Professional Identity, Bribery and Public Service Delivery: Evidence from a lab-in-the-field Experiment in Burundi

Transcription

1 Professional Identity, Bribery and Public Service Delivery: Evidence from a lab-in-the-field Experiment in Burundi Jean-Benoit Falisse Centre of African Studies, Edinburgh University Nastassia Leszczynska SBS-EM, ECARES, Université libre de Bruxelles and FNRS May 2017 ECARES working paper ECARES ULB - CP 114/04 50, F.D. Roosevelt Ave., B-1050 Brussels BELGIUM

2 Professional identity, bribery and public service A lab-in-the-field experiment in Burundi 1 Jean-Benoît Falisse 2 Nastassia Leszczynska 3 4 May 2017 Abstract Professional identity is deemed to be a key determinant of economic behaviour: the way an agent defines herself shapes her decisions. This paper explores the effect of professional identity reminders on corrupt behaviour. In a newly designed lab-in-the-field experiment, 527 public servants from Burundi were asked to allocate rationed vouchers between anonymous citizens, some of whom tried to bribe them. Public servants who were randomly exposed to a sensitization message that stressed professional identity and values behaved in a fairer manner than those exposed to a standard anti-corruption message or no message. This suggests that the moral cost of bribery increases when an agent reflects upon her professional identity, thereby prompting her to more equitable service delivery. However, another side of corrupt behaviour, bribe-taking, did not change with the messages, which highlights the importance of considering different components in corruption. The experiment provides fresh insights into the design of anticorruption strategies. JEL keywords: C91, D73, H40, O12, H83 Keywords: identity, corruption, public service delivery, public servants, Burundi In the last decades, a growing body of literature has been concerned with the monetary and non-monetary incentives behind corrupt actions (RoseAckerman 1975; Shleifer and Vishny 1993; Drugov 2010). However, few researchers have explored the relationship between identity and corruption, and when they have, their focus has been on culture, defined as shared values and beliefs among social groups (Banuri and Eckel 2012). The present research takes a different perspective and investigates the role of professional selfidentity, which is how a person perceives him/herself (Baumeister 1999a) as a professional, when facing bribery. The paper explores the idea that relationship to identity is a crucial determinant of how an agent will face corruption attempts. In addition to contributing to the 1 Favourable opinion of ULB Ethics committee regarding the experiment (ref 026/2013). This research was generously funded by , a coalition of NGOs, unions, movements and solidarity groups in Flanders (Belgium) and supported by the University of Burundi. We would like to thank Alix Nijimbere for his invaluable support, as well as Michel and Divine, our lab assistants. Y. Nindorera, Y.-A. de Montjoye, A. Nkurunziza, E. Ndayegamiye, and M. Jacobs provided us with precious help for designing the experiment. Thanks are due to A. Estache, G. Kirchsteiger, P. Verwimp, O. D Aoust, M. Kissine and R. Mulligan for feedback on previous versions. Nastassia Leszczynska acknowledges financial support from the Fonds National de la Recherche Scientifique (FNRS). 2 Centre of African Studies, Edinburgh University, jb.falisse@ed.ac.uk 3 SBS-EM, ECARES, Université libre de Bruxelles and FNRS, nleszczy@ulb.ac.be 2

3 fields of identity and corruption economics, there is also a pragmatic interest in such a study as it may give impetus to new and more efficient anti-corruption strategies. This study looks at bribery in the context of a laboratory game that required actual Burundian public servant participants to allocate vouchers representing access to a rationed public service. The scenario seeks to understand not only when public servants engage in bribery but also how a bribe offer affects their public service delivery. We find some evidence that reminding public servants of their professional identity and the qualities expected from it influences their delivery of the public service, i.e. vouchers allocation, but not their propensity to accept bribes. Furthermore, the participants seem to separate the dimensions of inequity and bribery-taking, which also correlate with different socio-demographic variables. The first section of this paper connects our research with the literature on the economics of corruption, provides explanations for the use of identity reminders, and gives background information on Burundi and its public service. The experimental design is explained in the second section and the results are presented in the third section. A last section discusses the potential mechanisms that underpin the results as well as their policy and research implications. 1 Background 1.1 The Economics of Corruption and Identity Corruption, which can be defined as the abuse of entrusted powers for private gain 4, can occur at different levels and take different forms (bribery, nepotism, regulatory capture, etc.). Among them, the bribery of low-level public servants is certainly the most frequently experienced by ordinary citizens, especially in low-income and fragile settings (Khan 1996; Noonan 1988; Van Rijckeghem and Weder 2001). There exists a widespread consensus that the practice of corruption generates negative externalities that jeopardize economic development and damage social life. 5 A large number of macro level surveys evidenced this point (Lambsdor 1999; Mauro 1995; Treisman 2007) and could not be disproved by the few empirical papers investigating the potential benefits of corruption (Aidt 2009). The corruption phenomenon not only involves societal and institutional but also individual dimensions (Guerrero and Rodriguez-Oreggia 2008). Personal and social characteristics (e.g. age, experience, religion or gender) constitute the identity of an individual, which are known to in uence behaviour related to corruption (Armantier and Boly 2013; Serra 2006; Swamy et al. 2001). The identity of an individual is always relative to the group she belongs to, and the way she relates to it (Baumann and Gingrich 2004). In social psychology, Tajfel 1978 defines the concept of social identity as the part of an individual s self-concept that derives from his knowledge of his belonging to a social group. One s identity is also a matter of self-definition, and this feature is captured by the concept of self-identity defined by Baumeister 1999b as: the individual s belief about himself or herself, including the person s attributes and who and what the self is. This relationship to identity is potentially crucial for understanding when an individual engages in corrupt behaviour. 4 Transparency International, World Bank. 5 Corruption hampers economic growth (Mo 2001; Fisman and Svensson 2007), jeopardizes social cohesion and welfare (Chan, To, and Chan 2006), and increases inequalities (Gyimah- Brempong 2002). 3

