The expectation of a possible punishment following a bribe makes bribing winners significantly more aroused than honnest winners. Finally, passive bribery yields no significant emotional responses, indicating that accepting a bribery is less stressing than offering it. JEL Classification: D03, L51, A13, B40, C92 Keywords: Corruption, auctions, experiments, bribery 2

3 1 Introduction Corruption is one of the most detrimental factors in an economy s functioning in both the developed and the developing world. It has been present during all times and has intruded all political systems. While it is impossible, given its secrecy, to calculate the actual losses streaming from this destructive element, estimations display that 1 trillion of US dollars is spent annually on bribes. But trying to approximate the monetary consequences of corruption is only half correct since the non pecuniary costs of this distortion are equally harmful and are most of the times missing from the recording attempts. The existence of corruption creates both an unhealthy economic and social environment as it is closely connected with the ethical and moral values of humanity. For economists, the reduction of corruption constitutes an objective that has not been accomplished yet. So far, the academic literature counts only with a few studies where possible solutions are suggested while, the main focus seems to be concentrated on the description of corruption s features and characteristics. It could be claimed that until recently corruption was approached only by macro and political economists. Yet, corruption is not an issue emerging only from the leaks of the institutional and legal framework. It is also a mentality and a human mind set. It was therefore a matter of time that corruption would be explored by micro and behavioral economists. If macroeconomics describe the conditions that foster corruption, then microeconomics are involved with the factors that initiate it. In this study we explore the effect of risk and other regarding preferences on bribery and monitor physiological reactions to different stages of the decision making process. To do this, we use a novel experimental framework in which bribery is used to alter the competitive outcome of an auction for the assignment of a project whose quality implies an externality 3

4 to the society. We aim at uncovering the connection between the economic preferences of agents and their propensity to bribe. This model posses two special characteristics: first there is an endogenous externality (social profit) which depends on the quality of a public project. Second this model has a specific Nash equilibrium, which is compared to actual behavior in order to elicit subjects preferences for quality and bribe. We also test the effect of a costly anti-corruption policy similar to intra-industry auditing. The scenario involves two firms bidding in qualities and bribes. Based on this, a town planner chooses the winner firm of the round. We study the effects of a passive receiver (citizen) of the public project s quality and the effects of an auditing option available to the losers. Finally, we monitor subjects emotional responses to different moments of the decision making and feedback receiving process. In the following section we provide a literature review over the topics of corruption and corruption related experimental studies. In section 3 we briefly describe the experimental design and discuss the methodology we followed in this study. Section 4 is dedicated to the results. We use section 5 to provide the concluding points of this study. 2 Related Literature Experiments on corruption are becoming an important trend of research. First experiments on corruption were run by Frank and Schulze (2000) and many experimentalists seek a design fitting as well as possible to reality. In fact, different models or contexts have been employed for the same topic. Frank and Schulze (2000) employ a particular environment (a students film club) where individuals have to decide in a situation with a trade-off between maximizing individual profit and maximizing the gains of the public interest. Frank and 4

5 Schulze (2000) find out that, apparently due to self-selection, economics students tend more to corruptibility than other groups of students. Abbink et al. (2002) employ as baseline a variation of a trust and reciprocity game in which the truster act as a firm and the trustee as a public official. They sequentially run more treatments, first an exogenous negative externality affecting the rest of players each time that a pair engages in bribe. Second, an exogenous probability of detection. No effects were found for the negative externality, however engaging in bribery decreases with the threat of detection. Then, Barr and Serra (2009) prefer to develop as framework a modified one-shot ultimatum game as a representation of petty corruption. Sender plays the role of citizen and receiver as public official (they consider this context more adapted for students). Each group is also composed by other members of society playing as passive receivers of the externality produced by engaging in bribery. They find that bribe acceptance is lower when external costs increase. Another aspect studied in this article is framing effects. In fact frame on corruption yields to two different outcomes. On the one hand this authors find that when instructions are presented as a petty corruption scenario instead of in abstract terms, the propensity to offer bribes is lower. No effects were found concerning bribe acceptance. On the other, Abbink and Hennig-Schmidt (2006) compare neutral versus loaded instructions and did not observe differences in subjects behavior. We consider that an other important issue to take into account is corruption experiments is the election of the pool. Many authors analyze different dimensions of the sample, for instance cultural, status, experience, educational background, are the most important. Frank and Schulze (2000) find that eco- 5

