• No results found

The relation between income inequality and crime rates on a municipality level in the Netherlands

N/A
N/A
Protected

Academic year: 2021

Share "The relation between income inequality and crime rates on a municipality level in the Netherlands"

Copied!
27
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

Bachelor thesis

“The relation between income inequality and crime rates on a

municipality level in the Netherlands”

Timo Pleus 6050166

(2)

Statement of Originality

This document is written by Student Timo Pleus who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its

references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Introduction

Tales and ballads dating back to the Middle Ages tell the story of an English outlaw that stole from the rich and gave to the poor (Hilton, 1958). The acts of this character, better known as Robin Hood, can be considered as a vigilante method to enforce a more equal distribution of wealth and income among citizens. In modern-day politics, the popular name of higher taxation levels for the so-called “super rich” to decrease income inequality that are up for discussion is the “Robin Hood” tax (Franko, Tolbert & Wilko, 2013). What is the driving force behind this need for a more equal spread of income? A reasoning that comes to mind is that by increasing the income level of individuals that live at or near the lower border of the distribution, the average quality of life of a society increases as a whole. The question that arises is if there exists any proof that equality in the distribution of income leads to an improvement of quality of life. A study that uses data from the European Quality of Life Survey with representative samples from 28 European countries couldn’t find a significant effect of the national level of income inequality on well-being, financial quality of life and health (Zagorski et al., 2013). In another study, where research is done on the relation between income inequality and quality of life in Germany, results are ambiguous (Schröder, 2016). No significant difference in satisfaction could be found between interviewed individuals during times of high inequality and low inequality (Schröder, 2016). The same individual appears to be less satisfied in years where inequality is higher (Schröder, 2016). If empirical studies cannot find clear indications that redistribution of income leads to an overall improvement of quality of life, the support for this kind of political measures should be searched for elsewhere. Another general endeavour of governing bodies is the reduction of crime within society. Increasing the police force and tightening the punishment for criminal acts are mere examples of obvious measures that could lead to the desired decrease of crime rates. The problem with these examples is that they only ‘scare off’ potential criminals.

(4)

That is one end of the spectrum to solve crime problems. A different way to approach this issue is to look at the incentive of individuals to resort to criminal behaviour. If this incentive is taken away, the need to scare off potential individuals that would resort to crime decreases as well. In this field of criminology, E. Britt Patterson (1991) researched the link between relative poverty, the discussed income inequality, and crime rates. In the study concerning 57 small social areas in Rochester, New York, St. Louis, Missouri and Tampa-St. Petersburg, Florida (Patterson, 1991). He finds that both absolute poverty and relative poverty relate to crime rates, but the relationship is conditional on the type of crime (Patterson, 1991). Even though these results don’t imply a clear relationship between income inequality and crime rates due to the conditionality, it is interesting to see if income redistribution could serve as an instrument to affect crime rates. This is exactly what the research of this thesis is about. In this study, we want to see what the relationship is between income inequality and crime rates within a different society than the previous discussed study of Patterson. Our region of choice is the Netherlands. To be more specific, we use the characteristics of a large sample of the municipalities of the Netherlands in the year 2014 to see if we can find a relationship between the two discussed variables. The research includes the examination of the relation between income inequality and overall crime rates, but also specific types of crime, to see if there are any differences in the relation. The main research question of this research is: Is there a significant relation between municipality-level income inequality and crime rates within the Netherlands? Results of the multiple analyses can contribute to the understanding of income redistribution effect on crime rates and serve as a motivation to Dutch governing measures towards the reduction of crime rates. As crime victimization impacts unemployment rates, occupational and parenting skills among others (Hanson et al., 2010), decreasing this likelihood should be a goal of the government. Potentially, through a relationship between income inequality and crime rates, a future empirical study might be able to link income inequality to quality of life. To be able to state anything about the relationship between income inequality and crime rates on a municipality-level a cross-sectional regression will be used. The crime rate in the municipality will serve as the dependent variable, while the Gini

(5)

coefficient, explained in the methodology section of this thesis, of each municipality will is our measurement of income inequality. Control variables for average income, municipality size, unemployment and province are used to explain differences in crime rates that are unrelated to the income inequality. The reasoning behind the choice and definition of these variables will be made clear in the following segments of the thesis. Firstly, the existing literature concerning the relationship between income inequality and crime rates is examined. A comparison between the similarities of this thesis and previous research helps in constructing a theory that motivates the choice of variables and hypotheses. It will be discussed in which manners the research of this thesis differs from previous studies and a more precise specification than as discussed before will be given towards the potential contribution of the findings. Subsequently, the manner of research is discussed. This includes the estimation model and its variables, which will be explained, including the assumptions and restrictions of the model. Following, the origin of the used data, and the manner it is used in the model, is discussed to contribute to the reliability of the research at hand. After the explanation of the model, an interpretation of the results that follow from the cross-sectional regressions is given. Results will be compared to findings from comparable literature to see the similarities and deviations. In the conclusions section of this thesis, implications of the results and the used methods are discussed. At the end, a suggestion towards future research that might lead to a better understanding of the topic is given.