4 Measuring and understanding corruption is challenging. Since people rarely and reluctantly share their experiences as bribers and bribees, surveys usually only gather opinions on and estimations of corruption rather than first-hand experiences (Kaufmann, Kraay, and Mastruzzi 2006). There is some evidence that this sort of perception data does not affect the actual level of corruption correctly (Olken and Barron 2009). Micro-level surveys combining direct and indirect observation of corruption constitute an improvement in the measurement issue (Duggan and Levitt 2002; Olken 2009; Reinikka and Svensson 2004; Sequeira and Djankov 2010), but they are scarce and costly. They also tend to focus on the briber and depend partly on potentially biased self-reporting bribes. Experiments, including the lab-in-the-field approach of this paper, constitute an interesting alternative approach to understanding corruption dynamics. They provide first-hand information on individual corrupt behaviour (Abbink and Serra 2012; Treisman 2007; Barr and Serra 2009a). Experiments on corruption started about 15 years ago 6, with Frank and Schulze 2000 designing a controlled field experiment where students faced the temptation of receiving money for taking a decision against public interest. A few years later, Abbink 2002 designed laboratory bribery games that involved a public official and a firm in three variations: pure reciprocity, negative externalities and possible penalty. The issue of service delivery was only touched upon through the channel of negative externalities. Scholars have since adapted the game to investigate other aspects of corrupt behaviour and test anticorruption strategies (Azfar and Nelson 2007; Jacquemet 2005). 1.2 Public service in Burundi Most corruption experiments are conducted in laboratories with students who play noncontextualised games or are instructed to act the part of public servants (Barr and Serra 2009a). As Serra and Wantchekon 2012 point out, a key methodological challenge is to find ways to articulate the advantages of laboratory and field research in producing contributions that can claim some level of external validity. These concerns are especially relevant in the present study, which is primarily concerned with the identity of the participants and intends to generate relevant insights for policy making. Hence our decision to organise a lab-in-thefield, or extra-lab experiment (Charness, Gneezy, and Kuhn 2013), where the selected participants are also the subjects of our study, i.e., public servants. The research is set in Burundi, a fragile country where the delivery of public services and the corruption of public servants are major problems and where there is an interest for new anti-corruption strategies. Burundi ranked 159 out of 175 in the 2014 Transparency International corruption perception index. In 2011, 74% of the Burundians reported to have paid a bribe in the last 12 months (Corruption Perception Index 2012) and the International Crisis Group denounced a worsening of the corruption crisis in Authority disruption and government failures in the last decades as well as the civil wars are believed to have fuelled corrupt behaviours. Reports, such as the one published in 2007 by International Alert and GRADIS, find feelings of restrained anger and resignation towards corruption in all regions of the country. They highlight the dramatic consequences of petty corruption in political, economic, and also social aspects of life in Burundi. A series of anti-corruption initiatives have been taken in recent years; they include a revision of the legal framework as well as stricter sanctions and measures to improve law enforcement. These include the creation of an independent tax revenue authority and a special anti-corruption police brigade (International Crisis Group, 2012). These measures 6 For an extensive review presenting recent advances in the study of corruption based on experiments, see Serra and Wantchekon

