Ethnocentric favoritism, ethnic polarization and
corruption tolerance – evidence from Bosnia
and Herzegovina
Bachelor thesis (15 hec) Spring term
Departement of Economics
University of Gothenburg
A
BSTRACT2
A
CKNOWLEDGEMENTS2
1.
I
NTRODUCTION3
2.
L
ITERATURE REVIEW5
3.
B
ACKGROUND9
3.1
L
INKING CORRUPTION TO THE EXPERIMENTAL SUBJECTS9
3.2
T
HEP
OLITICALC
ONTEXT11
3.3
M
ORE ON DATA FROMM
OSTAR12
3.4
W
HY IST
UZLA AN OPPOSITE OFM
OSTAR?
12
4.
D
ATA13
5.
E
XPERIMENTALD
ESIGN15
5.1
F
IELD STUDY AND SCENARIO15
5.2
D
ESCRIPTIVE STATISTICS16
5.3
T
REATMENT DEFINITION19
5.4
R
ANDOM ALLOCATION TESTS19
5.6
E
THNIC POLARIZATION21
5.7
E
XTENSION22
6.
E
MPIRICALS
TRATEGY23
7.
R
ESULTS24
7.1
G
ENERAL PERCEPTIONS ON ETHNICITY24
7.2
E
XPERIMENTAL RESULTS26
8.
H
ETEROGENEITY ANALYSIS28
9.
E
XTENSION32
10.
C
ONCLUSIONS37
R
EFERENCES39
A
RTICLES AND WORKING PAPERS39
Abstract
Earlier research on group identity has shown mixed effects of in-group favoritism (Rachel Kranton et. al. ; Helen Bernhard, Ernst Fehr and Urs Fischbacher 2006 ; Lorenz Goette et. al 2006 ; Yan Chen and Sherry Xin Li 2009 ; Nancy Buchan et al.) . A framing experiment with survey answers on attitudes towards corruption from students of two ethnic groups in Bosnia and Herzegovina was conducted in early 2015 through a field study. The experiment reveals three patterns. Firstly, individuals living in the ethnically polarized Mostar in Bosnia and Herzegovina display ethnocentric behavior by intensely favoring corrupt members of the own ethnic group ahead of corrupt members of the other ethnic group. Secondly, indirect evidence shows that ethnocentric favoritism might differ to a large extent in the more ethnically polarized Mostar as opposed to the less polarized Tuzla. Lastly, a priming experiment shows that individuals who are more aware of the consequences of corruption are also more intolerant to corruption and tend to display less ethnocentric favoritism.
Acknowledgements
The study would not have been possible to conduct if it weren’t for my supervisor Oana Borcan who provided guidance and encouragement throughout the entire
project. I also thank all experiment participants who sacrificed free time for answering the surveys. Furthermore, I am grateful to the SIDA organization (Swedish
1. Introduction
Since time immemorial, group affiliation and behavior has fascinated mankind. Various rulers and emperors have for millenniums used strategies to induce group affiliation, e.g. Alexander the Great, Genghis Khan, Fatih Sultan Mehmet and Adolf Hitler, typically in order to strengthen their military forces, both in size and
motivation. Group behavior is a topic that also has appealed to researchers, especially psychologists, who have investigated it for decades. Henri Tajfel and colleagues showed that people can, almost instantly, form self-preferencing in-groups even when the features are as trivial as preferences on Kandinsky vs. Klee paintings (Tajfel et al. 1971). The issue has recently started to attract economists, since the phenomenon can cause inefficiency, inequality and even economic crises (Predergast and Topel 1996 ; Barr and Oduro 2002 ; Fisman 2003 ; Bandiera et al. 2009 ; Anderson 2011).
In this study, I investigate in-group favoritism based on ethnicity and focus on the research question “Do people in Bosnia and Herzegovina have more tolerance for corruption by people of their own ethnicity than of other ethnicity?”. The other spotlight is an attempt to answer the question “Do people's attitudes to corruption (for own vs. other ethnicity corrupt individuals) differ in a more ethnically polarized as opposed to less polarized society?”. The strategy to address the questions includes a framing experiment with survey answers on attitudes towards corruption from students of the two ethnic groups “Bosniaks” and “Croats” in Bosnia and Herzegovina.
The experiment was carried out through a field study that was conducted in January and February 2015 with 239 students at three public universities in the cities of Mostar and Tuzla. The students were, after reading a scenario about a corrupt politician, asked to report their attitudes towards corruption behavior. The students were allocated randomly and the surveys were differently framed where the only difference was the ethnicity of the politician. Since the students at the two public universities in Mostar are basically divided based on their ethnicity, a simple linear regression analysis with treatment as the independent variable and corruption
intolerance as the dependent variable was carried out. Significant effects indicate that ethnocentric favoritism occurs to a large extent, since students in Mostar act
two cities, ordinary regression analysis cannot be conducted to causally answer the second question, and for this reason, a heterogeneity analysis has been made. By comparing attitudes of the students in polarized Mostar to those of ethnically united Tuzla, indirect evidence shows that ethnocentric favoritism might differ to a large extent in a more ethnically polarized as opposed to less polarized society. An
extension of the study includes a priming experiment where some of the students have received more information on corruption than the others. The outcome of the priming experiment is that individuals who are more aware of the consequences of corruption are also more intolerant to corruption and tend to report less ethnocentric favoritism.
There have been many interesting studies on in-group favoritism. Several studies show robust effects, but there are also those that show no or little effects, which by some researchers is assumed to be a consequence of similarity of
backgrounds among the subject pool. This study uses a subject pool of students and yet displays clear in-group favoritism effects. The study also focuses on ethnocentric in-group favoritism, which has been underinvestigated, particularly in connection to political processes and corruption. The study is somewhat different to related studies. Instead of an index, the study uses up to date and actual data, which in contrast to most studies, is acquired in Europe. Due to the experimental design, the data should eliminate some of the problems of biasedness and measurement error.
The paper is structured as follows: section 2 provides an overview of the literature, section 3 describes the context in which the study is made, section 4 explains the data more explicitly, section 5 provides the design of the experiment and includes descriptive statistics and random allocation tests, section 6 explains the empirical strategy with mathematical formulation, section 7 presents the results, section 8 provides the heterogeneity analysis, section 9 presents an extension with priming effects and section 10 provides conclusions.
2. Literature review
Sociologists and psychologists have for decades been interested in how members of groups are behaving to each other in comparison to non-members. In order to explain intergroup behavior, Henri Tajfel and John Turner introduced the terminology of in-group and out-in-group during work in formulating social identity theory in the 1970s and the 1980s. An in-group refers to a social group to which an individual
psychologically identifies as being a member. Language, ethnicity and religion were probably the most important features of distinct social groups historically, and even today, many people may find it emotionally meaningful to view themselves according to their ethnicity, religion, age or gender. An out-group, on the other hand, is a social group with which the individual does not identify. In-group favoritism refers to the phenomenon when people, under certain conditions, prefer and have affinity for the in-group over the out-group.