Literature review

In existing literature, there have been multiple researches in the field of relations between income and crime figures. One of the main discussions is if crime rates are more associated with absolute poverty, relative poverty or even both ( Patterson, 1991). A lack of sources to fulfill primary needs may give individuals an incentive to resort to criminal behavior (Patterson,1991); if for example medicine or rent is not affordable. People may feel the need to fall back to criminal activity when they aren’t able to reach levels of wealth that are considered to be normal in their community; what we could consider as their relative poverty compared to other community members (Patterson, 1991).

(6)

In his paper ‘Income inequality, race and place: Does the distribution of race and class within neighborhoods affect crime rates?’ from 2007, John R. Hipp discusses the theory behind the potential relationship between income inequality and crime rates (Hipp, 2007). Based on the routine activities theory, the amount of potential targets of property crime (the wealthy) and the amount of potential motivated offenders (the poor) and the absence of guardians (police officers or neighborhood watchers) together will result in an increase in the amount of crime in a neighborhood (Hipp, 2007). From this, it is concluded that an increase in income inequality leads to a higher number of potential targets and motivated offenders (Hipp, 2007). Based on this theory, a rise in (property) crime rates is to be expected as income inequality rises. Another theory that is discussed in the paper to explain a relation between income inequality and crime rate is the relative deprivation theory. The main aspect of this theory is that individuals compare themselves to others within their reference group (Hipp, 2007). The likelihood of an individual to resort to criminal behavior would increase if they feel that they are falling short compared to the people they compare themselves to (Hipp, 2007). An issue that arises is the identification process of this so-called reference group. It is possible that individuals compare themselves to all coresidents in their neighborhood, but it also considered plausible that others of their own racial or ethnic group forms the community that they mostly relate to (Hipp, 2007). In a paper by Kennedy et al from 1998, the relationship between firearm violent crime and income inequality is investigated. They find a strong relation between the inequality and violent type of crime (Kennedy et al., 1998). They discuss that a bigger gap between the rich and the poor leads to a loss in social capital, which in turn could lead to a rise in violent behaviour. The discussed study concerns an environment that is very different from the case of the Netherlands. For starters, we have to take into account that firearm access is relatively easy within the Unites States. Even though not all criminal activities are registered as firearm crimes, the fact that there is easy access to firearms could increase the likelihood of committing criminal acts. An individual could feel reassured by wearing a firearm, without even having to use it in the actual criminal act. This environment of easy access does not compare to the situation in the Netherlands, where stricter laws concerning firearms are enforced.

(7)

To see what can be found in the literature concerning the relationship between income inequality and crime figures in comparable countries, we turn to Europe. Comparable studies have been done in Sweden (Nilsson, 2004) and Belgium (Hooghe et al., 2011). Findings from the Sweden-based research state that the proportion of relatively poor relate strongly to overall and specific crime rates (Nilsson, 2004). A second important conclusion is that unemployment rates relate strongly positive to property crimes. Lastly, Nilsson not only considers the effect of income inequality on property crimes, but also on crimes that involve violence. It is concluded that inequality relates very differently to these types of crimes than to property crimes (Nilsson, 2004). The estimated coefficient is small, insignificant and even negative when the group that’s considered poor is enlarged (Nilsson, 2004). In the study that uses data on Belgium crime figures, the Gini coefficient is used to measure income inequality (Hooghe et al., 2011). The main findings are that income inequality has a significant positive effect on property crime rates, but a significant negative effect on violent crime rates (Hooghe et al., 2011). The same negative sign was found in the research done by Nilsson in 2004. The researchers state that the latter conclusion contradicts their expectations. They give a possible reason for this in the form of characteristics of the Belgian welfare system. Since the system provides the lower parts of the distribution with a certain income level, there is no way for the inequality to increase due to a fall of income at this lower limit. On the contrary, as income is not restricted by an upper limit, a larger inequality could only arise through income changes in the richest part of a community. It is indeed concluded that there is a positive correlation between median income and income inequality (Hooghe et al., 2011). This means that wealthier municipalities show larger inequalities that poorer ones. This is an interesting finding to take into account for the research in this paper, as the Dutch welfare system shows strong familiarities to the Belgian system. Another conclusion from the paper comes forth from the results concerning population density. A significant positive relation is found between crime and population density (Hooghe et al., 2011). The researchers state that this confirms the theory that criminal activity seems to significantly higher in cities. Addition to existing studies

(8)

The study in this paper is largely comparable to the studies held in Belgium and Sweden as discussed before. The manner in which this study differentiates itself from previous research is the expansive data that is used collected by the CBS (Centraal Bureau voor de Statistiek), a Dutch governmental institution that collects a divers set of statistics about the Netherlands. Information on income inequality, average income and unemployment is given on a municipality-level, which implies that the differences between these, relative, small groups of households can be investigated. Next to that, the research is done in an environment that hasn’t been used in existing literature review, potentially leading to new insights on the relation between income inequality and crime rates.