6 nomics students are the most corrupt. Alatas et al. (2009) are based on Abbink et al. (2002) model but they incorporate a citizen having the option to punish at last stage as a measure of tolerance of corruption. Their main result is that Indonesian public servants have less tolerance of corruption than Indonesian students. An interpretation is the experience effect, the more exposure to corruption makes people to feel more rejection to dishonesty. An interesting research is the one made by Cameron et al. (2009) in a corruption cross-cultural research between Australia, India, Indonesia and Singapore. As expected, people in India show more tolerance of corruption than people in Australia. Contrary to expected, in Singapore they are more tolerant and Indonesia less tolerant comparatively. Barr and Serra (2010) conducted experiments in Oxford and observed that among undergraduates students, subjects coming from more corrupt countries have more propensity to engage in bribery. In addition, the more time they have spent in UK, the propensity to bribery decreases. This results are not consistent for graduate students. We have briefly explored the literature concerning experiments on corruption. Different models have been used for the same topic with interesting results in each. Frame also has an important impact on corruption and finally the composition of the pool. Thus, we construct a simple game in the scenario of competition in auctions. Two firms have to face a trade-off between investing on quality and bribing the town planner. We believe from Mauro (1995) that corruption decreases investment and growth, thus this context could be an interesting starting point. Finally, we compare different matching protocols. As Abbink (2004) we study the differences between matching partners and strangers. 6

7 3 Experimental Design 3.1 Theoretical Framework Our model is inspired on the theoretical framework of Büchner et al. (2008) in which pairs of sellers bid to obtain a public contract and the bid of each seller could include a bribe to be paid by the public official as buyer. Consider the variation of the previous model. A game is played between two firms (i, j) bidding for a procurement contract which will be granted to one of the two firms by a public auctioneer (town planner). Like in the original framework, bidding takes place in two dimensions. However, in our version, the social externality of the winning project, net of social or private costs affects uniformly to all agents. Thus, rather than the price paid by the state to the winning firm, the first bidding dimension is a bidders quality net of costs (Q i < A, where A represents an exogenously given upper bound to quality). This allows us to model the first dimension of bidding as a simple linear, monotonically increasing function of the winning projects social quality benefiting unambiguously all agents, the auctioneer, the two firms and, potentially (as we do in a follow-up of this paper), the society surrounding them. The second dimension, like in the Büchner et al. s (2008) model, is a bribe (B i ) promised and finally paid by the winner of the auction to the auctioneer. Furthermore, in order to guarantee that a social dilemma emerges in this bribery game, we impose a restriction on bribes and qualities 1 : Q i + B i A. We use a linear specification of the three agents monetary (induced) utilities, adding a psychological cost parameter, γ, capturing an agent s aversion to bribe due to ethical reasons, expressed in monetary loss per monetary unit of bribe received by the town planner. Thus, 1 corresponding, for example, to a resource-driven budget constraint in a more general version of the model 7

8 the three agents utility levels are given by: (1) π tp = a tp Q w + (1 γ tp )B w (2) π w = a w Q w (c + γ w )B w + R (3) π l = a l Q w Where R is a fixed private profit obtained by the firm winning the procurement and c a per monetary unit of bribe cost, borne by a bribing winner, denoting that the bribe may yield further monetary costs on its way from the firm to the town planner. Assuming perfect information on the agents preferences, the resolution of the game depends on the hypothesis of continuous versus discrete strategies Continuous strategies In this section we study the theoretical prediction of the game assuming a continuos space of strategies. We first solve the case in which agents have nonmonetary (psychological) concerning, then the case in which agents purely maximize their payoffs. Non-monetary (psychological) payoffs 1. If a tp > 1 γ tp, the highest quality project will be chosen and firms will 8

9 bid only in qualities, leading to the equilibrium: (Q i, B i ) = (Q j, B j ) = (A, 0) independently of the firms preferences. 2. If a tp < 1 γ tp, the highest bribe will be preferred by the auctioneer. In that case, firms will bid with the maximum bribe they can, as long as the bribing (monetary and psychological) cost does not exceed the fixed amount R earned by the winner. Thus, in equilibrium, firm i will bid (Q i, B i ) = (A R c+γ i, R c+γ i ). Thus, we would expect town planners to choose the highest quality proposals if they are sufficiently bribery-averse, while they will choose the bidder with the highest bribe otherwise. Firms (believing that they are) faced with a quality-maximizing auctioneer will not bid with bribes, independently of their own preferences, whereas firms anticipating a bribery-maximizing behavior by the auctioneer will promise higher bribes, the more bribery averse they are. Monetary payoffs Finally, in absence of psychological, bribe-regarding considerations, a monetaryreward maximizing behavior would predict maximal quality bids, (Q i, B i ) = (Q j, B j ) = (A, 0), if a tp > 1, and (Q i, B i ) = (A R, R), if a c c tp > 1, in which case, the social dilemma leads to a socially suboptimal equilibrium. In the linear version, the following parameters are adopted, (a, A, R, c) = (1/2, 10, 10, 2), which guarantee the emergence of the social dilemma equilibrium bids: (Q i, B i ) = (Q j, B j ) = (5, 5) in the case of bribery-neutral agents with universal preference for the bribe-maximizing bids by the town planner, or top quality (Q i, B i ) = (Q j, B j ) = (A, 0), and quality maximizing auctioneer behavior if γ tp > 1/2. 9