5 however have proven largely insufficient and corruption has been an important theme in the 2010 and 2015 electoral campaigns. The despondency of Bujumbura residents towards rampant corruption reportedly contributed to fuelling the riots of At US$600 ppp 7, Burundi s GDP per capita is one of the lowest in the world and agriculture is the main source of income and employment for 90% of the Burundian population. The administrative system is largely inherited from Belgium, the former colonial power, and has been influenced and reformed with and by international aid support over the last decades. With few opportunities in the private sector, public service positions are highly regarded and are seen as an easy way to climb the social ladder and become wealthier. A recent survey conducted in Burundi on children professional aspirations finds that a majority of respondents - 35% - aspire to work in the public sector in the future (Jeusette & Verwimp 2017). The capture of resources and the politicization of the Burundi public service are very important and provide public servants with significant power and material advantages (International Crisis Group, 2012). The participants of the study were active public servants coming from health (clinics), education (primary and secondary schools), justice (non-judge staff at civil and criminal courts), and police sectors. These services having been qualified as central to the general citizenry (The East African Bribery Index 2013) and relating differently to corruption. According to the East African Bribery Index compiled by Transparency International (2013), policemen score first both in terms of the likelihood to ask for a bribe and in the share of total bribe they collect (24.7%, 52%). They are closely followed by the judiciary sector (21%, 27.8%), and the education sector (12.9%, 13.2%) with medical services appearing much less corrupted (3.1%, 0.4%). 1.3 Conceptual Framework and Rationale Donors and governments often use sensitization to improve the governance of fragile states. If sensitization campaigns are a non-expensive tool to try influencing citizens behaviour, their effectiveness is far from being established. Adequate attention rarely goes to the incentives and values the campaign aims at improving. In particular, sensitization campaigns based on identity need to be tested as a non-expensive approach with a potential impact on public service delivery. In economics, Akerlof and Kranton 2000 have stressed that agents make economic decisions not only on the basis of rational incentives but also in relation to their identity and self-image. They incorporate an identity component into the classic utility function in such a way that utility increases when a person s behaviour corresponds to their ideal self. This is consistent with experiments in social psychology that have shown that priming individuals on their identity influences ethical behaviour. Mazar, Amir, and Ariely 2008 argue that people who take a bad action deploy unconscious strategies to distance their self-identity from their unethical behaviour. People behaving unethically will typically disconnect their actions from their identity, minimizing or denying harmful actions. In the case of corruption, it could mean that the more blurred the negative externalities, the easier the denial. In line with this idea, behavioural psychologists Bryan, Adams, and Monin 2012 observed a decrease in cheating behaviour when asking the participants of a game not to be cheaters, rather than simply asking them not to cheat. Even though this experiment is not directly on corruption, it is an important insight that echoes theoretical findings on the importance of one s (perceived) identity in economic and social behaviour (Schlenker 1982). In a recent study on bankers social norms in Switzerland, Cohn, GDP per capita Purchasing Power Parity, World Development indicators. 5

6 Fehr, and Maréchal 2014 shed light on the power of professional self-identity, showing that reminding bankers of their profession has a negative impact on their ethical behaviour. In line with this literature, this article seeks to test whether such mechanism also plays out in the case of corruption attempts: are individuals who are reminded of their professional identity, in our case being a public servant, less likely to be corrupt? 2 Experimental Design 2.1 The game The study sought to reproduce a situation of petty corruption with a scenario involving a public servant, citizens, and the distribution of a basic public service (Barr and Serra 2009a). In our design, the public service was available in limited quantity only (rival good). This rationing feature was crucial to test for equity in terms of service provision. It also mimicked a real life situation, e.g. shortage of basic goods, limited availability of places in a service (schools), or production limitations to issuing administrative documents. The game was never presented to the participants as a bribery game: it was introduced as a game about the delivery of public services. After all the sessions were finished, debriefing was done by phone. The main participants, all actual public servants, were assigned the role of public servant in the game. They were expected to deliver a public service (in the form of allocating vouchers) to citizens. The citizens, recruited among the students of the University of the Great Lakes in Bujumbura, played in a different game session. 8 The game followed a structured sequence: 1. Preliminary stage: In a separate session, the citizens made requests for vouchers. They were informed that (1) the official price for requesting one voucher was BIF 500 9, (2) vouchers could be exchanged (at the end of the game) against BIF 1,250, and (3) there was a rationing situation which meant that civil servant might not be able to fulfil every request. The citizens could choose to make normal requests for vouchers, sending the public servant a BIF 500 request for one voucher, or special requests, sending the public servant a BIF 1,000 request for one voucher. The requests, along with game money, were then transferred to the public servants. The transfer was operated in such a way that each public servant would receive an identical distribution of requests. The citizens are passive players who are crucial for the credibility of the game: the public servants were informed that real people would be affected by their decisions in the game. 2. Stage 1: The context was explained to the participants. Public servants were given the task of managing the distribution of vouchers among citizens. It was clearly explained that (i) it is a public service that is limited in quantity (e.g. in a rationing situation) and that (ii) the public service is basic and equally deserved by each citizen. Public servants were therefore expected to distribute the vouchers equally among citizens 8 The game played by the citizens is not the focus of the present article and will not be discussed in length here, as our focus is on the identity of the bribee, i.e. the public servant. 9 $1 = BIF 1,600 at the time of the experiment 6

