• Abstract and Figures
• Public Full-text

RACIAL BIAS IN LETHAL POLICE ACTION IN BRAZIL1

1. Racial Prejudice in Brazil. As is widely known, Brazil is a racially mixed country, populated by the descendants of various groups: Portuguese and other European immigrants that colonised it from the XV century; black African immigrants that were brought as slaves over the centuries; non-European immigrants that arrived mainly in the XIX and XX century; and native inhabitants. The degree of racial intermix is relatively high. The official race categories, formally referred to as 'colours' by the Office of the Census (IBGE) and by other official documents are: black ('preto'), brown ('pardo'), white ('branco'), yellow ('amarelo') and indigenous ('indígena'). Out of the 146 million inhabitants registered in the 1991 Census, approximately 51% declared themselves to be white and 5% black, whereas 42% answered they were brown. If we consider that most people designed as 'brown' have some black African blood, this implies that Brazil is second only to Nigeria in terms of the size of its African origin population. Even though some sectors are using and demanding the use of the category negro ('negro') to substitute both black and brown, we will stick to the census categorisation given that this research will be based on official documents. As a consequence of slavery, the black population traditionally occupied the lowest strata of the social pyramid. Brazil was the last country in the continent to abolish slavery, in 1888. It is also considered to be the country with the most unequal wealth distribution in the world (World Bank, 19972). In a country with such an unequal income distribution and with limited opportunities for social mobility (see Scalon, 1999), it is not surprising that the black population is still today disadvantaged with respect to the white majority. In a society sharply structured by socio-economic differences, the debate on inequality and discrimination has often centred on whether these differences originate in class or in race. Several classic authors have traditionally defended the thesis that the

1 This research is part of a project conducted by the office of Human Rights in Rio de Janeiro thanks to a grant of the Ford Foundation to study the impact of race in the public security system in Brazil. The research team included James Cavallaro, Ariel Alves and Cristina Jakimiak, apart from the author. 2 The World Development Report published by the World Bank (1997) shows Brazil with the highest inequality coefficient (Gini coefficient) of all 64 countries who reported it. Likewise, it reveals that, out of 84 countries that report data, Brazil is the country where the richest 20% of the population have the greatest share of the national income (67,5%). 1 social structure in Brazil is determined by class rather than by race. Gilberto Freyre (1933) in his renowned book "Casa Grande e Senzala" contends that, despite the historical fracture between lords and slaves, the defining trait in Brazilian society is that of miscegenation, fostered in turn by the personal attachments between these opposing social strata that would be characteristic of Portuguese colonisation. Thus, Brazil could be defined as a "racial democracy" where social inequalities that did have a historic origin in slavery would have turned to be, at present, exclusively determined by social factors without a racial underpinning. Florestan Fernandes (1965), from a slightly different perspective, states that, since the disadvantages of the black population derive from their defective adaptation to a competitive economy after slavery was abolished, they could be overcome by a simple process of modernisation that incorporates blacks into the capitalist economy. In this scenario, the most important notion to address the question is that of bias. If there is racial bias, it means that people under the same conditions and with the same personal profile may be treated differently according to and because of their skin colour. Thus, highlighting that the blacks are disadvantaged would not prove per se the existence of racial prejudice in society, since these differences could also be explained as the perpetuation through class reproduction mechanisms of the racial divide that existed during slavery times. Even though racial discrimination might cease, blacks would transmit their low status to their children and so would the whites, thereby perpetuating the difference even in the absence of racial discrimination. This is not a hypothesis to be dismissed lightly considering the low degree of mobility of the Brazilian social structure (Scalon, 1999). As a result, if one is to show racial discrimination one has to demonstrate that blacks with the same characteristics as whites receive a worse treatment. Indeed, several studies in the last decades have revealed that blacks and browns earn less than whites even when they have the same degree of education, age and work experience (Hasenbalg, 1979; Silva, 1985). This is strong evidence against the notion of racial democracy, at least in the work market, as it underlines the existence of a racial bias on top of the prevailing inequalities between the racial groups. Not only do blacks and browns achieve a lower educational level than whites but, when they compare

2 themselves to whites with the same number of years of schooling, they also receive lower salaries than their white counterparts. So the difference cannot be explained simply in terms of an educational or otherwise human capital differential between the groups. Apparently, there is racial discrimination too. As a result of the clear correlation between race and class in the country, whereby blacks are poorer than whites, this methodological precaution of comparing blacks and whites in the same situation is necessary if one is to argue for the existence of racial prejudice or discrimination in Brazil. In fact, a public opinion poll conducted by the Folha de São Paulo in April 1995 on a sample of over 5.000 interviewees of the whole country showed that the vast majority of the population (80%) agreed that racial prejudice was a fact of life in Brazil.

2. Racial Prejudice in the public security system.

Brazil is a relatively violent country. Its homicide rate is listed as 29 per hundred thousand in a study carried out by United Nations in 19973, which is the third highest among 36 reporting countries, after South Africa and Jamaica. Yet, this rate can exceed 50 per hundred thousand in the major metropolitan areas of the country, like Rio de Janeiro and São Paulo. Over the last few years, only Colombia and some countries in Central America have worse rates in the continent. Public security is a major public concern in major cities. The prison population is relatively high and most inmates are members of the low class. There are wide allegations that police react with brutality (Human Rights Watch, 1997), especially when they intervene in the poorest neighbourhoods. This blends in with the notion of the "dangerous classes", which could also be read as the "dangerous races", the idea that certain social groups are inclined to crime and violence and must therefore be contained. The racial reading of this notion has been noted by several authors (Caldeira, 1996) and is reflected on the popular dictum that says "a white man running is an athlete, a black man running is a thief". If there is indeed racial prejudice and discrimination in Brazil, it would be hard to think of a better scenario for this prejudice to unfold than the public agencies that deal with

3 Crime Prevention and Criminal Justice Division. United Nations International Study on Firearm Regulation. United Nations Office at Vienna. 3 formal social control, in other words, with the preservation of order and the prevention and punishment of social deviance and crime. These agencies include police forces, prosecutors, judges and prisons. Altogether they form what we call the system of public security4. The reasons why any prejudice would have plenty of space to manifest itself are not difficult to imagine. These agencies deal particularly with people from the lowest socio- economic strata, where blacks and browns tend to be overrepresented as compared to the general population. As far as the police are concerned, they are accorded the use of legitimate violence and they do have to employ force or the threat of it in their work. This would open the possibility of venting their would-be prejudices against certain social targets through an excessive use of force beyond what is needed in the circumstances. The weak organisational control on daily police work and the prevailing impunity regarding police abuses (see Cano, 1999) would facilitate this course of action. The judges, for their part, have to make difficult and delicate decisions on individual cases where a wide array of factors has a bearing. It would not be far-fetched to think of the possibility that some individual and social biases on their perception, as for instance a racial bias, would contribute into the final decision. There is a long line of research on biases on juridical verdicts in other countries. For instance, some researchers in the US have gathered evidence pointing that the likelihood of a death sentence is higher if the victim is white than if he or she is black. (Radelet, 1981) As for the prisons, difficult environmental conditions (see Human Rights Watch, 1998), and the tradition of punitive actions against the inmates (see Varela, 1999) also leave ample room for bias and prejudice to appear. The greater the degree of arbitrariness in a certain context, the wider the door left for bias to enter. In fact, several authors have suggested that the institutions that deal with public security in Brazil are harsher against the black population (Paixão, 1983; Silva, 1988). Racial bias might occur at any of many moments in which the system of public security interacts with individuals, namely: a) Police approach to citizens. Policemen can approach more often members of a certain racial group if they consider it to be suspicious or can treat them in a rougher way when 4 Some people would refer to this as the criminal justice system. We prefer the notion of public security system since this includes not only issues of crime but also misdemeanors and the maintenance of public order and of the feeling of security of the population. In fact, most of the work carried out by policemen, particularly those belonging to the Military Police in Brazil, have to do more with these issues rather than with investigating or preventing crime.

4 they stop them in the street. Focus groups with young kids from slum areas carried out for the project "Serviço Civil Voluntário" (Voluntary Civil Service) in 1998 in Rio de Janeiro revealed that they feel specially targeted by policemen who allegedly pick on them on numerous occasions. And when the police deal with suspects, they could be more violent against suspects of a certain group. b) Police decision to register a complaint. Police discretionary work involves common decisions on whether somebody should or not be taken to the police station to register the facts he or she is accused of. Discriminated minorities could be taken more often to the police station for the same facts for which a member of the dominant group might simply be admonished or induced to pay bribery. c) Decision to open an enquiry. The Brazilian Civil Police, after registering a complaint, may decide to open an official enquiry ('inquérito') into it. For some crimes, as homicide, this is compulsory, but not so for others, where the decision may depend on the will of the victim, the severity of the facts, etc. All over the world, police have a wide margin to decide how to handle minor offences. In this respect, crimes committed by members of the discriminated group or against members of the dominant majority could lead to enquiries more often than the others. d) Decision to prosecute. Once an enquiry is open, it has to be sent to the prosecutors who then have to opt whether they will formally accuse somebody of a crime or whether they will file the enquiry without prosecuting anybody. If they decide on the former option, the enquiry turns into a legal process ('processo') and it has to go the courts. Race of the perpetrator and of the victim might play a role in this decision. e) Condition while awaiting trial. Member of racial minorities may be refused parole and kept in prison before trial more often than other groups. f) Sentencing. Judges might display a racial bias by, consciously or unconsciously, incurring in a higher probability of finding somebody guilty if he or she belongs to a certain racial group, or by pronouncing harsher sentences against its members. In both cases, it is obvious that any proof of racial bias has to respect the 'ceteris paribus' clause. In other words, it is not enough to find evidence that blacks are sentenced more often or to more severe penalties, it is necessary to show that they are worse treated than whites for the same crimes and under the same circumstances.

5 f) Prison treatment. Inmates of a certain race might be worse treated by the prison system. g) Prison benefits and sentence reduction. Convicts of a discriminated minority might find it harder to achieve certain prison benefits, like night-time incarceration, or a reduction of the time to be served.