10 3.1.2 Discrete strategies Our game s theoretical prediction differs when considering discrete strategies. The matrix in table 1 shows the payoffs of both firms for different combinations of bribe (and quality). In the discrete game there are two strong Nash equilibria (Q i, B i, Q j, B j ) = (7, 3, 7, 3) and (Q i, B i, Q j, B j ) = (6, 4, 6, 4) where the former is Pareto superior to the latter. The (Q i, B i, Q j, B j ) = (5, 5, 5, 5) is still a weak equilibrium and also Pareto dominated by both the other two. It is weak, because each firm is indifferent between this and biding lower bribes just to become a loser (it earns 12.5 e in both cases). Table 1: Matrix for discrete strategies 3.2 Experiment We recruited 77 subjects, 66 of them would be given the role of either firm or town planner, and the 11 remaining would participate in the game as citizens. 10

11 We then divided the 66 subjects into two groups of 33. These two groups were going to play a slightly different game. The first group (group 1) was going to play a game the outcome of which would be observed by the citizens. The outcome of the game of the second group (group 2) was not only going to be observed by the citizens but it was also going to affect their payoffs. For the whole duration of the game, the citizens were passive players, meaning that they are not decision-makers. Comparing the results of these two different games would let us seclude and study the effect of a passive agent that is affected by other agents decisions compared to when he is not affected by others decisions. Therefore, the experiment includes four treatments. Each treatment contains 15 sequential rounds. Treatment T0 and treatment T1 were run simultaneously in the lab by group 1 and group 2 respectively. In the end of these treatments, we keep the same division between group 1 and group 2 but we hand out new instructions and we redistribute the roles for each subject. Then we run treatment T0s and treatment T1s for group 1 and group 2 respectively. All the treatments are based on the following decision sequence: The two firms have an endowment of 10 units (that renews for each round) and the option to split it between quality and/or money transfer. After the firms split their resources between the two options, the town planner decides the winner Treatments In treatment T0 we divide the 33 subjects of group 1 into 11 clusters. In each cluster, there are two firms and one town planner. We informed the 11 clusters of group 1 that the amount each firm places for quality, as well as the winner of each round are presented with the use of a monitor to a group of 11 11

12 citizen players that are observing the evolution of the game from another room. In the same laboratory, and simultaneously with treatment T1, we divide the players of group 2 in 11 clusters, following exactly the same procedures as in treatment 1. The only difference is now that the firms and the town planners are informed that the outcome of each round affects the profits of the citizens. Specifically they are informed that there is one citizen corresponding to each cluster and that her payoff for each round will be exactly the quality placed by the winning firm. Specifically, (4) Π citizen = Q winner. As already explained, the alteration of this treatment will allow us when comparing with treatment T0 to investigate any impact that externality might have on the placement of bribery and eventually quality. The feature of a citizen as a passive receiver of the quality of the public project tries to illustrate how could firms and town planners could be affected by their consciousness that their decisions affect the welfare of the society. In the end of the 15th round of the first treatment we gave the 33 subjects of group 1 new roles, so now 11 new clusters of firms and town planners emerged while still the 11 citizen players observe through a monitor the evolution of the game (amount placed for quality by each firm and the winner of each round). 12

13 Continuing the same structure and scenario as the previous treatment, we now insert a new option for the firms; in this treatment, losing firms have the option to bring to suit the winning firm and the planner if a bribe placement is suspected. When a losing firm brings them to suit, the offer of the winning firm opens. If the winning firm has placed any amount of money to the bribe, then both the winning firm and the town planner get punished with a zero profit for the round. If the winning firm has placed no bribe, then the suing firm gets punished with a zero profit for the round. The option is available to the losing firm either it has placed a bribe or not. The rationale behind this option is that transparency and punishment can be used as an anti-corruption policy when the level of competition inside the industry is high. In real life such a policy would be costless for the government if the cost for the legal prosecution of a suspicious public project would burden any company that initiated such a procedure. In our experiment the risk faced by the losing firm (being punished with a zero payoff) if the lawsuit does not reveal any money transfer, simulates the cost that a company in real life would have to suffer to initiate a legal prosecution. By comparing the results of treatment T0 with those of treatment T0s, we can expose the effect that such a policy can have on the placement of bribe or quality. Similarly to treatment T0s, we assign the 33 players of group new roles, so that 11 new clusters emerge. The scenario is exactly the same as the one described for treatment T0, although (as in treatment T1) the outcome of each round affects the payoffs of the citizen players. The comparison of this treatment with treatment T1 will allow us to investigate the impact of the suing option in the presence of a citizen. Then, comparing treatment T1s with treatment T0s, we will be able to detect for a second time, whether the citizen as a passive receiver has an effect on our setting or not. A summary of our design is provided in table 2. 13

