FACULTY OF ECONOMICS AND BUSINESS
A Neighborhood Level Assessment of the Causal Relation Between
Income Inequality and Crime: A Case Study of Amsterdam
By Anna Wildeboer Bachelor Thesis Economics Supervisor: Prof. Dr. Erik Plug
27 February 2015
Abstract: This paper examines the relationship between income inequality and criminal activity for a sample of 75 neighborhoods in Amsterdam, for the period 2008-2011. As opposed to greater aggregation units, it considers neighborhoods as a more appropriate level of analysis as they constitute more meaningful frames of reference that are necessary for the experience of economic inequality. The economic theory of crime and the relative deprivation theory identify the mechanisms trough which income inequality aggravates crime. For the empirical analysis a panel-data based fixed effect OLS methodology is used that controls for other economic determinants of crime, unobserved neighborhood specific characteristics and trends over time. The analysis considers different income inequality measures and different samples of neighborhoods. The empirical findings indicate that overall crime rates are positively affected by income inequality. The findings related to the specific crime types are based on a different crime registration method and indicate no effect from income inequality on violent crime and robbery, but a positive effect on burglary rates.
TABLE OF CONTENT
Abstract i
Table of Content iii
Acknowledgements iv
1. Introduction 1
2. Literature Review 2
2.1 Background 2
2.2 Theoretical Foundations 4
2.2.1 The Economic Theory of Crime 4
2.2.2 The Relative Deprivation Theory 5
2.3 Inconsistencies, Types of Crimes and Aggregate Level of Analysis 7
2.4 Identifying Other Relevant Variables Affecting Crime 9
3. Data and Empirical Method 10
3.1 Dependent Variables 11
3.2 Independent Variables 12
3.3 Control Variables 12
3.4 Neighborhood Measurement 13
3.5 Sample Selection 13
3.6 Empirical Strategy and Model 14
4. Results 16
4.1 Basic Results 16
4.2 Alternative Measures of Inequality 19
4.3 Subsamples 20
5. Discussion 22
6. Conclusion 25
Appendices 27
Appendix A: Computation of the Inequality Measures 27
Appendix B: Tables and Figures 29
ACKNOWLEDGEMENTS
Hereby I would like to thank Hans van Wijk, working at the Economics Department of the municipality of Amsterdam, for our brainstorm sessions about a suitable topic concerning Amsterdam. I would also like to thank Cor Hylkema from Bureau O+S Amsterdam for his cooperation with regard to my search for data. I am grateful for the suggestions and feedback from prof. dr. Erik Plug that helped to improve the empirical section of my thesis. Lastly, I would like to express my gratitude to Sarah and Annemijn for their support and feedback.
1.INTRODUCTION
There is an ongoing international debate about economic inequality – the unequal distribution of economic resources – on all levels of society: researchers, governments, policy makers and the public are concerned with the causes and effects of this phenomenon. In his book Capital in the Twenty-First Century, Thomas Piketty (2014) argues that increasing inequality stemming from the returns on capital is an ongoing trend if it is not addressed by economic policy. Increasing rates of income inequality have a range of detrimental social consequences and reduce the social cohesion. According to Wilkinson and Pickett (2010) these negative effects are felt by all members of the society and not only by the poor. One of the argued effects of rising income inequality is increasing rates of violence and crime in society.
The concern with crime is well justified by its harmful effect on economic activity and its reducing effect on the quality of life of people who have to deal with a reduced sense of personal and proprietary security (Fajnzylber, Lederman, & Loayza, 2002a, p. 1324). The positive relation between inequality and crime rates is supported by the leading theories on crime of both the economic paradigm as well as of the sociological paradigm and these theories are rather complementary than exclusive (Kelly, 2000, p. 531). According to Becker’s (1968) economic theory of crime, engagement in criminal activity results from a rational decision based on a cost-benefit analysis. A more unequal income distribution - as opposed to an equal one – ceteris paribus would lead to lower opportunity costs and higher potential benefits for the poor. Under the assumption that economic agents are rational this results in an increase in crime rates. The relative deprivation theory (Runciman, 1966; Stack, 1984) developed in sociology and criminology states that income inequality results in social tensions and crime as the poor feel dispossessed when compared with the wealthier and therefore seek compensation and satisfaction by all means. Higher inequality would lead to a greater feeling of disadvantage and unfairness and thus results in higher crime rates
(Fajnzylber et al., 2002b, p. 2).
Researchers who have studied this relationship have conducted most empirical research with a focus on cross-sectional data on a national, county or city level. Outcomes of these studies have been ambiguous, although the main consensus is that relative higher income inequality is associated with higher levels of crime and violence (Hsieh & Pugh, 1993).
There is however a gap in the existing literature that covers the income inequality-crime relation on local level rather than on a higher aggregate level. By employing data on
neighborhood level, this thesis is able to include much more detailed information and nuances than cross-national and cross-county studies. It is highly interesting to consider the
consequences of increasing income inequality on a neighborhood level rather than on a national or regional level because cities can supplement national policies with effective local policies to decrease the incidence of crime and violence. This thesis is therefore concerned with the following research question: can support be found for the claim that income inequality has a causal positive effect when neighborhood level data is used? To study this question, Amsterdam is taken as a case study.
In order to answer this research question literature is studied and an empirical analysis is performed. The revised literature identifies the underlying theories and the mechanisms. Furthermore, the empirical research tests the causal effect of income inequality on crime rates. To do so, other economic control variables are taken into account. The properties of panel data are utilized and a fixed effect regression is employed to control for unobserved neighborhood specific characteristics and trends over time. The outcomes indicate that increased income inequality results in more criminal activity by the residents who experience this increased inequality. In addition, neighborhoods with higher income inequality are more prone to property crimes but not to violence.
The rest of this thesis proceeds as follows. The next section discusses the studied theoretical and empirical literature. Section 3 involves the methodology that describes the data and the empirical model. The empirical results are presented in section 4 and discussed in section 5. Section 6 contains concluding remarks.
2.LITERATURE REVIEW
2.1Background
Income inequality describes the phenomenon where income is unequally distributed across the population. The more unequal this distribution, the higher the concentration of the total income in the hands of a few. Inequality can be measured in various ways, but all express the dispersion or width of the income distribution. A simple statistical measure of dispersion that portrays income inequality is the variance (McKay, 2002). Another informative measure is the ratio of the income share of the poorest in relation to the richest quintile of the population.
However, despite its imperfections1, the most common measurement of income inequality is the Gini coefficient, which is mathematically based on the graphical representation of the Lorenz curve (Schutz, 1951, p. 108). The Lorenz curve plots the cumulative income distribution for the cumulative share of the population sorted from the poorest to the richest economic agent. The Gini coefficient ranges from zero to one. Zero indicates perfect
equality: a situation in which all economic agents have equal shares of the aggregate income. One represents perfect inequality, where one economic agent has all income and the rest have none.