Methodology

In order to research the relationship between income inequality and crime rates on a municipality-level within the Netherlands, an analysis has to be done on the gathered data from the CBS. For this, we will use a cross-sectional regression. This means that data is used on different entities for a single time period for our regression (Stock & Watson, 2015). In the case of this research, the entities are the municipalities of the Netherlands in 2014 as decided by the municipal division. It is important to specify this, as municipalities merge and arise throughout the years. This means that the amount of municipalities differ by year. In the case of this research, where time is not variable (as we only use data from the year 2014), there exist 403 municipalities within the Netherlands. Later, during the discussion of the data set, we will see that data isn’t available on all these 403 municipalities. This issue comes forth out of the fact that either information wasn’t shared with the CBS by the concerning municipality, or the CBS themselves decided that the data isn’t trustworthy (CBS, 2016). The reason behind choosing the data from 2014 is a practical one; the Gini coefficients of the municipalities is the only part that isn’t publicly available and was the only set on municipality-level income inequality that was received from the CBS on request. To give a clearer view on the variables that will be used in the regression, an explanation of each individual variable follows and what its addition is to our model.

Motivation of variables

(9)

First off, the dependent variable; which is the measure of the level of crime within a municipality. The variable is defined as the amount of registered crimes per 1000 inhabitants of the concerning municipality. The crime rate is taken, as absolute numbers of crime will most likely be larger in bigger municipalities. Therefore, the relative crime rate is used. It has to be noted that this number only embodies registered crime; possibly, not every act of crime is registered. This could be due numerous factors, of which a few could considerably impact the representation of the number. For instance, a victim of sexual assault might not go to the police to report the incident out of shame. Another case could be that the victim has a relationship with the culprit (family, partner or friend) and wishes to solve the problem on his or her own. It has to be taken into account that the numbers as collected from the database do not represent all actual crimes that are committed. The main explanatory variable that is used is the measure of income inequality. In the case of this thesis, this is the Gini coefficient. To clarify the meaning of this coefficient, a comprehensive explanation follows. The Gini coefficient is a measure of the spread of income among a population. The larger the income gap between the richer and poorer, the higher the coefficient is. To give a numerical value to this coefficient, the Lorenz curve is used, which represents the income spread. The Lorenz curve (represented by the red line) as above is based on artificial data, purely for clarification. It is easy to see how income is distributed among the population.

(10)

In this case, the poorest 20% of the population hold slightly less than 10% of the total income earned by this population. On the other hand, the richest 20% of the population receive about 40% of the population. The blue line represents the case of full equality. Every same-size part of the population holds an equally sized share of the total income. If a larger inequality would persist, the red line would bend outwards compared to the full equality line. This implies that the poorest part of the population holds an even smaller share, and the richer portion, obviously, a larger share. To assign a value to the Gini coefficient, the area between the line of full equality and the Lorenz curve is divided by the whole area underneath the full equality line. The larger the area between the lines, as inequality rises, the larger the Gini coefficient will be. From this, it is concluded that the Gini coefficient will always have a value between 0 and 1, as it represents the relative size of the blue area (as shown above) compared to the blue and red area combined. Closer to 0 implies a more equal distribution, closer to one means higher inequality. A limitation of using this variable is that the Gini coefficient only represents the income inequality for the whole municipality, but does not show the actual spread of income within the community. It is not possible to conclude in which part of the spectrum the inequality persists. A selected amount of independent variables is used to control for effects on the crime rate that isn’t explained by the income inequality.

(11)

The rate of unemployment is an important variable for the research. As is concluded from the literature review, unemployment seems to significantly impact crime rates across all discussed studies. In this research, the unemployment as a percentage of the labour force on a municipality-level is used as the variable. It has to be taken into account that the unemployment rate may relate to the income inequality, as unemployed individuals are more likely to persist near the lower border of the income distribution. Another variable that is used, is the number of households within the municipality. As concluded by Hooghe et al. (2011) and discussed in the literature review, crime mostly seems to be an urban phenomenon. To control for this, population density isn’t used, but the number of households are taken into account. Cities have a relative large amount of households compared to other municipalities, so the number of households as a variable is similar to the population density in a municipality. Next to that, the average standardized income of the municipality is included. Standardized income implies that the income has been corrected for differences in the size and composition of the households. This variable is part of the model, because there might be a difference in crime rates as municipalities are on average richer than other municipalities. Possibly, for the case of property crime, wealthier municipalities form a more interesting target. Concluding, the provinces of the Netherlands will act as binary variables to see the impact of the location of a specific municipality. The province of Zeeland will be excluded from the model, to avoid the so-called dummy trap. If all provinces would be inserted into the model, perfect multicollinearity arises. This means that each individual dummy regressor is a perfect linear function of the other dummy regressors. This would make it impossible to compute the estimator, as in the mathematical process, a division by zero is attempted, which is impossible (Stock & Watson, 2015).

Model and use of variables

For the actual data analysis, a OLS multiple regression model is used. The regressions are computed with the use of StataSE v.13, a statistical application that automatizes many calculations to make regressions less labour intensive.