The economics literature on group affiliation and social identity dragged behind, but many economists have by now investigated the issue (George A. Akerlof and Rachel E. Kranton 2000 ; Samuel Bowles and Herbert Gintis 2004 ; Helen Bernhard, Ernst Fehr and Urs Fischbacher 2006). In-group favoritism is often seen as undesirable since it creates blocks for reaching a meritocratic society with fair
competition for resources and economic opportunities. In the long run, it can result in inefficiency, income inequality and even economic crises (Predergast and Topel 1996 ; Barr and Oduro 2002 ; Fisman 2003 ; Bandiera et al. 2009 ; Anderson 2011). For investigating the minimal conditions that are required for favoritism to occur,
sociologists and psychologists often use a method called the minimal group paradigm. Researchers have, using this approach, showed that even arbitrary or virtually
game with payoffs in form of money that are given to the participants.
This study takes a different approach. In contrast to the minimal group approach, this study focuses on two ethnic groups from the ethnically diverse Bosnia and Herzegovina (henceforth BiH) as the subject pool. The subject pool in the experiment includes students, precisely to achieve comparability in terms of age and sex distribution. Contrary to previous studies that assumed that similarity of
backgrounds generated inconclusive results, this study displays clear in-group favoritism effects. Two hundred and thirty-nine members studying at the two public universities in Mostar and the public university in Tuzla participated in the
experiment. The participants in the study have similar personal characteristics, except from the main difference that they belong to different ethnic groups. Authors have believed that members of an ethnic group might act ethnocentrically by favoring members of their group over members in other groups (Berghe 1987). This tendency does not only include ethnic groups but religious groups as well (Vanhanen 1999). This experimental study examines these statements more closely. Moreover, Glaeser and Saks argue that if there are several ethnic groups in a society and “leaders tend to allocate resources towards backers of their own ethnicity, then members of one ethnic group might continue to support a leader of their own ethnic group, even if he is known to be corrupt” (Glaeser and Saks 2006). They present examples of American politicians to support their argument, and use a regression analysis with the help of a fractionalization index of American ethnic groups (via a census from 1980) to show that ethnic heterogeneity leads to more corruption. My study examines Glaeser and Saks’ statement with actual and up to date perceptions and adds an additional dimension to the analysis, namely the effect of ethnic polarization instead of heterogeneity.
different ethnic groups. Easterly and Levine (1997) used this index and a large sample of countries to show that GDP growth per capita is inversely related to ethnolinguistic fractionalization. They argue that this could explain Africa’s growth failure. The interest of such indexes has increased tremendously in the last two decades, and the ethnic fractionalization index has today become, more or less, a standard component in studies explaining cross-national differences in economic phenomena such as growth, GDP and investment.
There is, however, a different approach to measure ethnic conflict and its implications on development. Esteban and Ray (1994) constructed a theory behind ethnic polarization instead of ethnic fractionalization. They argue that polarization builds tensions, which in turn reduces trust and consequently growth. Montalvo and Reynal-Querol (2002, 2005) used this theory and provided a so-called RQ-index and empirical evidence of polarization’s negative effect on different aspects of growth, such as conflicts and investments. They showed that ethnic fractionalization does not have a significant effect on the likelihood of conflicts, while ethnic polarization, when using the RQ-index, is a significant explanatory variable for the incidence of civil wars. For this reason, I prefer the ethnic polarization index ahead of the ethnic fractionalization index. Ethnic fractionalization obviously increases the risk of ethnic polarization, but I would argue that ethnic conflict cannot be explained by how diverse a population is (ethnic fractionalization), but rather how antagonistic the ethnic groups are (ethnic polarization). An illustrative example of this might be the comparison of the two Bosnian cities Mostar and Tuzla. Mostar has an ethnically polarized population, which can explain why the city was deeply affected by the war in the 1990s, while Tuzla was more or less spared from war. Tuzla’s population was basically unified, even though it was heavily fractionalized. The purpose of the heterogeneity section in this study is to investigate whether and to what extent ethnic polarization increases corruption (more specifically corruption intolerance).
explanations in his comprehensive cross-national study. He showed, for instance, that corruption is lower in countries with long exposure to democracy, Protestant
traditions, histories of British rule, openness to trade and high development. Even though there are many interesting and observable explanations for corruption, the heterogeneity section of the study focuses only on the effect of ethnic polarization on corruption.
It is not simple to measure and judge the links between ethnic conflict and corruption, and sometimes there are problems with the direction of causality. O’Donnel (2006) and Bolongoita (2005) have, however, showed that post-conflict settings are likely to be categorized by high corruption levels and conflict probability. Their strategy was to compare post-conflict countries, e.g. Sudan, Somalia and Sudan, with countries with comparable income levels. More of this research is welcomed, but it is particularly important for cross-national studies to watch out for confounding factors and omitted-variable bias. This study neither includes cross-national data or indexes nor has a focus on an African country like most studies, but an experimental study performed on 239 individuals living in BiH. This path should eliminate some of the risks with misspecification, biasedness and measurement error.
3. Background
3.1 Linking corruption to the experimental subjects
There is evidence that bribes grease the wheels in public services and even industry by decreasing transaction costs and delays and therefore generating growth in economies (Leff 1964 ; Huntington 1968). However, there is scholarly and political agreement that corruption has a negative impact on both GDP per capita and
economic growth (Mustapha 2014 ; Kaufman and Wei 1998 ; Aidt 2007 ; Mauro 1995, 1997 ; Shleifer and Vishny 1993 ; Blackburn et al. 2009 ; Barro 1996 ; Tanzi et al. 2000).
Various organizations have presented statistics that show that ex-Yugoslav countries (situated in the southeastern Europe) are among the poorest and also most corrupt countries in Europe. In IMF’s World Economic Outlook Database-April 2014, for instance, Montenegro is ranked at place 58 of the countries in the world by GDP per capita, nominal prices (189 countries). Serbia is at place 62, the FYR of
Macedonia at 82 and BiH at place 84 with 4,597 US dollars/capita. There are only three countries in Europe with lower GDP levels. In Transparency International’s Corruption Perceptions Index from the same year, the FYR of Macedonia is ranked at place 64 on how corrupt their public sector is perceived to be (175 countries).
Montenegro is at place 76, Serbia at 78 and BiH sharing the place 80 with Benin, El Salvador, Mongolia and Morocco. The FYR of Macedonia, Montenegro, Serbia and BiH are among the few countries in Europe that are still considered to be developing
countries1
. The high corruption levels in these countries might be part of the reason why the countries are so poor and underdeveloped. Corruption in BiH is, according to a report from Transparency International, not restricted to one working group or one sector, but rather “affecting all sectors of society, including the judiciary, tax and custom administration, public utilities, procurement and privatization schemes as well as all major political processes.” (Corruption and Anti-Corruption in Bosnia and Herzegovina 2009).