14 3.3 Idiosyncratic attitudes In our study we employ risk attitudes and other-regarding preference tests, together with a number of other information about the subjects (age, gender, education status and intelligence). Risk attitudes are measured as a subject s willingness to choose a riskier option among a number of winning probability and prize pairs. Here, we adopt two lottery tests of risk aversion and risk aversion with losses, firstly introduced by?. The other tool we used in this experiment is two other-regarding preferences tests as these emerged from the study of?, specifically in the modified dictator game created by these authors. In the first of these tests, we measure the inequality aversion of our subjects by asking them to choose their favorite status between a more equal and a more unequal choice of wealth distribution for ten different levels of economic status. Using the same structure, the second of the other regarding preferences test measures the relative altruism of the subjects. While we classify and characterize the subjects within groups depending on their answers on these tests, we do not use these groups as variables for the econometric analysis of our results. For this purpose we use the original results of the questionnaires without any grouping as this might have produced misleading results, given that there are not any academic standards according to which we could classify safely the subjects answers to groups. We only use these categories when we describe the subjects in a later section, as we believe this would be more helpful for the reader to shape a general view of the subjects economic behavior. 14

15 Treatment Rounds Subjects Groups Stages Citizen observing Citizen payoffs Suing option Chronologie Risk + ORP 77 00:00 to 00:30 T0 1 to Firms + 11 T.P Firms auction Yes No 0:30 to 1:00 2-Planners decision T1 idem 22 Firms + 11 T.P Citz. 11 Idem Yes Q w No idem T0s 16 to Firms + 11 T.P Firms auction Yes Yes 1:00 to 1:30 2-Planners decision 3-Firms suing option T1s idem 22 Firms + 11 T.P Citz. 11 Idem Yes Q w Yes idem Table 2: Summary of treatments. 15

16 4 Results 4.1 The sample First we present a summary table with the composition of the sample by gender and treatment (table 3). Treatment Player Men Women Total Firm T0 Town Planner Firm T0s Town Planner Firm T1 Town Planner Firm T1s Town Planner Total All Table 3: Sample composition, by gender and treatment The sample is perfectly balanced across genders, treatments and roles played. Therefore, 66 people were playing as firm and town planner, 33 of whom were women and 33 men. 2 Concerning firms, in each treatment the share-out was equal (50% of each gender), while for town planners the sample was almost balanced. Figure 1 shows the distribution of the sample according to age. As we recruited undergraduate students we have a sample were the average age is 21 and most of subjects are between 20 and 24 years old. 2 The 11 citizens did not make decisions. Therefore, we do not analyze them at all. 16

17 Percent Age Figure 1: The distribution of the age Figure 2 shows the composition of the sample according to the topic they are studying. Subjects are recruited from the school of economics in Castellón. Most of them are students of economics (63.64%), business (27.27%) and from other schools (9.09%). 4.2 Treatment effects Our purpose in this section is to analyze the effect produced by adding new treatments to our baseline model. Actually we pretend to see how behavior is changing when implementing some new institutions in the abstract, small society of our setup Citizen effects Firstly we study the effect of having a fourth player who is the receptor of the potentially corrupt transactions. Thus, we make a between-subjects analysis. We compare firms and town planners decisions when they are playing with 17

18 Other Studies 9.09% Business 27.27% Economics 63.64% Other Studies Business Economics Figure 2: Sample composition according to undergraduate studies the new player ( citizen ) as receiver and observer of the quality of the wining project with respect to those where the citizen was just an observer. Firms: On table 4, we present decisions made by firms. Treatment Quality Bribe Total T T0s T T1s Table 4: Averages for quality and bribe. Let us focus on the net effect of the citizen. Firms playing treatment T1 and T1s are investing on average significantly more on quality and offering less bribe than those playing T0 and T0s. The average quality in T0 is 6.70 unit and 3.03 units of bribe while in T1 this values are 7.27 and 2.59 units respectively. In table 5 we observe that this differences are significant at 1% 18

19 (Mann-Withney p-value=0.00). In addition, we compare T0s and T1s where the difference is exactly the same. The mean quality and bribe in T0s are 8.57 and 1.20 units respectively, compared to 9.11 and 0.77 units in T1s. This differences are also significant at 1% (Mann-Withney p-value=0.00) as we see in table 5. These results give us a first overview of how decisions are changing on the firms side. Treatments Quality Bribe T0 vs T T0s vs T1s T0 + T0s vs T1 + T1s Table 5: Mann-Withney p-values comparing quality and bribe by treatments. We illustrate these results with some figures. Figure 3 presents two curves representing the evolution of quality. We see that quality investment is affected by the citizen along the 30 periods of the session. Except for few periods, quality for treatments with a citizen (T1 and T1s) is above the one with no citizen (T0 and T0s). Figure 4 clarifies our previous result. We present the distribution of quality for the four treatments. The main interest of these figures is the frequency with which firms invest 10 units of quality. In treatment T0, the frequency is almost null, whereas in treatment T1, it is 12%. Also, in treatments with the suing option this differ- 19

