Which are the Causes of Criminality in Brazil? Matteo Francesco Ferroni 1 September 2014 1 Il presente documento è di esclusiva pertinenza del relativo autore ed è esclusivamente riservato per l uso espressamente consentito dall autore medesimo, senza il cui preventivo espresso consenso scritto non può essere ulteriormente distribuito, adattato, memorizzato ovvero riprodotto, in tutto o in parte e in qualsiasi forma e tecnica.
Index Page Abstract 1 1. Introduction 1 2. Crime in Brazil 4 3. Becker s model 10 4. Explanatory variables 13 5. Data 18 6. Econometric model of criminality 25 7. Conclusion 30 References 32 Appendix 34 1. Data collection 34
Abstract The objective of this study is to better understand the determinants of criminality rate in Brazil, more specifically the determinants of homicides. I based myself on Becker s model of criminal rational behavior. After selecting some economical and sociological variables, I run a cross sectional regression using the data of 2010 from 608 Brazilian municipalities. My main result suggests that inequality has an impact on the homicide rate while poverty does not. Furthermore, there is evidence proving that urbanization and unemployment are positively related to the homicides, whereas education is negatively related. On the other hand the age composition of the population is positively related only until a certain level.
1. Introduction Criminality is a problem that has repercussions in different fields. Firstly it is a social problem that can directly and indirectly affects the life of the population. In fact it has an impact not only on the quality and the expectation of our lives but also on decisions that we make. When we have to choose in which neighborhood to live, which school to send our children or even in which country to spend our holidays, we are affected by criminality since we are concerned by our safety. Furthermore criminality is also a political problem because to fight it the government has to participate actively reallocating public resources. Moreover, and most importantly, it is an economic problem as it is caused by some economic variables and it directly affects the development of the countries. I am presenting a research on the causes of criminality rates that will enable us to better understand and consequently address the problem. For my work I will consider Brazil. I have chosen Brazil because it is an interesting country to analyze since it is one of the biggest developing countries and criminality is one of the biggest concerns. Considering the homicide rate, Brazil has the highest one between the BRICs and it belongs to the third region, South America, with the highest homicide rate after only Central America and Southern Africa. Furthermore, given the dimension of Brazil, its administrative division allows us to have a big sample and the IBGE (Brazilian Institute of Geography and Statistics) make available a big amount of data useful for this study. Most of the empirical studies on crime until now have been conducted on the US or on other developed countries. One of the main reasons is that the data is easier to collect. But this doesn t mean that the data is non existent or not reliable enough in the developing countries, which are very important to study. In fact if we limit our studies only to the developed countries, this will not be enough to understand the reasons of criminality. We can t expect to be able to apply correctly these results to other categories of countries, such as the developing ones. When two countries have different demographic, political and social characteristics is obvious that the criminal activities won t be affected by the 2
same factors. Or better, even if the factors are the same, their impact on the criminality level will be different. This is why is important to not restrict our sample of countries and run the same kind of studies on other categories. Even so some studies have been made also on other countries. To go back to Brazil, one of the first studies that analyses the causes of the criminality in this country was made by De Araujo Jr. and Fajnzylber in 2001. Since then more studies have been made on Brazil. One of the most recent ones has been conducted by Sachsida et al. in 2010. These two works have in common the fact that their analysis are made by panel data regressions which use as time series variable the period between 1981 and 1995 and as cross sectional data the Brazilian states. In their researches they don t use the same variables to explain criminality but we can still compare some of their results. They both conclude that unemployment is positively correlated to the criminal level and that the public security spending, such as the police force, is negatively correlated. On the other hand their works don t fit with each other in determining the impact of inequality. In the work of De Araujo Jr. and Fajnzylber it seems that an increase in inequality will lead to a decrease of the criminal level, while in Sachsida s work it will lead to the opposite result. Furthermore De Araujo Jr. and Fajnzylber conclude that an increase on the income has a negative impact on criminality and Sachsida et al. conclude that the level of urbanization has a positive impact while the poverty level seems to have none. My work will provide some evidence in support of some of these results but in some cases it will go against them. Furthermore I will introduce and use different variables from them. Besides the variables that I will use, my work is different from the previous ones for two main reasons. The first one is that the data that I will use is more recent as it is taken from the Demographic Census of 2010 of the IBGE. Since any study has been done considering the data from the last twenty years, is important to check if the patterns are the same or if they have changed for some reason. There is no any particular reason to believe that this is the case but if we think how much our society has changed and how 3