The country proclaimed independence from the former Socialist Federal Republic of Yugoslavia in 1992, which was followed by the Bosnian War, lasting from 1992 to late 1995. The war and the complex institutional structure might also
help to explain the low GDP and high corruption levels. BiH has a multiethnic
population of almost four million with three “constitutional” ethnic groups – Bosniaks (often called Bosnian Muslims), Bosnian Serbs (also Christian Orthodox) and
Bosnian Croats (or Catholics)2
. Some of the people also call themselves just “citizens
of Bosnia and Herzegovina” or Bosnians3
, due to stronger multicultural feel. Since the demographical situation is severely complex, there are some factors that need to be taken into account when studies are made in the country. There are several individuals who characterize themselves as “Yugoslavs” or “Others” or do not know who to characterize themselves, which could bring complications into studies. Even though the term Bosniak is frequently referred to Bosnian Muslims, there are others who characterize themselves as Bosniaks. The term Bosniak actually comes from the term Bosnjanin, which denoted the people of the medieval Bosnian Kingdom, a country that existed before Islam came to the territory. As a consequence, there are individuals who, for example, characterize themselves as not only Bosnian Catholics but also Bosniak Catholics (instead of Croats), which could affect studies and their
conclusions. The complex institutional structure in the country has deeply affected the educational system. The educational system in BiH is not centralized but administered through three different systems, depending on the language students speak (Bosnian, Serbian or Croatian). In some parts of the country, education is under jurisdiction of cantons (member states of the Federation of BiH), and a consequence is that students attend different programs, not only varying in language but also in views on history and religion. The educational system in BiH is intensely studied by the Transparency International. A 2012 report shows that 56% of two thousand students surveyed at higher education institutions in BiH believe that corruption is widespread in schools (only 10% believe that there is no corruption), and almost a third of the five hundred surveyed teachers and other staff members in schools hold the same belief. Almost half of the asked students were willing to pay a bribe to pass the exam if there was no other way to pass it (Analiza percepcije korupcije u visokom obrazovanju u BiH –
Sazetak - 20124
). As I will show later in the Data section, results from my study show similar results.
2https://www.cia.gov/library/publications/the-world-factbook/geos/bk.html 3 http://www.ba.undp.org/?id=882
3.2 The Political Context
The experiment of this study included two surveys, one with a scenario of a corrupt Bosniak and the other with corrupt Croat politician, which were given to the survey respondents. The surveys can be found in Appendix (page 43-46). In summary, the scenario is, due to ethnical reasons, made up and is about a presumptive candidate for a member in the Council of Ministers of Bosnia and Herzegovina. This politician is accused in court for corruption for buying a degree in management from an unknown university in the country. In October 2014, just three months before the study was made, general elections were held in BiH. The Party of Democratic Action, which is often described as a Bosniak nationalist party, was, with 18,74% of the votes, the largest party in the House of Representatives. A couple of months later, the party negotiated an alliance to rule the country with other parties, including the Croatian Democratic Union of Bosnia and Herzegovina, which by many people is seen as a Croatian nationalist party. The House of Representatives confirmed the members of the Council of Ministers on 31 March 2015.
3.3 More on data from Mostar
Two of the three universities in the sample are located in the city of Mostar. Mostar is
the sixth largest city and situated in southern BiH5
(about 113.000 inhabitants) and is together with Belfast, Nicosia and Mitrovica considered to be one of the mostly divided larger cities in Europe. The city is sometimes referred to as two cities in one, with a western part in which a vast Croatian majority lives and an eastern part which
is mainly inhabited by Bosniaks6
. The city is perhaps best explained as a modern version of Berlin between 1961 and 1989, except form that there is no wall between the two parts but a river (Neretva) and the main boulevard (M6.1) instead. This borderline was the front line during the war in Mostar, which was a conflict between the Republic of Bosnia and Herzegovina and the self-proclaimed Croatian
Community Herzeg-Bosnia and lasted from mid 1992 to early 19947
. Since Mostar is divided into two entities, there are two electricity companies, two phone networks,
two postal services, two football clubs and two public universities in the city8
. University of Dzemal Bijedic is located in the eastern Bosniak side of Mostar, and University of Mostar in the western Croatian side.
3.4 Why is Tuzla an opposite of Mostar?
The whole country was in a horrific war between 1992 and late 1995, so it is very difficult to find a unified city in BiH, especially large enough to be compared to Mostar and to have a public university. Tuzla, however, is the next best example of relative ethnic and cultural homogeneity in BiH. Tuzla is the third largest city located
in northeastern BiH9
(with about 120.000 inhabitants). The Greek researcher Ioannis Armakolas (2011) presents plenty of evidence of Tuzla’s multi-ethnicity and
tolerance in his book The Paradox of Tuzla City. After the elections in 1990, Tuzla was, for instance, the only city in the country that was led by an anti-nationalist and multi-ethnic government (this alliance had almost 70 % of the votes). While a horrible war took place in the whole country, the city of Tuzla was, besides a couple of
Tuzla included individuals from all three ethnic groups, with a commander, Ilija Jurisic, who is a Bosnian Croat. The city has been famous for multi-ethnicity long before the war. The Husino rebellion in 1920 was, for instance, an anti-nationalist and revolutionary uprising of miners, and took place in the Catholic village Husino, just
outside of Tuzla10
. In his book Bosnian Muslims in the Second World War, Marko Attila Hoare (2013), a British historian, presents another remarkable event from the same place but in 1941, when an entire Ustasha force (Croatian fascist, terrorist
movement1112
) changed side and went over to fight for the Yugoslav Partisans (a
socialist, multi-ethnic resistance force13
). All but one of these Croatian Husino soldiers died fighting as Partisans during the war. For these reasons, I compare the corruption attitudes of students in Mostar universities with those of students in Tuzla.
4. Data
As mentioned earlier, 239 randomly chosen students at three universities in BiH (students of different study areas) were surveyed. Similarly to Transparency International’s study from 2012, the respondents reported high willingness and experience of corruption. Only 51% answered “fully disagree” to the statement “If an exam was so hard that I would not pass it by studying, I would be prepared to pay for passing the exam”.
Table no. 1
Variable Obs Mean Std. Dev. Min Max
To the question “Have you or do you know someone who has bought an exam/diploma?” almost 60% of the students answered “Yes”.
There are also similarities in the students’ personal characteristics in the two studies. From table no. 2, we can see that Transparency International has a larger sample, but the sex distribution is more even in my study, and the same goes for the distribution between the years of university of the students. We can also see that both studies show similar results in how students finance their schooling.
Table no. 2
Transparency International’s results My results
Student Frequency % Frequency %
5. Experimental Design
5.1 Field study and scenario
Due to the fact that the higher education institutions are rife with corruption warrants, an investigation with 239 randomly chosen students at three public universities in two cities (about 80 students per university) has been made. Online opt-in sample is not representative of the entire population, but predictably over-samples younger, male, educated, politically interested, liberal and richer respondents. Since non-probability online surveys may give biased results compared to more traditional representative surveys (Malhotra & Krosnick (2007), Yeager er al. (2011), the strategy of the data collecting includes that the sampling is done directly in the field through face-to-face interviewing and traditional survey answering, performed through a field study made in January and February 2015. The sampling was done completely randomly, since the surveys were mixed and given to random students at different campus buildings.
The universities are: University of Dzemal Bijedic in Mostar, University of Mostar and University of Tuzla. As presented in Appendix (page 43-46), the students have been introduced to a scenario with a male politician who has been accused in court for corruption for buying his degree. The students are asked to select an appropriate sentence to the politician of the four alternatives:
1. The politician should be free of all charges of corruption, and should be able to continue his political career immediately.
2. A fine penalty of 13.500 KM (minimum sentence for corruption). NO suspension from political positions.
3. Seven months of prison sentence and a fine penalty of 27.250 KM (medium sentence for corruption).
As mentioned earlier, there is a fundamental difference in the two surveys that are presented randomly to the students. The surveys are the same in all aspects except from that one survey is with a corrupt Bosniak (Bosnian Muslim) politician and the other is with a corrupt ethnic Croat (Bosnian Catholic). About half of the chosen students the University of Dzemal Bijedic have received surveys with the Bosniak politician, and the other half has received a survey with the Croatian politician. The same goes for the chosen students at University of Mostar and University of Tuzla.