An assessment of the presence of economic inequality sheds light on possible positive and negative effects of the phenomenon. On the one hand, Persson and Tabellini (1994) argue that economic inequality is related to lower economic growth and Bénabou (1996) perceives reduced human capital formation to be a negative consequence of this inequality. Yet, on the other hand inequality can be seen as a necessary incentive to work, invest and/or engage in entrepreneurial activity and therefore it leads to higher productivity and national output (Okun, 1975; Mankiw, 2013).
A much more immediate social cost of inequality is its impact on crime. This negative effect of economic inequality is the focus of this research. The concern with crime in this study is well justified since it hampers economic development (Mehlum, Moene, & Torvik, 2005) and imposes significant costs both on society (Lederman, Loayza, & Menendez, 2002) and on individuals (Atkinson, Healey, & Mourato, 2005). Costs on the society firstly include a significant share of government resources being allocated to crime prevention and secondly the loss of foreign direct investment and tourism. On an individual level crime induces loss of income and property, medical expenses, traumatization and costs incurred in order to avoid being victimized.
There is a general consensus that economic conditions, such as income inequality and real income, are associated with crime rates (Hsieh & Pugh, 1993). Increasing levels of economic inequality have been linked to increasing crime rates. Therefore, and inspired by the public discussion on inequality, this empirical research considers the causal relation between income inequality and crime rates. Below the theoretical foundations are discussed, followed by an evaluation of previous empirical research on the subject of economic inequality and crime and its findings. Next, light is shed on the aggregate level of study and lastly, an assessment is made of other relevant variables that affect crime.
1 Objections to the crudity and ambiguity of the Gini coefficient arises from the fact that the shapes of the areas where this measure is based on may be infinitely varied, stemming from different income distributions underlying the Lorenz curve, without it resulting in changes in the coefficient (Schutz, 1951).
2.2 Theoretical Foundations
According to the leading economical and sociological theories on crime there is a positive relation between inequality and crime. These theories explain variations in crime rates through different incentives, deterrents and surroundings experienced by individuals in their specific environments (Kelly, 2000, p. 530). The theories reflect the individual decision making level. However, they are also useful for aggregated levels of study when averages are taken of all individuals in a specific area (Fajzylber, Lederman, & Loayza, 2002a, p.1330).
2.2.1 The Economic Theory of Crime
In the economic theory of crime developed by Becker (1968) crime is treated as the outcome of a rational decision process. In his Nobel lecture (1993, p. 390) The Economic Way of Looking at Human Behavior he formulated it as follows: “Rationality implies that some individuals become criminals because of the financial and other rewards from crime
compared to legal work, taking account of the likelihood of apprehension and conviction, and the severity of punishment”. The first assumption of the economic theory of crime is that potential criminals act rationally; they base their decision to engage in crime or not on a cost-benefit analysis. Furthermore, the theory assumes that individuals are risk neutral, in such that they commit a crime whenever it is expected that net benefits are large enough. Whether the net benefits are large enough depends on a personal or cultural (for group aggregation) threshold level based on moral and ethical beliefs. Criminal activity is not only determined by the rationality and preference for crime but also by the economic and social environment, both the natural and by public policy shaped environment.
Fajnzylbar et al. (1998; 2002a) presented a simple model that helps to organize these ideas and that motivates variables that affect the criminal decision. This model is depicted in equation (1). The model states that for a particular individual the expected net benefit (nb) of committing a crime is equal to the probability of not being caught (1 – pr) times the loot (l), minus the total costs accompanying the planning process and the execution of the crime (c), minus the sacrificed wages from legitimate activities (w), minus the expected punishment for the committed crime - which is determined by the probability of being caught times the punishment (pr*pu).
nb = (1-pr)*l – c – w – pr*pu (1)
To incorporate the moral and ethical beliefs in the model, the net benefits should exceed a specific threshold before the individual commits the crime. To model this, the moral and
ethical beliefs (m) should be expressed in a pecuniary value so that it can be compared to the other values in the equation. Equation (2) establishes the relation between the decision to commit the crime (d) and the net benefits given the moral stance. Where d = 1 and d = 0 represent the decision to commit the crime or not to commit the crime, respectively.
d = 1 if nb ≥ m (2)
d = 0 if nb< m
With this model in hand the next step is to explain the possible relation between income inequality and crime. According to Fajnzylber et al. (1998; 2002a) the effect of income inequality depends on the relative income position of an individual. It is argued that increasing income inequality will not induce the rich to commit more crime because their income increases. Nevertheless, for the poor an increase in income inequality might result in an increase in the criminal activity, as the income gap widens. Stated in the terms of the model, the income from legal activities (w) decreases and the possible loot (l) increases because the rich get richer. Both terms increase the net benefit of crime and therefore an increase in income inequality ceteris paribus results in more criminal activity.
The theoretical model of Chiu and Madden (1998) find the same relation according to a similar reasoning: “If income inequality increases so that low incomes become lower and high incomes become higher, then the level of crime is driven up from two sources: the alternative to crime is less attractive for criminals and the potential proceeds from crime are greater”.
Furthermore, a rise in the inequality level might also induce more crime by reducing the moral threshold (m) trough the so-called ‘envy effect’ (Fajzylber et al., 1998; 2002a). This envy effect follows a similar argument as the sociological reasoning and this is explained in detail below.
2.2.2The Relative Deprivation Theory
Scholars in the field of criminology and sociology also support this suggested positive relation between income inequality and crime rates. Their beliefs are however based on the leading theory of the sociological paradigm; the relative deprivation theory.
The first thorough treatment of this concept can be found in Runciman’s work (1966). According to this scholar, relative deprivation occurs whenever four conditions are met. Firstly, the potential relative deprived person, person A, does not have a good called X.
Secondly, person A perceives some other person(s), which may include his past or future self, possessing X. The third condition is that person A wants good X. And lastly, (s)he sees it as feasible to obtain X.
A critical contingency in the emergence of feelings of relative deprivation is the selection of “frames of reference” and “reference groups”, because individuals can only develop this feeling in relation to others with whom they compare themselves (Messner & Tardiff, 1986, p.300). Relative deprivation produces feelings of envy and injustice originating from a discrepancy between the membership reference group and a comparative reference group (Runciman, 1966, p. 14). Furthermore, Runciman (1966, p. 10) argues that relative deprivation at a group level (which is the level of focus from income inequality) has three dimensions. It varies in magnitude, frequency and degree, and all of these dimensions increase the perceived feeling of relative deprivation.
Vold (1986, p. 138) related the concept of relative deprivation to economic conditions and observed a similar relation. He states that poverty is always in part a subjective condition rather than a simple objective fact; it is not just the presence or absence of a certain amount of wealth but also relative to what others have. The economic condition associated with relative deprivation is economic inequality. If income inequality increases because the group of people who lack average income grows then this leads to a higher frequency of relative deprivation. If income inequality increases because the gap between the below average incomes and the above average incomes increases then this results in a greater magnitude of relative deprivation.
Thus, according to the relative deprivation theory, income inequality generates feelings of envy and unfairness, which ultimately induce two different types of crime. On the one hand, feelings of envy stimulate aggressive impulses that are expressed in violent crime (Messner & Tardiff, 1986, p. 299). While on the other hand, the senses of injustice results in property crime as a primitive form of income redistribution (Stack, 1984, p. 231).