20 Quality citizen NO citizen Suing option Period T0 + T0s T1 + T1s Figure 3: The evolution of quality: citizen effect ence persists. In T0s, the frequency of firms investing all their endowment on quality is 55.15%, while in T1s this is true in 76.97% of the transactions. Town Planners: We define three types of town planners according to their revealed preferences. Rational town planners are choosing according to their monetary payoffs. Pro-social town planners are those choosing the firm investing more on quality despite the sacrifice that this may imply in monetary terms. Antisocial planners choose the firm investing less in quality even if this yields them a lower monetary reward. Figure 5 represents the distribution for each town planner profile, by treatment. 20

21 T0 T0s Quality (%) T T1s NO citizen Citizen Figure 4: The distribution of quality: citizen effect In the baseline treatments we see that the majority of town planners are acting rationally. In fact, in treatment T0 there are more pro-social town planners than in T1 (24% in T0 vs 15% in T1). However, in treatments with a suing option, there are more pro-social town planners in the presence of a citizen (69% in T0s vs 80% in T1s). Moreover, in T0s w observe more anti-social behavior (8.30%) than in T1s (1.70%). It seems that the citizen affects town planners in a pro-social way, but just when the suing option is present Suing option effects We observe a dramatic effect on the values of table 4. As a consequence of the suing option, the investment on quality increases by 28% (from 6.70 to 8.57 units) and bribes decrease by 60.40% (from 3.03 to 1.20 units ) (in T0 and T0s). Concerning the treatments with a citizen, quality increases by 25.31% and bribes decrease by 70.27%. As we observe on table 6 all this differences 21

22 T0 T0s T1 T1s Rational Pro-social Figure 5: Town Planner s preferences: differences among treatments are significant at 1% (Wilcoxon p-value=0.00). Treatments Quality Bribe T0 vs T0s T1 vs T1s Table 6: Wicoxon p-values comparing quality and bribe by treatments. Figure 6 is useful to illustrate the effectiveness of the suing option. In the first part of the experiment, that is in first 15 rounds, the average bribe is between 2 and 3 units. However, in the second part, the average bribe drops to below 1 unit. We also notice that bribe decreases over time. Actually, this institution is causing a sharp increase in bribe-free behavior, depicted on figures 7. 22

23 Bribe Bribe Suing option Period Figure 6: The evolution of the bribe From all the transfers made by firms to town planners, 1.82% were null in Treatment T0 and 12.42% in T1. However, this amount increases significantly in the second treatment. In 63.33% of the interactions firms did not offer any bribe in T0s, whereas 81.12% of the transactions in T1s were bribe-free. That is, the suing threat is generating a more bribery-averse society. Up to now we observe a clear decrease on bribery due to the suing threat but we have not examined how often are loser firms bringing to suit other players. In fact, this audit-and-punish system could have been inefficient if the resources destroyed to reveal and punish corrupt exeeded the social gains from a decrease in bribery and an increase of quality. On figure 8 the evolution of suing is depicted defining a dummy equal to 1 when a loser firm brings suit and 0 when it does not. While the frequency of suing starts around 70%, we observe a clear decreas- 23

24 T0 T0s NO Citizen Percent T1 T1s Citizen Percent Figure 7: The distribution of the bribe, by treatment ing trend to less than 10% in last period. Furthermore, from all the lawsuits, just 25% were effective or right (75% of audits revealed no bribe). Therefore, it is the threat and not the actual adoption of the suing institution that drastically reduces bribery. Figure 5 demonstrates how the suing option affects town planners choices. In treatments T0 and T1 pro-social choices occur in 23.77% and 14.68% respectively, whereas in T0s and T1s town planners are making pro-social decisions in 69% and 78.33% of times respectively. Figure 9 illustrate how town planners choices progress over time. We aggregate data ignoring the citizen feature to evaluate the net effect of the suing option on town planners. In the first 15 periods there is an increasing trend on rational and a de- 24

25 % % suing Suing option Period Figure 8: The evolution of bringing suit creasing trend for pro-social decisions. In addition, the percentage of rationals is always above that of pro-social ones. This profiles diverges over time. However, when the suing option is available, the picture is completely different. The dominant choice is pro-social, while monetary-maximizing rationality declines over time Social efficiency To conclude, we have observed how the two institutions affect agents decisions: a fourth player as receiver of the social externality as well as the audit-andpunish institution reduce bribery. However, we should also study the effect of these institutions on economic efficiency. We define social efficiency as the total profits earned by each group. Figure 10 represents the evolution of the average per group profit for each treatment. 25