These are levels in which the state apparatus could display some racial bias. However, they do not encompass all possible discriminations that society at large may dispense to those suspected, accused or sentenced. For example, former black convicts might have more difficulties in finding a job than white ex-convicts but we could not attribute this to the state. In all these instances, bias would have to be proven while controlling for other influences. Most important, socio-economic status or class is considered to be an important determinant of the treatment accorded by the public security apparatus. For instance, Brazil has a penal law that allows for special, better, prisons for those prosecuted who have a university degree, in clear contradiction with the basic principle of equality before the law. As a result, comparing the treatment accorded to people of different races who belong to the same social class can only test racial discrimination. As we mentioned before, discrimination is more likely to show up in instances where there is more room for discretionary decision on the part of the public agent. For instance, there are no set rules to prescribe when somebody should be stopped and asked to identify him or herself or what personal characteristics are considered suspicious. Hence, the stereotypes of the agent can act unrestrained. On the other hand, judges have specific rules, like aggravating and mitigating circumstances, on which to base the penalties imposed and the verdict of guilt itself. In this context, the influence of possible prejudice will probably be limited. Yet, data are more abundant and more detailed precisely on those levels that are more regulated and where the impact of bias is probably smaller. The best example refers to sentences. A judicial process contains many documents to justify the final sentence. All this information may be difficult and tedious to process, as it is not usually presented in a form that allows an easy analysis of a set of cases, but is still there for the scholars to research.

6 On the other extreme, there are no registers of the many instances when policemen decide to stop somebody and request identification. Likewise, there is no way to recover the kind of treatment that policemen accord to people they interact with in the streets. It is possible to think of a research design to measure racial bias in these contexts, but the costs and the methodological difficulties involved are enormous. Hence, research often tends to focus just on the instances where racial bias is probably more infrequent and less intense. Several studies have addressed the existence of racial bias in judicial sentences in Brazil. Costa-Ribeiro (1995) analysed a sample of the judicial proceedings related to 'blood crimes' (homicide and attempted homicide) in the First Jury Tribunal ('Primeiro Tribunal do Juri') of the city of Rio de Janeiro, between 1900 and 1930. He concluded that black defendants showed an apparently higher likelihood of being found guilty than brown or, particularly, white defendants. Further, defendants accused of attacking white victims had also a higher probability of being considered guilty than those who allegedly attacked brown or black victims. He also quotes several Brazilian authors of the end of the XIX century and the first half of the XX century as openly defending the thesis that blacks were more inclined to crime, be it out of an innate tendency (Rodrigues, 1894; Cunha, 1936) or be it due to cultural lag and maladjustment (Hungria, 1956). Thus, the increased likelihood of condemning blacks would just be an application of prevailing stereotypes and prejudices against them. More recently, Adorno (1995) reviewed a sample of sentences for cases of robbery with the help of accomplices ('roubo qualificado com concurso de outros agentes') issued in the city of São Paulo in 1990. He found that negroes (including 'pretos' and 'pardos') enjoyed private attorneys and presented defence witnesses less often than whites, and that their likelihood of being found guilty was 9% higher than that of whites. Thus, he concluded for the existence of racial bias in the judiciary, and he pointed out that what distinguished Brazil was the tolerance towards discrimination rather than the absence of it. On the other hand, Kahn (1999) analysed sentences for crimes of homicide, robbery and theft in the state of São Paulo, controlling only for the number of crimes committed, and was unable to find a differential racial outcome in the sentences of any of these three types of crime.

7 One way to overcome the lack of data in the official records on police action according to the race of the people involved is to resort to information from the victims as, for example, in victimisation surveys. Mitchell and Wood (in press) analysed the victimisation supplement of the PNAD (National Residential Sample Survey) carried out by the Bureau of the Census (IBGE) in 1988, and found that blacks were more likely to be assaulted by policemen than whites, even after controlling for region, urban status, age, education and income. Blacks were also more likely to be assaulted in general than whites, but the imbalance was particularly strong in relation to assaults by policemen: whereas the odds of a black person being assaulted were 1.4 times the odds of a white person, the odds of a black person being assaulted by a policeman were 2.4 times higher than those of whites. Some other studies gathered similar evidence. Kahn (1998), analysing opinion and victimisation surveys for São Paulo from 1995 e 1997 (carried out by Datafolha), demonstrates that blacks are more afraid of the police than whites (20% vs. 11%), to the point that in 1997 blacks were the only group who tended to be more afraid of the police than of the criminals. Nearly half the black citizens interviewed had been searched by the police (47%) compared to a third of the white citizens (34%). In another poll with over a thousand interviewees in São Paulo in 1997, 6% of whites and 14% of blacks claimed to have been physically assaulted by policemen. However, a survey by CPDOC/ ISER in the Rio de Janeiro metropolitan area in 1996 did not reveal the same clear pattern of race bias in police action (CPDOC-FGV/ISER, 1997). In any case, most of this research points in the direction of blacks being a preferential target of police action, both legal and illegal. The present article also deals with the question of racial bias in police action and, in particular, in actions that result in the death of civilians. It sets out to test whether police are more violent against blacks or browns than against whites. It would be interesting to analyse police action as a whole, rather than just those actions that result in fatalities, since any racial bias in the latter will probably be a small sample of a more general problem. However, data on police actions which include both details of police approach and race of the persons involved in large scale are simply not available. As a result, we believe that the analysis of possible bias in the use of lethal force by police is extremely relevant, both because it may represent a more general tendency of racial bias in policing which is otherwise very hard to research, and because of its dramatic consequences.

8 3. Different Manifestations of Racial Bias.

Basically, racial bias can be expressed in two forms: a) The most obvious way is by granting a better treatment to members of one racial group than to members of others, even when they both share the same characteristics and conditions. For example, blacks would be considered as suspicious more often, be treated more harshly and violently by police, be sentenced more often and suffer worse conditions in the prison system. The state agents would be following and applying their racial stereotypes and prejudices in their daily work. The origin of this prejudice need not be in the institutions themselves but may be just a result of prevailing stereotypes among the citizenry. In this case, race can be conceived as the independent variable that determines the treatment by the state apparatus: the type of police approach, sentencing, and other dependent variables. Thus, race is treated as something fixed and race variation is the element that determines the outcome. b) The second possibility is that race itself, or rather, the perception of race, is determined by other factors such that the negative association between crime and certain racial groups is maintained. Hence, it is not that, for instance, blacks are worse treated, but that people viewed in a negative light, say criminals, are seen as darker -and therefore more likely to be classified as black- than law- abiding citizens of the same skin colour. So it is the negative characteristics of the person that trigger the perception of its colour as darker than it would otherwise be. People who are considered undesirable, and therefore treated in a worse way, tend to be perceived as darker. In this way, the connection between race and a negative label (e.g. crime) is sustained but the causal link is reversed: the negative traits determine race classification rather than vice-versa. Under this hypothesis, race would not be conceived as an independent variable but as dependent on other elements. Further, race would not be a fixed variable with several possible categories, but a fluid continuum which is crystallised in a category as a response to a specific question or situation. When the answer has

9 to come as one of a few pre-established categories, as in the census, the person chooses that which most resembles his position in the continuum. Moreover, this classification is not consensual among different people nor stable for the same person over time, and it responds not only to physical characteristics (the person's phenotype) but also to socio-demographic variables and even to the situational context of the question. In other words, the same person may be defined as white by some people and as brown by others, and the person him or herself may change his own racial category over the course of his life. Many scholars subscribe to this flexible and even situational notion of race (Harris, 1964; Barth, 1969; Sansone, 1997; Wade, 1997). Indeed, there are several pieces of research that call into question the idea of a consensual, objective and stable racial classification. Wood (1991) compared the same cohorts in the 1950 and 1980 censuses and concluded that the change of racial distribution could only be explained by the fact that many people classified as black in 1950 reclassified as brown in 1980, a move he attributed to the will of upwardly mobile blacks to 'whiten' their race. Based on the study of a village, Harris (1964) also mentioned a tendency towards the 'whitening' of more affluent and better educated non- whites. This process of 'whitening' could be occurring for the observers and for the person him or herself. In the already mentioned survey conducted by Folha de São Paulo in 1995, people were classified in terms of race by the interviewers and were also asked to define their own race. Agreement of self and interviewer classification was predominant (79%) but that still leaves 21% of respondents with a discrepant race categorisation. Another opinion survey carried out by CPDOC/ ISER in 1996 in the Rio de Janeiro metropolitan area (de Carvalho et al., 1998) revealed that 76% of those classified by interviewers as white defined themselves to be white, whereas only one third of those classified by the interviewer as browns attached this label to themselves (40% of them preferred the alternative category 'moreno' which can be translated in the same way). People categorised as blacks by the interviewers, for their part, defined themselves accordingly in less than two out of three cases (37% used ‘negro’ and 29% 'preto'). This inconsistency seems to be more pronounced among the

10 black. Both these surveys showed that people asked to classify their race or their colour tend to use a wide array of nuanced categories on top of those included by official documents. In short, these results reveal that when people are asked to classify themselves at least one out of four persons defines him or herself racially in ways different than an external observer. There are many possible theories to explain why racial classification may shift or how it can be mediated. Some of them are of a more structural, sociological nature, like the queuing theory (Lieberson, 1990) that states that when the dominant group does not have enough members to fill in important positions in the social pyramid, new opportunities arise for minorities to reclassify as members of the dominant group, that is, to 'whiten' themselves. Or like competition theories, that predict that in places where there is more economic competition, societal rewards will be allocated according to more universalistic criteria (Fernandes, 1965), thus reducing the importance and the salience of class distinctions and the influence of racial classifications. Other theories are more psychosocial, like the prediction that the white versus non-white differentiation will become more pronounced if the dominant white majority feels threatened. Cognitive theories of perception can also contribute to explain race attribution: contrast theory will predict that the same person with just a moderately dark skin and a few African traits will be considered black in an environment where the percentage of whites is very high, but may be seen as brown or even white in communities with a large proportion of black people. The thresholds between black and brown and between brown and white would be determined by the relative distribution of races in the local context. Telles and Lim (1998), analysing the data of the above mentioned 1995 survey by the Folha de São Paulo, conclude that, after controlling for education and other labour market variables, whites earn 26% more than browns with interviewer racial classification but only 17% more with self-classification. They explained these results by suggesting that interviewers tend to ‘whiten’ those with higher socio-economic status and ‘darken’ those with low status. In a later article on the same data, Telles (1999) points out that self-identified browns

11 with higher education tend to be classified as white by interviewers, whereas education does not seem to have an effect on blacks' consistency in racial categorisations. Thus, he argues that well-educated browns are 'whitened' by interviewers while no such process operates for blacks. In the absence of an 'objective' measure of race, it is difficult to distinguish between people being 'whitened' by interviewers or 'blackening' themselves, since both processes can lead to the same results. Blackening and whitening have been only inferred empirically by comparing one classification criterion with another. Given that both are subjective, its is hard to determine with clarity what process is exactly at work.