Following a similar argument the strain theory developed by Merton (1938) comes to the same conclusion. The strain theory entails that when faced with the relative success of others, less successful individuals feel frustrated by their situation. Consequently, they become alienated from society and engage in criminal activity in response. Since income inequality is a possible cause of this strain it induces the low-status individuals to commit crimes.
These theories are complementary rather than exclusive since different types of crime are better explained by different reasoning. That is, property crime is better explained by the economic theories based on rational behavior, and violent crime is better explained by the sociological theories based on emotional behavior. It is not the aim of this research paper to distinguish between the different theories explaining the relation between income inequality and crime. Rather, an attempt is made to test this theoretical link on a neighborhood level. Since both approaches indicate a positive relation between income inequality and crime the main hypothesis of this research is as follows:
Hypothesis 1: Income inequality has a positive effect on criminal activity.
2.3Inconsistencies, Types of Crimes and Aggregate Level of Analysis
Three important conclusions become apparent when previous empirical studies are analyzed. That is 1) there are mixed results considering the effect of income inequality on crime, 2) it is important to make a distinction between different types of crime and 3) there is a wide variation in the aggregate level of analysis. These three conclusions are explained separately below.
The majority of the studies that consider the relation between income inequality and crime rates have found a significant positive relation. A meta-analysis conducted by Hsieh and Pugh (1993) identified a positive relation between income inequality and (violent) crime, although their findings indicate a stronger link with violent crime than with robbery rates. Patterson’s (1991) study on the small residential neighborhood level found no evidence for a meaningful association between income inequality and violent crime while it does indicate an expected positive association between income inequality and burglary rates. Reversely, Kelly’s (2000) empirical study on a urban county level found that income inequality does aggravate violent crime and robbery rates while there is no effect on burglary rates.
Fajnzylber et al. (2002b) studied the phenomenon on a national level and indicate a positive effect from income inequality on violent crime and robbery rates. Additionally, Choe (2008) found a strong and robust effect from relative income inequality on burglary and robbery when the relation was analyzed on a US state level.
However, other empirical studies failed to support the suggested hypothesis of a positive effect from income inequality on crime rates. Messner and Tardiff (1986), focusing on violent crime and a neighborhood scale, found no effect from income inequality on homicide rates. Bourguignon, Nuñez and Sanchez (2003) only found insignificant results
regarding the income inequality and overall crime rate relation. Lastly, Neumayer (2003; 2005) found no proof for the link between income inequality and violent and property crime on a national scale.
As the previously mentioned research indicates, income inequality might have different effects on different types of crime. This suggests that it is necessary to differentiate between violent crime and property crime. The examined theories on crime support the relation between income inequality and both types of crime, thus leading to the following theoretically supported hypotheses:
Hypothesis 2: Income inequality has a positive effect on violent crime.
Hypothesis 3: Income inequality has a positive effect on property crime.
Concerning the aggregate level of study, many researchers have chosen to employ nations and states as the size of the sample unit. Some have shifted the focus from the
national to regional level by utilizing county or city data. However, the most appropriate level of analysis for a study on the effects of income inequality on crime rates is a smaller
aggregate sample unit. The relative deprivation theory entails that acts of crime result from a feeling of deprivation in comparison to a reference group. Therefore the entities according to which inequality is assessed should encompass meaningful frames of reference for social comparison (Messner & Tardiff, 1986, p. 300). For the larger aggregate sampling units the theoretical and empirical framework are not corresponding because it is difficult to conceive how residents become aware of the level of income inequality within states, counties and cities (Nettler, 1984, p. 233; Williams, 1984, p. 285). As neighborhoods reflect the social community to which people are most often exposed they represent a much more appropriate unit of aggregation (Patterson, 1991, p. 761).
Only few studies have employed this insight (Hsieh & Pugh, 1993)2, however many
researchers have recommended this aggregate level of analysis (o.a. Kelly, 2000, p. 537; Neumeyer, 2005, p.110). Hence, the aggregate level of analysis that is used in this empirical study is the neighborhood level.
2 Fourteen out of the thirty-four studies considered by their meta-analysis used national sample units, eighteen focused on the regional level and two carried out their research on a local scale. The research on neighborhood level includes the studies of Messner and Tardiff (1986) and Patterson (1991).
2.4Identifying Other Relevant Variables Affecting Crime
As mentioned above, both the economic and the social environment influence crime. These environments encompass an extensive number of factors that can be of influence on the hypothesized relation. Economic determinants include average income, education, unemployment and poverty. Sociological and demographical aspects that could be of influence are among others cultural stance, police activity, origin and age and sex composition.
An economic condition that has often been considered in the theoretical and empirical literature in determining the incidence of crime is the real income level (Fajnzylber et al., 1998, p. 3), shifting the focus from the relative to the absolute income level. Fleischer (1966, p. 122) indicates two conceptual independent influences that income can have on crime rates, which operate in opposite directions. Accordingly, and seen in the light of the economical model above, low income might increase the tendency to commit crime because the possible benefits of legitimate activities and therefore the opportunity cost from criminal activity are low (w). While high income might also increase the tendency to commit crime by the less fortunate since the possible loot increases (l). For both effects also empirical support is found (Fleischer, 1966; Ehrlich, 1973). Taken together, these mentioned mechanisms suggest there is an ambiguous effect from the average income on criminal activity on the aggregate level.
An individual’s education level may influence the decision to engage in criminal activity through several channels. Higher levels of educational attainment are associated with higher expected returns from legal activities (w), therefore increasing the opportunity cost of criminal activity. Additionally, education may have a civilization effect, increasing the individual’s moral stance (m) and therefore also the threshold level in the decision process to engage in criminal activity (Fajnzylber et al., 2002a, p. 1328; Buonanno & Montolio, 2008, p. 92). Both mechanisms suggest a negative relation between education level and criminal activity. However, education can also reduce the costs of executing a crime (c), reduce the possibility of being caught (pr) or raise the possible loot (l) by engaging in criminal activities with higher returns, such as fraud instead of burglary. Taken together, the effect of education is, a priori, ambiguous. It can be argued however, that if legal economic activities are more education intensive than illegal activities, then the expected effect will be negative. On the aggregate level this implies that higher educated groups will have lower crime rates.
To distinguish the effect of income inequality from that of poverty, Kelly (2000, p. 530) suggests to include several measures of deprivation as control variables, specifically unemployment and poverty rates. A high unemployment level may be associated with high
crime rates since it reflects the low income opportunities available in the legal labor market (w) (Fajnzylber et al., 1998, p. 4). Furthermore, being employed is like an “occupational therapy”, reducing the possible time and thought left for engagement in criminal activity. Poverty rates, in turn, control for the argument made by Vold and Bernard (1986, p. 138). They argue that being poor will induce criminal activity to flourish because income is insufficient for survival and minimum basic needs.