5.2 Descriptive statistics
As shown in table no. 3, the sex distribution of the survey respondents is quite even. The average student in the sample is 22 years old, has finished a little bit more than two and a half years of university and has a pretty good grade. A third of the students commute to the university from rural surroundings, and about 60 per cent were raised in rural areas. Only two thirds answered the questions on parental income, but the average monthly salary after tax of the fathers is 1086 BAM (roughly 550 EUR), while an average mother earns about the half of that amount.
Table no. 3
Variable Obs Mean Std. Dev. Min Max
Male 239 .464 .500 0 1
Age 239 22.08 2.43 18 31
Year of uni 239 2.69 1.26 1 5
Grade 219 8.47 1.11 6 10
Living rurally 239 .347 .477 0 1
Raised rurally 239 .607 .490 0 1
Father’s income 164 1086 1189 0 10000
Mother’s income 165 562 595 0 3000
Table no. 4
Variable Obs Mean Std. Dev. Min Max
Corruption intolerance 239 3.36 .764 1 4
As presented below, about half of the surveyed students at the three universities received a survey with a Bosniak politician and the other half with a Croatian politician.
Table no. 5
Variable Obs Mean Std. Dev. Min Max
Suspect Bosniak 239 .481 .501 0 1
Suspect Croat 239 .519 .501 0 1
The mentioned data can be used for regression analysis to answer the question: Do people in Bosnia and Herzegovina in general have more tolerance to corruption for Bosniaks or for Croatians, or do they judge them the same?
The question will be examined and answered in section 7. Results (7.1 General perceptions on ethnicity).
Table no. 6
Panel A-‐University of Dzemal Bijedic Panel B – University of Mostar
the students at University of Mostar have been converted to the most frequently used grading scale among universities in the country (6 to 10).
As shown in table no. 7 below, about half of the chosen students at the University of Dzemal Bijedic have received surveys with the Bosniak politician, and the other half has received a survey with the Croatian politician. The same goes for the chosen students at University at Mostar, which is also presented below. The ethnicity shares at each university are also obtainable. About 83 per cent of the students at University of Dzemal Bijedic declared that they are Bosniak/Bosnian, and 90 % of students at University of Mostar declared that they are Croatian.
Table no. 7
University of Dzemal Bijedic
Variable Obs Mean Std. Dev. Min Max
Suspect Bosniak 77 .468 .502 0 1 Suspect Croat 77 .532 .502 0 1 Student Bosnian 77 .208 .408 0 1 Student Bosniak 77 .623 .488 0 1 Student Croat 77 .091 .289 0 1 Student Serb 77 .078 .270 0 1 Table no. 8
University of Mostar
Variable Obs Mean Std. Dev. Min Max
5.3 Treatment definition
The data allows for treatment definition. Treatment definition is necessary in order to answer the question on whether or not students in BiH have more tolerance for
corruption by people of their own ethnicity than of other ethnicity, or in other words if the students report ethnocentric favoritism. To address this issue, the treatment group must include students in Mostar sharing the same ethnicity as the corrupt politician. Because of the distinctive ethnicity shares, I make the assumption that students at University of Dzemal Bijedic are Bosniaks and that students at University of Mostar are Croats. The control group includes students in Mostar not sharing the same ethnicity as the politician. The control group includes, just to clarity matters, students at University of Dzemal Bijedic who have received a survey with the corrupt Croat politician and students at University of Mostar a survey with the corrupt Bosniak politician. Even though the ethnicity shares are distinctive, data shows that the two universities are slightly mixed, and therefore, it is a strong assumption to assume that all students at University of Dzemal Bijedic are Bosniaks and all students at
University of Mostar are Croats. However, additional regression analysis where only answers from Bosniaks at University of Dzemal Bijedic and Croats at University of Mostar are included show that the effects move in the direction where they get even stronger. For this reason, it is plausible to make the specified assumption.
5.4 Random allocation tests
characteristics should not differ. Only two of the 21 tests had statistically significant values (p-values lower than 0.10), and these two characteristics were income and education of mother. Mothers were more educated and earned more in the control group, but that would not have been true if it weren’t for four outliers in the control group. There is also a possibility that the fact would not have been true if all
respondents answered all survey questions and the two groups had an equal number of respondents. Income of mother was the survey’s most skipped question, and missing values increases the data volatility and probability of biasedness.
Table no. 9 Male -‐ Two-‐sample t test with equal variances
Group Obs Mean Std. Err. Std. Dev. [95% Conf. Interval] 0 83 .482 .055 .503 .372 .592
1 77 .468 .057 .502 .354 .582 t = 0.1811 Ha: diff ! = 0 degrees of freedom = 158 Pr(T > t) = 0.8565
Table no. 10
Mother’s income -‐ Two-‐sample t test with equal variances
Group Obs Mean Std. Err. Std. Dev. [95% Conf. Interval] 0 58 758 96.9 739 564 952
1 50 513 73.5 520 365 661 t = 1.9621 Ha: diff != 0 degrees of freedom = 106 Pr(T > t) = 0.0524 *
The treatment mentioned above can be used for regression analysis to answer the question:
Do people in Bosnia and Herzegovina have more tolerance for corruption by people of their own ethnicity than of other ethnicity?
5.6 Ethnic polarization
The study has up to now focused on ethnocentric favoritism, which also is the main objective. The second part of the study includes ethnic polarization with the
comparison of students in Mostar and students in Tuzla. As will be explained in later sections, by just putting a dummy in regressions for students in Tuzla, the results are subject to selection bias and no causality can be established. For this reason,
regression results cannot be interpreted causally but only indirectly, and therefore, a heterogeneity analysis is the next best strategy (more details in section 8.
Heterogeneity analysis).
The heterogeneity analysis strategy is to compare a randomly allocated group of students in the unified city of Tuzla with the students from the ethnically polarized Mostar. The public university in Tuzla (University of Tuzla) has roughly the same number of students as the two public universities in Mostar. As aforementioned, even though a large share consists of Bosniaks, Tuzla is famous for its multi-ethnicity and unity.
In order to test if the allocation of respondents at University of Tuzla was truly random, a treatment can be defined here. Since there is plenty of evidence for Tuzla’s unity, the treatment group consists of students at University of Tuzla who have received a survey with the Bosniak politician. The control group includes students with a survey with the Croatian politician. A total of 21 tests showed that the
allocation of these students is truly random as well. In fact, there were no statistically significant values, and the majority of the tests had p-values higher than 0.50. This means that even the unobserved characteristics should not differ between the groups.
Table no. 11
t-‐test age -‐ Two-‐sample t test with equal variances
Group Obs Mean Std. Err. Std. Dev. [95% Conf. Interval] 0 42 22.3 .417 2.70 21.5 23.2 1 37 22.4 .425 2.59 21.5 23.2
t = -‐0.1154 Ha: diff != 0 degrees of freedom = 77
Pr(T > t) = 0.9085
Table no. 12
t-‐test Mother’s income -‐ Two-‐sample t test with equal variances
Group Obs Mean Std. Err. Std. Dev. [95% Conf. Interval] 0 29 385 69.5 374 243 528 1 28 427 89.1 472 244 610 t = -‐0.3679 Ha: diff != 0 degrees of freedom = 55
Pr(|T| > |t|) = 0.7144
Results of all 21 tests with means and p-values are presented in Appendix page 49.