much Brazil has growth in the last two decades this is a licit doubt. Furthermore is always important to verify the results of previous studies and it would be even better to do so analyzing a different sample so we will be able to affirm if the results are still correct for a different population. The second difference is that I will use a different sample. My study will be based on the Brazilians municipalities that in 2010 had more than 50,000 inhabitants. I have made this decision to avoid taking under consideration too small municipalities where the criminality level can be easily misrepresented, as I will explain better below. In this way I will use the data from 608 out of 5,565 municipalities. Even if this means that I am analyzing only 11% of the municipalities, I m considering 66% of the whole Brazilian population and 80% of the homicides that happened in that year. Furthermore the sample is big enough to give us results whit good statistic significance. The objective of this study is to better understand the main determining factors of criminality in Brazil, more specifically the determinants of homicides since it is considered as the most serious crime, it is the easiest to measure and is less underreported than other crimes. To do this I will base my work on the model of criminal rational behavior introduced by Gary S. Becker in 1968. I will then select some macroeconomic and sociological variables that the model suggests could have an impact on the criminality level. With the data from the Demographic Census of 2010 I will run a cross sectional regression for the 608 selected municipalities in order to identify the magnitude and the significance of the variables taken under consideration. 4
2. Crime in Brazil Is criminality a major problem in Brazil? To try to answer this question let s consider some data about it, especially regarding homicides. In 1980 Brazil s homicide rate was of 15 homicides per 100,000 people, in 1990 it has risen to 22.2, in 2000 to 26.8 and in 2009 it reached 27.2 homicides per 100,000 people. In thirty years Brazil s homicide rate almost doubled. This is without doubt an alarming growth. But are these rates particularly high or are they at the same level of other countries? Thanks to the studies of the UNODC (United Nations Office on Drugs and Crime), we can compare these rates to the other countries, and we find out that Brazil s homicide rate is high for the international standards. The worldwide rate in 2010 has been of 6.9 per 100,000 inhabitants, which means that Brazilian s rate is 3.9 times the global average. Even only considering Latin America it is still high. In fact it is 1.36 times the mean for Latin America. From the UN s intentional homicide rate, Brazil results to be the 16 th country out of 209 for intentional homicide rate and, excluding African and Caribbean countries, it is preceded only by Venezuela and Colombia. Stated that homicides are a major concern in Brazil, we can now try to analyze more in details how this problem appears. Homicides in Brazil affect primarily young men, like in almost every country. In 2009, 92 percent of homicides victims were men and 55 percent of them were aged between 15 and 29. If we also include the population between 30 and 39 years old, we cover 77 percent of the homicides victims. It is interesting to notice that the increase of the homicide rate between 1980 and today was almost entirely driven by a growth of male homicides. Focusing more on the youth problem we can see that in 1980, 23 percent of deaths among young men between 15 and 24 years old were caused by homicides and 36 percent were due to natural causes. Since then youth violence has risen in Brazil. In 2002 the percentage of deaths caused by homicide exceeded 50 percent and the deaths caused by natural causes dropped to 20 percent. 5
Comparing with other countries as Canada, where the youth homicide rate is only 1.7 per 100,000, or Chile (3.0) we can argue that a so high rate in Brazil is avoidable. From different studies it has emerged that the likelihood of delinquency between Brazilian youths is raised by 21 to 35 percent by some risk factors as substance abuse, having committed general offenses, having antisocial parents, low family socioeconomic status, poor school performance and aggressive behaviors. Intervening on some of these risk factors may reduce youth homicides. Let s now analyze the problem from a geographical point of view. To do so we can divide Brazil in macroregions, as shown in Figure 2, and easily see how the homicide rates have varied during the last twenty years. The development of the rates can be seen in Figure 1. In 1990 the South-East started with the country s highest regional homicide rate but today it has been surpassed by the Center-West and the North-East, which have the same rate. During this period the regional homicide rate rose in every region but especially in the North-East. We can also see that there has been a convergence between the regional rates and, from their trends, it seems a phenomenon that will continue in the future. Figure 1 - Homicide rate by macroregions 45 40 35 30 25 20 15 10 5 0 1 9 9 0 1 9 9 2 1 9 9 4 1 9 9 6 1 9 9 8 2 0 0 0 2 0 0 2 2 0 0 4 2 0 0 6 2 0 0 8 North North-East South-East South Center-West Brazil Source: Ministe rio da Sau de - Subsistema de Informac a o sobre Mortalidade 6
Figure 2 Map of Brazil s macroregions North It could be seen that there is quite a difference between the macroregions. Let s move now to a lower level and look at the Brazilian states. In 2009 state-level homicide rates ranged from 12.2 per 100,000 inhabitants in Piauí to 59.3 in Alagoas. States with rates above 40 included Espírito Santo (56.9), Pernambuco (45.0) and Pará (40.2). Other states with homicide rates below 20 included Minas Gerais (18.7), São Paulo (15.8) and Santa Catarina (13.4). To understand the magnitude of these numbers we just have to consider that the world average in 2010 was just 6.9 homicides per 100,000 inhabitants. Comparing the Brazilian states to the other countries of the world, even the state with the lower homicide rate is still 7