Furthermore, population characteristics can be important in explaining differences in crime rates. Racial composition is often considered, as geographical areas with more foreigners are associated with higher rates of criminal activity (Buonanno and Montolio, 2008, p. 91). Additionally, numerous studies indicate that the male population is more prone to engage in criminal activity, especially considering the young males (Nettler, 1982, p. 17). Finally, Bourguignon, Nuñez and Sanchez (2003) argue that unobserved factors might simultaneously affect crime rates and income inequality. Similarly Glaeser, Sacerdote and Scheinkman (1996) argue that income inequality might merely pick up the effect of
unobservable factors, given the enormous variation in the crime rate across space and the fact that measurable characteristics account for little of this variation. This highlights the
importance of the employment of fixed effects regression analysis. These unobserved factors most importantly cover cultural differences, directly related to the moral threshold in the model (m). To control for these unobserved factors and observable independent variables that lack useable data – such as police interaction, family stability, etc. – neighborhood fixed effects are incorporated in the analysis.
In conclusion, there are many observable and unobservable factors that are specific to the neighborhood unit of observation and possibly correlate with income inequality as well as criminal activity. The methods used to control for these factors are discussed in the next section.
3.DATA AND EMPIRICAL METHOD
This research paper uses data provided by the local research and statistics department Bureau Onderzoek en Statistiek (O+S) from the municipality of Amsterdam. Their yearly publication ‘Stadsdelen in cijfers’ includes information on the neighborhood level that is related to the topic of this thesis. From this publication, useable data of 75 neighborhoods are obtained after the sample selection process. Due to inconsistencies in the yearly-published data only four
years are comparable and therefore included in the analysis, respectively the years 2008-2011. The panel data set that has been constructed contains 300 observations. A summary of the descriptive statistics from the complete sample is presented in table 1. The next
subsections discuss the dependent, independent and control variables. This is followed by an explanation of the sample selection process and finally the empirical method is presented.
3.1 Dependent Variables
This research examines the question whether income inequality positively impacts crime rates. To study this question in detail different types of crimes are considered. To this end, the dependent variable crime rate is split up in multiple dependent variables that represent overall crime, violent crime and property crime. For the total crime rates the overall crime index and youth crime index are employed. Furthermore, the violent crime index, robbery crime index and burglary crime index are taken into account. The violent and robbery crime indices represent violent crime and the burglary crime index represents property crime. Index numbers are used to make it possible to compare neighborhoods of different sizes with each other and over time. The 2003 city averages are used as a base year for each of the different indices, with the exception of the youth crime index. The base year for this index is 2007. Index numbers are determined by the percentage deviation from the city average in the base year and high index numbers indicate high crime rates.
The data is constructed based on police reports and victimization surveys. A combination of these two sources lowers the downward bias that is often induced by
underreporting3. The overall crime index is compounded from a weighed average of several
subgroups, including: burglary, robbery, violence, nuisance, drugs, vandalism and traffic. The youth crime index is based on the number of juvenile suspects that fall in the age group of 12-24 years. An important distinction of the youth crime index as compared to all other indices is that it is based on the living address of the juveniles instead of on the location where the crime was committed. The violent crime index is based on homicide, sexual offense, assault, abuse, threat and other violent offenses. The robbery crime index includes offenses like raid, pickpocketing, theft and fraud. Lastly, the burglary crime index takes into account residential burglary and burglary at companies and institutions.
3 Underreporting in police reports might be present due to lacking commitment or underestimation of the severity from some event (Kelly, 2000, p. 532). In addition victims might not report a suffered criminal offense to the police, especially when it’s considered to be shameful.
3.2 Independent Variables
The independent variable of interest is income inequality as measured by the Gini coefficient. The Gini coefficient ranges from zero to one and the higher this value the more unequal income is distributed across an entity. Additionally, the standard deviation of income is considered as an alternative income inequality measure and again a higher value is associated with more inequality. Both variables are constructed based on O+S grouped data that reports the number of income recipients according to five income classes (see Appendix A for a detailed description on this computation). A limitation of this measure concerns the imprecise data that constitutes the foundation for the computations. The statistics regarding the income distribution are based on just five distinctive income classes. In addition, the numbers are rounded to hundreds of people and this is problematic because some neighborhoods only contain a couple of thousand residents. Therefore, the estimations might be imprecise as well.
3.3 Control Variables
There are several other independent variables to be considered, as identified by my theoretical framework presented in section II. The economic control variables include
absolute income, poverty, education and unemployment. Absolute income is accounted for by
the neighborhood average real income per capita, in prices of 2007.4 For the poverty rate the
national methodology of the Centraal Bureau voor de Statistiek (CBS) is followed. This measure considers households to be poor if their income represents <110% of the legally
TABLE 1–DESCRIPTIVE STATISTICS
All Neighborhoods (n=75)
Crime indices Observation Mean Standard Deviation Minimum Maximum Youth Crime Index 300 91.76 33.00 20 173 Overall Crime Index 300 77.58 20.89 37 229 Violent Crime Index 300 78.35 26.75 28 161 Robbery Crime Index 300 73.00 22.88 22 164 Burglary Crime Index 300 70.48 37.12 22 268
Explanatory variables
Gini coefficient 300 0.333 0.020 0.244 0.372 Gini coefficient Morgan 300 0.351 0.023 0.250 0.395 Standard Deviation income 300 18175.46 2282.99 9047 22454 Average income 300 30950.89 9147.78 19805 66341
Poverty rate 300 15.89 7.28 2 30.4
Education (% low educated) 300 28.71 13.58 5 59 Unemployment rate 300 6.63 2.78 1.5 14.5
Notes: For the crime rates indices are used such that they are comparable between neighborhoods of different sizes. In the base year (2007 for the youth
crime index and 2003 for the other indices) the average of each index was set on 100. Average income indicates the real average income in prices of 2007. Education is measured by the percentage that is low educated.
determined social minimum. Education is controlled for by including the percentage residents
aged >15 year with a low educational attainment5. The last economic control variable that is
employed in this research concerns the unemployment rate as registered by the Centrum
Werk en Inkomen (CWI).6
The social and demographic characteristics are not directly included in the model but are accounted for by the inclusion of neighborhoods fixed effects. Therefore this empirical study employs a panel data set.
3.4 Neighborhood Measurement
The aggregate level of analysis in this thesis is the neighborhood level. One obvious difficulty with the use of this level of aggregation is the sorting of the crime data. All crime indices of consideration, except for the youth crime index, are classified in terms of where the criminal incident occurred and not where the offender resided. This makes it questionable whether the crime data can be matched to the neighborhood characteristics. Inferred conclusions based on these indices are therefore subject to the strong assumption that crime occurred in a specific neighborhood is committed by one of its residents. Although this statement might be doubtful by nature, there is however some research supporting this claim. Nettler (1984, p. 74) indicates that violent crimes are committed near to the residences of the offenders. Furthermore, Patterson (1991, p. 762) argues for the residential proximity of criminal activity, which implies that “offenders commit crimes in their own backyards”. Additionally, Chiu and Madden (1998, p. 124) states that burglars tend not to travel too far to commit crime, typically not more than a few miles. Yet, as the youth crime index is based on the living address of the offenders, the conclusions drawn from this index can be asserted with more certainty and therefore constitutes the variable of consideration in the robustness analysis. In addition, data on smaller units of aggregation provide more profound measures of the dependent and explanatory variables. It is capable of portraying more nuances in
between-group variation, as the unit of analysis is smaller (Patterson, 1991, p. 762).