The collapse of the notion of race as objective, stable and consensual introduces a caveat in research on racial bias. Since race is not an objective classification, every time we report it we have to be specific as to how it was measured and by whom. Furthermore, if one is to show bias by comparing the proportion of race in two sources of data, say proportion of each race in people arrested by police and in the population as a whole, both sources should, in principle, measure race in the same way and in moments not too far apart, for we have seen that the same persons may change their racial attribution over time. The census and other similar public surveys measure race through self-attribution. However, there are experts who believe that some interviewers do not always ask that question but often fill it in themselves, because of the will to end the questionnaire quicker when they think they already know the answer or because of the embarrassment involved in asking the question (Pinto, 1996). To the extent that this is true, we will be facing census data that would be the result of a mixture of self-attribution and interviewer-attribution. Probably, interviewers would tend to answer the item themselves particularly in cases where the race category of the person seems more obvious to them (very white or very black persons) but there is no way to assure this. While the census relies on self-classification, many other official documents seem to be filled in by the officials involved, often in the presence of the person but presumably without asking. For example, police reports refer to race of suspects or victims as perceived by the policemen, and there is some logic in this since police use these documents for

12 internal communication. For instance, if policemen are searching for a suspect, it is more important for them to note the race as defined by other colleagues, whose criterion would presumably be known to them, than the self-categorisation of the subject. Yet there is no guarantee that different policemen, and particularly policemen belonging to different racial groups, will classify race in a consensual manner. In short, it is not obvious from most documents how the race question was answered, whether with the participation of the person or not. One cannot even be sure that each and every document of the same source was filled in following the same procedure. In this study, we will try to compare the race classifications originating from different sources.

4. Objectives and Methodology.

As was explained before, the main objective of this research is to test whether there is a racial bias in police use of lethal force. In other words, we will put to the test the general hypothesis that police may be more violent against racial minorities, notably black people, than against whites. If the hypothesis is confirmed, it will be interesting to verify what the outcome is with respect to the intermediate category: browns. We will address this general hypothesis by analysing police actions that lead to civilian deaths, both because there is an official record of persons killed by police which does not exist in the case of people ill-treated by the officers, and because of the tragic consequences of these actions. Hence, the question could be rephrased more specifically as: does the police kill more black people than white people? Ideally, the question should be even more specific: does the police kill more black people than white people in comparable circumstances? The last three words are of crucial importance if we are to prove the existence of racial bias. Every major police force has to make use of lethal force at some point or other in order to defend themselves or the life of others. In this area, the issue is how much force is justifiable5, so as to separate the legitimate use of lethal force from the abuses and from the

5 The United Nations, during its Eighth Congress on the Prevention of Crime and Treatment of Offenders in August and September 1990 in Havana, Cuba, drafted the Basic Principles on the Use of Force or Firearms by Law Enforcement Officials, which state in point number nine: 13 executions committed by policemen under the disguise of fighting crime. Brazilian police have repeatedly been accused of an excessive use of force (Chevigny, 1991; Human Rights Watch, 1997; Cano, 1998). Nevertheless, this is not the concern of this article. The objective here is not to determine whether police kill too much or not but whether, in the same circumstances, they are more likely to kill members of some racial groups than others. Therefore, we will not dwell too much on the question of how many of these deaths were justified and how many of them could or should have been avoided. Rather, we will stick mostly to the comparisons between racial groups. On the other hand, it cannot be denied that, if some racial groups are killed more than others in comparable circumstances, this implies an excessive use of force in itself since lethal force should always be exerted in the minimum degree possible. As a result, the level of victimisation of the groups that were most targeted should be reduced to, at most, the level of the other groups.

5. Data Sources. Two cases were examined: Rio de Janeiro and São Paulo. Different sources were used in each of them: a) In São Paulo, a double strategy was followed. First, the research team visited all five Jury Tribunals of the city of São Paulo, where intentional crimes against life are tried. Four out of the five Tribunals, all except the Fifth Jury Tribunal, agreed to allow access to their archives for this research. The registry books of each Tribunal were read to seek out the cases of intentional homicide where the profession of the indicted person was military or civil police. All cases that had occurred between January 1996 and the moment of the research (June-July 1999) were included, both if the policeman was on duty and off duty. The cases whose proceedings had come to an end were filed in the archives of the tribunal, so they had an archive number in the registry book and it was easy to have access to them. Cases whose proceedings were still pending were much harder to trace and many of them could not be found. Despite this, three out of the four Jury Tribunals were able to come up with a few pending cases. “Law enforcement officials shall not use firearms against persons except in self-defence or defence of others against the imminent threat of death or serious injury, to prevent the perpetration of a particularly serious crime involving grave threat to life, to arrest a person presenting such a danger and resisting authority or to prevent his or her escape, and only when less extreme means are insufficient to achieve these objectives. In any event, intentional lethal use of firearms may only be made when strictly unavoidable in order to protect life. “ (not underlined in the original) 14 As a complementary strategy, we requested from the Police Ombudsman ('Ouvidoria'6) the number of the judicial process of all cases of intentional homicides committed by policemen that they had collected for the period of interest. We then requested the cases corresponding to these judicial process numbers from the Jury Tribunals, as a way to double check that we would not be missing any cases. All in all, we collected information on a total of 215 civilians intentionally killed by policemen in the city of São Paulo. For each of these cases, the research team filled in a form that gathered information on: a) registry number and archive number (if the case had been filed, either because it was dismissed or because it reached a final verdict); b) victim's name and identity number; c) date of the facts; d) documents that mentioned 'colour' of the victim and the corresponding colour category attributed to the victim. Researchers attempted to collect one document with information about race from each of the following types: 1. Criminal records ('Boletim Individual' or 'Relato Vida Pregresa'); 2. Police documents referring to the arrest ('Flagrante', 'Registro de Ocorrência' or 'Mandado de Prisão') 3. Autopsy Reports 4. Other documents.

The intention was to collect one document from each different type rather than all the possible documents in order to reduce the amount of time researchers spent browsing over each case. The assumption behind this is that documents of the same source would not show major inconsistencies in the racial classification of the victims. In fact, most of the cases included more than one official document that registered the person's colour, which will enable us to test the consistency of the racial classification of different sources.

6 The Ouvidoria is an institution recently created in serveral states of Brazil. Its mandate broadly consists in: a) collecting citizens testimonies of police abuses and crimes while protecting the identity of the witnesses; b) following up the investigations of such cases carried out by Internal Affairs Units; and c) informing society on these issues. 15 b) In Rio de Janeiro, we worked with the data from the research carried out by ISER (Instituto de Estudos da Religião) on the use of lethal force by police in the city of Rio. This research, conducted at the request of the Human Rights Commission and the Public Security Commission of the State Assembly, analysed all registered cases of civilians killed or wounded by gunshot in police actions in the city of Rio between January 1993 and July 1996. It was based primarily on Civil Police Incident Reports ('Registros de Ocorrência') and on Autopsy Reports of the Office of the Forensic Pathologist but consulted also some Military Police Inquiries and Inspections and Reports of the Rio de Janeiro Military Police Command’s Special Commission of Summary Investigation. The results of this investigation (Cano, 1998) showed that the police in Rio killed a very high number of people, and that this lethal outcome doubled after the introduction of a "bravery reward" that offered pay bonuses primarily to policemen who engaged in shootouts against suspects. Further, the 'lethality index' (number of civilians killed divided by the number of civilians wounded in these encounters) also doubled after the introduction of these policies, jumping from 1.7 fatal victims per person wounded to 3.5, thus revealing that policemen displayed in many cases an intent to kill rather than capture the opponent. The excessive use of force and the presence of cases of summary executions were confirmed by forensic data: 61.5% of the fatal victims presented at least one shot to the head and 65% of them showed at least one shot on the back; one third of these corpses had also injuries unrelated to gunshot wounds; 40 victims had been killed with shots at point-blank range. The data set included information on the incident, on each of the victims and on the policemen involved. The main advantage of these data is the presence of information on people wounded but not killed by police interventions, which is otherwise not easy to obtain. Thus, we will be submitting these data to a reanalysis from the perspective of the race of the victims. Also, in the original data set, 'colour' of the victim originated from Civil Police Incidence Reports. However, our research team also coded and processed all documents from the Office of the Forensic Pathologist that mentioned the race. This will make it possible to test the consistency of the racial classification of medical and police sources. The data include information on 991 civilians killed and 726 civilians wounded by gunshot.

6. Results.

16 6.1 Results of São Paulo. Of the total of 215 fatal victims of police actions, 179 (83%) corresponded to judicial processes that had already been filed (most of them dismissed) at the time of the research, while the 36 remaining victims (17%) belonged to judicial processes that were still pending. The question that arises is whether filed cases might have a different profile from those that are pending. If so, we may be in front of a relatively biased sample, since most of our cases are dismissed. The first and most obvious difference between both types of cases is that pending cases are more recent, for they did not have enough time to be judged or dropped. The crosstabulation of type of case by year proves that the proportion of pending cases is lowest in the first year and highest in the last two years. This also means that the number of total cases decreased over the years, probably not because fewer cases occurred but because they did not reach a legal conclusion. TABLE 1 Present Status of the Case by Year Present Status of the Case Pending Filed YEAR Total 96 7 88 95 97 14 63 77 98 6 24 30 99 9 4 13 Total 36 179 215

Nevertheless, pending cases may be different from dismissed ones not just because they are more recent but, probably, because they contain more evidence or proof that the homicide committed by the policeman may not have been justified. The cases where the evidence points to the possibility of guilt will most likely take longer since both the prosecution and the defence will devote more time to them. Cases with clear indications that the behaviour of the police was justified or cases without any evidence at all will probably be dismissed sooner. In fact, several of the cases that had deserved special attention by the Police Ombudsman and by the press were still open. Hence, filed cases may not be a completely representative sample of the total universe of relevant incidents after all. One methodological option would be to work just with the most distant years,