7 3. Stage 2: Each public servant received an endowment of twelve vouchers. Public servants were instructed to distribute these twelve vouchers among three citizens (vouchers can be cashed by citizens only). 4. Stage 3: Each public servant received three seemingly similar envelopes from the citizens; they contained anonymous written requests for vouchers as well as their payment. One of the three envelopes contained a bribe. The bribe consisted of a special request for more vouchers than what a citizen was entitled to receive as well as more money than the vouchers normally cost. Each public servant received two normal requests and one special request. Each public servant received: 12 vouchers giving access to the public service. Each voucher costs BIF 500 to the citizen and can be cashed in against BIF 1,250 at the end of the game by citizens only. 3 envelopes containing citizen requests: o 2 envelopes containing a request for 4 vouchers and BIF 2,000 in game money. o 1 envelope containing a special request for 6 vouchers and BIF 6,000 in game money. 5. Stage 4: Each public servant distributed their vouchers among the requests by putting them into the corresponding envelope. In practice, public servants were free to keep whatever amount of game money the citizens sent them; although the instructions indicated that only the amount of game money corresponding to the cost of the vouchers should be kept. 6. Stage 5: Each public servant exchanged game money against real money at the end of the game. 10 Two dimensions of corruption were studied and linked to the non-respect of two rules: 1. Bribery (price): the amount of money taken by the public servants that does not correspond to the delivered voucher(s). 2. Equity in public service delivery (resource allocation): the number of vouchers distributed by the public servant to each citizen. People were thus given the opportunity to claim the money they were not entitled to at the citizens expense. The briber had to decide whether and how much to offer as a bribe without knowing whether the bribee would be willing to grant him the favour he asked for in return and the bribee was free to reject the bribe, accept and grant the favour, or accept but not grant the favour (Barr and Serra 2009a). In the cases where the favour asked by the briber was delivered, other citizens lost voucher(s), which represent the social and economic cost of corruption. In order to investigate the role of professional identity reminders, two message treatments with change in phrasing were introduced in the fashion of Bryan, Adams, and Monin 2012 s cheating game. Our messages, however, do not use negative and imperative wording and come in a booklet that contains practical information before introducing the game s rules and instructions. Messages were delivered at the start of the session; they were positive reminders about good governance behaviour, notified through a written message in the native language of the participant before giving the instructions of the game. Public 10 The citizens were matched with what corresponds to the average response of public servants to their request, and were paid accordingly. 7

8 servants were randomly assigned a no message treatment, a general message treatment, and a professional identity message treatment. The general message was based on a classical good governance statement that is often used in Burundi and translates as: Good governance is the pillar of an equitable and uncorrupted society. The professional identity message contains a subtle linguistic change reminding them their professional identity before they make a decision and reads: A real public servant is equitable and incorruptible. The content of the general message and the professional identity message do not differ ethically and they come from the same external source. They are, however, expected to have different moral loading, which means that they affect the participant s own relationship to what they de ne as right or wrong. As underlined in the background section, the hypothesis is that the professional identity message makes it harder for the participant to disconnect their actions from their identity. The abandonment of professional values is therefore less easy and this should makes participants less prone to engage in bribery. 2.2 Recruitment The importance of recruiting participants currently working in the public administration has already been highlighted. Accordingly, the participants were all public servants. They all had low-level qualified positions and were deliberately not recruited from managing or senior positions (such as school principal or police officer). In order to recruit participants, leaflets and posters advertised the academic study in workplaces (health centres, schools, ministries). They guaranteed the participant s anonymity and a participation fee that included travel expenses. Registration was done by phone, and was conditional on whether the candidate was indeed a public servant and could read and write the national language Kirundi. Throughout each step of the study, recruitment, experiment, survey, and payment, total anonymity of the participants was respected. The participants played in individual cubicles and were paid individually after the experiment was finished. At no point were the participants asked for their names. The post-conflict context of the country, as well as the sensitive nature of the experiment required this condition to be fulfilled in order to encourage participation and obtain spontaneous behaviour during the game. Anonymity was also important to diminish the feeling of being observed which could influence (moral) choices during the experiment. In order to link the results of the game with the socio-demographic survey, each participant was identified with a unique number during the study. 2.3 Setting The experiment took place in December 2013 and January 2014 in a classroom provided by the University of Burundi in Bujumbura. The game was one-shot and played in paper and pencil form. After the game, participants were asked to fill in a questionnaire about their socio-demographic background. It also entailed questions about their opinions regarding personal and professional values but did not explicitly mention corruption. The experiment manager and his lab assistants were Burundians; at no point were foreign experimenters in contact with participants. The whole session was run in Kirundi to make sure that all the rules were understood and that a foreign language would not reduce the participant s emotional response (Costa et al. 2014). Participants received their pay-off at the end of the session after having filled in the post-experiment questionnaire. In addition to a lump sum show up fee of BIF 5,000 ($3.25), the participants received between BIF 6,000 to BIF 10,000 ($3.75 to $6.25), depending on their decisions during the game. The minimum amount that could be won during the game (BIF 11,000 - $6.875) corresponds to two days of daily wage for the less well-paid category of public servant participating in our game. 8