3.5 Sample Selection
The city of Amsterdam is divided in 8 districts and encompasses 97 neighborhoods in total. The sample selection process consisted of three phases. First the neighborhoods with too little
5 The standard education classification from the CBS is followed (SOI 2006). Accordingly, the low educated group consists of persons that at most completed the secondary education, first phase. This includes the lower vocational education and the first three years of the higher general secondary education and pre-university education.
6 These statistics are based on the non-working and job-seeking persons registered by the CWI as a percentage of the total labor force (aged 15-64), with a 12-hours a week minimum.
residents were left out because no meaningful income inequality measures could be
constructed on basis of the rough income distribution statistics available. This accounted for the exclusion of 9 neighborhoods. Second, the neighborhoods with no crime data available were omitted, many of which coincided with the neighborhoods identified in the first phase, but this additionally led to a drop of 10 neighborhoods. Finally, each neighborhood was analyzed and judged on usefulness separately, resulting in a drop of 3 additional
neighborhoods.7 Eventually useable data is available for 75 neighborhoods. In addition to this
main sample, three subsamples were created for a more detailed analysis of the relation between income inequality and crime. The neighborhoods included in the full sample and created subsamples are reported in table B1 in Appendix B, for descriptive statistics of the subsamples consult table B2.
3.6 Empirical Strategy and Model
In order to test the hypothesized relations empirical research is conducted in which the causality and robustness of the link between income inequality and crime are investigated. The empirical strategy is staged: First the correlation between the Gini coefficient and the different types of crime is considered. Second, other potential crime determinants are added to the empirical model resulting in a multiple ordinary least square regression. Next the empirical model is completed with the addition of neighborhood fixed effects to control for unobserved omitted variables and to employ all information contained in the panel data set. Time fixed effects are included in all regressions. After the causality is examined the subsequent parts inspect the robustness of the results. This is done by a reassessment of the fixed effects model for different income inequality measures and various subsamples.
The empirical models employed in this research are based on the variables indicated by the economic theory of crime and from which data is available. The base model will be an ordinary least square regression with income inequality and the year dummies as the
explanatory variables of crime rates. The multiple regression model is build up according to theory and adds average income, poverty, education and unemployment as explanatory variables to disentangle the effect of income inequality from the other possible economic explanations. Univariate analysis of these variables indicates that average income displays a positively skewed distribution. Therefore the natural logarithm of average income is
7 From which two for the nonexistence during the period of consideration and one for having a different geographical boundary for the dependent and independent variable of interest.
employed as it reduces the skewness of the distribution and improves the statistical fit in the regression analysis.
Common sense suspects imperfect multicollinearity between these additional
explanatory variables since they are all associated with one another. This notion is confirmed when the correlation matrix between all explanatory variables is consulted in table B4, indicating correlations of the control variables between 0.7403 and 0.9204. As a result one or more of the regression coefficients of the control variables could be estimated imprecisely. Nevertheless they al remain included in the model as expert judgment and economic theory identify them as to be of importance and because they are not the variable of interest of this research.
To account for unobserved omitted variables that can cause a bias in the estimated coefficients both neighborhood and time fixed effects are included in the model. The neighborhood fixed effects control for omitted variables that vary across neighborhoods but do not change over time (Stock & Watson, 2012, p. 396). These may include unobservable or unavailable variables. Unobservable factors can be cultural stance and sociological
characteristics – such as family stability – that differ between neighborhoods. Unavailable omitted variables may include the previously indicated population characteristics and police activity. Just as neighborhood fixed effects can control for variables that are constant over time and differ between entities, so can time fixed effects control for variables that are constant across entities but evolve over time (Stock & Watson, 2012, p. 400). This may encompass citywide public programs related to criminality or changes in classifications at some period in time. The complete fixed effects regression model contains n*T distinctive intercepts – one for every neighborhood-year combination.
Resulting from the discussion above the main empirical model regarding the crime-inequality link of consideration in this research looks as follows:
CRIMEit = β1INEQit + β2lnINCit + β3POVit + β4EDUit + β5URit + αi + λt + uit
Where CRIMEit characterizes the dependent variable crime rate from neighborhood i
observed in year t, as represented by the different crime indices of consideration – youth crime (YCI), overall crime (OCI), violent crime (VCI), robbery (RCI) and burglary (BCI).
The explanatory variable of interest is denoted by INEQit and covers the different income
inequality measures including the two Gini coefficient measures (GINI and GINIM) and the standard deviation of income (SDINC). The independent control variables are denoted by the
second until the fifth variable and their corresponding abbreviation. Included are the natural logarithm of income (lnINC), the poverty rate (POV), the percentage lower educated (EDU)
and the unemployment rate (UR). αi contains the unknown neighborhood specific intercepts
to be estimated that allow for the permanent differences across neighborhoods (neighborhood fixed effects). Any citywide trend will be captured by the time fixed effects as denoted by the year intercepts in λt. Finally uit represents the error term. Note that there is no constant added
to the model as every neighborhood-time combination has its own specific intercept for which is accounted by the entity and time fixed effects intercepts (αi and λt).8 The coefficient
β1 is the primary focus of this research and portrays the causal effect of an increase in the
income inequality on crime rates. In the next section the primary results are presented and subjected to multiple robustness tests.
4.RESULTS
4.1 Basic Results
This empirical analysis is concerned with estimating the causal effect from income inequality on crime rates within the neighborhood specification. The empirical strategy as outlined in subsection 3.5 is followed.
Table 2 reports the results concerning the relationship between different crime types and the economic conditions of neighborhoods. For every crime type the three previously mentioned estimation techniques are depicted. The simple ordinary least square regression from crime indices on the Gini measure of income inequality suggests positive relations between all of them, justifying the principal thought underlying this research (Appendix B, figure B2-B6). However, controlling for the possible heteroskedasticity and autocorrelation of the error term with clustered standard errors leads to the insignificance of all coefficients on income inequality estimated by simple OLS. When time fixed effects are included the estimates are similar except for the burglary crime index, which coefficient on inequality turns insignificant negative. These estimates are depicted in every first column.
In the multiple ordinary least square regression the other economic control variables are added. The coefficients on income inequality remain insignificant and in addition of them turned sign. As was highlighted in section 3, the coefficients on the other economic variables are somewhat difficult to interpret as they might be estimated imprecisely.
8 Alternatively, expressed with entity and time dummies the model looks like this:
CRIMEit = β0 + β1INEQit + β2LNINCit + β3POVit + β4EDUit + β5URit + γ2D2i + … + γnDni + δ2B2t + … + δTBTt + uit Where D2 to Dn represent the neighborhood specific dummies and B2 to BT are the time specific dummies.