17 those in which the processes had more time to be resolved, thus diminishing the problem by reducing the number of pending cases. However, this is not a good idea in our situation since we would lose a substantial part of an already limited sample. On the other hand, eliminating the pending cases altogether would leave us with a consistent set of cases, those that have been filed, but would at the same time increase the bias with respect to the universe of interest. One should note that, if the main hypothesis that police are more violent against the blacks turns out to be true, then the cases with more evidence of excessive and unjustified use of force will include a higher proportion of black victims. And this is precisely the kind of cases that we have the least, for many of them will probably be still pending and hence much more difficult to find. Consequently, the test of our hypothesis on the present data set will be a conservative one, given that the bias introduced by the unfound cases will, if anything, sway the results against our hypothesis. Unless we are prepared to accept that police are more violent against whites, we can rest assured that, if we find a racial bias against the black population in these data, the magnitude of this bias in the real world will be at least as high if not higher than our results. Specifically, the information about ‘colour’ was extracted from the following documents: TABLE 2

18 Number of Subjects with Information on Race extracted from each of the following documents

Sum R.O. 203 Laudo IML 189 Hospital 35 Atestado de Óbito 10 Boletim de Ocorrência PM 10 Guia Recolhimento Cadáver 7 IPM 7 Laudo Residuográfico 4 Recognição Visuográfica 3 Relato Vida Pregressa 2 Folha Antecedentes 1 Plantão PS 1 Relatório de Identificação 1 Auto de Qualificação em Interrogatório 1 Requisição Exame Corpo Delito 1 Flagrante 1 Boletim Individual 1 Laudo Perinecroscópico 1

Nearly all the victims (203 out of 215) had a Police Incident Report (R.O.) with information on race. And most of them also had race extracted from Autopsy Reports. The rest of the documents are seldom used. We will analyse the colour according to the Incident Report because of two main reasons: a) nearly all victims had an Incident Report so the loss of cases is minimal; and b) the Incident Report is the first document issued by the police, just after the facts, so it will tend to reflect the policemen's perception of the victim race at the time of their intervention better than any other document. We will compare the racial composition of the persons killed by the police with two groups of reference: the general population and the convict population. For the general population, the last official figures for the city of São Paulo belong to the 1991 census.

19 Even though it may seem distant from the dates of the present research, the race composition is relatively stable over shorts periods of time and could not have changed much in a few years. A small percentage of the interviewees, 116.724 out of the total of 9.646.185 people registered in the census, did not have information on 'colour'. Since the percentage of missing data is small (1,21%), these 116.724 people were removed from the table so the results could be compared directly to the other groups which had no missing data. The source for the racial composition of the convict population is the State Secretary of Prison Administration. The data correspond to the people who had been convicted and were under the custody of the prison administration system7 of the State of São Paulo in 1997, a moment which is in the middle of our research period. Ideally, we would like to have the data just for convicts who are residents of the city of São Paulo, which is the geographical area of our research, but it would be extremely difficult to separate inmates in each state prison according to their race and place of residence. As for the people who were killed by the police, the source is the Incident Report, as indicated before. TABLE 3 Race Composition of the People killed by Police, compared to the Convict and the General Population of the city of São Paulo Colour Persons Convict General killed by Population* Population police 1996-1999 % 1997 % 1991 % White 107 52.71% 19,325 54.74% 6,665,960 69.95% Brown 67 33.00% 10,561 29.91% 2,281,067 23.94% Black 27 13.30% 5,272 14.93% 371,042 3.89% Others ------147 0.42% 211,392 2.22% White & Brown 2 0.99% ------

TOTAL 203 100.00% 35.305 100.00% 9.529.461 100,00% (*) The convict population corresponds to the State of São Paulo

7 There are prisoners in the prison system who have not been convicted since they are awaiting trial and there are cases of convicts in the police stations who have not been transferred to the prisons yet. Both these groups are absent from these data. 20 It should be noted that two Incident Reports defined a victim simultaneously as white and brown, thus reminding us of the fuzzy nature of race categorisations. The race distribution of the fatal victims of police actions is very similar to that of the convict population, with the exception of a slightly higher proportion of brown victims. However, the sample of victims of police interventions is relatively small and minor differences in the distributions should not be overinterpreted. In principle, the main difference between the fatal victims and the convict population is that the latter is composed of both major and petty criminals whereas the former should be composed only of dangerous criminals since police should only shoot against opponents that are armed and dangerous. In any case, the racial composition of both groups is very similar. Had we found the victim group to be darker than the prisoners, this could have been interpreted as a sign that police might be acting more violently against the black criminals than against the other suspects, but it could also be argued that the race composition of the more dangerous criminals may not be the same as that of the petty criminals. Conversely, the fact that both racial distributions resemble one another might be interpreted to mean that police show no racial bias in their use of lethal force as compared to their overall repression of criminality, but the caution of the possibility of a different racial composition of small and violent criminals is still valid. Also, we must be reminded that placing somebody in jail is the result of the actions of several institutions that belong to the system of public security (police, prosecutors, judges, etc.) and not just of the police. Therefore, any resulting racial bias of the convict population cannot be attributed automatically to the police. Moreover, the fact that the fatal victims and the convict population share a common race distribution cannot be interpreted as showing an absence of bias on the use of lethal force, since the racial composition of the prisoners may itself be subjected to various and powerful biases. It just shows that they both seem to follow a similar pattern, but this pattern may itself be strongly biased. In fact, when we compare both distributions to the general population, we see that the proportion of white inhabitants in the city is close to 70% whereas less than 55% of the victims or the prisoners are white. Likewise, browns represent only 24% of the population but 30% of the convicts and 33% of the victims. The

21 contrast is particularly strong in the case of blacks that make up 4% of the population but rise up to 15% of the convicts and 13% of the victims. Hence, the likelihood of a black person being killed by police is over three times that of the general population. These results, particularly in the case of blacks, seem to indicate that both the police and the public security system in general are strongly biased against racial minorities in São Paulo. Nevertheless, two caveats are in order if we are to grant this conclusion. This interpretation is dependent on two major assumptions: a) the race categorisations in all three sources (the population census, the prison system and the Police Incident Reports) must be carried out in a homogeneous way, that is, according to common criteria. However, as we have explained, census data are based on self-attribution while prison and police data depend on the perception of the officers. Thus, it is difficult to guarantee the equivalence between the different sources, especially between the population distribution and the other two. b) all racial groups must engage in crime in the same degree. If the different races have a different probability of committing crimes, the comparison between the population distribution and the fatal victims' distribution will be meaningless. In other words, police may be killing proportionally more members of a certain racial group because they engage more in violent criminal activity. Thus, the differential racial proportion might be a consequence of different base rates rather than the result of bias on the part of the police. Fighting against the traditional prejudice that associated blacks to crime and danger, several authors have questioned that such an association exists at all (Adorno, 1995). There is in fact no convincing evidence that some racial groups are more prone to crime than others. However, given the possible link between poverty, marginality and crime on the one hand (see for instance Fajnzylber et al., 1998), and the strong correlation between race and poverty (blacks are much poorer) on the other, the possibility that there exist a higher proportion of people who commit crimes in some racial groups than in others cannot be summarily discarded. While rejecting the old prejudice that saw some races as more inclined to crime due to their nature or their culture, it might still be true that some racial groups live in conditions that make their members more likely to enter a criminal life. This point has not been proven but neither has it been rejected.

22 Hence, even though these results are indicative of the existence of racial bias in police use of lethal force, the possibility of some groups being involved more frequently in criminal activities and the possibility of a difference in the racial classification criterion between census and police data remain legitimate alternative hypotheses. So caution should be taken while interpreting these results.

Next, we analyse the consistency between the racial classifications of the different sources. Penitentiary documents are practically non-existent in this data set. Police documents refer usually to the Incident Report and medical documents refer to the Autopsy Report. Thus, we may decide to check only these two specific documents. If we were to consider all possible police documents we would gain 7 subjects who have no Incident Report and if we included all medical and technical documents, we would gain another 5 subjects who do not have an Autopsy Report. It is a small gain in both cases, which would be reversed by the fact of using documents of different origins, even though categorised by the same general source. As a result of these considerations, we opted to work only with Incident Reports and Autopsy Reports. The Incident Reports works with the categories 'white', 'brown' and 'negro' but does not use the category 'black', whereas the Autopsy Reports employ the labels 'white', 'brown', 'negro' and also 'black'. In order to compare both registries we collapsed the Autopsy Report categories of 'black' and 'negro' into a single one. The crosstabulation of their respective race categorisations yields the following result.

TABLE 4

Crosstabulation of Colour of the Victim according to the Incident Report and the Autopsy Report

Count a Colour according to the Autopsy Report White and White Brown Brown Black/Negro Total Colour White 80 16 96 according to White and Brown 1 1 2 the Incident Brown 14 43 1 58 Report Black 1 6 15 22 Total 95 1 65 17 178 a. Consistent classifications are marked with shading

23 The consistency of both classifications is moderately high since 138 out of the 178 victims (77%) are categorised in the same manner. Yet this means we have over 20% inconsistent cases. The number of people considered white in both sources is practically the same. There are 16 victims classified as white by the police and as brown in the Autopsy Report and there are 14 who show the opposite pattern. This highlights the relatively fuzzy nature of racial attributions. On the other end, 6 out of the 22 people described as negro by the police are seen as brown by the forensic doctors while only one victim presents the opposite pattern (brown for the police, black/negro in the Autopsy Report). These results seem to demonstrate that whereas both sources seem to share similar criteria in the 'white- brown' transition, thus yielding a similar number of errors in either direction (white towards brown, brown towards white), the criteria used in the 'brown-negro' frontier seem to be slightly difference. More people shift from brown in the autopsies to negro in the police than vice-versa. The few victims that are registered as both white and brown in the same document have a strange pattern since they can appear as black/negro in the other document. As a result, the proportion of negroes in the police is slightly higher than in the Autopsy Reports (13% vs. 10%) but the numbers are too small to allow clear conclusions. Thus, there is a mild indication that the police might tend to slightly 'darken' suspects when compared to the doctors of the Office of the Forensic Pathologist. It should be noted that, even though the proportion of blacks/negroes is smaller when using the classification of the Autopsy Reports, it is still much higher than the percentage of blacks/negroes in the population. This reinforces the possibility that there may be a racial bias in police use of lethal force, since the racial imbalance holds even when it is another institution that performs the racial classification.