at the same level of Mozambique and Ecuador and Alagoas is second only to Honduras, the country with the highest homicide rate in the world (90.4 in 2012). We reach now the lowest administration level in Brazil: municipalities. Dividing the country in this way we are able to identify where the rates are higher. The municipality that had in 2011 the highest homicide rate was Simões Filho, a city in the state of Bahia with a population of 119,760, with 141.5 homicides per 100,000 inhabitants. To give an idea of the dimension, since 1995 any country in the world had a higher homicide rate, the most closer was El Salvador, which in 1995 had a rate of 139.1 per 100,000 inhabitants. From these data on the homicide rates, we can clearly say that criminality is a big problem that is widespread in the whole Brazil. But should we be concerned by this? How can we be affected by criminality? Most of the times we don t realize the consequence that the criminal activity can have on the whole population and on the whole country. From an economic point of view, the first effect of crime is the reallocation of resources among agents and especially from legal to illegal activities. This will lead to a reduction of the consumption level. Crime will also undermine the authority of the law and weaken the perception of the property right. These effects will increase the level of inefficiency and uncertainty in the economy leading to a reduction of domestic and foreign investment. The direct consequence will be a fall in the economic growth. There are two mains channels through which the effects of crime can spread. The first one is the process that I introduced above that will end reducing the growth rate of the per capita income. The second one is the damage of properties and the reduction of the population able to work, which will lead to a decrease of the output level. Both these results will have a negative impact on the growth potential of the economy in the short run. On the other hand, in the long run, while a reduction of the output level can be temporary and come back to its original level, a decrease of the growth rate of the per capita income will cause more permanent damage to the economy. It is important to notice that these effects are stronger in the developing countries since for their population it is harder to support these consequences. 8
After explaining the theory behind the criminal impact on an economy, I will now talk about more concrete causes and consequences. In Brazil crime is considered a major constraint to business growth. First of all there are the costs of all the goods and services used to prevent criminality, to treat its victims and to capture and punishing its perpetrators. They include expenditure on police, judiciary system, medical treatment, psychological counseling, housing and social services. These costs are called direct costs of criminality. In 2006 the World Bank estimated the direct costs of crime for Brazil to be between 3 and 5 percent of the GDP per year. To have an idea of what this means we just have to consider that the entire agricultural sector represent 3.5% of the Brazilian GDP. The World Bank also estimated the effects of crime on the Brazilian economy between 1991 and 1995. During these years the average homicide rate was 28.7 murders per 100,000 people. If the homicide rate had been just 10 percent lower, meaning 25.8 murders per 100,000 people, ceteris paribus the per capita income would have increased between 0.2 and 0.8 percent during the next five years. Using 1996 population, this would be equivalent to an increase of 2.2 billion of dollars in the total income of Brazilian population. But these costs are only a lower bound of the real impact that criminality has on the society, in reality the costs are much larger and the impact of crime on the economic growth is potentially more important. In fact Buvinic, Morrison and Shifter (1999) suggest that the effects of criminality can be divided in four categories: direct costs, non-monetary costs, economic and social multiplier effects. These last three costs and effects are more difficult to measure but they surely exist and their impact is probably larger than the impact of the direct costs on the economy. Health impacts that do not necessarily generate a demand for more medical services such as increased morbidity, increased mortality through homicide and suicide, abuse of alcohol and drugs and depressive disorders are part of the non-monetary costs. The economic multiplier effects are the ones that decrease the accumulation of human capital, lower the rates of labor market participation, reduce the productivity on the job, increase absenteeism, lower earnings, impact inter-generational productivity and lower the levels of saving and investment. Finally the social multiplier effects include the inter-generational transmission of violence, the erosion of social capital, a reduced quality of life and a lower 9
participation in democratic processes. Since Brazil is a country with an economy still in development these multiplier effects can have a more enduring impact on the Brazilian society and economy. Criminality is therefore a problem that does not only affect the single individuals and their families but also damage the entire population and society of a country and we are all involved. This leads us to the next question: what can we do to fight it? From the World Bank report, mentioned above, emerged some messages that allow us to better understand which is the current situation of crime in Brazil and which are the more efficient ways to deal with the problem. Since criminality is affected by factors that operate at multiple levels, there is no single solution, no matter how well planned and implemented, to reduce levels of crime and violence. For this reason it is important that the public policies for the prevention of crime and violence, whatever is the administration level, must be constructed on a solid base of empirical information about crime levels, trends and spatial distribution. In order to achieve this is important to reach a standardization of definitions, to have a better data collection and an integrated information system between the institutions. To address the problem it is important to implement both preventive actions and responses to crime since the two interventions are complementary. A more efficient and professional criminal justice system, as police forces, is essential to lower the levels of criminality but prevention activities are generally more cost-effective than control actions. Another important point is that effective institutions must exist for public safety strategies to work. The institutions need to coordinate their public safety initiatives at the federal, state and municipal level. States play a key role in the prevention of crime and violence not only by controlling the judicial and police apparatus but also by implementing many of the prevention programs. Even though the municipal level is an important entry-point for the prevention of crime and violence. In fact, since the smaller dimension, it is easier to collect information and implement effective prevention programs. But if the municipalities do not have the capacity to implement the programs, there will be the need for a state intervention along side the municipality authorities. 10