Moreover, these coefficients have rather different outcomes for the different crime types. If
the youth crime index is considered, the coefficients on poverty, education9 and
unemployment all have the expected positive sign, although the coefficient on unemployment is not significant. These findings change dramatically when the variables are considered for the other crime indices. Poverty and education most often indicate a negative effect, which is inconsistent with the expectations. The positive coefficient on unemployment is confirmed by the other regressions and in addition they are significant. Furthermore, the coefficients on
income estimated with the multiple regression model are all negative – except for the youth
crime index – and some are significant. Because income is transformed to its natural
logarithm its interpretation is as follows: a 1% change in income is associated with a change in the crime indices of 0.01 β2.
Every third column in table 2 represents the fixed effects regression, including both neighborhood and time fixed effects. Year dummies are depicted to assess the presence of a trend over time in the crime data. The estimated coefficients suggest very different reactions from violent and property crime to changes in the income inequality within a neighborhood. The youth crime index and burglary crime index indicate a significant positive relation while the robbery crime index signals a significant negative relation. Furthermore, the overall crime index and the violent crime index find insignificant positive and negative relations,
respectively. These results support the main hypothesis of this research; they indicate that the general crime rate is positively affected by income inequality. In addition, following the crime classification from section 2, it can be concluded that income inequality does not affect violent crime or even negatively impacts violent crime. Therefore, hypothesis 2 is not accepted. The effect from income inequality on property crime is, as expected, strongly positive and significant. This confirms hypothesis 3.
The outcomes of the fixed effects regression furthermore suggest a negative time trend in the crime rates caused by unobserved omitted variables. This can be seen from the significant negative and increasing estimated coefficients on the year dummies. These coefficients refer to the intercept as compared to the year 2008. This negative trend seems to be present for all but the robbery crime index. Finally, none of the estimated coefficients on the control variables are of statistical significance apart from the negative poverty coefficient
9 As was discussed in section 2, higher educational attainment discourages criminal activity. However in this study, education is controlled for by the percentage lower educated. An increase in this rate would predict an increase in crime rates and therefore the expected coefficient is positive.
TABLE 2–ESTIMATES OF THE EFFECT OF INCOME INEQUALITY ON DIFFERENT TYPES OF CRIME
Dependent variable
Youth Crime Index Overall Crime Index Violent Crime Index Robbery Crime Index Burglary Crime Index Estimation technique Simple Multiple Simple Multiple Simple Multiple Simple Multiple Simple Multiple
OLS OLS FE OLS OLS FE OLS OLS FE OLS OLS FE OLS OLS FE Explanatory variables
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] Income inequality (Gini coefficient) 280.70 82.59 186.82** 229.10 -65.73 41.53 197.69 -66.36 -113.34 203.49 139.02 -156.28** -7.29 -245.43 359.73**
(172.18) (219.65) (71.27) (131.31) (126.15) (86.66) (119.76) (127.35) (79.87) (122.76) (127.06) (72.45) (245.29) (343.80) (160.30) Average income (log of real average income) 23.55 -4.72 -69.42** 10.59 -44.04 -34.45 -9.87 38.38 -72.92* 58.35
(25.68) 35.33 (27.47) (25.05) (30.55) (37.66) (27.46) (26.15) (41.60) (65.35) Poverty rate 1.95** -4.20** -1.78** -2.58 -1.70* -1.78 -0.91 -1.20 -3.38*** -1.99
(0.91) (1.66) (0.75) 2.47 (0.95) (1.63) (0.98) (1.04) (1.16) (4.63) Education (% low educated) 0.90*** 0.24 -1.13*** -0.23 -0.61 -0.11 -1.19*** -0.89 0.44 0.85
(0.33) (0.73) (0.36) (0.81) (0.38) (0.62) (0.36) (0.63) (0.55) (2.61) Unemployment rate 2.47 -0.78 4.59*** 0.09 6.76*** 0.82 4.51** 0.30 5.41** -0.16 (2.11) (1.15) (1.38) (0.99) (1.98) (1.66) (1.86) (0.81) (2.69) (2.73) 2009 (year dummy) -0.62 -1.48 -0.13 -8.30*** -8.59*** -8.14*** -6.57*** -7.58*** -6.32*** -2.45 -3.24 -2.55* -5.84 -6.31 -6.19 (1.75) (1.73) (1.76) (1.65) (1.82) (1.64) (2.22) (2.36) (2.20) (1.55) (1.81) (1.48) (4.92) (5.06) (4.78) 2010 (year dummy) -9.08*** -10.95*** -7.41*** -12.45*** -14.69*** -11.42*** -16.50*** -20.81*** -15.71*** 1.60 -1.65 2.84* -4.71 -7.75 -5.76 (2.33) (-2.84) (2.41) (2.41) (2.82) (2.23) (2.27) (2.96) (2.64) (1.66) (2.45) (1.54) (4.83) (6.05) (4.96) 2011 (year dummy) -15.63*** -13.96*** -14.61*** -14.72*** -12.62*** -13.65*** -17.84*** -14.82*** -16.25*** 1.65 2.80 2.76 -11.33** -7.94 -11.70** (1.88) (2.33) (2.00) (2.40) (2.45) (2.18) (2.49) (2.55) 2.52 (1.71) (1.80) (1.89) (5.05) (5.32) (4.51) constant 4.48 -245.13 148.78 10.04 853.95*** 9.98 22.64 564.73 506.56 4.93 147.64 -228.44 78.38 914.44 -636.46 (58.01) (332.92) (368.46) (43.02) (317.22) (276.69) (40.06) (355.43) (395.24) (40.44) (317.76) (266.00) (85.10) (553.95) (681.54) R2 0.0618 0.6525 0.9093 0.1116 0.2432 0.8246 0.0905 0.2402 0.8237 0.0386 0.2746 0.8860 0.0119 0.2172 0.6527 No. neighborhoods 75 75 75 75 75 75 75 75 75 75 75 75 75 75 75 No. years 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 No. observations 300 300 300 300 300 300 300 300 300 300 300 300 300 300 300
Notes: Clustered standard errors are in parentheses. * Significant at the 10-percent level. ** Significant at the 5-percent level. *** Significant at the 1-percent level.
in regression [3]. The results are robust when controlled for outliers, except for the coefficient on income inequality on burglary, which remains similar in magnitude but turns statistically insignificant.
Notwithstanding their importance, the specific crime regressions are not included for further statistical analysis, because these results should be interpreted with the notion of a different and less preferred crime data registration method, this will be further discussed in section 5. As the youth crime index produces the most reliable estimates and the complete model is believed to be the most realistic, the results concerning the main hypothesis obtained from regression 3 are subjected to a sensitivity analysis.
4.2 Alternative Measures of Inequality
To assess the robustness of the relation between income inequality and crime rates the Gini coefficient is replaced by alternative measures of inequality. If hypothesis 1 is true, this should lead to the same results as obtained before. The alternative measures of income inequality include the standard deviation of income and another Gini coefficient, computed
according to the formula and assumptions developed by Morgan10 (1962, p. 281). Table 3
presents the results.
The first regression of table 3 coincides with regression 3 from table 2 and is included for further discussion and comparison. The coefficient on GINI is positive and significant at the 5% level. As the Gini coefficient ranges from 0 to 1 the estimate should be divided by 100 to be interpretable, meaning that a 0.01 increase in the Gini coefficient results in increase of almost 2 in the youth crime index. The other significant coefficients in the regression comprise the negative coefficients on poverty and 2 out of the 3 year dummies.