6.2 Results in Rio de Janeiro. The research carried out by ISER on all incidents of police use of weapons that resulted in civilian casualties between January 1993 and July 1996 yielded the following number of victims.

24 TABLE 58 VICTIMISATION IN POLICE ACTIONS WITH CIVILIANS KILLED OR INJURED(*) NUMBER NUMBER NUMBER OF NUMBER OF NUMBER NUMBER NUMBER NUMBER OF OF OF OF OF OF OPPONEN OPPONENT CIVILIANS CIVILIANS PERSONS PERSONS POLICE POLICE TS S KILLED INJURED KILLED INJURED OFFICER OFFICER (**) NOT NOT S S YEAR KILLED INJURED ACCIDENTAL ACCIDENTAL CLASSIFIE CLASSIFIE KILLED INJURED LY LY D D

1993 155 104 7 48 6 5 5 33

1994 220 129 8 80 9 21 6 25

1995 358 134 10 91 4 3 10 40

1996*** 209 50 5 58 0 3 5 24

TOTAL 942 417 30 277 19 32 26 122

(*) Two more victims were left out of the table since it was unclear from the documents whether they ended up wounded or dead. (**) We call opponents those against whom police shot intentionally. This does not imply that they were necessarily opposing armed resistance to police, since there are cases with clear signs of summary executions. (***)The year 1996 includes only cases through July 31.

Race will be analysed according to the Incident Report because of two main reasons: a) it reflects the perception of colour of police in the moment nearest to the episode and often by the agents who were themselves involved in it; b) it is the only document currently available for all types of victims, since we have medical documents, such as Autopsy Reports and others, only for the deceased but not for the wounded. Out of the total 1,719 victims, 514 did not have their 'colour' registered in the Incident Report. A further two victims were registered as having 'other colour', that is, different from 'white', 'brown' or 'black'. These 2 persons were eliminated from the analysis in order to simplify tables and statistics. In short, we have a rate of missing data on race corresponding to approximately 30% of all cases. This missing rate is much higher for the

8 There are a few minor differences between this table and the one published in Cano (1988) due to corrections on the data that came out after the original publication. 25 wounded (51%) than for the dead (14%), since Incident Reports tend to contain more specific and detailed information on the fatal victims. Also, if a person is dead its external characteristics, such as colour, are a way to identify the corpse, whereas many wounded victims can provide a name or an identity number. This high missing rate among the wounded is true both for opponents (50%) and for accidental victims (53%). Consequently, we suppose that the high proportion of missing data among the wounded does not correlate with 'colour', in other words, it is approximately the same for all races. This is an important assumption for many of the analyses that will follow. The next table presents the race composition of victims of police actions whose race is known, separating victims by type. The race distribution of the general population of the city of Rio, obtained from the 1991 official census, and that of the convict population are also added for reference. The data of the convict population were obtained from the Department of Penitentiary Affairs (DESIPE) of the state of Rio de Janeiro and correspond to December 19969. Unfortunately, they also refer to convicts of the whole state since, for reasons similar to those already explained in the case of São Paulo, it is not possible to separate inmates who were residents of the city from the rest.

9 In the census data, the 0,81% of people who did not provide their colour and the 0,26% who attributed themselves to colours different from white, brown or black, were omitted for the sake of comparability with the victim data. Likewise, the 1% of the convict population who belonged to other races (114 inmates in total) was also eliminated from the analyses for the same reasons. Other racial groups, like the Asians (who report colour as 'yellow') are more important in the city of São Paulo but represent a very small proportion of the population in the city of Rio. 26 TABLE 6

Race Composition of Civilian Victims of Police Actions, of the Convict Population and of the General Population in the City of Rio de Janeiro COLOUR TYPE OF VICTIM CONVICT GENERAL POPULATION POPULATION Dead Wound Accide Oppon ed ntal 1996 1991 White ent240 29.8% Oppone89 42.6% Victim80 50.6% 4,578 40.4% 3,252,517Census 60.0%

Brown 325 40.4% 64 30.6% 38 24.1% 2,975 26.3% 1,715,178 31.6%

Black 240 29.8% 56 26.8% 40 25.3% 3,780 33.4% 454,609 8.4%

TOTAL 805 100.0% 209 100.0% 158 100.0% 11,333 100.0% 4,967,695 100.00%

Sources: Civil Police Incident Reports 1993-1996 (ISER); IBGE Population Census, 1991; DESIPE (Department of Penitentiary Institutions, State of Rio de Janeiro) December 1996. (*) Accidentally dead victims are too few to allow for a meaningful decomposition by race.

Several interesting conclusions can be inferred from this table. Accidental victims represent, in principle, a sample of people from the areas where police and suspects10 make use of their weapons. They were not engaged in armed confrontations, they just happened to live or pass by places where police used their firearms. As can be seen, the proportion of whites and browns among the accidental victims is approximately 8% lower than in the general population, whereas the proportion of blacks is three times higher. In other words, police use their guns much more frequently in areas where black people live. This should come as no surprise since the research of ISER demonstrated that more than half the fatal victims of police interventions were killed in slums ('favelas'), areas where the blacks are highly overrepresented. According to estimations based on a sample carried out by the Official Bureau of the Census (IBGE) together with the census itself in 1991, whites comprise 64% of the population in the rest of the city (so called 'asphalt' areas) but only 38% in the favelas (Preteceille & Valladares, 1999).

10 Accidental victims may have been wounded or killed from police as well as from suspects' bullets. They are all together in the same category since it is extremely difficult, with the available information, to determine which side the bullet came from. 27 Analysing the wounded opponents, we see that the proportion of whites is smaller than in the accidental victims' population. Conversely, blacks and specially browns have a bigger share. Last, the percentage of whites among the dead opponents comes down dramatically to only 30%, while the proportion of browns and blacks continue to increase: 40% and 30% of the total, respectively. Hence, there is no doubt that civilian victims of police interventions are significantly darker than the general population. Whites, for instance, make up 60% of the inhabitants of the city, 51% of the accidental victims, 43% of the opponents wounded and only 30% of those opponents killed by police. On the other extreme, blacks represent only 8% of the citizens in the city, a quarter of the accidental victims, 27% of the opponents wounded and 30% of the opponents killed. Thus, blacks are more than three times more likely to be wounded or killed by police than would be expected by their share in the population. As for the convict population, it is clearly darker than the general population but still darker than the accidental victims. Its racial distribution is closer to that of the wounded opponents though with a moderately higher proportion of blacks and lower proportion of browns. The convict population includes proportionally more blacks but also more whites than the dead opponents, but if we add browns and blacks together, the latter are still darker than the convicts. This may provide further support for the existence of a police bias in the use of force, even when compared to the prison population, not just when contrasted with the general citizenry. Notwithstanding this, we should again underline the warning that the convict population probably has a different profile than the opponents since the former includes also petty criminals unlikely to be involved in shootings against police. Initially, two caveats could be argued against the conclusion that police show a racial bias against the blacks and browns in their use of lethal force. The first is that they use their weapons more often in areas where the white population is more scarce and, as a result, blacks and browns could be expected to be overrepresented among the victims of police interventions. Thus, the imbalance could be the result of a geographical bias or a difference in base rates rather than the result of a purposeful intention of the agents.

28 However, people against whom the police shoot intentionally are markedly less white and more brown and black than the accidental victims who presumably live in the same areas. Therefore, there is a strong claim that police use lethal force intentionally more against blacks and browns than against whites, since the comparison can now be established using a reference group of residents of similar areas. Second, returning to the problem of the various race classifications, it could be argued that census data originate from self-attribution and, accordingly, their comparability with the race categorisation in the police documents could be questioned. Once again, the intentional victims are darker than the accidental victims and in both cases the race categorisation is carried out by the same source and in the same context: the Police Incident Report. So this caveat could be relevant for the comparison with census data and, to a lesser degree, for the comparison between police and prison documents (which are not based on self-attribution but originate from different sources) but it cannot be applied to explain away the differences between the accidental and the intentional victims. Last but not least, it could be said that the likelihood of the different racial groups fighting the police might be different. Blacks and browns, being poorer, might engage more in criminal activities and end up confronting the police with a gun more often. This would explain why blacks and browns are more often intentional victims of police shootings than the general population and also than their fellow residents. The fact that blacks are more often sent to prison might seem to lend some support to this possibility, even though, as indicated above, the prisoners' racial composition may itself be the result of similar racial biases. Since the number of people of each racial group that become involved in crime is unknown, the only way to address this alternative hypothesis is to compare opponents that are wounded to those killed by police. In principle, they were both supposed to be confronting police with weapons before they were shot by the agents. Some were wounded and some were killed, but the difference should lie only in the outcome, not on the circumstances that gave rise to the shootings. Thus, the likelihood of being wounded versus being killed should be the same for all races, unless some factor that mediated a lethal result could be shown to correlate with race. In other words, the probability of dying among those individuals shot by police should be the same for all races even though the number of

29 people from a certain race who become involved in shootings against police may be different. In statistical terms, the marginals of the distributions (the number of people of each race that get shot by police) may be different, but that will not by itself produce a different joint distribution of race and lethal outcome. Thus, we might have more blacks who commit crimes and more blacks who shoot policemen. This will result in more blacks being shot by police but that still does not justify a higher chance of death for blacks who get shot by police in comparison to whites shot by police. Unless there is a bias on the part of the police that make them act more violently against some races than against others. In order to test the alternative hypothesis that blacks and browns may be killed more due to their higher involvement in violent crimes, we submit the data on opponents wounded and killed by race to a log-linear model. The model for the value of every cell is the following:

Ln (mij)=  + i + j + ij

th th where, mij is the value of the cell of the i row and the j column,  is the constant for all cells, i is the effect of row i (race: white, brown and black), j is the effect of column j (outcome: wounded or killed) and ij is the interaction effect of row i and column j (race and outcome). When we try to eliminate the interaction term from the model, the one that is interpreted as indicating that different races show different probabilities of dying after being shot, the likelihood ratio test yields a significant result (Likelihood Ratio Chi-square test= 12.66; degrees of freedom=2; p=0.0018). This indicates that the interaction term is significant and cannot be removed from the model. We can see the departures from an independence model if we see the residuals from the model with only the main effects, without the interaction. TABLE 7 Expected Number of Victims by Outcome and Race according to the hypothesis that race has no impact on the outcome, and actual number of Observed Victims Factor Expected Observed Count % Count %