Having stated the need for more empirical information about crime and its causes, and the importance of the municipal level in the fight against crime thanks to its effectiveness, its is clear that a better understanding of this phenomenon on a municipal level is fundamental. To better analyze the problem it is important to be able to simplify the reality in a way that allows us to discern the different causes and their respective impact. This is why to deal with my work I will use an economic model that I m going to introduce in the next paragraph. 3. Becker s model In order to understand the causes of the crimes we usually rely on criminology, an interdisciplinary field in the behavioral sciences, which has been studied during centuries by sociologists, psychologists, psychiatrists and social anthropologists. Only recently this field was approached from an economic point of view. Gary S. Becker extended the domain of microeconomic analysis to a wide range of human behavior and interaction, including nonmarket behavior, such as criminality. This is an important innovation as thanks to his work we are able to analyze someone s propensity to commit a crime not from a psychological neither sociological point of view but analyzing his economic environment. In 1968 Becker introduced a model of rational criminal behavior based on the comparison between how much a person can earn on the labor market and the expected return of committing a crime. In Becker s opinion a person commits an offense if the expected utility to him exceeds the utility he could get by using his time and other resources at other activities (Becker, 1968). This means that when a person chooses between the legal and the illegal sector of the economy, he considers the benefits and the costs (which can be economic but also psychological) of participating in the two kinds of activity. 11
In this work I will use a simplified model based on Fajnzylber, Lederman and Loayza (2002). This model will help us to organize the ideas furnished by Becker s model in an inequation and motivate the explanatory variables that I will use in the following sections. Firstly I will introduce a model explaining criminal behavior from the perspective of the individual and then aggregate it to the population level. I will assume three basic hypotheses regarding the agents to work with this model. The first one is that the agents are rational and able to calculate the benefits of committing a crime. The second one is that the agents are risk-neutral. The last one is that the agents have moral values. With these assumptions we have that: where Where is the expected utility of committing the crime, is the monetary value of the crime, C is the cost of organizing and executing the crime, Oc is the opportunity cost of the agent and M his moral values. E(p) is the expected monetary value for the agent of the punishment, Pr(p) is its probability and p the monetary value of the punishment. This means that the agent commits a crime only if his expected utility of the crime, given by the monetary value of the crime minus the cost of doing it, the opportunity cost and the expected monetary value of the punishment (equal to the product of the monetary value of the punishment and its probability), is bigger than his moral values. The inequation can be rewritten as: It shows that someone will commit a crime when the monetary value of the crime is bigger than the sum of the moral values of the subject, which can be influenced by the education that he received, the cost of organizing and executing a crime, the opportunity cost of the subject, one of its possible determiner is if he has a good job, and the expected monetary 12
value of the punishment, which is for example affected by the number of policeman in the population. As we can see this is a microeconomic model that predicts the behavior of a single individual taking under consideration variables that vary from person to person and for this reason they are unobservable for the whole population. Nevertheless this model can still be used to predict the behavior of the whole population analyzing some macroeconomic variables that have an impact on the coefficients, like the ones that I have talked above. For example we are unable to calculate the opportunity cost of committing a crime for each individual and not even have a measure of the average opportunity cost for the whole population. But we know that the opportunity cost is affected by other variables such as education and the education level of the population can be easily calculated. In the next paragraph I will present the macroeconomic variables that I will take under account and explain how they can have an impact on the coefficients of the model and therefore affect the decision of committing a crime and, in other words, have an effect on the criminality level. 4. Explanatory variables To choose the possible determinant causes of criminality I will base myself on the recent literature (de Araújo Junior and Fajnzylber (2001), Aguiar de Oliveira (2005), Dills et al. (2008), Scorzafave and Soares (2009), Yearwood and Koinis (2009), Sachsida et al. (2010)) and on the simplified model above. The determiners are variables concerning economic, social and demographic conditions of the units studied and they are education, urbanization, age, unemployment, average wage, poverty and inequality. 13