9 3 Empirical analysis 3.1 Descriptive statistics Table 1 Basic socio-demographic indicators, by profession (1) total (2) nurses (3) teachers (4) clerks (5) police mean sd mean sd mean sd mean sd mean sd Gender (female) Age (in years) Education: primary post-secondary Internally Displaced Returnee Exp. of violence (1-4) Ex-combatant Fish/meat per week Never accessed internet Observations The sample consists of 527 public servants whose basic socio-demographic characteristics are presented in table 1. Participants of all professional categories are more likely to be male, except in the health sector. The level of education is remarkably lower in the police sector. Other indicators are typical of low-income contexts: large household (6.21 members on average), limited access to the internet and computers (which justifies the pen and paper form of the game), and an important level of material poverty as shown by the low proportion of households that have a car or motorbike as well as the absence of fish and meat in daily meals. The level of violence is high and affects the fragile political situation of Burundi: 62% of the participants declared that their life and the life of their families was disturbed by violence in the last year. The Burundi police forces were partly made up of exrebels and it is therefore not a surprise that it counts most of the internally displaced persons, returnees, and ex-combatants. In the recent years, court of justice officers could earn (monthly) BIF 55, ,000, policemen BIF 20, ,000, teachers around BIF 50,000, and nurses between 50, ,000. A new policy introduced at the end of 2012 has tried to harmonize the salaries between BIF 110,000 and 300,000 a month, according to clear criteria of qualification and experience. Table 2 shows the three groups are relatively balanced but not perfectly similar, which means that controlling for socio-demographic indicators will be important. Similarly, work experience (table 3) and exposure to corruption (table 4) will also need to be controlled. The participants have been in their position for variable amount of time (see table 3), with the police force and the nurses usually having more experience than teachers and court of justice employees. This may explain why those categories also display higher rates of promotions and salary increases. The satisfaction of the participants with their job is moderate, and about 30% of the public servants have another job on the side (shop, taxi, etc.). This figure is much lower in the case of the policemen who are often stationed in barracks and therefore enjoy less flexibility to engage in a complementary professional activity. 9

10 Table 2 Socio-demographic indicators by group, by group Control Message 1 (standard) Message 2 (prof. identity) mean sd mean sd mean sd Age (in years) Gender ( female) Only primary school Finished secondary school Catholic Size of the household member religious group cultural / sport group self-help group Cell phones in the house TV set Radio Car/motorbike Catastrophic expenditure Ex-combatant Internally Displaced (IDP) Returnee Exp. of violence (1-4) Nurse Teacher Justice Police Observations When asked their opinion about the main problems of the country in our post-game questionnaire, corruption and good governance tops the list of the three main problems according to the participants. It comes ahead of violence and leadership, with the issues of ethnic division, human resources, and infrastructure mentioned in one third of the cases or less and rarely ranked top priorities. When asked who sets the example in terms of good governance in the country, public servants disagree: the president and his entourage are cited by 39% of the respondents, especially the policemen who are in part ex-rebels who fought alongside the president, but the church(es) are a more frequent choice among teachers. Table 3 Work experience, by group (1) total (2) nurses (3) teachers (4) clerks (5) police mean sd mean sd mean sd mean sd mean sd Work experience (year) Promoted? Had salary increase Job satisfaction (0-4) Has a second job Observations

11 Finally (table 4), on average our public servants affirm that they are exposed to corruption from time to time and mention an important level of control of their activities particularly in health centres. Policemen (and to a lesser extent court staff) received more training on good governance. This was mostly as part of the Disarmament, Demobilisation, and Reintegration (DDR) and capacity-building programmes supported or implemented by international aid. Table 4 Experience with corruption, by group (1) total (2) nurses (3) teachers (4) clerks (5) police mean sd mean sd mean sd mean sd mean sd Governance training Sensitization session Bribery exposure (0-4) Controlled work (0-4) Observations Identification strategy Experimental results are primarily analysed with OLS, Logit, and Count (Poisson) 11 regressions, using the following models, with robust standard errors clustered per game session: B it = c + β 1 T 1 + β 2 T 2 + S t + ε i (1) F it = c + β 1 T 1 + β 2 T 2 + S t + ε i (2) Bribery (B it) is the public servant s behaviour towards money. The instructions specified that the money taken should correspond to the value of the delivered vouchers. Out of BIF 10,000 in total, a public servant is expected to take only BIF 6,000. Bribe amounts can then take values from BIF 0 to BIF 4,000, BIF 4,000 representing the maximum bribe amount. In the Logit model, Bit is a binary variable that takes the value 1 if the participant took more money than BIF 6,000. Fairness (F it) is the public servant s distribution of vouchers. The instructions specified that every citizen has an equal right to access the limited public service. Out of 12 vouchers, an equitable public servant should deliver 4 vouchers giving access to the public service to each citizen. Inequity level can then take values ranging from 0 to 16, according to the number of vouchers misattributed. These values are necessarily even, as reattributing a voucher means both depriving one player whilst favouring another. In the Logit model, I it is a binary variable that takes the value 1 if the participant allocated her vouchers in any other way than the equitable T 1 is a dummy indicating whether the individual was exposed to the first treatment, the standard message, and T 2 a dummy for the identity message. S t are fixed effects capturing session-level invariants: sessions were organised on different days but in the same room and by the same assistants. In alternative specifications, we also added, X it, a vector of extra personal covariates that cover basic sociodemographic indicators (age, education, gender, and wealth), religion, experience of war and violence, profession and work experience, and exposure to corruption and anti-corruption. 11 We did not use zero-inflated models as there is no reasonable hypothesis of a process explaining the excess zeros. 11