26 % Rational Pro-social Suing option Period Rational Pro-social Figure 9: The evolution of town planner preferences In this figure we observe some interesting results. First, in treatments with a lawsuit (T0s and T1s), efficiency starts at lower levels than in baseline treatments (T0 and T1). Nevertheless, as subjects are playing more periods, efficiency increases significantly in treatments with a suing option and remains above baseline treatments from period 8. This pattern fits with a society where the implementation of new institutions needs some time to adapt. Finally treatments with a citizen are more efficient. Thus a more transparent society is more efficient. 4.3 Idiosyncratic features Risk Aversion Figure 11 represents the distribution of the sample according to risk aversion. We calculate per-subject average probability chosen in the first 4 lottery panels and in the last 4. Therefore, we distinguish risk aversion without losses, from 26

27 Period Treatment T0 Treatment T1 Treatment T0s Treatment T1s Figure 10: The evolution of the efficiency of groups, by treatment risk aversion with losses (see Sabater and Georgantzis (2002)). Both samples follow a normal distribution where decisions are concentrated around a probability of 0.5. We hence classify from now on as highly risk averse subjects those choosing on average a probability higher than 0.5 and low risk averse those choosing a lottery with a higher winning probability than 0.5. Figure 12 shows the evolution of bribe distinguishing between high and low risk aversion. We observe some features in this figure. In first treatment (period 1 to 15) there are not differences between high and low risk averse subjects. However, in the second treatment we notice a divergence between this two groups. Contrary to our expectation, high risk averse subjects are offering higher bribe than low risk averse. This is in line of auction games where high risk averse subjects 27

28 Risk aversion Risk aversion with losses Percent winning lottery probabilities Percent winning lottery probabilities Figure 11: The composition of the sample according to risk aversion bid higher prices Other Regarding Preferences Figure 13 gives us an overview of the sample which helps us to understand how subjects other regarding preferences are distributed. The value we take for each variable is the switching point from left to right. We exclude subjects making double crossing. The interpretation we give for these variables is the following: for altruism, the higher is the value the less altruistic is the subject. For envious, the lower is the value the more envious is the subject. Regarding the altruism variable, the most frequently chosen option is not changing from left to right, that is the lowest altruistic option. Then, the majority of the sample switches between row 5 and

29 Bribe Suing option Period High risk aversion Low risk aversion Figure 12: The evolution of bribe according to subjects risk aversion levels Econometric models In this section we aim at finding some hints explaining what causes subjects to make some decisions related to the bribery game. Thus, we employ as potential explanatory variables some personal characteristics as well as previous decisions on the experiment. We regress these variables on the main variables extracted from subjects decisions such as bribes, suing decisions and town planners pro-social preferences. We therefore construct three models for panel data. For the whole econometric analysis we separate into two groups: baseline treatments corresponding to treatments T0 and T1, and suing treatments corresponding to T0s and T1s. We believe that this separation is reasonable because we deal with very different treatments where subjects decisions are not expected to follow related patterns. Model 1: Bribe 29

30 Other regarding preferences Altruism Envious Percent Switching point Percent Switching point Figure 13: The distribution of the sample according to other regarding preferences Model 1 is a Minimum Least Squares regression where the dependent variable is bribe. The personal characteristics included in this regressions are Altruism, Jealous, Risk Aversion, and Gender. We also added lagged variables such as winning or not in previous period, the bribe amount offered in the previous period, and being accused for bribe or not in the previous period. Table 7 reports the coefficients and the corresponding standard errors. Bribe = α + β 1 Altruism + β 2 Jalous + β 3 Riskaversion + β 4 Gender +β 5 P eriod + β 6 W in[t 1] + β 7 Bribe[t 1] + β 8 Besuit[t 1] (5) 30

31 Quality Suing option Period High altruistic Low altruistic Figure 14: The composition of the sample by altruism We observe that personal characteristics are significant to understand bribe in suing treatments, while they are not in the baseline treatments. In suing treatments, we find several significant variables. The less altruistic subjects are, the more they offer a bribe (in other words, more altruistic subjects invest more on quality). The more risk averse subjects are, the more they bribe. If firms have won in the previous period they bribe less. On the contrary, in suing treatments, bribe in the previous period does not explain current bribes, whereas having been audited for bribery in the previous round, increases one s bribe in the current round. 31

32 y=bribe M.L.S.(1) M.L.S.(2) M.L.S.(3) M.L.S.(4) Baseline Suing Baseline Suing Baseline Suing Baseline Suing Altruism 0.31 * * 0.31 * * 0.33 * * (1.64) (-1.64) (1.63) (-1.65) (1.69) (-1.64 ) (0.73) (-1.22) Envious ** 0.56 ** ** * (-1.94) (0.63) (-1.93) (0.62) (-1.80 ) ( 0.61 ) ( -1.36) (0.76) Risk aversion *** *** *** *** (0.04) ( 3.99) (0.05 ) (4.04) ( 0.15) ( 4.03 ) ( 0.29 ) ( 3.46) Gender (1.52) ( 0.55 ) ( 1.51) (0.54) Bribe [t-1] 0.15 *** *** ** (3.15) (-1.62) (3.07) (-1.50) ( 2.78) (-1.51) Win [t-1] *** *** *** *** (0.3) (-4.09) (0.26) (-3.94) (0.13) (-3.98) (-1.23) ( -3.98) Citizen ** ** ** * (-2.12) (-0.65) (-2.11) (-0.69) (-2.10) (-0.69) ( -1.70) (-0.45) Period *** (-0.27) (-4.56) Constant *** *** *** (6.45) (0.8) (6.91) (-0.09) (7.34) (0.10) (7.12) (-0.15) R2 within R2 between R2 overall Observations Groups Table 7: M.L.S. regression with dependent variable: y= bribe 32