OUTCOME Wounded RACE White 67,81 ( 6.69) 89.00 ( 8.78) RACE Brown 80,18 ( 7.91) 64.00 ( 6.31) 30 RACE Black 61,01 ( 6.02) 56.00 ( 5.52)

OUTCOME Dead RACE White 261,19 (25.76) 240.00 (23.67) RACE Brown 308,82 (30.46) 325.00 (32.05) RACE Black 234,99 (23.17) 240.00 (23.67)

We can see that, if race did not exert any influence on the final outcome, there would be less whites wounded and more whites dead. Likewise, there would be more browns and blacks wounded and less killed. Thus, as predicted, blacks and browns are more likely to end up killed than whites after being shot by police. Yet, before we can conclude the existence of bias, there is one more element that could explain these results other than racial bias. Police use their weapons more often in slum areas, favelas, but the lethal outcome of their intervention is also much more intense in the favelas (Cano, 1998). For instance, the lethality index (number of opponents killed for each opponent wounded) of police actions in favelas is 3.6, while the lethality index outside favelas is 1.6. This means that, in the slums, more than three out of four people who are shot by police end up dead, which is an extremely high level. In the other areas of town, it does not get to two out of three. Given that police are more lethal in favelas and there are more black and brown people in the favelas, this could all add up to a higher likelihood of being killed among the black and brown simply because they live in the areas where police tend to kill more. In other words, these results could be attributed to an artefact of the area of town rather than to a racial bias on the part of the police. In order to test this hypothesis, we added a new factor to the log-linear model: area of town (inside our outside favela). The model could be now expressed as such:

Ln (mij)=  + i + j + k+ ij + ik+ jk+ ijk

th th where mij is the value of the cell of the i row and the j column,  is the constant for all cells, i is the effect of row i (outcome: wounded or killed), j is the effect of column j (race: white, brown and black), k is the effect of the third dimension (area), ij is the interaction of outcome and race, ik the interaction of outcome and area, jk the interaction of race and area, and ijk the triple interaction (race, outcome and area).

31 The likelihood ratio test shows that the three-way interaction is non-significant and can be removed from the model (Likelihood Ratio Chi-square test= 0.476; degrees of freedom=2; p=0,788). Once we remove it from the model, we test the three 2-way interactions. The area x race interaction is also non-significant (Likelihood Ratio Chi- square test= 4.364; degrees of freedom=2; p=0,1128). This may be interpreted to mean that the race distribution of the victims is not significantly different inside and outside the favelas. Since we know that the race distribution of the population is in fact very different inside and outside the favelas, this lack of significance is actually indicating that police pick especially on blacks outside favelas so that the proportion of black victims outside (28%) comes close to the proportion of black victims inside favelas (31%). However, the race x outcome interaction is significant (Likelihood Ratio Chi-square test=12.66; degrees of freedom=2; p=0,0018) and so is the area x outcome interaction (Likelihood Ratio Chi- square test= 42.86; degrees of freedom=1; p<0,0001). The latter means that there is more likelihood of dying inside than outside the favelas. The former, which is the ultimate question in this test, can be interpreted as such: blacks and browns have a higher probability of being killed versus wounded as compared to whites. The following table decomposes the crosstabulation of race and outcome according to area. TABLE 8

Crosstabulation of Outcome by Race according to Area of Town

RACE AREA White Brown Black Total Outside Favela OUTCOME Wounded Count 63 43 39 145 % within 34,4% 24,0% 28,1% 28,9% RACE Dead Count 120 136 100 356 % within 65,6% 76,0% 71,9% 71,1% RACE Total Count 183 179 139 501 % within 100,0% 100,0% 100,0% 100,0% RACE Inside Favela OUTCOME Wounded Count 26 21 17 64 % within 17,8% 10,0% 10,8% 12,5% RACE Dead Count 120 189 140 449 % within 82,2% 90,0% 89,2% 87,5% RACE Total Count 146 210 157 513 % within 100,0% 100,0% 100,0% 100,0% RACE

32 Outside the favelas, 66% of the whites shots by police end up dying compared to 76% of the browns and 72% of the blacks. Inside the favelas, 82% of the whites shot die versus 90% of the browns and 89% of the blacks. In other words, when police shoot inside a favela it is much more probable that their targets will die and not just be wounded. But both inside and outside favela blacks and browns are more likely to die than whites. Whites die approximately 8% less than blacks and browns. It is noteworthy that browns fare even worse than blacks but this difference is small and non-significant11. One other way to realise the magnitude of this bias is to calculate the lethality index (number of opponents killed divided by the number of opponents wounded in these confrontations) by race and by area. In normal armed confrontations, both military and police, it is expected to find on average more people wounded than killed. This corresponds to a coefficient of value less than 1. On the other hand, coefficients significantly above 1 indicate an excessive use of lethal force revealing that it is more likely to be killed than wounded after being shot. TABLE 9 Lethality Index of Opponents in Police Interventions Area of Town Whites Brown Black

Outside Favela 1.9 3.2 2.6

Inside Favela 4.6 9.0 8.2

It is apparent that the lethal rates are far worse in the favelas, but whites fare better than non-whites both inside and outside. The lethality index for whites is half that of the other races inside the favela and around two thirds outside the favelas. Interestingly, browns are worst, even worse than blacks though this latter difference is not very marked. It may seem strange to observe such different coefficients with just 8% difference in the likelihood of dying. Yet when few people survive, a relatively small increase in percentage of survivals may be enough to double the lethality index.

11 Applying the model just to blacks and browns, the race by outcome interaction yields the following results: Likelihood Ratio Chi-square test= 0.705; degrees of freedom=1; p=0.401. 33 One last factor that could mediate the lethal outcome of these encounters is the armament presented by the opponents. If opponents of one racial group were more heavily armed than others, this could force police to shoot more repeatedly and with more firepower, thus increasing the possibility of killing these opponents. The data indicate that white opponents had an average of 1.42 weapons apprehended in the episodes when they were shot, brown opponents had an average of 1.77 weapons apprehended in their episodes and black opponents an average of 1.51 arms12. This difference is non-significant (F=2.78; d.f.=2 & 1001; p=0.063) and, in any case, does not go in any direction that could help explain the differential mortality of the various races. Likewise, the average number of shots that had been fired from the weapons confiscated in the episode is also similar for victims of the three races: 3.26 shots for the white victims, 3.06 for the brown victims; and 2.83 shots for the black victims. These differences are again non-significant (F=0.22; d.f.=2 & 1005; p=0.803). Last, the more powerful arm apprehended in the episode is usually a gun or a pistol rather than long or more destructive weapons (Cano, 1998).This is true for 70% of the white victims, 68% of brown victims and 67% of black victims). The difference between the racial groups is once again non-significant (Chi-square= 6.14; d.f.=6; p=0.407). Hence, there is no evidence in the data that some racial groups might have been more or better armed than others so as to explain the difference in the mortality rate of opponents of different races in terms of the need for the police to use more firepower against some groups than others.

This is very strong evidence in favour of a police racial bias in the use of lethal force against the black and brown populations. The results are particularly supportive of this thesis because they: a) compare the final harm (fatal or not) of people who are shot by police and who are supposedly in a similar situation; b) the racial classifications are all carried out by the same source: police themselves;

12 Actually, the unit of registry for weapons apprehended is the episode, not the opponent. Thus, all the victims of one episode will be reporting the same guns here. Also, one episode can obviously have victims of different races involved. So the data for each race group are not completely independent, but they are still valuable in order to have an idea of the situation. 34 c) the fact that the analysis is executed and holds true both inside and outside favela controls for the racial composition of the particular area (the racial composition may be different inside and outside the favela but the effects holds true in both cases) and also eliminates a particularly powerful rival hypothesis (that blacks get killed more because they often live in favelas, where people in general get killed more). The only limitation of these results is that they are restricted to a racial bias occurring among those targets of police fire, that is, once the police decided to open fire, revealing that police act more violently against the browns and the blacks. The evidence is not conclusive about a possible racial bias operating at the time when police take the decision of whether or not to fire. For that, we only have the comparison between population data and victim data, but in this case there might be other rival explanations as we have seen.

There is strong evidence that some of these deaths may have been the result of summary executions rather than of genuine armed confrontations (Cano, 1998). It is highly likely that the chances of escaping alive from an attempted execution are much lower than the chances of escaping alive from a real armed confrontation where both sides are shooting. Hence, one of the elements that could drive up the likelihood of dying among those shot by police is a higher proportion of executions over the total number of fatalities. So if blacks that are shot present a higher likelihood of dying, this may be the product of police picking especially on blacks to execute them. This would be one way in which a bias against the black race could be exerted. In order to explore whether the bias occurs because of a higher likelihood of blacks and browns being executed, we will cross the signs of executions with race. Only dead opponents for whom we have a Forensic Report will be considered for these analyses, a total of 658 people in all. The average number of entry wounds per corpse due to shots is the same for all three races: 4.33 for whites, 4.36 for browns and 4.32 for blacks. The average number of entry wounds on the back is also very similar for the three racial groups: 1.63 for whites, 1.61 for browns and 1.69 for blacks. Likewise, the average number of entry wounds on the

35 head is similar in all three groups: 1.18 for whites, 1.11 for browns and 1.04 for blacks. All these differences are non-significant13. As for the existence of injuries other than those produced by gunshot, another sign of the possibility of an execution, 34% of white, 35% of brown and 26% of black corpses present this kind of injuries. The difference, which seems to go in favour of blacks is, in any case, non-significant (Chi-square=5.06; d.f.=2; p=0.08). Out of the 40 corpses that presented shots at point-blank range, the strongest forensic sign of an execution, 16 were classified as white, 17 as brown and 7 as black. The figures are too low to enable final conclusions but there is no sign of a strong racial imbalance. In fact, it is the blacks who seem to present this problem less frequently (4% of all black corpses had this kind of shots versus 6% of all brown corpses and 8% of all white corpses). In short, we found no evidence among the fatal victims of signs of executions that clearly correlated with race. Therefore, we cannot assume that the racial bias in the use of lethal force against browns and blacks occurs because of a higher proportion of executions among these groups but as a result of a generally more intense use of firearms.