12 For 75 observations, there is at least one missing answer to the questionnaire, which translates as missing values for some socio-demographic variables. We chose to replace those missing values by the mean value of the variable in the sample. 4 Results Experimental games were a novelty for all the participants and, despite an effort to design a game that was as simple as possible and recruit only participants who would read and write the national language of Kirundi, 35 participants (6.6%) displayed behaviour that may indicate that they had not understood the principles of the game: they did not take part of the money corresponding to the vouchers they distributed, and therefore essentially gave vouchers at a discounted price (or for free). 12 These 35 participants were only included in the analysis at the time of performing robustness checks. The allocation of vouchers relative to the amount of money taken beyond what was expected and instructed, suggesting a corrupt behaviour, is represented in the figure below (figure 1). The most frequent case is, by far, to take the bribe but fairly distribute the vouchers. Figure 1 Bribe-taking and voucher allocation 12 Out of these 35 outliers, 11 did not take any money out of the envelopes. This behaviour will be discussed in the next section. 12

13 4.1 Bribery The overall level of bribe-taking is very high: 87.95% of the participants took part of or the entirety of the bribe. On average, 67.4% of the bribe (BIF 3,373) was taken. When looking at the data in table 5, there is only a very slight and non-significant difference between the amount taken in the professional identity message and standard message groups. Table 5 Effects of the messages on bribe-taking (1) (2) (3) (4) OLS OLS Logit Logit Any message (182.1) (0.497) Message 1 (standard) (157.8) (0.426) Message 2 (prof. identity) (233.4) (0.723) constant *** *** *** *** (125.5) (129.6) (0.0232) (0.0200) Session FE yes yes yes yes N The outcome variable is the amount of extra money taken (between BIF 0 and 4,000). Standard errors in parentheses, standard errors clustered per session. * p < 0.10, ** p < 0.05, *** p < 0.01 The distributions of the three groups are quite similar, showing again that the attitudes of the different groups are not very different (figure 2). 90% 85% 80% 75% 16% 14% 12% 10% 8% 6% 4% 2% 0% no message standard message professional identity message Figure 2 distributions of levels of corruption (in BIF of bribe taken), by group 4.2 Public service delivery 76.21% of the participants chose to allocate their 12 vouchers equitably (four vouchers for each citizen request). On average 0.51 vouchers out of 12 were misattributed. The level of inequality is computed as 4 V V V 3, where V i is the number of vouchers in an 13

14 envelope. Very clearly, the scenario is the one preferred by the participants across the three treatment groups. As indicated in table 6, participants who received the standard or the identity message are less unequal in their voucher distribution. The effect of the standard message is only borderline significant and valid only for the logit model. The effect of the professional identity message is, significantly, negative and of a size close to the one of the standard message, meaning that the added-value of the professional identity may be on the distribution (more homogenous behaviours than in the case of the standard message). This result holds in all the aforementioned robustness checks. The Logit effect corresponds to a marginal effect of Table 6 Effects of the messages on equality (1) (2) (3) (4) OLS OLS Logit Logit Any message * ** (0.234) (0.220) Message 1 (standard) * (0.311) (0.266) Message 2 (prof. identity) ** ** (0.185) (0.208) constant 1.293*** 1.292*** *** *** (0.161) (0.155) ( ) (0.0107) Session FE yes yes yes yes N Standard errors in parentheses, standard errors clustered per session. * p < 0.10, ** p < 0.05, *** p < 0.01 The difference between the distributions is noticeable, especially between the control and two intervention groups. 85% 80% 75% 70% 25% 20% 15% 10% 5% 0% no message standard message professional identity message Figure 3 distributions of levels of inequality (0 = equal), by group 14

