THE CRIME KUZNETS CURVE

SERIE DOCUMENTOS DE TRABAJO No. 155

Abril de 2014

THE CRIME KUZNETS CURVE Paolo Buonanno Leopoldo Fergusson Juan F. Vargas

The Crime Kuznets Curve

Paolo Buonanno∗ Leopoldo Fergusson† Juan F. Vargas‡

Abstract

We document the existence of a Crime Kuznets Curve in US states since the 1970s.As income levels have risen, crime has followed an inverted U-shaped pattern, firstincreasing and then dropping. The Crime Kuznets Curve is not explained by incomeinequality. In fact, we show that during the sample period inequality has risenmonotonically with income, ruling out the traditional Kuznets Curve. Our findingis robust to adding a large set of controls that are used in the literature to explainthe incidence of crime, as well as to controlling for state and year fixed effects. TheCurve is also revealed in nonparametric specifications. The Crime Kuznets Curveexists for property crime and for some categories of violent crime.

Keywords: Crime Kuznets Curve, Income inequality, Non-parametric Kernel anal-ysis.

∗Universita degli Studi di Bergamo, Department of Economics. Via dei Caniana 2 - 24127, Bergamo.E-mail: [email protected].†Universidad de los Andes, Department of Economics, Cra 1 No 18A–12, of. W812, Bogota. E-mail:

[email protected]‡Universidad del Rosario, Department of Economics. Calle 12c No. 4–69, of. 315, Bogota. E-mail:

[email protected].

mailto:[email protected]

1 Introduction

This paper documents for the first time the existence of an inverted U-shaped relationship

between crime and income within US states for the period 1970-2011. Crime increases

with per capita income until it reaches a maximum and then decreases as income keeps

rising. We provide compelling robust parametric and nonparametric support for this

“Crime Kuznets Curve” (CKC). The curve survives controlling for the standard socioe-

conomic and demographic determinants of crime, as well as including state and time

fixed effects that account respectively for potential time-invariant omitted variables and

for time shocks in crime that are common to all states (e.g. federal criminal law changes).

We rule out the obvious candidate explanation, namely that crime responds monoton-

ically to changes in income inequality and in turn inequality has an inverted U-shaped

relationship with income. This non-linearity is in fact the essence of the traditional

Kuznets Curve (KC), named after the seminal paper by Simon Kuznets (1955) who noted

a non-monotonic association between income inequality and economic growth. However,

we show that the CKC survives the inclusion of economic inequality as control. More-

over, we also establish that during our sample period inequality has been monotonically

increasing with income in most US states. Hence, the KC is not the underlying factor

explaining the CKC.

The relationship between income and crime has attracted the attention of social sci-

entists, notably economists and criminologists. The basic version of the economic model

of crime (Becker, 1968 and Ehrlich, 1973) suggests that rational agents would engage in

crime or illegal behavior as long the expected benefits offset the expected costs. Indeed,

for a given probability of apprehension and expected punishment, higher levels of income

increase the opportunity cost of crime of potential criminals, thus reducing the total time

devoted to criminal activities. However, this logic also implies that wealthier individuals

are more attractive criminal targets. This predicts a higher victimization of the rich by

the relatively poor. However, as suggested by Allen (1996) and Chiu and Madden (1998),

the rich may implement private protection strategies to tackle crime (e.g. alarm systems,

private security, etc.), thus offsetting the criminal incentives implied by the cost-benefit

calculation of the disadvantaged.

In short, the theoretical prediction regarding the relationship between income and

crime is ambiguous, and the answer is ultimately empirical. However, most theoretical

models fail to predict a non-monotonic relationship between income and crime, and thus

cannot account for the CKC. While the paper’s objective is to document the existence

and robustness of the CKC at the macro level, and not to uncover the features creating

this relationship, we discuss possible avenues for future research that could help uncover

these microeconomic underpinnings. Moreover, we show that the CKC exists for property

crime and for categories of violent crime that can be related to economic appropriation,

1

like robbery. Instead, the inverse-U pattern is less robust for violent crimes not nec-

essarily connected with economic incentives. This distinction, which is less salient in

the nonparametric analysis, may be important to help identify which causal mechanisms

underlie the CKC.