33 Model 2: Bring to suit Model 2 tries to explain what determines a subject s decision to bringing suit. The endogenous variable is dichotomous, where 1 corresponds to the loser firm bringing suit, and 0 otherwise. Independent variables are Altruism, Jealous, Loss Aversion 1, Loss Aversion 2, Audited (being suit) in previous period, Gender and Bribe. Table 8 reports coefficients and the corresponding standard errors. Bringsuit = α + β 1 Altruism + β 2 Jalous + β 3 Lossaversion1 + (6) β 4 Lossaversion2 + β 5 Besuit[t 1] + β 6 Gender + β 7 Bribe From our mode two variables seem to explain the decision to bringing suit. The less jealous subjects are, the less likely they are to bring suit. We also created two variables concerning subjects loss aversion. Loss aversion 2 is significant with a negative sign. Thus, the more averse subjects are to losses, the less likely they are to bring suit. 33

34 Probit (1) Probit(2) Probit(3) Probit(4) Probit(5) Altruism (-1.32) (1.35) (-1.30) Jalous -0.26* -0.28** -0.28** -0.25* -0.22* (-1.92) (2.63) (-2.07) (-1.84) (-1.71) Loss aversion (0.87) (0.96) (0.98) (0.95) Loss aversion2-2.28* -2.07* -2.11* -2.24** -1.78* (-1.90) (-1.84) (-1.89) (-2.02) (-1.80) Be Suit [t-1] (0.00) (0.00) (0.00) (0.00) (0.00) Gender 0.16 (0.49) Bribe (0.53) (0.42) Constant (0.15) (0.38) (0.37) (-1.85) (-2.56) Table 8: Probit regression with dependent variable: y= Bring suit 34

35 Model 3: Pro-social town planner Model 3 aims at characterizing pro-social town planners. The endogenous variable is dichotomous and takes value 1 if the town planner s decision was pro-social and 0 otherwise. Bribe = α + β 1 Altruism + β 2 Jalous + β 3 Riskaversion +β 4 Bribewinner + β 5 Qualitywinner + β 6 Gender + β 7 Subject (7) 35

36 Probit(1) Probit(2) Probit(3) Probit(4) Baseline Suing Baseline Suing Baseline Suing Baseline Suing Altruism -0.22* 0.44*** ** *** *** (-1.69) (2.98) (-1.52) (1.99) (-1.54) (3.83) (-0.81) (4.00) Jalous * ** (-0.56) (1.95) (-0.08) (1.33) (-2.05) (1.03) (-1.42) (0.75) Risk aversion -3.01* 6.50** -2.70** 8.53* (-1.79) (2.23) (-1.66) (1.90) (-1.17) (-0.38) (-0.45) (-0.58) Bribe winner -2.14*** -0.50** -2.18*** (-3.06) (-2.11) (-3.08) (-1.26) Quality winner -1.41** 0.91** -1.30*** 1.32*** (-2.23) (2.38) (-2.31) (2.67) Gender (0.44) (0.24) (0.27) (-0.32) Subject * (0.66) (-1.80) (1.57) (-0.87) Constant 18.10** ** 17.96*** ** (2.73) (-2.34) (2.67) (-2.56) (0.31) (-0.25) (-0.35) (-0.39) Table 9: Probit regression with dependent variable: y= town planner pro-social 36

37 4.3.4 Gender Finally we elaborate a brief analysis concerning potential gender differences. We explore the main decisions to observe if some clear effects appear when comparing women and men. Hence, on the firms side we concentrate on divergences on bribe and bringing suit and for town planners we examine differences on preferences. Figure 15 represents the evolution of bribe for women and men along the whole experiment. Bribe women men Suing option Period Men Women Figure 15: The evolution of bribe, by gender Though we do not appreciate clear differences in this figure, we notice that women have bribed more than men in most of periods. To formalize this finding we calculate averages and run a non-parametric Mann-Withney test. Table 10 recapitulates all these results. 37