(12)

As the collected data is of the cross-sectional type, the usage of the OLS multiple regression model fits for the research of this thesis (Stock & Watson, 2015). The model is as follows. 𝐶𝑟𝑖𝑚𝑒𝑅𝑎𝑡𝑒 = 𝛽!+ 𝛽!𝐺𝑖𝑛𝑖 + 𝛽! 𝑈𝑛𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡 + 𝛽!𝐴𝑣𝑒𝑟𝑎𝑔𝑒𝐼𝑛𝑐𝑜𝑚𝑒 + 𝛽!𝐻𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑𝑠 + 𝛽!𝐷𝑟𝑒𝑛𝑡ℎ𝑒 + 𝛽!𝐹𝑙𝑒𝑣𝑜𝑙𝑎𝑛𝑑 + 𝛽!𝐹𝑟𝑖𝑒𝑠𝑙𝑎𝑛𝑑 + 𝛽!𝑂𝑣𝑒𝑟𝑖𝑗𝑠𝑠𝑒𝑙 + 𝛽!𝑁𝑜𝑜𝑟𝑑𝐻𝑜𝑙𝑙𝑎𝑛𝑑 + 𝛽!"𝑍𝑢𝑖𝑑𝐻𝑜𝑙𝑙𝑎𝑛𝑑 + 𝛽!!𝑁𝑜𝑜𝑟𝑑𝐵𝑟𝑎𝑏𝑎𝑛𝑡 + 𝛽!"𝐿𝑖𝑚𝑏𝑢𝑟𝑔 + 𝛽!"𝑈𝑡𝑟𝑒𝑐ℎ𝑡 + 𝛽!"𝐺𝑟𝑜𝑛𝑖𝑛𝑔𝑒𝑛 + 𝛽!"𝐺𝑒𝑙𝑑𝑒𝑟𝑙𝑎𝑛𝑑

To test for misspecification of the model, the Ramsey RESET test is used. It tests whether non-linear combinations of the fitted values help explain the dependent variable. When this is the case, the model is mis-specified. Next to that, we will test for the heteroscedacity of the error terms. Heteroscedacity implies that the deviations from the fitted values are different in sub-populations. We investigate this by computing the Breusch-Pagan test. If heteroscedacity seems to be case, robust standard errors are used for our regression model to further the trustworthiness of the results. In the previous segment, it was already motivated why and which variables are used. Now, we will take a closer look at how the variables are mathematically designed for usage in the model. CrimeRate: This is the amount of registered crime for every 1000 inhabitants of the municipality. Data is collected from the online database StatLine of the CBS. Gini: These are the Gini coefficients on a municipality-level for 2014. The values are multiplied by 100, so all values will be between 0 and 100. This is done to simplify the interpretation of the estimated coefficient 𝛽!. This variable is used as the measure of income inequality in the model. This data is received from the CBS on request. Unemployment: This is the unemployment rate (as a percentage) in the municipality in 2014, compared to the total labour force of the specific municipality. This data is as well collected from the StatLine database of the CBS. AverageIncome: The standardized income of households in the municipality in 2014, divided by 1000. The logarithm of this value is taken, as relative income changes are more interesting for our research than absolute changes. The data is gathered from CBS. Households: This is the amount of households in a municipality in 2014, divided by 1000. For similar reasons as in the case of the average income, the logarithm of all values is taken. Again, data is received from the online database StatLine.

(13)

The provinces: These are binary variables, so-called dummy variables. If a specific municipality lies in a certain province, the binary variable of the corresponding province variable has the value 1. If this doesn’t hold, it has the value 0. Lastly, this data is gathered from the online database of the CBS as well.

Assumptions of the model

There are certain least squares assumptions concerning the econometric model used that need to hold. The four most important ones, as given by Stock & Watson (2015) are as followed. Briefly will be discussed how they relate to the used data in this thesis. 1. X is exogenous, which means that the expected conditional mean of the error term is 0 for every given X. 2. The second assumption of the least squares model assumes that all individual variables are independently and identically drawn. As we have data on almost the full set of municipalities, due to spatial relations between municipalities (as crime activity in one municipality spills over to a municipality in the proximity), the independency might not hold. We control for provinces to correct this somewhat, though the limitation of this is that provinces are rather large and might not be considered the direct proximity for an individual municipality. 3. Large outliers are unlikely. As will follow from the dataset that is used in the upcoming discussion, it will be clear that all municipality crime rates are within the range of 0 to 125 for all crime rates (rates are given as a per 1000 inhabitants ratio) and municipality Gini coefficients all are between 0.20 and 0.45. 4. No perfect multicollinearity. This holds, as there is no full linear relation between one variable and another. By excluding the province of Zeeland as a dummy variable from the model, further risk of perfect multicollinearity is avoided.

Hypotheses

As discussed before, the aim of this research is to test for the influence of income inequality on crime rates on a municipality-level within the Netherlands. From literature, the following hypotheses are formed: 𝐻!: 𝛽! = 0 𝐻!: 𝛽! > 0

(14)

Where 𝛽!embodies the coefficient as specified in the model, i.e. of the income equality effect. A t-test is used to see whether or not the coefficient of the Gini coefficient is significantly different from zero. Results on this will follow in a later segment. To see whether the impact of the concerning provinces is significant, a F-test is used on joint hypotheses; this gives an indication if there exists any significant effect among the provinces.

Dataset

All collected data is retrieved from the online database StatLine of the CBS. In the upcoming segment, a brief summarization of the data is discussed. As will be made clear, there are some differences in the number of observations between the different types of crime. Certainly in the case of violence crime rate, this will limit in how well interpretations of the results are applicable on all municipalities.