The same pattern is revealed when GINI is replaced by SDINC or GINIM, as shown by regression 2 and 3 of table 3, respectively. The coefficient on the standard deviation of income is positive and significant at the 1% level, indicating that an increase in the standard deviation of income of 1.000 euro increases the youth crime index with 2.79. It should be noted here that the standard deviation of income portrays high correlations with the other economic explanatory variables, indicating possible multicollinearity. However, when these control variables are excluded from the fixed effect regression the coefficient on the standard deviation of income remains similar. The coefficient on the Morgan’s Gini coefficient is also positive and of similar significance and magnitude as the original Gini coefficient. The
standardized effects are similar, although the effect from SDINC is somewhat stronger.11 This is accompanied by a larger negative, but still insignificant estimated effect from average income on crime rates in regression 2. All other coefficients correspond to the estimates of the principal regression. Hence, it can be concluded that the results are robust to alternative measures of inequality.
TABLE 3–ROBUSTNESS TEST OF THE COEFFICIENT ON INEQUALITY:ALTERNATIVE INEQUALITY MEASURES Dependent variable
Youth Crime Index
Estimation technique FE FE FE Explanatory variables [1]a [2] [3] Gini coefficient 186.82** (71.27)
Standard deviation income b 2.79***
(0.96)
Gini coefficient Morgan 168.15**
(75.69) Average income (log of real average income) -4.72 -31.32 -1.21 (35.33) (36.38) (35.89)
Poverty rate -4.20** -4.37** -4.13**
(1.66) (1.67) (1.66) Education (% low educated) 0.24 0.25 0.23 (0.73) (0.73) (0.73) Unemployment rate -0.78 -0.70 -0.80 (1.15) (1.14) (1.15) 2009 (year dummy) -0.13 -0.55 -0.17 (1.76) (1.72) (1.77) 2010 (year dummy) -7.41*** -7.75*** -7.42*** (2.41) (2.37) (2.44) 2011 (year dummy) -14.61*** -14.70*** -14.60*** (2.00) (1.98) (2.01) constant 148.78 436.56 114.92 (368.46) (374.25) (377.53) R2 0.9093 0.9094 0.9091 No. neighborhoods 75 75 75 No. years 4 4 4 No. observations 300 300 300
Notes: Clustered standard errors are in parentheses. * Significant at the 10-percent level. ** Significant at the 5-percent level. *** Significant at the
1-percent level. The constant in the fixed effect regressions represent the year 2008. The fixed effect model includes both neighborhood and year fixed effects.
a
Included for comparison, identical to the results from Youth Crime Index in table 2 (regression [3]).
b
measured in thousands for clearer representation of the coefficient.
4.3 Subsamples
In addition to the previous robustness check the results are also evaluated for different subsamples of neighborhoods. The sample that encompasses all neighborhoods is first replaced by a subsample consisting of 31 neighborhoods that are characterized by high youth crime rates, next it is replaced by a subsample of 21 neighborhoods that were indicated as most problematic by respondents of a survey and last it is replaced by another subsample of
11 The magnitude of the effect from the different inequality measures can be compared by looking at the standardized effects. Per standard deviation of the income inequality measures the effects are 186.82*0.020=3.74, 2.78*2.26=6.28 and 168.15*0.022=3.70, for the GINI, SDINC and GINIM respectively.
21 neighborhoods that are characterized by low youth crime rates. Table 4 reports the results of these regressions. Again, regression 1 in this table is the equivalent of the principle regression and is included for comparison. Regression 2 corresponds to the first subsample, regression 3 resembles the second subsample and regression 4 covers the third subsample.
It can be seen from the regression output that the coefficient on income inequality remains positive and of the same magnitude, but that there is a difference in the statistical significance between them. The complete sample and the second subsample both demonstrate a significant coefficient at the 5% level. The third subsample displays a weak significant coefficient, while the effect from income inequality of the first subsample is statistically insignificant. This might be caused by the increased standard error that results from the lower sample size. The negative significant coefficient on poverty remains similar in the first subsample but becomes insignificantly positive in the two last. Moreover, there also seems to be a negative but less obvious time trend in the crime rates within the smaller subsamples.
TABLE 4–ROBUSTNESS TEST OF THE COEFFICIENT ON INEQUALITY:SUBSAMPLES Dependent variable
Youth Crime Index
Sample All neighborhoods Neighborhoods with Perceived problematic Neighborhoods with heighest youth crime ratesb Neighborhoodsc lowest youth crime ratesd
Estimation technique FE FE FE FE Explanatory variables [1]a [2] [3] [4] Gini coefficient 186.82** 168.57 180,33** 199.49* (71.27) (119.84) (75.90) (113.12) Average income -4.72 119.46 62.47 -60.19 (log of real average income) 35.33 (81.62) (81.14) (48.36)
Poverty rate -4.20** -4.99* 1.71 0.57
(1.66) (2.56) (2.88) (2.30) Education (% low educated) 0.24 0.16 1.20 1.08 (0.73) (0.99) (1.00) (1.62) Unemployment rate -0.78 -3.31 -1.18 -2.40** (1.15) (2.00) (1.77) (0.89) 2009 (year dummy) -0.13 -1.15 3.46 5.23 (1.76) (3.36) (4.28) (3.77) 2010 (year dummy) -7.41*** -5.71 -1.64 -7.74** (2.41) (4.90) (4.49) (2.93) 2011 (year dummy) -14.61*** -18.38*** -11.38** -8.68*** (2.00) (3.52) (4.13) (2.77) constant 148.78 -999.25 -653.11 594.98 (368.46) (822.55) (834.85) (522.65) R2 0.9093 0.6845 0.9290 0.6637 No. neighborhoods 75 31 21 21 No. years 4 4 4 4 No. observations 300 124 84 84
Notes: Clustered standard errors are in parentheses. * Significant at the 10-percent level. ** Significant at the 5-percent level. *** Significant at the 1-percent
level.
a
Included for comparison, identical to the results from Youth Crime Index in table 2 (regression [3])
b This subsample is based on the criteria that the average youth crime index from the neighborhood is ≥ 100 (the base year average). c
This subsample is based on a small survey where the respondents indicated neighborhoods which they perceived as problematic. Neighborhoods are included if they received ≥ 5 votes.
Furthermore, many coefficients of the control variables changed either in magnitude or from sign but this concerns mostly insignificant estimates and variables that are not the primary focus of this research. Notable is however the changed sign and magnitude of the coefficient on average income. This hints at the ambiguous effect from absolute income on crime rates as discussed in the literature review.
5.DISCUSSION
The statistical analysis performed in order to test the hypothesized relations between income inequality and crime rates on a neighborhood level have resulted in some interesting findings. Firstly, in accordance with the economic theory of crime and the relative deprivation theory, it was found that areas with greater income inequality in Amsterdam tend to have higher overall levels of crime. This statement, which supports the first hypothesis, is based on the results obtained from the fixed effect regression that employed the youth crime index as dependent variable. This regression presents the most reliable outcomes as the youth crime index is based on the residence neighborhood of the juvenile offenders and as the fixed effect model controls for neighborhood specific characteristics and time trends caused by
unobserved variables. The results confirm that there are indeed unobserved omitted variables specific to neighborhoods that are of importance, as can be seen from the difference in the
estimated coefficients when neighborhood fixed effects are added last and the increased R2.