As saw above the economic model introduced by Becker explains that the decision of taking part in an illegal activity is positive correlated with the monetary value of the crime and negative correlated to the opportunity cost, the punishment and the moral cost, and all these variables have some kind of impact on the reasons behind the decision of committing a crime. The level of the education in a population is an essential element for the prosperity of a country and as we can imagine it also has effects on crime. First of all a higher education increases the opportunity costs of committing a crime because we expect that someone with an high level of education can have an easier access to a profitable job. For this reason when a person is deciding whether to commit a crime or not it is more plausible that he will start an illegal activity if his alternative is a non-profitable job caused by a lower level of education. Education has an impact also on the moral values of the individuals. We can easily imagine that a boy who spent his youth in a school will have acquired more solid moral values than someone who sadly had to grow in the streets. These two effects caused by education have a negative impact on criminality but there is still one thing to consider. We can think that a person can employ what he has learned in school to achieve an illegal objective. This can be interpreted in the model as a reduction of the costs of organizing a crime. For this reason education can also have a positive impact on criminality. Nevertheless not every person uses the knowledge acquired in school for a bad cause whereas every one will be able to have a more profitable job thanks to a higher education. Furthermore higher rates of education have positive externalities on the moral values of the whole population and everyone will be discouraged to commit a crime. For these reasons we expect that the first two effects are stronger than the third one and this means that a population with a higher education rate is expected to have a lower level of criminality. We are used to think that the majority of the crimes happen in cities, especially in metropolises, rather than in rural areas. This belief has been confirmed by Aguiar de 14
Oliveira (2005), who demonstrated that the city size is positively correlated with the criminality level. The fact of living in an urban area can have effects on the cost of organizing a crime. In fact living in a big agglomeration allows you to easily enter in contact with other criminals, to find the equipment needed (as fire-arms) and it is also easier to find the victim of the crime. For these reasons the effect of urbanization on criminality is expected to be positive, in other words in a population who lives in a more urbanized area the criminality will be higher. Since the costs of organizing the crime are lower, according to Becker s model a person to commit a crime will now need, ceteris paribus, a lower profit to decide to start an illegal activity and this will increment the number of crimes committed. We can also consider that the fact of living in close contact with other people can amplify the effects that inequality has on the population, but I will come back to this later. The age of the population can affect the criminality level as well. As I will explain below, the majority of the crimes are committed by a certain range of the population, the one that is in an age when it is easier to commit a crime. Generally we expect that is easier for someone in his twenties to commit a crime than for someone who is over sixty or some kid who is not even ten years old. Not only a young man has a better physical condition to commit a crime but he also is psychologically more inclined to this kind of activity and he is often in contact with other people of the same age with the same characteristics. Connecting it to the model, we can say that the fact of being in a criminal age can be seen as a reduction of the costs of committing a crime and also the probability of getting caught. This means that, as explained for the variable urbanization, ceteris paribus, the people will now need a lower monetary value of the crime to decide to commit it. Therefore we expect that when a big percentage of the population is in the criminal age the criminality level will be higher. The unemployment has a strong impact on the opportunity cost of the individuals. In fact, if he is unemployed, he has a simple decision to make: either he commits an illegal activity or he stays unemployed. We can easily understand that the expected monetary value of the first action is higher than the second one. For this reason when we have a population 15
with a high rate of unemployment it is more plausible that we will have a bigger criminal activity. But also in this case the unemployment rate can have an effect in the other direction. If the unemployment is so high it can have an impact on the average income of the population and we can interpret this as a reduction of the expected monetary value of committing a crime. Without knowing which one of these two effects is stronger we will expect that the unemployment will have an ambiguous effect on the criminality. Poverty is frequently associated with criminality and in the literature it is often used as an explanatory variable. But, using Becker s model, we can reach another conclusion. The principal effect of the poverty is a reduction of the opportunity cost. In fact poor people has not a lot to loose and it is easier for them to decide to commit a crime (we can see it also as a reduction of the moral values caused by desperation and necessity). This implies a positive correlation between poverty and criminality but there is also another effect. The other effect, that is often neglected, is the impact of the poverty on the whole population. If the entire population is poor, a criminal will not have an incentive to undertake an illegal activity because there will be a large number of people in his same economic situation. This can be seen as a decrease of the monetary value of the crime and for this the criminality is expected to decrease. Combining the positive and the negative effect that poverty has on criminality we find that it has an ambiguous effect and it is not excluded that it could even have no impact. The same reasoning can be made for the average wage of population. If the average wage of the population is high that means that the opportunity cost of committing a crime will be high as well and this will have a negative impact on criminality. At the same time it can also have a positive effect because, for the part of the population who has a wage much lower than the average one, will mean an increase in the monetary value of the crime. Differently from the variable poverty, where the attention is put on the dimension of the fraction of the population that has a low income, for the variable wage we consider the actual value of the average wage of the population. Even if using two different methods, we are analyzing the same situation and therefore is not surprising that we reach the same 16