Variable Observations Mean Standard

Deviation Minimum Maximum

Overall Crime rate 392 48.3 18.1 0 124.4 Unemployment 392 6.5 1.14 4.5 12.6 Gini coefficient 392 0.262 0.032 0.22 0.44 Average income 392 24.7 2.54 18 38.4 Households 392 18.9 33.1 0.5 428.1 Recall that unemployment is given as a percentage, while average income and households is divided by 1000. It helps to confirm what the overall crime rate, as stated by the database of the CBS, means. It includes any type of registered crime, ranging from burglary to sexual assault and from traffic offences to drug crimes. There are a few interesting values to be taken note of from the table. As is clear from the row of the overall crime rate, the minimum lies at 0. This implies that there exists at least one municipality where there was no registration of any crime activity during 2014. This is not an error in the data, as there actually do exist municipalities for which this holds. For example, on the isle of Schiermonnikoog, which forms a municipality on its own, there was no official registration of any case of crime activity at all.

(15)

The highest value, an average of 124.4 registered crimes per 1000 inhabitants, is assigned to the municipality of Amsterdam. This is the biggest city in the Netherlands, with over 400.000 households. As was discussed in the literature review, this relates well to the theory that criminal acts are mostly seen in urbanized areas. In the next case, an overview is given of the data on property crime rate.

Variable Observations Mean Standard

Deviation Minimum Maximum

Property Crime rate 381 24.5 11.6 0 80.1 Unemployment 381 6.5 1.14 4.5 12.6 Gini coefficient 381 0.261 0.032 0.22 0.44 Average income 381 24.7 2.51 18 38.4 Households 381 19.2 33.3 0.5 428.1 Obviously, the values of the independent variables aren’t that different from the dataset involving overall crime rate. The most important difference is that this collection of data focuses solely on property crimes. This number does not only include registered offences of theft and burglary, but also scamming, money laundering, forgery, extortion and fencing.

Variable Observations Mean Standard

Deviation Minimum Maximum

Violence Crime rate 290 7.6 4.4 0 72 Unemployment 290 6.7 1.17 4.5 12.6 Gini coefficient 290 0.262 0.031 0.22 0.44 Average income 290 24.5 2.35 18 37.8 Households 290 23.46 37.3 2.6 428.1 The term violence crime rate, as used in the table, is a bit more ambiguous. The data that is used from the CBS includes acts of arson, damaging public and private goods and crimes against public order and public authority. It does not include assaults, sexual offences or homicide against persons. Most noteworthy from this table is that the number of observations of significantly lower than in our other cases. This comes forth out of the fact that there

(16)

were many, mostly smaller sized, municipalities on which no data was available. As the number of observations falls, and the municipalities that are removed aren’t randomly chosen, but are mostly smaller municipalities, it has to be taken into account that this influences how trustworthy the results are that we gather from this dataset.

Results

The results section consists out of three parts. In all sections, a regression is done with the usage of crime rate, Gini coefficient, unemployment rate, number of households, average income and all provinces as dummies, except for Zeeland. In the first part, the results from the regression on property crime rate data is interpreted, in the second part the same is done for violence crime rate data, and in the concluding part a similar interpretation is given for overall crime rate data. It should be notified that these are merely interpretations of our results. The level of significance of variables and the implications of the estimated coefficient of the independent variables for the dependent variable is given and how we compare this to literature. On the other hand, any concluding answers towards the research question will be given in the conclusion section of this thesis. As the interpretation of estimated coefficients will be extensively discussed in the results section on property crime, and this discussion also applies for the interpretation of the other two sections, the discussion will be more briefly in the results section of overall and violence crime rate.

Property crime rate

Initially, we will look at how the crime rate moves as income inequality rises. As this is purely for an indication how crime rates behaves compared to different of our variables, we will look at the graphical representation of the property crime rate.

(17)

From the graph, it is clear that there is a cluster of observations around the area where the Gini coefficient, our measure of income inequality, is between 0.22 and 0.27. What this implies, is that many municipalities have a comparable income inequality within their community. The more extreme values seem to arise as income inequality rises. There are a few observations of significantly higher income inequality, and the corresponding crime rates seem to be at a higher level than the average of the cluster that we find in the middle.

(18)

When we plot the same property crime rate data with the average income of households in a municipality, the highest crime rates are seen in municipalities that are in the lower segment of the distribution. On the other hand, municipalities with average income levels above the 30,000 euro, crime rate seems to be higher compared to the cluster. This might connect to the fact that when a municipality is richer, there are more targets of property crime, as stealing from the rich has a potential higher pay off than stealing from the poor. Comparing property crime rates with the unemployment in municipalities shows an even clearer trend. As unemployment rises, crime rates tend to be higher as well. The outlier when it comes to unemployment is Rotterdam, with an unemployment percentage of 12.6% of the total labour force. Before we come to the regression results, we can state that that from a graphical analysis there are some indications that a relation might exist between the dependent and independent variables.

Regression results for property crime rate

First off, robust standard errors are used in the regression, as the result from the Breusch-Pagan test rejects the null hypothesis of homoscedacity. This means that heteroscedacity is in effect. In the results table of the regression, robust standard errors are used for estimations.

(19)

The result from the Ramsey RESET test, which tests if nonlinear combinations of the fitted values help explain the dependent variable is that we do not reject the nul hypothesis that combinations do not help explain the dependent variable. The F-statistic has a value of 1.46, and a p-value (probability) of 0.2252. This implies that we should not reject it, so we will not include further combinations of independent variables.