Both official country-level crime statistics and victimization surveys show that crime

rates have systematically fallen over the past decade in most of the developed world.

According to The Economist (2013), in the US the fall began around 1991, in the UK

around 1995, and in France and other Western European countries crime rates have

fallen since about 2001. While this pattern holds for either property and violent crime,

the decline stage has been relatively more pronounced for property crime.1 The fall in

crime rates in developed countries has followed a relative long period of steadily increasing

rates, as documented by Buonanno et al. (2011).

A crucial concern in interpreting the evolution of crime rates from a cross-country

perspective is related with the comparability of crime data. While in other critical pol-

icy sectors data are collected systematically and according to uniform standards across

countries, the same is not true for crime rates. This has to do in part with the fact that

crime is a social phenomenon that is hidden by its illegal nature and often goes under

denounced. This complexity has been recognized by several scholars, among others Aebi

(2004), Dills et al. (2008), Goldberger and Rosenfeld (2009) and Durlauf et al. (2010).

We focus in this paper in the case of the US. This has several advantages like mitigating

some of the concerns regarding the institutional heterogeneity and data comparability

present in cross-country studies. Standards for crime measurement in the US are held at

the federal level, making the data homogenous. Besides, we believe that the US provide

a very interesting framework to study the relationship between income and crime. It is

well known that the US experienced an unexpected drop in crime rates starting in the

1990s, after a period of dramatic growth in both violent and property crimes (Levitt,

2004).

The rest of the paper is organized as follows. Section 2 describes the data. Section 3

discusses the empirical strategy both for the parametric and the nonparametric analyses.

Section 4 reports the results that lead us to claim the existence of a CKC in the US.

Finally Section 5 concludes.

1For instance, in New York City the annual number of car thefts has fallen by 93% over the last 20years. Similarly, in England and Wales reported car thefts drop from 400,000 in 1997 to just 86,000 in2012 (The Economist, 2013).

2

2 Data

We constructed a balanced panel with annual observations at the state level for 50 US

states over the period 1970-2011.2 As for the dependent variable, we consider the seven

felony offenses recorded in the FBI’s Uniform Crime Reports. In particular, we distinguish

between different forms of property crime (burglary, larceny and car theft) and violent

crime (murder, assault, robbery and rape). The distinction between property and violent

crime is motivated by the observation that property crime is more likely to depend on

economic motivations than violent crime.

Panel A of Table 1 reports the descriptive statistics of the crime outcomes, for which

rates are computed in terms of 100,000 inhabitants. Property crime rate is on average

almost 10 times as prevalent as violent crime. Within property crimes the most prevalent

is larceny and the least prevalent is car theft. In the case of violent crimes the highest

incidence corresponds to assault and the lowest to murder.

The main independent variable of interest is income. Real GDP per capita at the

state level for the entire sample period is obtained from the US Bureau of Economic

Analysis, and the base year is 2009.

We also collected a large set of standard socioeconomic and demographic variables that

are likely to be correlated with crime rates. Of these, the most important is the income

Gini coefficient.3 It is important to control for inequality for two reasons. One the one

hand inequality has been theoretically and empirically linked with criminal behavior (see

for example Becker, 1968 and Buonanno and Vargas, 2013 respectively). On the other

hand, the traditional KC features an inverted U-shaped relationship between income

and inequality and hence inequality is an obvious confounder when exploring a potential

non-linear relationship between income and crime.

We also control for population density, the state employment rate and the distribution

of the state male population across several age categories. Controlling for population

density is important as it is well documented that the incidence of crime is higher in

densely populated areas than in sparsely populated areas (Glaeser and Sacerdote, 1999).