The treatment can, if adjusted, be used for regression analysis to help answer the question:
Do people's attitudes to corruption (for own vs. other ethnicity corrupt individuals) differ in a more ethnically polarized as opposed to less polarized society?
An examination and indirect answer to the question can be found in section 8 (Heterogeneity analysis).
5.7 Extension
As presented in section 9 (Extension), an extension of the study included an experiment to examine what psychologists call priming effects. In summary, forty-two randomly chosen respondents of the 239 students have received a survey with an information sheet on corruption. The information sheet is presented in Appendix (page 47). The sheet is presented before the scenario about the corrupt politician and briefly explains how bad and widespread corruption is in BiH. The students are, after reading the info sheet, asked to read the scenario and select sentences for the
politician like all other students. This experiment was made to see if individuals who are more aware of the consequences of corruption display the same degree of
6. Empirical Strategy
The specification that I use to analyze the ethnocentric favoritism effect is the following equation:
𝑦!" = 𝛼 + 𝛽 𝑇!"+ 𝛾´!" + 𝜃! + 𝜀!" (1)
where 𝑖 indexes students at university 𝑢, which includes three dummies for the universities in BiH (University of Dzemal Bijedic, University of Mostar and University of Tuzla). The first regressions only focus on ethnocentric favoritism effects and here only the universities in Mostar are included. The regression in the heterogeneity section uses the same model as above, but with a different sample and treatment (data from University of Tuzla is included). 𝑦 is the dependent variable “corruption intolerance” which is based on four punishments that the students can choose for the corrupt politician. 𝑇 is my treatment variable where the treatment group includes students who share the same ethnicity and control group includes students who do not share the same ethnicity as the corrupt politician (remember assumption described on page 19). 𝛾 are controls that are divided into three types: basic personal controls (gender, age, grade, parental education etc.), socio-economic controls (parental income), and attitudinal controls (questions taken from surveys conducted by the Transparency International on attitudes towards corruption, e.g. “Can ordinary people make a difference in the fight against corruption?” (variable called “People can change”) and “How would you assess your government’s actions in the fight against corruption?” (variable called “Interventions effect”)). 𝜃 are
university fixed effects. In all the regressions, I report to heteroscedasticity-consistent standard errors.
In simple linear regression analysis, there are some conditions that have to hold to make the coefficient of interest deliver the unbiased estimate of the treatment effect. The first assumption requires that my function is linear in parameters, which it clearly is in this case. The second assumption requires that the sampling is random. The t-tests presented in section 5 showed that the sampling was truly random and thus, this assumption is not violated. The third condition is that there is variation in the
the fourth condition called Zero conditional mean is most important. It requires that the explanatory variable does not correlate with the error term, and if that is the case, the variables will be overestimated. The regressions include plenty of controls, both personal, socio-economic and attitudinal. Section 5 illustrated the randomization and showed that the characteristics did not differ remarkably between the treatment group and control group. This means that even the unobserved variables (variables in the error term) should not differ between the groups and thus, the condition should not be violated.
7. Results
7.1 General perceptions on ethnicity
Do students in Bosnia and Herzegovina in general have more tolerance to corruption for Bosniaks or for Croatians, or do they judge them the same?
The Regression table no. 1 below shows that students in BiH in general punish
corrupt politicians in a similar way irrespective of the politician’s ethnicity, indicating that the students have equivalent tolerance to both ethnicities. Since there is no
randomization yet, there is no treatment. The table gives a mean difference (M.D) between Bosniak and Croat politicians in terms of corruption perceptions. We can see that the M.D coefficient is insignificant and close to zero. The coefficient remains insignificant and close to zero even when basic personal controls (column 2), fixed effects (column 3), more socio-economic controls (column 4) and attitudinal controls (column 5) are included. The conclusion is that there is overall no significant
Regression table no. 1
corruption intolerance (1) (2) (3) (4) (5) min (6) max (7)
M.D -0.149 -0.161 -0.156 -0.153 -0.190 0.115** -0.176** (0.098) (0.108) (0.107) (0.131) (0.117) (0.056) (0.075) Male -0.015 -0.005 0.196 0.275** -0.028 0.181** (0.111) (0.110) (0.139) (0.127) (0.059) (0.085) Age 0.036 0.011 -0.015 0.005 -0.017 -0.024 (0.028) (0.028) (0.034) (0.030) (0.013) (0.022) Year of uni -0.031 -0.006 -0.010 -0.027 -0.000 0.000 (0.052) (0.052) (0.057) (0.055) (0.020) (0.041) Grade 0.036 0.070* 0.099** 0.128*** -0.010 0.091*** (0.035) (0.037) (0.047) (0.041) (0.017) (0.030) Living rural 0.072 -0.047 0.156 0.131 0.042 0.120 (0.134) (0.138) (0.159) (0.146) (0.061) (0.105) Raised rural 0.011 0.080 0.020 0.140 -0.098 0.017 (0.144) (0.143) (0.187) (0.171) (0.068) (0.106) Financed by family 0.022 -0.075 -0.019 -0.085 0.055 -0.079 (0.171) (0.167) (0.198) (0.219) (0.082) (0.121) Father higher education -0.011 -0.032 0.032 -0.011 -0.050 -0.055 (0.149) (0.142) (0.183) (0.165) (0.069) (0.106) Mother higher education -0.036 0.002 0.159 0.320 -0.090 0.156 (0.170) (0.167) (0.230) (0.211) (0.089) (0.116) Voting multiethnic 0.096 -0.033 0.053 0.096 -0.033 0.035 (0.129) (0.129) (0.169) (0.145) (0.071) (0.106) Father income -0.000 -0.000 0.000 -0.000* (0.000) (0.000) (0.000) (0.000) Mother income -0.000 -0.000 0.000 0.000 (0.000) (0.000) (0.000) (0.000) Court verdict 0.231*** -0.024 0.157*** (0.067) (0.028) (0.043) Would buy -0.013 0.037 0.008 (0.080) (0.042) (0.043) Interventions effect 0.024 -0.001 0.039 (0.093) (0.038) (0.054)
People can change -0.293*** 0.081*** -0.165***
(0.063) (0.026) (0.042) Bosnian university -0.093 -0.252* -0.373*** 0.150** -0.204** (0.117) (0.150) (0.123) (0.068) (0.091) Croatian university -0.409*** -0.368** -0.513*** 0.150** -0.230** (0.148) (0.158) (0.156) (0.064) (0.113) Observations 239 214 214 153 153 153 153 R-squared 0.010 0.028 0.065 0.091 0.288 0.210 0.270
7.2 Experimental results
Do students in Bosnia and Herzegovina have more tolerance for corruption by people of their own ethnicity than of other ethnicity?
The Regression table no. 2 displays the estimates of model (1). In column one, I report the results from a simple linear regression with treatment as the independent variable; in column two, I include basic personal controls, such as gender, age, grade, parental education etc.; in column three, university fixed effects are included; in column four, I include socio-economic controls, i.e. parental income; in column five, I add attitudinal controls, i.e. answers to questions regarding attitudes towards
corruption; in column six, I look only on the coefficient estimate on the least intolerant punishment (sentence 1); in column seven, I look only on the coefficient estimate on the most intolerant punishment (sentence 4).