Property Crime Est. Coefficient Std. error F-test (provinces) Gini coefficient 1.236*** (.230) Unemployment 2.231*** (.572) Av. Income (log) -28.681*** (9.367) Households (log) 4.835*** (.831) Drenthe -4.969** (2.061) Flevoland -1.736 (3.255) Friesland -7.769*** (2.507) Overijssel -2.031 (2.562) Noord Holland 5.867*** (2.182) Zuid Holland 0.599 (2.081) Noord Brabant 6.340*** (1.942) Limburg 8.078*** (2.379) Utrecht 3.841* (2.293) Groningen -6.528*** (2.217) Gelderland 0.121 (1.819) R-squared: 0.5953 11.55 * = significant at 10% level **= significant at 5% level ***=significant at 1% level Very high significance is found for the effect of the income inequality, unemployment, average income and number of households. With over 99% certainty the effect is different from zero. Interpreting the effect of the Gini coefficient, we find an estimated value of 1.236. The sign of the estimation is positive, which implies that as income inequality rises, crime rates tend to rise as well. The meaning of this value is that as the Gini coefficient itself rises with 0.01 (recall that we multiplied by 100 for interpretation purposes), we

(20)

expect a rise of the property crime rate by 1.236. This compares to results found in comparable studies, like the one conducted in Belgium and Sweden. For unemployment a similar interpretation can be given. The value of 2.231 gives an indication that as unemployment rates rise within a municipality, property crime rates are expected to be higher. As the unemployment is given as the percentage number, the correct interpretation of this number is that property crime rates are expected to rise with 2.231 when the unemployment rate rises by one percentage point. The result supports the theory that when a higher share of individuals is in a jobless situation, the likelihood of criminal activity rises as well. It has to be stated that in our analysis we cannot say anything about who is the actual culprit of the rise of the crime rate, but the relation between the two variables is clear nonetheless. Looking at average income, we see a strongly significant and negative estimation. Since we took the logarithm of the income and not of crime rate, the estimation is slightly harder to interpret. When the average income rises with a fixed percentage, for example 10%, the expected rise in property crime rates is -28.681 times the logarithm of 1.1, i.e. -1.19. So, to summarize, a rise of 10% in income would come with an expected fall of -1.19 in crime rate. We do not find in our case that higher average income leads to a higher rate of property crime. This conflicts with some parts of the discussed literature. As the log of the number of households is taken as well, a similar interpretation of this estimation is at hand. The value of 4.835 implies that at 10% rise in income, an expected rise of the crime rate by 4.835 times the log of 1.1, i.e. 0.2 would follow. This result does compare well with the discussed literature, as in the paper by Hooghe et al. (2011) it was stated and reconfirmed through empirical research that crime seems to be an urban phenomenon. Lastly, we combine the interpretation of the provinces. We can see that there is a considerable difference in the significance of the effect of the numerous provinces. The F-test used to see if the null hypothesis that all of the coefficients of the provinces are zero shows that we should reject this, as we would expect from the high significance seen among some of the provinces. Most interestingly to pick out of the set are the strongly significant and negative estimated coefficients of Friesland and Groningen. Municipalities tend to have lower crime rates when residing in these provinces, just due to the effect of their location. This

(21)

could be connected to the fact that the population density in these provinces is vastly lower, and there aren’t that many big cities. Remember that literature already suggested that crime tends to manifest itself mostly in cities. A more precise definition of the spatial relation in the model between different municipalities could add to the trustworthiness of these results, which should be considered for further research.

Regression results on overall crime rate

We use the exact same methods for the regression on overall crime rate as on property crime rate, which means that the Ramsey RESET test and Breusch-Pagan test are computed to check for combinations of explanatory variables that predict the model better and homoscedacity respectively. In this case, we get similar results to the test conducted on the data of property crime rates. A p-value of 0.6131 for the Ramsey RESET test gives us little reason to reject that the current model should be adjusted and the Breusch-Pagan test gives us a p-value of practically zero, which implies heteroscedacity again. So, again, in Stata, the option robust is used for a better estimation of the model. The reason overall crime rate and property crime rate are discussed in sequence, is because of their similarity in results (as will be clear from the following table)

Overall Crime Est. Coefficient Std. error F-test (provinces) Gini coefficient 2.046*** (.212) Unemployment 5.854*** (.705) Av. Income (log) -37.852** (9.146) Households (log) 5.712*** (.876) Drenthe -15.495** (4.386) Flevoland -8.937** (6.374) Friesland -18.994*** (3.980) Overijssel -12.001*** (3.725) Noord Holland 4.584 (3.515) Zuid Holland -4.325 (3.416) Noord Brabant 2.023 (3.366)

(22)