38 Treatment Bribe Men Bribe Women Mann-Withney p-value T T0s *** T * T1s Total *** Table 10: Bribe differences across genders. From this table we extract some interesting results. Women offer higher bribes than men in each treatment and hence in the whole experiment (see Jaber et al. (2011)). This differences are significant at 1%. We also find differences in treatment T0s (Mann-Withney p-value= 0.00), and in treatment T1 (Mann-Withney p-value=0.06). Figure 16 represent the evolution of bringing suit, by gender. We do not find significant difference on the frequency of bringing suit for women and men. Both denounce on average 25% of the occasions in which they have the possibility. Figure 17 illustrates the differences on town planners preferences among gender and treatment. It seems that no differences are found on baseline treatments while in treatments T0s and T1s some divergences appear. In T0s male town planners are more pro-social than women. (81% for men vs 59% for women). Indeed 15% of women are anti-social (0% for men). In T1s the opposite occurs, there are slightly less pro-social men than women (73% for men vs 82% for women). However, again 5.3% of women are anti-social while no men are found as antisocial. In fact, when the suing option is available women are acting in a more 38

39 Percent men women Suing option Period Men Women Figure 16: The evolution of denouncing, by gender anti-social manner as a way to punish other agents. This part overviews gender differences. We found women bribing more and acting more in an anti-social way as town planners when the suing option is available. No other gender differences were found. 5 Emotional responses In an ongoing series of experiments we have also controlled for skin conductance reactions associated with psychological arousal levels in specific moments of the decision-making and feedback-receiving process. We report here the results obtained from two specific instances associated with the arousal that subjects feel when engaging in bribery (or not) and when waiting for others to bring them to suit. 39

40 Men - T0 Men - T0s Men - T1 Men - T1s Women - T0 Women - T0s Women - T1 Women - T1s Rational Pro-social Figure 17: Town Planners preferences according to gender Figure 18 shows that firms who choose not to bribe exhibit on average an increased skin conductance response compared to those who did bribe in baseline treatment. It appears significant differences between both types of profiles. Thus, those subjects who are honest but against their monetary maximization demonstrate an internal conflict exposed in emotional responses. Figure 19 represents the same event than the previous one but in whistleblowing treatment. Surprisingly in this treatment the result is completely the opposite. Firms who bribe exhibit higher arousal than those not bribing. In this case we incorporate the risk of losing the profits, thus the threat of being denounced yields to an emotional stress for firms while bribing. Then in figure 20 we illustrate the emotional response of firms winning the license while they could be denounced by loser firms. Those having bribed display significantly higher emotional response than those not bribing. Again, 40

41 No Bribe Bribe 0.1 SCR (μsiemens) p< Time (sec) Figure 18: Firms skin conductance response during post-decision seconds (Baseline treatment) the risk of losing the monetary profits with no possibility of rectify make subjects to feel an emotional aversion. Figure 21 shows town planners physiological reaction in the treatment with a suing option at the moment of their project assignment decision. We compare those giving the license to a briber with those preferring bribe-free bids. No differences are found. Figure 22 shows town planners emotions while waiting for the loser firm to decide whether to audit the winning proposal or not. Comparison between planners choosing a bribe-free project with those obtaining a bribe shows no difference. Therefore, results of the skin conductance measurement of emotional arousal show an interesting overall pattern concerning bribers. They feel a higher 41

42 No Bribe Bribe 0.1 SCR (μsiemens) Time (sec) Figure 19: Firms skin conductance reaction during post-decision seconds (Whistle-blowing treatment) arousal when they do not bribe in the baseline treatment and when they do in the denouncement one, namely, when they opt for the least common strategy in each case. On the contrary receivers of bribery seem not to react, confirming the fact that passive bribery carries a moral excuse because it is indirect as opposed to active bribery, which is direct. Finally, bribing winners waiting for losers to decide whether to audit them or not exhibit significant degrees of emotional arousal. During the waiting time for losers deciding on the suing option, passive bribery does not cause offenders any additional emotional pressure compared to that experienced by non offenders. This result completes our argument concerning the lower emotional pressure of passive bribery as opposed to active one. To summarize, subjects seem to feel aroused when acting in dissonance to their neoclassical utility function, whereas the consequences of breaking an explicit or implicit social norm are revealed in the expected way, namely active offenders are more aroused than passive ones when threatened 42

43 No Bribe Bribe 0.1 SCR (μsiemens) p< Time (sec) Figure 20: Winners skin-conductance while waiting for loser to decide whether to blow the whistle with a punishment for their action. 6 Concluding Remarks In this work we propose a model of corruption in a context of an auction of a public project to one of two firms competing via project quality and a bribe sent to the town planner deciding on the winner of the auction. Using experimental methods we have shown that people are intrinsically attracted to prosocial behavior. Furthermore, we focus on the role of two design conditions: (i) a highly risky option available to the losers of the auction to investigate the deal between the planner and the winner of the licence. If the investigation reveals no bribe, the accusing firm loses all its profit, otherwise the accused firm and the planner lose their profits. Despite the few times that the option is actually used, bribes fall dramatically; (ii) a citizen who is a passive receiver of the consequences of the potentially corrupt transaction. We observe that 43