conclusion as we did for the poverty level: the average wage of the population has an ambiguous effect on the crime rate. We can notice that there is a potential collinearity problem here since the variable wage and poverty are very related. I will address this problem later on but it is important that this problem could be present also between other variables. The last variable that I will take in consideration is the inequality. Differently from poverty, inequality is the result of the economic situation of the whole population. This means that we have an unequal population when the difference between the income of the rich and the poor is elevated. For this reason inequality has a double effect on the coefficients of Becker s model: an elevated income for the richer part of the population means a high monetary value of the crime and a low income for the poorer part of the population can be interpreted, as we did with poverty, as a decrease of the opportunity cost. These two effects go on the same direction and for this reason we can be sure that inequality has a positive impact on crime. It is interesting to notice that, for the same causes, we can say that in case of an economic growth that affects the whole population equally, the criminality level will rest unchanged since the increase of the monetary value of the crime will be compensated by an increase of the opportunity cost of the same size. But we know that this is an unlikely hypothesis and that in reality the higher-ups classes, in absence of a strong government intervention, will benefit more from the economic growth than the lower classes. This implies that the inequality will rise and we expect that the criminality will increase as well. Another consideration that can be made, which goes in the same direction as the other effects discussed above, is that the inequality can affect the moral values of the population. If there is a diffuse feeling of unfairness caused by an accentuated inequality, the people will not fell guilty for committing a crime, indeed they will fell justified and this can lead to even higher criminality rates. These are the explanatory variables that I will use in my work. As we can see I have introduced at least one variable for each coefficient of the Becker s model except for the expected monetary value of the punishment. 17
The reason why I have not consider any variable for the monetary value of the punishment is because I should have analyzed criminal law of each municipality and assign to each punishment a monetary value and it is not the objective of this work. We can consider it has an omitted variable but I am not expecting strong consequences on the validity of the work because the law between each municipality it is not very different therefore we can normalize it. I have neither considered any variable for the probability of the punishment due to the lack of data on the municipality level. I could have used as variable the percentage of police per person or the expenditures on law enforcement but neither of the two data was available for every municipality. It is important to notice that the variable law enforcement is affected by the problem of reverse causality, which means that the independent variable is potentially caused by the dependent variable. In fact when we increase the law enforcement (and the probability of receiving the punishment is therefore higher) we should observe a reduction of the criminality, but it may be the case that we increase the level of law enforcement because we have a high level of criminality. For this reason we may find a positive correlation between law enforcement and criminality even if we are expecting the opposite effect and this means that the effect of this variable is ambiguous and omitting it is less problematic than excluding other ones. We have to keep in mind that some of these variables are correlated with each other. For example education surely has an impact on unemployment and poverty, urbanization on education, the average wage and poverty are the result of the same data, and so on. This could be a problem since if the correlation is too high we will face the problem of multicollinearity. This means that two or more independent variables are highly correlated and one can be predicted from the others, leading to invalid results about the individual predictors. The analysis of the correlation between the variables will be left for a further phase of the work. 18
5. Data My work will be based on a cross-sectional regression between the data of Brazilian municipalities in 2010. My main source of data will be the Demographic Census of 2010 of the IBGE (Instituto Brasileiro de Geografia e Estatística) but I will also use the data collected by DATASUS (Ministry of Health). The data set that I will use to analyze the criminality in Brazil is therefore composed by Brazilian municipalities. Brazil has 5,570 municipalities (in 2010 they were 5,565) with an average population of 34,361 inhabitants. Crime rate is a variable that can easily change from a year to another especially when considering a small population. In fact it is sufficient a few homicides more than the normal average to falsify the crime rate and this distortion is stronger the smaller the population is. To avoid this problem my sample will be made only by the municipalities of a certain dimension, specifically the ones that in 2010 counted a population of more than 50,000 residents. In this way I will use the data collected from 608 municipalities across all the 27 Brazilian