The main result is robust to the measure of income inequality employed and partly robust when tested for three additional subsamples.
Whether this effect is big or small can be seen from the standardized effect. Per standard deviation of the Gini coefficient the youth crime index positively changes by approximately 3.75 index numbers. This is a rather small effect when compared to the average and standard deviation from the youth crime index and this should be considered when possible policy implications are evaluated.
The results are external valid for the overall crime rate in so far that the population aged 12-24 years (included in the juvenile data) perceives income inequality to the same extent and responds to it in a similar manner as the total population. As income inequality can become apparent from the simple observation ‘what have others that I do not have?’, it is likely that it is indeed perceived similarly for adolescents and adults.
The coefficient on poverty presents a puzzling picture as it is prevailing negative and
significant.12 The negative coefficient implies that neighborhoods with a greater proportion of
poor residents tend to have lower crime rates. This outcome is contrary to the expectations and previous studies which all suggest a positive effect from poverty on crime rates. One possible explanation might be the difference in the employed poverty criterion by the main body of research and this research paper. Usually, the population with income below 70% or 75% of the poverty line is labeled as poor. The choice of this lower bound is motivated by the findings that criminal offenders are disproportionally extracted from the very lowest levels of the socioeconomic hierarchy (Messner & Tardiff, 1986, p 304). However, for Amsterdam only statistics about the population below 110% of the poverty line are available. Moreover, the Netherlands, as welfare state, has an extensive social services system as compared to the United States. Therefore the measure employed for poverty in this case study might not truly reflect the level of resources needed for survival and minimum basis needs, as was indicated by the identified mechanism in the literature review. Even though this might account for the insignificance of poverty in determining crime rates, it does not explain a negative relation.
Another explanation that might confirm the found negative coefficient may lie in the relation between the poverty in neighborhoods and the social programs for youngsters offered by city authorities and voluntary initiatives. Poor neighborhoods are often the focus of (subsidized) programs, such as youth counseling, youth centers, sport facilities and other initiatives to limit their exposure to harmful influences and to keep them of the streets. The negative coefficient possibly indicates the effectiveness of these efforts.
In addition, the other economic explanatory variables do not seem to be of significant importance. This might be caused by the possible multicollinearity between these variables as identified in section 3. Moreover, for the average income and education it can depict the contradicting mechanisms that keep the overall effect in balance. Wilkinson and Pickett (2010) support the null-effect of average income. They state that as long as it concerns the developed world the absolute income levels contrary to the relative income distribution do not affect crime rates.
The negative trend in crime rates over time13 can be attributed to multiple causes. One
explanation specific to Amsterdam is the project ‘Top600 approach’ as part of the safety policy of the municipality. This project is a collaboration between more than 30
organizations, including the city council, police, public prosecution and care and youth
12 This negative coefficient on poverty persist even when all other economical variables are excluded, see Appendix B table B3. 13 See figure B7 in Appendix B.
services. It aims at reducing the overall crime by addressing the 600 heavier habitual offenders. The approach consists of three crime-reducing channels: fast and consistent sentencing, assistance and (after) care and influx restriction of family members. Other national prevailing factors that lower the criminal offenses include the aging population, better security measures taken by individuals and businesses and increased chances of penalties and sentencing (de Rechtspraak, 2012).
The statistical regressions that employed one of the specific crime indices as dependent variable resulted in some remarkable results. Violent crime does not seem to be affected by income inequality and property crime strongly increases if income inequality increases. Therefore hypothesis 2 is rejected and hypothesis 3 is accepted. Although these findings are not fully supported by the revised theory it does correspond to the empirical findings of two studies that investigated the inequality-crime link on a neighborhood scale. These two studies were conducted by Messner and Tardiff (1986) and Patterson (1991). Their
research also relied on the strong assumption concerning the crime registration. As was stated
in subsection 3.4 it is however questionable whether crime data registered on the location of the criminal offense instead of the offender’s residence can be linked to neighborhood characteristics.
This concern is aggravated by an inspection of the correlation matrix between the different crime indices in table B2 in Appendix B. Points of focus are the correlations between the youth crime index (which is based on the residence location of the juveniles) and the other crime indices. It can be seen that the overall crime index barely correlates with the youth crime index (0.1038) and that the crime type that is most often committed in the own neighborhood is violent crime (0.3520). Furthermore, the factsheet on juvenile delinquency published by O+S (2014) indicates that even though most juveniles reside in the district Amsterdam South-East, most criminal offenses are committed in the city center district. This observation is probably applicable to all criminal offenses.
These observations might indicate that the specific crime type regressions measure whether neighborhoods with higher levels of income inequality experience more criminal activity (instead of measuring if residents who experience more income inequality commit more crime). In this context, an increase in inequality might refer to an increase in the returns of the loot (l). If it is assumed that loot value is less related with violent crime and robbery than with burglary, this possibly explains why not any positive effect from income inequality is found on violent crime and robbery crime, but on burglary crime.
Overall, outcomes with respect to data obtained on criminal residence relate to criminals weak economic background and frustration while outcomes with respect to the data on the location of a criminal offense relate to the possible returns of the loot and feelings of envy.
6.CONCLUSION
This empirical research considered the criminogenic effect of income inequality. In order to examine this link a case study on Amsterdam has been conducted. In contrast with previous research on this topic the aggregate unit of analysis has been the neighborhood level. As opposed to greater aggregation units, neighborhoods are considered more appropriate units for social comparison and therefore for the experience of income inequality. In addition, local insights are useful for municipalities in their attempt to supplement national policy and to tackle city specific problems.
The main conclusion of this thesis is, congruent with the economic theory of crime and the relative deprivation theory, that income inequality, as measured by the Gini
coefficient, has a positive and significant effect on the incidence of crime. This result is based on the main empirical model, which controls for economic determinants of crime and
neighborhood and time fixed effects. Moreover, this result is robust to alternative measures of income inequality employed and partly robust to the sample of neighborhoods considered. When specific crime types are considered it is found that neighborhoods with higher income inequality are more prone to property crimes but not to violent crimes. In addition, during the process of arriving at the main conclusion some other interesting results were found. First, this research indicates an unexpected negative effect from poverty on crime rates. Second, the crime rates portray a significant and negative trend over time. And last, the absolute income, the educational attainment and the unemployment rate are not related to crime rates in a significant, robust or consistent way.
The results of this case study are particularly useful for the municipality of
Amsterdam. Their efforts to reduce criminality can be accompanied by addressing the income inequality. At first glance this might look like a task for national policy given that the
government has the tax system as major instrument. However, on a local scale municipalities also have various (indirect) tools at hand to influence the income distribution of their inhabitants. Possible policy instruments include property tax, social housing, education provision or a citywide minimum wage (as was implemented by Seattle before). Concerning