15 4.3 Relation between bribe-taking and public service delivery The bribe-taking and public service delivery aspects of the experiment are, to a large extent, independent from each other. They are correlated with a set of variables that are quite different. Even when controlling for covariates in an OLS model, there is no statistically significant correlation between them at a p<0.1 level. Looking at the situation using a binary angle, there is still no association between the two dimensions (table 7). The most frequent combination is accepting a bribe while being equitable in voucher distribution % of the participants adopt this behaviour. Only 9.64% of the participants are both insensitive to the bribe and equal in distribution, thus respecting the two rules of the game and being a model public. Only a fifth of the sample chooses to give some extra vouchers away when taking the bribe. Table 7 Association between voucher allocation and bribe-taking bribe-taking (%) unequal allocation (%) No Yes No Yes Pearson s Chi 2 p-value Characteristics associated with bribe-taking and fairness The data allows to explore further the profile that are most commonly associated with corrupt and unfair behaviour. To do this, we simply added the vector of covariates X it to the models 1 and 2. In terms of corruption/bribe-taking (table 8), Catholics as well as people who are more experienced in their job and have been promoted are more likely to take the bribe. The rest of the results is not consistent across the two models. Controlling for a set of variables, teachers and court clerks take in average less money than nurses (the control group) and far less than police but they are not more or less corrupt when considering whether they take any money at all (the binary model). In terms of equal distribution of vouchers, being a public servant in the judiciary sector has a significant negative impact on inequity that holds across the different models. Exposure to campaigns of sensitisation again corruption, exposure to corruption, and training on good governance principles also bring more equal behaviour, but these results do not totally hold in the logit model. The robustness of the results was checked using a series of different specifications. In both the bribery taking and inequity in public service cases, we tested the models again (i) gradually adding groups of covariates on personal characteristics, (ii) using the professional category means rather than general means when imputing missing variables, (iii) inserting a dummy controlling for the observations whose covariate dataset is incomplete and including outliers, and (iv) squaring the outcome variable. The message effects are consistent across the different models. Although they do loose significance they keep the expected sign when dividing the sample by professional groups. 15

16 Table 8 Characteristics associated with bribe-taking and fairness (1) (2) (3) (4) Bribe-taking (logit) Bribe-taking (OLS) Inequality (logit) Inequality (OLS) Treatment Message (standard) (0.439) (170.2) (0.310) (0.324) Message ** ** (prof. identity) (0.657) (227.1) (0.213) (0.174) Basic Age demographics (0.0274) (13.17) (0.0263) (0.0208) Gender (0.437) (159.2) (0.197) (0.160) Education * (0.123) (40.85) (0.108) (0.115) Wealth (1.541) (626.0) (0.962) (0.918) Catholic * ** (0.356) (139.6) (0.281) (0.223) Experience of Ex-combatant * violence/war (0.604) (172.0) (0.445) (0.400) IDP (0.483) (212.8) (0.311) (0.255) Returnee * (0.565) (280.0) (0.243) (0.147) Exp. violence Professional identity Professional exp. with corruption (0.111) (37.68) (0.0919) (0.0653) Teacher *** (2.321) (271.6) (1.020) (0.335) Court clerk *** *** *** (0.901) (132.7) (0.206) (0.139) Police *** (2.359) (238.0) (1.097) (0.559) Experience ** ** (0.0394) (16.13) (0.0250) (0.0215) Vocation * (0.194) (57.03) (0.114) (0.102) Promotion *** (0.448) (126.0) (0.305) (0.202) Satisfaction * (0.177) (63.93) (0.178) (0.120) Work control (0.129) (57.18) (0.121) (0.0971) Sensitisation * ** (0.411) (157.2) (0.296) (0.233) Exposure ** (0.173) (68.92) (0.0909) (0.0839) Training ** (0.269) (134.3) (0.303) (0.241) constant *** *** (1.752) (768.8) (1.087) (0.777) Session FE yes yes yes yes N Standard errors in parentheses, standard errors clustered per session. * p < 0.10, ** p < 0.05, *** p < Wealth is a mean index made of the z-scores of the following indicators: weekly consumption of fish or meat, ownership of a mobile phone, car, TV set, and radio, all weighted by the size of the household. All dummies: belonging to a religious, cultural or self-help group All dummies except Experience of violence which is the answer to the question how much has violence disrupted your family life in the past year, on a scale from 1 to 5, 5 being very disrupted. 2. Teacher, Justice, and Police are professional categories, with 16

17 respect to nurses, which is the reference category. 3. Experience is the number of years of experience. 4. Vocation and promotion are dummies that take the value one if the participant had been promoted or does a job they consider their vocation. 5. Satisfaction and Work Control are each on a scale from 1 to 5, with 5 the most satisfied / controlled in their work. 6. Sensitization and training are whether the participant received a sensitization or a training session on good governance. 7. Exposure ranges from 1 to 5, with 5 being participants who say they are very often the subject of bribery attempts. 4.5 Heterogeneous effects The results presented under 4.4 as well as the literature and Transparency International s East Africa Bribery index suggest potential heterogeneous effects between the different socio-professional categories. In the table below, the following interaction models are estimated: 3 B it = c + β 1 T 1 + β 2 T 2 + (β 1m J m T 1 + β 2m J m T 2 ) + S t + ε i (3) m 3 F it = c + β 1 T 1 + β 2 T 2 + (β 1m J m T 1 + β 2m J m T 2 ) + S t + ε i (4) m where all the variables are unchanged from models 1 and 2 at the exception of J 1, J 2, and J 3, which are dummies for teachers, court clerks, and the police with health workers as a reference category. Table 9 heterogeneous effects associated with socio-professional category (1) (2) (3) (4) Bribe-taking OLS Bribe-taking Logit Inequality OLS Inequality Logit Message 1 (standard) (190.7) (0.703) (1.151) (0.519) Message 2 (prof. identity) *** *** *** *** (101.7) (0.935) (0.614) (0.788) Teacher * (571.5) (2.332) (0.752) (1.072) Teacher * M (281.2) (0.950) (1.159) (0.541) Teacher * M *** *** *** (223.3) (1.667) (0.618) (0.800) Court clerk (556.5) (1.525) (0.750) (0.845) Clerk * M (482.1) (1.522) (1.269) (1.167) Clerk * M *** ** *** (671.5) (1.817) (0.709) (0.936) Police *** (380.4) (2.100) (0.430) (1.055) Police * M (425.6) (1.122) (1.229) (0.553) Police * M *** ** *** (352.3) (1.067) (0.669) (0.818) constant *** *** *** *** (199.7) (0.0926) (0.478) (0.0383) Session FE yes yes yes yes N Standard errors in parentheses, standard errors clustered per session. * p < 0.10, ** p < 0.05, *** p <