states. As I am analyzing the causes of violent crime in particular, the proxy that I will use for the variable criminality is the homicide rate per 100,000 person. Homicide rate is often used as an index of serious crime and violence and, like every approach, it has its limitations and its advantages. One limitation is that considering only homicides we don t include all the other forms of violence such as assaults, sexual assaults, robberies and abductions. To have an idea of the dimension of the other crimes we can see the results of a survey conducted by ILANUD (United Nations Latin American Institute for the Prevention of Crime and the Treatment of Offenders) in 2001 for Rio de Janeiro, São Paulo, Recife and Vitória. In 2010 the average of the homicide rates of these four cities was 42 per 100,000 person, which correspond to 0,042%. The victimization rate for sexual assaults was 1,4%, for assaults 2,5% and for robberies 5,5%. Not including all these other crimes undoubtedly affects the evaluation of the level of violent crime. For example, in a country where the gun control is higher, the homicide rate 19
will probably be lower than in a similar country but this doesn t mean that the level of all the other types of violent violence will be lower as well. On the other hand homicide is considered as the most serious crime and for this reason it makes sense to use it as the only determinant of violent crime. Most importantly as it is the easiest to measure, we can easily evaluate the homicide rate but we can t say the same thing for crimes such robberies (should we consider all kind of robberies the same or differentiate them in function of the value of the goods that have been stolen?) and is less underreported than other crimes, we just have to think at how many sexual assaults are not reported. The Brazilian Ministry of Health furnishes the number of deaths causes by homicide for every municipality and state. With this data is easy to build the homicide rate per 100,000 person for each municipality. As we can see in Figure 1, the average rate of homicides in the all country has slightly increased in the last twenty years in a constant way. Also the variation in the five macroregions, divided as shown in Figure 2, has been essentially linear. On the other hand, if we analyze each state individually, we can notice that in some states the homicide rate had noteworthy fluctuations. This means that in some cases the variation of the violence in one state could be due, among other things, to the reallocation of the criminals between the states. We can take the example of Rio de Janeiro and Espírito Santo, two bordering states. As we can see in Figure 3, from 2006 the homicide rate in Rio de Janeiro started to fall and at the same time the homicide rate of Espírito Santo started to increase of almost the same amount. This fact could be seen as a reallocation of the criminal activity. The decrease of the criminality in Rio de Janeiro is the result of a program started in 2007 by the Brazilian government with the aim of reducing the criminality and violence in the favelas, the slums, of Rio and other big cities. From 2008 a new military police force, called Unidades de Polícia Pacificadora, started to be deployed in the favelas of Rio de Janeiro to keep with the fight against criminality. Since Espírito Santo doesn t have big cities with favelas, this program has not been implemented with the same intensity and this has led to weaker law 20
enforcement in Espírito Santo in comparison to Rio de Janeiro. This could have driven some of the criminals to move to the state where the law enforcement is weaker. This is also a good example of the problem of reverse causality that affects the variable law enforcement that I presented in Section 4. In fact even if in the long run, lets say in 2009, a higher level of law enforcement is related to a low level of criminality as we can see in Rio de Janeiro, in the beginning (2006-2007) the homicide rate in Rio de Janeiro was slightly lower than in Espírito Santo, but the law enforcement level was much higher suggesting that it is not true that an higher level of law enforcement lead to low criminality. Figure 3 - Homicide rate of Rio de Janeiro and Espírito Santo 65 60 55 50 45 40 35 30 25 20 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 Espírito Santo Rio de Janeiro Source: Ministe rio da Sau de - Subsistema de Informac a o sobre Mortalidade From this example we can see that in the short run the criminality level could be more affected by external events, such as new policies, than by the variables that I have introduced. Therefore we have to be careful with the homicide rate because, beyond being very variable, it can be distorted in the short run by some event that are very different between them and much harder to identify. A solution to avoid this problem would be use a panel data with different years in order to remove the short run distortion. As it is not possible due to the low frequency of the census made in Brazil, I will try to minimize this problem using the average number of homicides that happens between 2009 and 2011. In this way I will resolve in part the possible problem of having a number of homicides in 21
2010 higher or lower than usual, problem that could affect the results of the regression. Even if they will be presented in the next section, I will quickly compare the results of the regressions of the single years with the regression of the three years average. The coefficients are just slightly different, meaning that the results of the work wouldn t have been different if I would have considered only one year. But the regression with the highest R 2 and adjusted R 2 and the more significant coefficients is the one done with the three years average. This justifies my methodology. The variable education will be based on the ratio of the population which is or has been to school to the total population (the sum between the population that has never been to school and the population that has been to school). Been to school in this case is someone that has been enrolled at school during a certain period of his life. This could mean that even someone that has been to school only one year will be included in the educated