Limburg 2.204 (3.667) Utrecht -3.844 (3.765) Groningen -6.528*** (4.021) Gelderland 0.121* (3.422) R-squared: 0.6662 15.99 * = significant at 10% level **= significant at 5% level ***=significant at 1% level Interpretations are, like stated before, comparable to the situation of the case of property crime rates. The estimated coefficient for the Gini coefficient, unemployment, average income and number of households are again strongly significant. The biggest difference is that average income is only significant until a 95% confidence level, whereas for the case of property crime rates, it was even significant for the 99% level. The effect of income level does seem to be considerable higher than solely for property crimes, comparing the estimated 2.046 in the overall crime rate situation compared to the 1.236 coefficient for property crime rate. This would imply that as the income inequality rises, as defined by the Gini coefficient, overall crime rate would rise even faster. This could possibly imply that there are more types of crimes included in this overall crime rate that positively relate to the income inequality on a municipality level. This, too, would be an interesting topic for further research. For average income, we see a strong significant negative sign again. Crime rates, overall and of the property type, seem to fall as the average income of a municipality rises. Possibly, richer municipalities as a community or the individuals themselves may take more preventive measures like neighbourhood watches and installation of cameras to prevent crime. It has to be noted that this is just a thought, and isn’t controlled for in the model. A possible further research could take this into account. Lastly, the set of provinces again have mixed significances and sign of their estimation. Comparable to our previous discussion, Friesland, Groningen and Drenthe show strong significance and negative relations to the crime rate. It appears to be that crime rates tend to be lower here in multiple cases. We do not find any strong significant and positive relationships between a province and the crime rate. Only in the case of Gelderland, there is a significance at the 90% confidence. This effect only implies that if a municipality resides within Gelderland, crime rates tend to be 0.121 higher. Even though other estimated coefficients are higher,

(23)

like in the case of Noord-Holland and Limburg, since there is no clear reason to reject the hypothesis that the estimated coefficient is different from zero (because of the high p-values) we cannot say there is a clear relationship.

Regression results on violence crime rates

In the last case, we discuss the results on the violence crime rates. The reason we discuss this one last, is because results deviate strongly from the other two discussed crime rates. We already looked at the limited amount of observations that are available, and how it might impact the trustworthiness of the results. As will become clear from the table of results, the used model has a very hard time to predict the crime rates. Again, same results from the Ramsey RESET test and the Breusch-Pagan test, with p-values of 0.0593 (F value of 2.51) and 0.000 respectively. This implies we will not adjust the model, and use the robust option because of heteroscedacity.

Violent Crime Est. Coefficient Std. error F-test (provinces) Gini coefficient -0.271 (.053) Unemployment 0.419 (.367) Av. Income (log) -3.183 (2.311) Households (log) 1.094 (0.806) Drenthe -.940 (.946) Flevoland -.288 (.962) Friesland -1.895 (1.097) Overijssel -1.239 (3.856) Noord Holland .670 (.768) Zuid Holland -.959 (.695) Noord Brabant -1.245 (.703) Limburg -.650 (.855) Utrecht -.686 (.767) Groningen -.968 (1.306) Gelderland -1.391 (.745) R-squared: 0.1204 2.58 * = significant at 10% level **= significant at 5% level ***=significant at 1% level

(24)

As becomes clear from the table above, none of the estimated coefficients are significant at the used levels of confidence. Next to that, we see that R-squared is far smaller than in the case of property crime rate and overall crime rate, implying that the model has a hard time to predict the variances. So, in this situation, our used model does not do a sufficient job to predict the observed outcomes. Most likely, the model doesn’t fit the data very well, so stating anything based on these results is relatively hard. What is does give us, is a nice comparison with the results received from the two previous regressions. The fact that results differ so strongly, implies that the used independent variables might be able to predict crime rates very well for certain types of crime, but not for all. Apparently, when it comes to violent crime activities, different variables should be used to try and explain how these rates differ from municipality to municipality. This is somewhat comparable to the study by Nilsson (2004), where it was found that the estimated coefficient is small, insignificant and even negative for violent criminal acts. There is a resemblance with our results, as in this case we find a small, insignificant and negative estimated coefficient as well. Possibly, a different model should be used to make trustworthy estimations for these types of violent crime rates.

Conclusion

In our conducted research, we found similar results to the literature, concerning the research on overall and property crime rates. Strong significant relations are found for income inequality, unemployment and average income. This is as predicted by studying the existing literature. Next to that, we also see reconfirming results regarding the theory that crime is mostly an urban phenomenon, as was found by Hooghe et al. (2011). Our use of provinces give us a somewhat ambiguous result, though we can see significant relationships for certain provinces, mostly Friesland and Groningen, the most northern provinces of the Netherlands. On the other hand, if we consider the effects of the income inequality, unemployment, average income, number of households and the provinces for violent crime rates, it appears that the model isn’t able to predict the dependent outcomes of

(25)

the violent crime rates very well. This suggest that the model should be changed, as perhaps different variables do a better job at explaining how violent criminal activity arises. We do have to take into account that our sample of municipalities for this type of crime rate was significantly smaller and was influenced by decisions of the database builder (i.e. the CBS). This limits the trustworthiness of our results. Combining these two most impactful results gives us reason to believe that there is indeed a relation between crime and income inequality, but it has to be clearly stated that not all types of crime have a similar or clear relationship. The limitations of the research have to be kept in mind. The most important limitations are the reduced sample size of the violent crime rate and the fact that provinces used to find clusters of observations is a rather crude method to control for spill over effects. Nonetheless, results from this research might add to understanding how and where crime arises. Possibly, it helps political bodies to adjust regulations so that crime reduction can be achieved. Because of the use of data on a municipality level and the unique environment of the Netherlands, this research differentiates itself from previous studies. For further research, the most important suggestion this is to look into have a better way to consider spill over effects of crime in the proximity of the municipality. Next to that, it could be interesting to research how violence crime rates are better explained by an empirical research and econometrical model.