Since the attitudinal questions were asked after the scenario and the scenario can affect the students’ answers to these questions, it is possible that these controls correlate with the treatment coefficient and therefore are misleading (the coefficient in column 5 can be overstated in one direction or another).
The Regression table no. 2 shows that students at the two universities in Mostar in general act in an ethnocentric way and punish the politician of the other ethnicity harsher than the politician of the own ethnicity. The variable T refers to the treatment, which is the scenario with the politicians sharing the same ethnicity as the
Regression table 2 corruption intolerance (1) (2) (3) (4) (5) free (6) max (7) T -0.380*** -0.368** -0.343** -0.447** -0.279* 0.079* -0.097 (0.130) (0.147) (0.145) (0.193) (0.165) (0.043) (0.092) Male -0.025 -0.030 0.169 0.117 -0.028 0.076 (0.146) (0.144) (0.195) (0.158) (0.033) (0.103) Age 0.061 0.032 0.014 0.030 -0.013 -0.037 (0.038) (0.039) (0.058) (0.049) (0.012) (0.029) Year of uni 0.018 0.035 -0.006 -0.003 0.026 0.055 (0.072) (0.071) (0.089) (0.077) (0.018) (0.046) Grade 0.062 0.103 0.169* 0.083 -0.030 0.028 (0.067) (0.068) (0.098) (0.082) (0.020) (0.055) Living rural -0.055 -0.127 0.077 0.110 -0.061 0.036 (0.179) (0.177) (0.223) (0.192) (0.050) (0.124) Raised rural 0.131 0.188 0.193 0.342 -0.003 0.155 (0.170) (0.165) (0.237) (0.216) (0.062) (0.122) Financed by family 0.051 -0.010 0.168 -0.121 -0.042 -0.112 (0.203) (0.202) (0.253) (0.286) (0.088) (0.148) Father higher education 0.021 0.017 0.144 0.189 -0.017 0.083 (0.193) (0.188) (0.283) (0.233) (0.072) (0.118) Mother higher education -0.058 -0.029 0.178 0.526** -0.045 0.229* (0.203) (0.199) (0.287) (0.241) (0.088) (0.120) Voting multiethnic 0.098 -0.064 0.026 0.073 -0.032 -0.050 (0.153) (0.164) (0.255) (0.207) (0.040) (0.136) Father income -0.000 0.000 -0.000 -0.000 (0.000) (0.000) (0.000) (0.000) Mother income -0.000 -0.000** 0.000 -0.000 (0.000) (0.000) (0.000) (0.000) Court verdict 0.243** -0.035 0.146*** (0.093) (0.023) (0.054) Would buy 0.031 -0.026 -0.002 (0.101) (0.026) (0.052) Interventions effect -0.117 0.018 -0.049 (0.114) (0.043) (0.066)
People can change -0.445*** 0.026 -0.269***
(0.066) (0.021) (0.042)
Croatian university -0.330* -0.120 -0.144 0.037 -0.104
(0.171) (0.221) (0.152) (0.033) (0.115)
Observations 160 147 147 102 102 102 102
R-squared 0.052 0.080 0.103 0.142 0.426 0.188 0.399
Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
8. Heterogeneity analysis
Do students' attitudes to corruption (for own vs. other ethnicity corrupt individuals) differ in a more ethnically polarized as opposed to less polarized society?
In order to answer the question above, it is crucial to get a causal estimate of ethnic polarization, for which it is required that the treatment of polarization is allocated randomly. Since many students self-select the universities and cities where they will study, they are, to some extent, exposed to different degrees of ethnic polarization by their own choice. By just putting a dummy in regressions for students in Tuzla, the results are subject to selection bias and no causality can be established. Since the question above cannot be answered causally but only indirectly, a heterogeneity analysis is the next best strategy.
Table no. 13
Variable Obs Mean Std. Dev. Min Max
Corr. intolerance Bosnian uni 77 3.42 .695 1 4
Corr. intolerance Croat uni 83 3.14 .926 1 4
Corr. intolerance Tuzla uni 79 3.52 .574 1 4
The table no. 13 shows the average sentence at each university (also called corruption intolerance). We can see that students at the University of Tuzla were most intolerant to corruption, with the average sentence of 3,52. This could support the argument that ethnic polarization might have a negative effect on corruption intolerance.
This supports the argument that the students in Tuzla give similar sentences
regardless of the ethnicity of the corrupt politician. This table should be compared to Regression table no. 2 for students in Mostar where the coefficients were significant and large. The conclusion of this comparison is that attitudes on corruption might differ in an ethnically polarized society as opposed to a less polarized society (although these differences might be the expression of some other characteristics of Tuzla inhabitants that also protected the city from war and make cohabitation of various ethnicity more harmonious).
Regression table no. 3 corruption intolerance (1) (2) (3) (4) min (5) max (6) T -0.163 -0.093 -0.120 -0.251 -0.000 -0.251 (0.129) (0.127) (0.152) (0.167) (0.029) (0.161) Male 0.152 0.326** 0.394** -0.077 0.317* (0.157) (0.157) (0.174) (0.051) (0.159) Age -0.003 -0.035 -0.026 0.001 -0.025 (0.037) (0.038) (0.043) (0.007) (0.044) Year of uni -0.121* -0.069 -0.047 0.022 -0.026 (0.068) (0.074) (0.092) (0.018) (0.087) Grade 0.045 0.093** 0.127** -0.019 0.107** (0.041) (0.038) (0.052) (0.014) (0.046) Living rural 0.389** 0.423** 0.344** 0.024 0.368** (0.152) (0.172) (0.166) (0.032) (0.161) Raised rural -0.529** -0.601*** -0.425* 0.025 -0.399* (0.211) (0.221) (0.233) (0.049) (0.221) Financed by family -0.468*** -0.560*** -0.321* 0.007 -0.314 (0.150) (0.194) (0.190) (0.066) (0.187)
Father higher education 0.088 0.028 -0.086 0.008 -0.078
(0.183) (0.195) (0.186) (0.033) (0.188)
Mother higher education -0.293 -0.376 -0.466* 0.220* -0.246
(0.266) (0.301) (0.275) (0.128) (0.207) Voting multiethnic -0.052 -0.143 -0.210 -0.006 -0.217 (0.202) (0.233) (0.183) (0.051) (0.185) Father income -0.000 -0.000 -0.000 -0.000* (0.000) (0.000) (0.000) (0.000) Mother income 0.000 0.000 -0.000 0.000 (0.000) (0.000) (0.000) (0.000) Court verdict 0.191** -0.002 0.189*** (0.073) (0.018) (0.066) Would buy 0.056 0.079 0.136 (0.123) (0.051) (0.104) Interventions effect 0.113 -0.004 0.109 (0.078) (0.019) (0.077)
People can change -0.092 0.045 -0.047
(0.093) (0.029) (0.083)
Observations 79 67 51 51 51 51
Another way of comparing the students in the different cities is to include an interaction term between the students’ selected punishments and the Bosnian
university in Mostar (University of Dzemal Bijedic). This is necessary because most students at the University of Tuzla might be Bosniaks, but they self-report as
Bosnians because of the stronger multicultural feel. By including more variables into the specification, I hold these variables explicitly constant. This specification only includes students studying at the Bosnian university in Mostar and the Tuzla University, whose ethnicity shares are similar (both universities are predominantly represented by Bosniaks/Bosnians).