population. This will not generate the optimal value of the variable but is the best thing that could be done with the data taken from the Demographic Census of 2010. For the urbanization rate I could choose between the ratio of the urban domiciles to the total number of domiciles or the urban population to the total population. I have opted for the second one because using the domiciles can create distorted data, as we know that the rural domiciles count a larger number of people for domicile. This data is also taken from the Demographic Census of 2010. Thanks to the data collected by DATASUS I was able to divide the homicides victim by age group as it is possible to see in Figure 4. I will assume that the majority of the homicides are connected to the criminality. Since I am interested in the age of the ones who commit the crime I will also have to assume that the victim and the perpetrator share the same age group. This is because it is the age of the perpetrator and not the age of the victim who affect its propensity to commit a crime. From this I can find the range of age where most of the homicides happen. I will choose all the groups that have their average homicide rate, calculated with the data from 2000 to 2009, above the average national 22
homicide rate calculated with the data from the same period. For this reason the range that I will take under consideration is between 15 and 39 years. Thanks to this conclusion I calculated the ratio of the population between 15 and 39 to the whole population using the data from the Demographic Census of 2010. This ratio will be use as a proxy of the variable age of population. 70 Figure 4 - Average homicide rate from 2000 to 2009 divided by age group 60 50 40 30 20 10 0 0 to 9 10 to 14 15 to 19 20 to 24 25 to 29 30 to 39 40 to 49 50 to 59 more than 60 Source: DATASUS The unemployment rate for every municipality is easily calculated making the ratio between the population over 10 years old that had an occupation during the week when the census was made and the whole economically active population. To calculate the average wage I made a weighted wage between the average wage of the economically active population and the average wage of the economically inactive population. This data concerns only the population over 10 years old. For the variable poverty I will use the ratio of the population which income is under half of the minimal wage (510 R$ in 2010) to the total population of the municipality. The Demographic Census of 2010 furnishes all this data. I excluded the part of the population who doesn t receive an income in calculating the poverty index, because it represents a huge part of the population since the work off the books is very common in Brazil but this 23
doesn t necessarily mean that their income is low. However the ratio will be calculated to the total population, so also considering the population under 10 years and without an income. Using the average exchange rate of 2010 between the Real and the Dollar, receiving less than 255R$ per month means receiving less than 5$ per day. Even if this amount is over the poverty line, I will use it as a proxy for the poverty level of the municipalities. The Gini coefficient is the more current measure of income distribution of a nation s residents and it can easily give an idea of the level of inequality among a population. Since in my work I need to understand the impact that inequality, meaning the difference of income, has on the opportunity cost and on the monetary value of the crime, I don t consider the Gini coefficient as the best instrument to use in this situation for two main reasons. Firstly none of the coefficients of the Becker s model is directly affected by the Gini coefficient and this means that we can t clearly see the effect that inequality has on criminality. Secondly the Gini coefficient can sometimes give us a wrong interpretation of the level of inequality, especially when we want to compare it between two different populations. This is because the area between the line of equality and the Lorenz curve, even if it is shaped in a different way, could be the same. This means that two populations with the same Gini coefficient can have in reality a different level of inequality. To avoid these two problems, my solution is to build a new measure of the inequality that has not been use in any of the other works. For each municipality, I divided the respective populations (the data available concern only the persons older than 10 years with an income) in quartiles. Then I calculated the weighted average income of the first and fourth quartile. With these two results, instead of calculating the difference between the weighted average income of the richest quartile and of the poorest, I considered their ratio since it shows how many times the income of the richest part of the population is higher than the income of the poorest one. This is a better proxy for inequality since it gives us a clear level of inequality, which can t be distorted, and we can see the direct effects that the inequality has on the coefficient of the Becker s model. In fact an increase of the income of 24
the highest quartile means an increase of the monetary value of the crime and a diminution of the income of the lowest quartile means a decrease of the opportunity cost. In Table 1 it is possible to see a summary of the variable introduced and in Table 2 some statistics. Table 1 Explanatory variables for homicides (criminality) Terminology Explanatory variables Expected sign Education Population been to School / Total Population Negative Urbanization Urban Population / Total Population Positive Age 15 to 39 Population / Total Population Positive Unemployment Pop. with an Occupation / Ec. Active Population Not defined Average wage Average wage Not defined Poverty Income < ½ Minimal Wage / Total Population Not defined Inequality Income Richer Quartile / Income Poorer Quartile Positive Table 2 Statistics of the variables Variable Mean Median Minimum Maximum Homicides* 28.721 23.441 0 150.787 Education 90.078% 91.190% 68.312% 97.863% Urbanization 86.511% 93.201% 16.902% 100% Age 42.950% 42.759% 34.735% 56.884% Unemployment 8.169% 7.724% 1.900% 28.090% Average wage 1,060 R$ 1,054 R$ 398 R$ 2,925 R$ 25