One other way in which the racial bias could manifest itself might be that blacks, after having been shot, could be taken to the hospital less often than whites, thus increasing the likelihood of death. However, the data show that 16.8% of the white victims were not taken to hospital, compared to 21.1% of brown victims and 16.9% of black victims. These differences are non-significant (Chi-square=2.91; d.f.=2; p=0.233) and, in any case, do not go in a direction that could clearly explain the differential mortality rate of the three racial groups, given that blacks and whites are taken to hospital in the same proportion. Thus, we have no specific definition of the mode in which the racial bias operates beyond a general conclusion that police act more violently and produce a more lethal result when shooting blacks and browns than when shooting whites.

The research of ISER demonstrated how the lethal outcome of police interventions increased after the introduction of so called 'bravery awards' to policemen involved in

13 The tests values for the three comparisons are, respectively: F=0.01; d.f.=2 & 644; p=0.990. F=0.07; d.f.=2 & 644; p=0.933. F=0.57; d.f.=2 & 645; p=0.567 . 36 shootouts and other related policies. We decided to carry out an analysis to see whether these new policies of the former government in Rio might have increased or otherwise affected the racial bias. In order to do this, we added the new factor 'administration' (previous administrations vs. the one lead by General Cerqueira, the mentor of these policies) to our log-linear model. The four-way and the three-way interactions as a whole are non-significant14. More specifically, the interaction of race by outcome by administration is not significant (Likelihood Ratio Chi-square test= 2.286; degrees of freedom=2; p=0.3188). Notwithstanding this, the interaction of administration by area is significant (Likelihood Ratio Chi-square test= 64.66; degrees of freedom=1; p<0.0001) and so is the interaction of administration by outcome (Likelihood Ratio Chi-square test= 12.77; degrees of freedom=1; p=0.0004). All this can be interpreted in the following way: the bravery awards and related policies had an significant impact of increasing the lethal outcome of police interventions and of increasing particularly the number of incidents in which police used firearms in the favelas (Cano, 1998). Yet it did not affect the racial bias which remained more or less constant over the period. Logically, the increased lethal nature of police actions, and the growing number of armed incidents in the favelas, resulted in a higher number of blacks being killed than before. However, the likelihood of the victims being killed rose uniformly for all racial groups leaving the racial bias intact. The fact that this bias was not apparently altered by the policies that promoted armed confrontation fits in with the notion that it is a structural problem rather than the result of contextual or specific events in the recent past.

Finally, we will test the consistency of the racial classifications of the police and of other sources, to see whether any process of general darkening or whitening of these victims by police may be at work. As a reminder, all the documents other than the Police Incident Report that refer to race were taken from the Autopsy Report Files in the Institute of Forensic Medicine and are, therefore, of a predominantly medical nature. We only have other information on race in the case of dead victims for whom the Autopsy Report could be found. TABLE 9

14 4-way interaction: Likelihood Ratio Chi-square test= 1.88; degrees of freedom=2; p=0.3900. 3-way interaction: Likelihood Ratio Chi-square test= 10.16; degrees of freedom=9; p=0.3308. 37 Number of Subjects with Information on Race extracted from each of the following documents

N Laudo Cadavérico 521 Termo Reconhecimento 496 Guia de Remoção 440 Boletim Infor. Médico-Hospitalar 165 Guia Inform. Hospital para IML 89 Guia de Comunicação de Óbito 50 Declaração Óbito Hospitalar 39 Minuta Declaração de Óbito 34 Solicitação IML para Hospital 13 Solicitação Hospital para Remoção 11

The documents from which race information was collected most often were the Autopsy Report ('Laudo Cadavérico'), the Corpse Recognition Form ('Termo de Reconhecimento') and the Corpse Removal Form ('Guia de Remoção de Cadáver'). The latter is a document issued in the police station while all the others are medical or technical (forensic) documents. Victims whose Autopsy Reports were found have an average of 2.6 documents mentioning 'colour'. Before we contrast the colour categorisations of both sources, the police and the Forensic Institute, we start by checking the equivalence of both documents in each of the sources. For example, both the Corpse Removal Form and the Incidence Report are police documents issued by the police districts. Of the total of 409 fatal victims that had race information in both of these documents15, 324 were classified in the same manner (79,2% of the total). The remaining 21% of mismatched cases are distributed in a symmetric fashion. As a result, the proportion of each racial group is very similar in both documents: 31,1% whites, 35,5% browns and 33,5% blacks in the Incident Reports, compared to 30,8% whites, 35,7% browns and 33,5% blacks in the Corpse Removal Forms. This implies that there seems to be no differential criteria for racial categorisations behind both documents, just random differences in individual cases. In other words, one could not speak of a relative whitening or darkening of the victims in one document with respect to the other.

15 Victims with a colour other than the three main ones (white, brown and black) were ignored for this analysis. 38 Next, we compare the colour categorisations of two documents that originate in the Forensic Institute: Autopsy Reports and Corpse Recognition Forms. Out of 281 cases with valid information, 200 (71,2%) are classified consistently. The remaining 19% are not categorised symmetrically. The Autopsy Reports tend to classify a higher proportion of people as brown, compared to the Corpse Recognition Forms. This occurs at both ends of the continuum: browns vs. blacks and browns vs. whites. As a result, the proportion of both whites and blacks is lower in the Autopsy Reports: 30,2% whites, 42% browns and 27,8% blacks in the Autopsy Reports, compared to 35,2% whites, 33,5% browns and 31,3% blacks in the Corpse Recognition Forms. In short, the agreement rate of racial classifications between documents originating from the same institutions varies between 70 and 80%, thus underlining the relatively fuzzy nature of these categorisations. When racial categories according to police (Incident Reports) and to the Forensic Institute (Autopsy Reports) are contrasted, the result is the following. TABLE 10

Crosstabulation of Colour of the Victim according to the Incident Report and the Autopsy Reporta

Colour according to the Autopsy Report White Brown Black Total Colour White Count 114 37 4 155 according % of Total 22,6% 7,3% ,8% 30,8% to the Brown Count 29 128 33 190 Incident % of Total Report 5,8% 25,4% 6,5% 37,7% Black Count 5 57 97 159 % of Total 1,0% 11,3% 19,2% 31,5% Total Count 148 222 134 504 % of Total 29,4% 44,0% 26,6% 100,0% a. Consistent classifications are marked with shading

The consistency rate is equal to 67,2% of the cases, lower than in São Paulo. The main difference is the higher proportion of blacks in the Incident Reports. This is due to the fact that there are 57 cases of browns in the Autopsy Report classified as blacks by the police, versus only 33 in the opposite situation. In other words, the police tend to consider black some people that the doctors consider as brown, thus revealing a lighter threshold between both groups in the eyes of the police.

39 The contrast of the racial categorisations of the Incident Reports with that of the other main forensic document, the Corpse Recognition Form, yielded the following outcome. TABLE 11

Crosstabulation of Colour of the Victim according to the Incident Report and the Corpse Recognition Forma

Colour according to the Corpse Recognition Form White Brown Black Total Colour White Count 82 11 4 97 according to % of 28,3% 3,8% 1,4% 33,4% the Incident Total Report Brown Count 16 72 12 100 % of 5,5% 24,8% 4,1% 34,5% Total Black Count 5 14 74 93 % of 1,7% 4,8% 25,5% 32,1% Total Total Count 103 97 90 290 % of 35,5% 33,4% 31,0% 100,0% Total a. Consistent classifications are marked with shading

The consistency rate is 78,6% of the total number of cases. The mismatches are more evenly distributed than in the previous crosstabulation, but there is still a slightly lower proportion of whites and a slightly higher proportion of blacks in the police. So far, the evidence both in Rio and São Paulo points at a mild darkening of the racial categorisation by the police as compared to the Forensic Institute, reflected in a wider proportion of blacks in the police data. However, the magnitude of this difference is small and its significance is further limited by the fact that there are occasional discrepancies of comparable intensity also between documents elaborated by a single source, as is the case of the Autopsy Report and the Corpse Identification Form. In short, there is an indication that the police might be perceiving its victims as darker than other institutions but it is hard to think of this process as completely accounting for the racial bias in the use of lethal force. Indeed, if we were to use race as coded by the Autopsy Report instead of by the Incident Report in the analyses, the proportion of blacks among the fatal victims would come down from 29% to 27%, which is still far above its share in the general population16. Hence, even though there might be a slight process of

16 A much more powerful test of this alternative hypothesis could be carried out by applying the log-linear models on fatal and non-fatal victims with racial categorizations defined by the Forensic Institute. However, this possibility is precluded by the fact that we only have forensic reports on fatal victims, but not on non-fatal 40 victims' blackening in the racial categorisation of the police, it cannot explain the racial imbalance in the victimisation rates, which should be attributed instead to a racial bias in police action.