(26)

Bibliography

CBS. (2016). CBS Statline database. Retrieved on 20 05 2016 from http://statline.cbs.nl/Statweb/dome/?LA=nl Franko, W., Tolbert, C. J., & Witko, C. (2013). Inequality, self-interest, and public support for “Robin Hood” tax policies. Political Research Quarterly, 66, 923-937. Hanson, R., Sawyer, G., Begle, A., & Hubel, G. (2010). The impact of crime victimization on quality of life. Journal of Traumatic Stress, 23(2), 189-197. Hilton, R. (1958). The Origins of Robin Hood. Past & Present, (14), 30-44. Hipp, J. (2007). INCOME INEQUALITY, RACE, AND PLACE: DOES THE DISTRIBUTION OF RACE AND CLASS WITHIN NEIGHBORHOODS AFFECT CRIME RATES?.Criminology, 45(3), 665- 697. Hooghe, M., Vanhoutte, B., Hardyns, W., & Bircan, T. (2011). Unemployment, Inequality, Poverty and Crime: Spatial Distribution Patterns of Criminal Acts in Belgium, 2001–06. The British Journal of Criminology, 51(1), 1-20. Kennedy, Kawachi, Prothrow-Stith, Lochner, & Gupta. (1998). Social capital, income inequality, and firearm violent crime. Social Science & Medicine, 47(1), 7-17. Nilsson, Anna. (2004). Income Inequality and Crime: The Case of Sweden. Stockholm University, Department of Economics. 2004:3. Patterson, E. (1991). POVERTY, INCOME INEQUALITY, AND COMMUNITY CRIME RATES.Criminology, 29(4), 755-776. Schröder, M. (2016). How Income Inequality Influences Life Satisfaction: Hybrid Effects Evidence from the German SOEP. 32(2), 307-320. Stock, J. H., & Watson, M. W. (2015). Regression Analysis of Economic Time Series Data. In J. H. Stock, & M. W. Watson, Introduction to Econometrics (pp. 568-717). Essex: Pearson Education Limited. Zagorski, K., Evans, M., Kelley, D., & Piotrowska, R. (2014). Does National Income Inequality Affect Individuals’ Quality of Life in Europe? Inequality, Happiness, Finances, and Health.Social Indicators Research, 117(3), 1089-1110.

(27)

(51.37) (21.50) (7.99) _cons 67.731 55.383* 13.235 (3.59) (2.43) (0.70) ZHolland -4.325 0.599 -0.959 (3.67) (2.70) (0.77) Utrecht -3.844 3.841 -0.686 (3.24) (2.64) (3.86) Overijssel -12.001*** -2.031 1.239 (3.82) (2.50) (0.77) NHolland 4.584 5.867* 0.700 (3.31) (2.39) (0.70) Brabant 2.023 6.340** -1.245 (3.69) (2.60) (0.86) Limburg 2.204 8.078** -0.650 (3.54) (2.89) (1.31) Groningen -17.815*** -6.528* -0.968 (3.23) (2.42) (0.74) Gelderland -6.032 0.121 -1.391 (4.72) (2.89) (1.10) Friesland -18.994*** -7.769** -1.895 (4.33) (3.81) (0.96) Flevoland -8.937* -1.736 -0.288 (3.96) (3.11) (0.95) Drenthe -15.495*** -4.969 -0.940 (17.09) (6.80) (2.31) lavinc -37.852* -28.681*** -3.183 (1.45) (0.66) (0.81) lhuis 5.712*** 4.835*** 1.094 (0.97) (0.51) (0.37) nunemp 5.854*** 2.231*** 0.419 (0.42) (0.15) (0.05) bgini 2.046*** 1.236*** -0.027 b/se b/se b/se m1 m2 m3

Attachment

Data on all three regressions, where m1 is the regression for overall crime rate, m2 for property crime rate and m3 for violence crime rate.

Referenties

GERELATEERDE DOCUMENTEN

 Natalia Vladimirovna Chevtchik, the Netherlands, 2017 ISBN: 978-90-365-4384-2 DOI: 10.3990/1.9789036543842 Printed by Gildeprint, Enschede, the Netherlands, Cover design by

These voltages, given by G & C C , will be relayed back to the power supply (depending on the switching topology) source via an intrinsic body diode that is present inside

This will be broken down into a series of smaller investigation, where we set out to establish (1) whether subjects show evidence of having learned (or segmented) words

Overall, 2D-LC-UV/MS is a very powerful technique for the characterization of mAbs and ADCs since it allows automatization, a high resolving power can be achieved, it involves

Aus der Umfrage geht hervor, dass Schüler mit Deutschunterricht erwartungsgemäß mehr Kenntnisse über Deutschland haben, mehr Kontakt mit Deutschland und den Deutschen haben und

DNA samples were analysed for the presence of pathogens, using both the MagicPlex Sepsis Test (Seegene) and SepsiTest (Molzym), and results were compared to blood cultures..

The comparison of the simulated snow albedo evolution with the in situ measurements shows that the parameterizations adopted by Noah, BATS, and CLASS are only able to simulate an

Although such impacts were re- ported in the 1923 Great Kant¯o earthquake and the 1999 Chi- Chi earthquake, careless reconstruction in hazard-prone ar- eas and consequently huge