18 As expected, there are clear heterogeneous effects. They are robust to introducing additional covariates and using a count model. These effects are also visible when simply representing the mean value for the variables bribe-taking and fairness by socio-professional category and intervention group (Figures 5 and 6). In terms of bribe-taking, the only category reacting as expected are the court clerks who take less money after being exposed to the standard message and even (slightly) less when exposed to the professional identity message (the difference between the two treatment groups is not significant). The other groups display less coherent behaviour, in particular the nurses of the standard intervention who take significantly more money than both the control and second intervention groups no message standard message professional identity message health workers teachers court clerks police Figure 4 Effects on bribe-taking by socio-professional category and intervention group In terms of vouchers allocation, the behaviours of the clerks and health workers are driving the effect observed at the aggregated level, with a similar relative decrease in the level of unfairness for both categories but a much higher starting point for the health workers. The effect of the messages is also visible on the police, but interestingly it is the standard message rather than the professional identity message that has the strongest effect no message standard message professional identity message health workers teachers court clerks police Figure 5 Effects on fairness by socio-professional category and intervention group There is no significant interaction effects between either message and the exposure, sensitisation, and training of the civil servant has received to and on corruption. This result is robust to a series of alternative specifications and grouping the two interventions. Combined with the results of table 6, it could suggest a long-lasting effect of initiatives that seek to curb corruption. 18

19 Four key results emerge from the experiment: (1) none of the messages influence the propensity to accept a bribe, (2) only the message that calls attention to professional identity has an impact on the way public servants allocate vouchers, i.e. public service delivery; (3) the bribery and public service delivery dimensions are not correlated; and (4) the type of professional identity matters: nurses, teachers, court clerks, and police react differently to anti-corruption messages. 5 Discussion This section discusses the results in terms of bribery, then in terms of public service delivery, before commenting on the interaction between these two dimensions as well as methodological caveats and prospects. 5.1 Bribery The professional identity message is negatively correlated with bribe-taking, but this correlation is not statistically significant. After excluding measurement issues, three main categories of hypotheses remain for explaining this insignificant result: Firstly, the experiment disentangled the bribe-taking and public service dimensions, allowing the public servant to take the bribe without delivering the favour asked by the briber. This means that, unlike in many bribery games, the amount of bribe taken does not necessarily affect public service delivery. In other words, bribe-taking in the game is compatible with an equitable allocation of vouchers. The latter, which was an option chosen by a majority of the participants, might be enough for a participant to think of themselves as a good civil servant, especially in a context whose social norms may not be the same as those of high-income industrialised countries (Sardan 1999). Secondly, the situation of public servants is often precarious in Burundi, and the high level of bribe-taking and the insignificant result of the professional identity message may be need rather than greed-induced. The BIF 4,000 extra is a non-negligible amount, especially for teachers and policemen whose salaries still remain comparatively low. Need-induced corruption is invoked when institutions are weak and salaries are so low that any opportunity to augment household budget is grasped, which could be the case of some of the participants. Yet, it remains likely that for other civil servants corruption is more greed-induced, in the sense that people have enough to satisfy their basic needs but cannot resist bene ting from a tempting situation (Ndikumana 2006). Thirdly, the design of the game creates a peculiar environment in which the participants are totally anonymous and may feel under little or no scrutiny. This design was important to make sure the game would take place smoothly in the aforementioned context. The context created by the experiment may, however, not be disconnected from real-life situations in developing countries: citizens with limited social capital often have little bargaining power when attempting to obtain extra illegal services from poorly monitored public servants. This being said, scrutiny of public servants is likely to be higher in an office rather than in a lab and well-connected citizens may be in a position to monitor the public servant s behaviour. The form of the prompts that were used to remind the participant of their identity is also peculiar: the best available option was to use written messages about the professional identity of the participant inside a leaflet that included basic information about the experiment (timing, practical information, etc.). Even though messages were carefully composed in Kirundi and tested several times, it is possible that the participant saw the message, which reminded them of the social norms associated with their profession, but failed to change her behaviour having a different opinion of what constitute a public servant s typical behaviour. In a recent study on bankers social norms in Switzerland, Cohn, Fehr, and Maréchal