𝑦!" = 𝛼 + 𝛽 𝐵𝑈 × 𝑇!"+ 𝛽!𝑇!"+ 𝛽!× 𝐵𝑈 + 𝛾´!"+ 𝜀!" (2) The Regression table no. 4 displays the estimates of model (2). In this regression, I report to heteroscedasticity-consistent standard errors as earlier. In column one, I report the results from a multiple linear regression with an interaction term of Bosnian university (BU) and T the independent variable and hold BU and T constant; in column two, I include basic personal controls; in column three, I include socio-economic controls; in column four, I add attitudinal controls; in column five, I look only on the coefficient estimate on the second least intolerant punishment (sentence 2); in column six, I look only on the coefficient estimate on the most intolerant punishment (sentence 4).
OLS gives us unbiased estimated of the underlying population model if four conditions are fulfilled. Linear in parameters and random sampling was discussed earlier and the same reasoning can be applied here. None of the independent variables is constant and there are no exact linear relationships among the independent
variables, which means that the condition of no perfect collinearity holds. The earlier discussion on the study’s strong randomization implies that biasedness due to the condition of zero conditional mean should not exist.
In Regression table no. 4, we can see that the coefficient of the interaction term is significant and large again (almost a half a standard deviation), even when personal controls, socio-economic controls and attitudinal controls are included (it actually increases when controls are included). This shows that there is a big
suggests that it might be the case that attitudes to corruption (for own vs. other ethnicity corrupt individuals) differ in a more ethnically polarized as opposed to less polarized society. Regression table no. 4 corruption intolerance (1) (2) (3) (4) min (5) max (6) T x Bosnian university -0.357* -0.460** -0.627** -0.474* 0.279** -0.141 (0.199) (0.230) (0.303) (0.260) (0.120) (0.214) T -0.163 -0.139 -0.195 -0.351** 0.061 -0.320** (0.129) (0.139) (0.152) (0.149) (0.044) (0.146) Bosnian university 0.063 0.124 0.035 -0.116 0.009 -0.141 (0.123) (0.135) (0.165) (0.156) (0.043) (0.145) Male -0.131 0.202 0.309** -0.067 0.258*** (0.112) (0.130) (0.119) (0.053) (0.096) Age 0.008 -0.040 -0.052** -0.005 -0.048** (0.026) (0.028) (0.025) (0.011) (0.024) Year of uni -0.014 0.019 0.064 -0.009 0.035 (0.055) (0.068) (0.068) (0.023) (0.060) Grade 0.011 0.055 0.069* -0.001 0.078** (0.036) (0.038) (0.035) (0.012) (0.035) Living rural -0.012 0.261 0.241* 0.061 0.326*** (0.156) (0.168) (0.144) (0.060) (0.116) Raised rural -0.056 -0.222 -0.196 -0.059 -0.254* (0.142) (0.191) (0.157) (0.062) (0.137) Financed by family -0.333* -0.272 -0.089 0.079 -0.014 (0.169) (0.180) (0.184) (0.064) (0.171)
Father higher education 0.136 0.105 0.034 -0.119** -0.068
(0.132) (0.171) (0.134) (0.055) (0.131)
Mother higher education -0.300* -0.123 -0.038 0.053 0.035
(0.176) (0.229) (0.222) (0.119) (0.157) Voting multiethnic -0.075 -0.056 -0.030 0.016 -0.026 (0.124) (0.169) (0.133) (0.064) (0.113) Father income -0.000 -0.000 -0.000 -0.000 (0.000) (0.000) (0.000) (0.000) Mother income 0.000 0.000 0.000 0.000 (0.000) (0.000) (0.000) (0.000) Court verdict 0.310*** -0.057* 0.254*** (0.057) (0.032) (0.045) Would buy 0.109 -0.027 0.090 (0.082) (0.050) (0.054) Interventions effect 0.145** -0.011 0.118* (0.069) (0.027) (0.063)
People can change 0.023 -0.071
(0.025) (0.059)
Constant 3.595***
(0.091)
Observations 156 137 89 89 89 89
R-squared 0.097 0.146 0.290 0.469 0.342 0.443
In summary, the heterogeneity section provided data results from Mostar and Tuzla that indirectly showed that ethnocentric favoritism might largely differ in a more ethnically polarized as opposed to less polarized society. At first, a comparison of two regression results of data from the universities in Mostar and University of Tuzla was provided. Subsequently, a specific regression analysis with an interaction term of the treatment and data from University of Dzemal Bijedic was made. Both approaches showed similar results.
9. Extension
It has been showed that in-group favoritism might largely differ between societies. Other work has showed that in-group favoritism might be harmful for economic substantiality, and thus, it is imperative for regulators to reduce polarization and deal with societies depending on their context. Therefore, an extension of the study included an experiment to examine what psychologists call priming effects. A share of 17.6 % of the sample, i.e. 42 randomly chosen respondents of the 239 students (14 at every university), has received a survey with an information sheet on corruption. The info sheet is presented in Appendix (page 47). The sheet consists of three sentences on how widespread corruption is in BiH, how it is bad for growth and fosters inequality and that it creates blocks for a future membership in the European Union.
Table no. 14
Corruption info -‐ Two-‐sample t test with equal variances
Group Obs Mean Std. Err. Std. Dev. [95% Conf. Inter]
0 197 3.284 .056 .789 3.173 3.395 1 42 3.690 .080 .517 3.529 3.85
t = -‐3.1898 Ha: diff != 0 degrees of freedom = 237
Pr(T > t) = 0.0016 ***
The t-test above shows that there is a big difference in the students’ punishments,
Regression table no. 5 shows that students in BiH in general punish the corrupt politician harsher if they are more aware of the consequences of corruption. The coefficient “T” refers to the treatment group. We can see that the treatment coefficient is significant and not close to zero (column 1). The coefficient is a bit larger than a half a standard deviation, which is high. The coefficient remains significant and similar even when basic personal controls (column 2), fixed effects (column 3), socio-economic controls (column 4) and attitudinal controls (column 5) are included. This supports the argument that the students punish the politician harsher if they are more aware of the consequences of corruption, in other words: priming effects are robust.