7. Summary and Conclusions. Numerous pieces of research have confirmed that blacks and browns are worse off in Brazil. Also, a few studies have addressed the fact that blacks and browns are worse treated by the public security system in the country (police, judges, prisons, etc.). Yet in order to show bias it is necessary to show that members of some racial groups receive a worse treatment than members of other groups under the same circumstances. Unfortunately, bias is more likely in the cases where there is more discretion on the part of the public agent (e.g. police approach to citizens) but data, and hence research, are more abundant in the cases where proceedings are more regulated and better recorded (e.g. sentences). This article attempts to test the hypothesis that there is a racial bias in the use of lethal force by police, whereby they employ it more frequently or more intensely against some racial groups (blacks and browns) than against others. Racial bias could be exerted in two ways: a) granting a worse treatment to people perceived as black or brown; b) perceiving as 'darker', and therefore more likely to be categorised as black or brown, those who are socially stigmatised (criminals, etc.). Data referring to civilians killed by police were obtained for the cities of Rio de Janeiro and São Paulo. In São Paulo, records of four of the five Jury Tribunals were reviewed in search for civilians killed by police since 1996. The search was complemented with information from the Ouvidoria (police ombudsman). A total of 215 cases was gathered. The race distribution of these fatal victims was close to the race composition of the convict population and both were considerably darker than the general population. For instance, blacks make up 4% of the population, 15% of the convicts and 13% of the people killed by police. This evidence points at the existence of racial bias both in the penitenciaries and in police action, but with two caveats. The first is that race categorisations had been carried out by different sources (by the police or the prison agents ones. 41 and by the citizens themselves in the census data) and may therefore not be directly comparable. The second is that in order to demonstrate bias from these data we have to assume that all racial groups engage in crime and in armed confrontations with the police in the same degree. To the extent that there is a link between crime and poverty, it may be conceived that blacks and browns, being poorer and living in the marginal areas of town, might enter criminal activities more often. Data for Rio de Janeiro originate from a research by ISER which collected all cases of civilians wounded or killed by gunshots in police interventions in the city of Rio between January 1993 and July 1996. These data included 991 civilians killed (942 of which had been intentionally shot by police) and 726 civilians injured (417 of which had been intentionally shot by police). Accidental victims, which represent a sample of the people who live in areas where police make use of their weapons, are darker than the general population. This confirms that police make use of lethal force particularly in areas where the blacks and browns live, that is, in the slums (favelas). Wounded opponents are darker than accidental victims and dead opponents have the darkest profile of all, comparable to the convict population. Whites make up 60% of the population, 51% of the accidental victims, 40% of the convicts, 43% of the wounded opponents and just 30% of the dead opponents. Blacks, for their part, make up 8% of the population, 25% of the accidental victims, 33% of the convicts, 27% of the wounded opponents and 30% of the dead opponents. As a result, there is a disproportionally high number of blacks imprisoned and victims of police action. Exactly as in São Paulo, blacks are more than three times more likely to be shot dead by police than would be expected from their weight in the population. In order to conclude that there is indeed bias, we have to mention the two previous caveats: the difference in the racial categorisations; and the fact that members of some racial groups, being poorer, may engage in criminal activity more often. However, there is one more argument that might explain away these results. If police make use of lethal force particularly in favelas, where the proportion of blacks and browns is highest, members of these racial groups may be victims of police interventions more often not because police intentionally pick on them but merely due to the fact that they live in the areas where police shoot more frequently. In order to test these alternative hypotheses, we submitted the data

42 of opponents killed and wounded to a log-linear model with three factors: outcome (wounded or killed), race and area (inside and outside favela). Results indicate that the chance of being killed versus wounded is higher when the incident occurs in a favela. More importantly, the chance of being killed versus wounded is significantly superior (around 8% higher) for blacks and browns than for whites, and this result holds true both inside and outside a favela. This turns into lethality indexes (number of people killed divided by number of people wounded) which are 37% to 100% higher for the blacks and the browns as compared to whites. It is very difficult to interpret these data in any way other than a racial bias on the part of the police. The evidence is particularly solid since: a) racial categorisations are performed by the same source, police themselves, and in the same moment; b) the analysis controls for the area of town where the incident happened and hence discards the alternative hypothesis that the racial imbalance may be due to the different racial compositions of these areas; c) wounded and killed opponents were in principle in the same situation, confronting police with weapons, and differed only by their outcome. There were no differences in the armament that was apprehended from opponents of each race. This rejects the possibility that police might have had to shoot more intensely against blacks because they were better armed. Likewise, there was no difference in the proportion of victims that were taken to hospital by race, indicating that the differential mortality could not be explained due to lack of assistance after the wounds. More importantly, signs of summary executions, which were abundant, did not show a different pattern by race, so the evidence does not confirm the idea that policemen pick on blacks to execute them summarily. Hence, we can only conclude that police are more violent in their use of lethal force against blacks and browns than against whites. This is evidence of racial bias that would occur at the time when police decide to use their weapons. Yet it is not conclusive on the possibility of racial bias on the decision of whether or not to use their weapons. Both in Rio and in São Paulo the data suggest a mild darkening in the perception of these victims in the eyes of the policemen compared to the perception of the Forensic Institute, as revealed by the facts that policemen tend to classify more victims as black.

43 Nevertheless, this darkening is too small to explain the imbalance in the racial distribution of the victims, which should be attributed, for the most part, to a bias in police action.

44 8. References.

Adorno, Sérgio (1995) "Discriminação Racial e Justiça Criminal em São Paulo" Novos Estudos. CEBRAP. n. 43 November pg. 26-44

Barth, Fredrik (1969) "Introduction" in Barth, Fredrik (ed.) Ethnic Groups and Boundaries. Boston: Little, Brown and Company.

Caldeira, Teresa P.R. (1996) "Crime and Individual Rights: Reframing the Question of Violence in Latin America" in Jelin, Elizabeth & Hershberg, Eric (eds.) Constructing Democracy: Human Rights, Citizenship and Society in Latin America. Boulder, Colorado: Westview.

Cano, Ignacio (1998) The Use of Lethal Force by Police in Rio de Janeiro. Rio de Janeiro. ISER.

Cano, Ignacio (1999) Letalidade da Ação Policial no Rio de Janeiro: A Atuação da Justiça Militar. Rio de Janeiro. ISER.

de Carvalho, José Murilo; Pandolfi, Dulce; Carneiro, Leandro Piquet & Grynszpan, Mário (1998) Lei, Justiça e Cidadania. Cor, religião, acesso à informação e serviços públicos. Rio de Janeiro. CPDOC-FGV/ISER.

Chevigny, Paul (1991) "Police Deadly Force as Social Control: Jamaica, Brazil and Argentina." Série Dossiê NEV, n.2, p. 9, Núcleo de Estudos da Violência, USP, São Paulo.

Chevigny, Paul "Police Deadly Force as Social Control: Jamaica, Argentina and Brazil" Criminal Law Forum: An International Journal , n.1 (3) p.389-425, Rutgers University, NJ.

Costa-Ribeiro, Carlos Antônio (1995) Cor e Criminalidade. Estudo e Análise da Justiça no Rio de Janeiro (1900-1930). Rio de Janeiro. Editora UFRJ.

CPDOC-FGV/ISER (1997) Lei, Justiça e Cidadania. Direitos, vitimização e cultura política na região metropolitana do Rio de Janeiro. Rio de Janeiro: CPDOC-FGV/ISER

Cunha, Euclides da (1936) Os sertões, Campanha de Canudos. Rio de Janeiro, Livraria Francisco Alves.

Fernandes, Florestan (1965) A Integração do Negro na Sociedade de Classes. São Paulo: Dominus Editora.

Fajnzylber, Pablo; Lederman, Daniel & Loayza, Norman (1998) Determinants of Crime Rates in Latin America and the World. An empirical assessment. Washington, D.C.: The World Bank.

45 Freyre, Gilberto (1933) Casa Grande e Senzala. Rio de Janeiro: Maia & Schmidt.

Harris, Marvin (1964) Pattern of Race in the Americas. New York: Walker.

Hasenbalg, Carlos (1979) Discriminação e desigualdades raciais no Brasil. Rio de Janeiro. Graal.

Human Rights Watch (1997) Police Brutality in Urban Brazil. New York: Human Rights Watch/ Americas.

Human Rights Watch (1998) Behind Bars in Brazil. New York: Human Rights Watch.

Hungria, Nelson (1956) "A criminalidade dos homens de cor no Brasil" in Comentários ao Código Penal vol.III. Rio de Janeiro. Revista Forense.

Kahn, Túlio (1998) "Os Negros e a Polícia: Recuperando a Confiança Mútua" Boletim Informativo do Grupo de Pesquisa da Discriminação da USP, num. 1, year III. São Paulo: USP.

Kahn, Túlio (1999) "Justiça e Discriminação Racial" Boletim Informativo do Grupo de Pesquisa da Discriminação da USP, num. 1, year V. São Paulo: USP.

Lieberson, Stanley (1990) A Piece of the Pie: Blacks and White Immigrants Since 1880. Berkeley, California: University of California.

Mitchell, Michaels J. & Wood, Charles H. (in press) "Ironies of Citizenship: Skin Color, Police Brutality and the Challenge to Democracy in Brazil" Social Forces

Paixão, A.L. (1983) "Crime e criminosos em Belo Horizonte. 1932-1978" in Pinheiro, P.S. (ed.) Crime, Violência e Poder. São Paulo: Brasiliense.

Pinto, Regina Pahim (1996) "Classifying the Brazilian Population by Color: Underlying Problems" in Sá Barreto, Elba Siqueira de & Zibas, Dagmar M.L. (eds.) Brazilian Issues on Education, Gender and Race. São Paulo: The Carlos Chagas Foundation.

Preteceille, Edmond & Valladares, Licia (1999) "Favela, Favelas. Unidade ou diversidade da favela carioca. Seminário Impactos Metropolitanos da Reestruturação: desigualdades sociais e governança urbana"

Radelet, M. (1991) "Racial characteristics and the imposition of the death penalty." American Sociological Review, num. 46, pg. 918-927

Rodrigues, Nina (1994) As raças humanas e a responsabilidade penal no Brasil. Rio de Janeiro, Guanabara.

46 Sansone, Livio (1997) "The New Politics of Black Culture in Bahia, Brazil" in The Politics of Ethnic Consciousness pg. 277-309 New York: St. Martin's Press.

Silva, Jorge da (1988) Violência e Racismo no Rio de Janeiro. Rio de Janeiro. EDUFF. Comissão Justiça e Paz, 1994).

Silva, Nelson do Vale (1985) "Updating the cost of not being white in Brazil" in Fontaine, P.M. (ed.) Race, Class and Power in Brazil, chapter 4. Los Angeles: UCLA Center for Afro-American Studies.

Scalon, Maria Celi (1999) Mobilidade social no Brasil: padrões & tendências Rio de Janeiro: Editora Revan.

Telles, Edward (1999) Patterns and Constraints in Inconsistency between Racial Categorization and Identification in Brazil . Unpublished document.

Telles, Edward E. & Lim, Nelson (1998) "Does it matter who answers the race question? Racial Classification and income inequality in Brazil" Demography, vol. 35 n. 4 pg. 465- 474

United Nations Crime Prevention and Criminal Justice Division. (1997) United Nations International Study on Firearm Regulation. United Nations Office at Vienna.

Varela, Drauzio (1999) Estação Carandiru. São Paulo: Companhia das Letras.

Wade, Peter (1997) Race and Ethnicity in Latin America. Chicago: Pluto Press

Wood, Charles "Categorias Censitárias e Classificações Subjetivas de Raça no Brasil" em Lovell, Peggy A. Desigualdade Racial no Brasil Contemporâneo. Pg. 93-113. Belo Horizonte: Universidade Federal de Minas Gerais: Centro de Desenvolvimento e Planejamento Regional.

The World Bank. (1997) World Development Report 1997. The state in a changing world. World Bank-Oxford University Press.

47

Load more

  47

Copy Link