Regression table no. 5 corruption intolerance (1) (2) (3) (4) (5) min (6) max (7) T 0.406*** 0.379*** 0.379*** 0.343** 0.335** -0.098 0.186* (0.097) (0.106) (0.109) (0.156) (0.148) (0.062) (0.104) Male -0.013 -0.003 0.181 0.263** -0.025 0.175** (0.109) (0.107) (0.139) (0.128) (0.060) (0.087) Age 0.030 0.005 -0.017 0.001 -0.016 -0.027 (0.026) (0.026) (0.033) (0.031) (0.014) (0.022) Year of uni -0.010 0.015 0.004 -0.012 -0.005 0.009 (0.051) (0.050) (0.055) (0.055) (0.020) (0.041) Grade 0.040 0.075* 0.111** 0.140*** -0.016 0.101*** (0.036) (0.038) (0.045) (0.040) (0.016) (0.031) Living rural 0.034 -0.085 0.078 0.044 0.071 0.067 (0.133) (0.136) (0.157) (0.150) (0.062) (0.112) Raised rural -0.001 0.070 0.009 0.124 -0.096 0.011 (0.139) (0.138) (0.181) (0.163) (0.067) (0.104) Financed by family 0.041 -0.057 -0.022 -0.101 0.062 -0.091 (0.168) (0.163) (0.196) (0.213) (0.081) (0.119) Father higher education 0.001 -0.020 0.056 0.013 -0.060 -0.038 (0.146) (0.140) (0.179) (0.157) (0.068) (0.103) Mother higher education -0.061 -0.023 0.131 0.282 -0.074 0.129 (0.167) (0.165) (0.228) (0.205) (0.091) (0.113) Voting multiethnic 0.101 -0.032 -0.001 0.052 -0.022 0.013 (0.138) (0.139) (0.187) (0.161) (0.074) (0.119) Father income -0.000 -0.000 0.000 -0.000* (0.000) (0.000) (0.000) (0.000) Mother income -0.000 -0.000 0.000 0.000 (0.000) (0.000) (0.000) (0.000) Court verdict 0.239*** -0.025 0.161*** (0.069) (0.030) (0.045) Would buy -0.031 0.047 -0.007 (0.080) (0.042) (0.044) Interventions effect -0.005 0.011 0.019 (0.091) (0.037) (0.056)
People can change -0.281*** 0.076*** -0.157***
Another way of examining the priming effects is to see if the information on corruption has affected the ethnicity of suspect effect. Two additional regressions were estimated. In the first regression table, only students who have received
information on corruption are included. In the second regression table, only students who have not received information on corruption are included. “T” refers to the treatment group in which students from all three universities who have answered a survey with a corrupt Bosniak politician belong. The control group refers to students who have answered a survey with a corrupt Croatian politician. By comparing the coefficients “T” in the two tables, we can see that the coefficients are somewhat smaller when students have read an information sheet on corruption before stating a sentence for the politician. The easiest way to see the effect of the information is by comparing column 6 and 7 in the two regression tables (significance is lost when information on corruption is included). This means that students who are more aware of consequences of corruption tend to “forget about ethnic tensions” and give
With info on corruption Regression table 6 corruption intolerance (1) (2) (3) (4) (5) min (6) max (7) T -0.048 -0.088 -0.099 -0.078 0.532 -0.081 0.451 (0.161) (0.207) (0.213) (0.427) (0.477) (0.140) (0.374) Male 0.183 0.214 0.662 0.891* -0.163 0.728 (0.197) (0.195) (0.579) (0.422) (0.171) (0.408) Age 0.018 0.028 -0.132 -0.122 0.041 -0.081 (0.031) (0.035) (0.119) (0.102) (0.041) (0.093) Year of uni -0.069 -0.081 -0.048 0.032 0.010 0.041 (0.113) (0.122) (0.200) (0.130) (0.071) (0.146) Grade 0.031 0.082 0.363 0.296 -0.063 0.232 (0.118) (0.121) (0.278) (0.269) (0.111) (0.224) Living rural -0.446 -0.482 -0.351 -1.709 0.180 -1.528 (0.281) (0.295) (0.773) (0.876) (0.405) (0.928) Raised rural 0.074 0.014 0.179 1.107 -0.017 1.090 (0.288) (0.331) (0.772) (0.798) (0.395) (0.806) Financed by family 0.060 0.090 0.270 0.353 -0.118 0.235 (0.245) (0.248) (0.493) (0.555) (0.232) (0.609) Father higher education -0.060 -0.069 -0.137 -0.060 0.069 0.009 (0.227) (0.228) (0.390) (0.424) (0.153) (0.512) Mother higher education -0.413 -0.433 -0.693 -0.961* 0.327 -0.635 (0.352) (0.354) (0.662) (0.446) (0.263) (0.424) Voting multiethnic -0.342 -0.323 -0.833 -0.402 -0.094 -0.496 (0.286) (0.332) (0.457) (0.413) (0.165) (0.484) Father income 0.000 -0.000 -0.000 -0.000 (0.000) (0.000) (0.000) (0.000) Mother income 0.000 -0.000 -0.000 -0.000 (0.000) (0.000) (0.000) (0.000) Court verdict 0.648 -0.038 0.610 (0.330) (0.124) (0.347) Would buy -0.279 0.151 -0.128 (0.292) (0.125) (0.237) Interventions effect -0.050 0.076 0.026 (0.367) (0.133) (0.389)
People can change 0.048 -0.085 -0.036
(0.258) (0.101) (0.249) Bosnian university -0.219 -0.644 -0.523 0.091 -0.431 (0.252) (0.555) (0.484) (0.169) (0.485) Croatian university 0.016 -0.525 -0.981** 0.011 -0.970* (0.292) (0.445) (0.344) (0.217) (0.432) Observations 42 39 39 25 25 25 25 R-squared 0.002 0.282 0.308 0.571 0.874 0.782 0.829
Without info on corruption Regression table 7 corruption intolerance (1) (2) (3) (4) (5) min (6) max (7) T -0.177 -0.198 -0.188 -0.152 -0.236* 0.133** -0.226*** (0.111) (0.123) (0.122) (0.147) (0.130) (0.063) (0.081) Male -0.089 -0.080 0.100 0.181 0.004 0.102 (0.128) (0.125) (0.156) (0.138) (0.066) (0.088) Age 0.036 0.004 -0.007 0.014 -0.024* -0.028 (0.034) (0.034) (0.036) (0.034) (0.014) (0.024) Year of uni -0.011 0.026 0.000 -0.015 -0.008 0.012 (0.060) (0.058) (0.060) (0.064) (0.023) (0.046) Grade 0.025 0.065 0.076 0.119*** -0.008 0.078** (0.037) (0.041) (0.046) (0.043) (0.020) (0.031) Living rural 0.115 -0.031 0.167 0.199 0.034 0.173 (0.157) (0.156) (0.175) (0.165) (0.069) (0.118) Raised rural 0.054 0.122 0.051 0.115 -0.091 0.001 (0.165) (0.162) (0.205) (0.182) (0.074) (0.111) Financed by family 0.015 -0.105 -0.029 -0.079 0.040 -0.119 (0.205) (0.200) (0.225) (0.243) (0.093) (0.135) Father higher education -0.004 -0.027 0.039 -0.028 -0.043 -0.067 (0.179) (0.165) (0.212) (0.184) (0.083) (0.111) Mother higher education -0.005 0.032 0.244 0.367 -0.137 0.130 (0.203) (0.195) (0.270) (0.231) (0.085) (0.119) Voting multiethnic 0.214 0.060 0.195 0.161 -0.004 0.129 (0.141) (0.145) (0.196) (0.159) (0.083) (0.108) Father income -0.000 -0.000 0.000 -0.000* (0.000) (0.000) (0.000) (0.000) Mother income -0.000 -0.000 0.000** 0.000 (0.000) (0.000) (0.000) (0.000) Court verdict 0.254*** -0.037 0.165*** (0.074) (0.032) (0.046) Would buy 0.009 0.028 0.013 (0.085) (0.042) (0.046) Interventions effect -0.004 0.012 0.039 (0.114) (0.048) (0.067)
People can change -0.307*** 0.089*** -0.162***
(0.072) (0.031) (0.045) Bosnian university -0.139 -0.309* -0.371*** 0.158** -0.196** (0.133) (0.164) (0.134) (0.072) (0.099) Croatian university -0.515*** -0.462*** -0.578*** 0.183** -0.255** (0.160) (0.173) (0.172) (0.076) (0.125) Observations 197 175 175 128 128 128 128 R-squared 0.013 0.046 0.101 0.131 0.334 0.279 0.291