Several reasons may explain this fact: in a dense area the pool of potential victims

is larger, criminal networks are more developed and criminal activities may experience

economies of scale due, for example, to lower search costs. Hence we expect larger

population density to be associated with higher crime rates.

In turn, controlling for the employment rate accounts for labor market opportunities

that, according to the benchmark economic model of crime (Becker, 1968 and Ehrlich,

2Consistent with previous empirical research on crime (e.g. Donohue III and Levitt, 2001; Raphaeland Winter-Ebmer, 2001 and Lin, 2008) we exclude District of Columbia from our analysis because itconstitutes an outlier.

3Inequality data are available for download at Mark W. Frank’s website(http://www.shsu.edu/∼eco mwf/inequality.html) and the data available represent an updatewith respect to the data used in Frank (2009).

3

1973), reduce the amount of time devoted to criminal enterprises. This occurs for two

reasons. First, the expected returns from legal activity increase if the probability of

being employed is higher. Second, given a downward sloping labor demand curve, more

employment is associated with a higher wage rate. In both cases the opportunity cost of

crime increases, thus discouraging this activity. We then expect higher employment rates

to be associated with lower crime rates. This prediction has been confirmed by recent

empirical scholarship that employs panel data techniques at the state or regional level

(e.g. Raphael and Winter-Ebmer, 2001; Gould, Weinberg and Mustard, 2002; Lin, 2008;

and Fougere, Kramarz and Pouget 2009, Oster and Agell, 2007).

Finally, controlling for the state-level share of male population across different age

brackets serves the double purpose of accounting for the well-documented age-crime pro-

file as well as the male crime bias. These controls ensure that estimates are not contam-

inated by age and gender differences across states and overtime.

Panel B of Table 1 reports the descriptive statistics of the independent variables. The

real income per capita, measured in thousands of dollars of 2009, is on average $28,553.

The average value of the Gini coefficient of income inequality is 0.54. On average the

employment rate over the state population is 55% and there are about 169 individuals

per squared mile. The fraction of males is decreasing with age.

Since it is impossible to control for the myriad of factors that may potentially affect

crime our list of controls is likely to be incomplete. For this reason, in all regressions

we include both state and year fixed effects. The former account for any remaining

heterogeneity across states as long as it is time invariant. The latter absorbs any shock

that may affect all states simultaneously at any moment of time. The empirical approach

is explained in more detail in the next section.

3 Empirical strategy

Kuznets (1955) based his inverted U hypothesis on the observed time series evolution of

three developed countries (US, UK and Germany), as well as on cross sectional obser-

vations of three developing nations (India, Ceylon and Puerto Rico). Since the work of

Ahluwalia (1976a and 1976b) the empirical relevance of the KC has been investigated in

a more systematic way by several scholars, leading however to little consensus. The exis-

tence or not of a KC depends on the nature of the empirical investigation, whether based

on cross-sectional country-level variation or based on the time series evolution within

individual countries (see Gallup, 2012 and Kanbur, 2012 for recent reviews of the KC

literature).

To estimate whether there is an inverted U-shaped relationship between income and

the incidence of crime we run the following benchmark specification in which we include

4

a quadratic polynomial of income:

Crimeit = αi + γt + β1Incomeit + β2Income2it + δXit + εit, (1)

where Crimeit is either of the measures of the crime described in section 2 in state i at

time t; Incomeit it real GDP per capita in state i at time t; Xit is a vector containing the

state-specific time-varying controls discussed in section 2 and αi and γt are respectively

state and time fixed effects. Finally εit is the error term.

A more recent empirical debate, surrounding the existence or not of an Environmen-

tal Kuznets Curve (EKC) has taken place among scholars from across a wide range of

disciplines.4 Several papers contributing to the EKC debate use however a higher order

polynomial of income to parametrically estimate its effect on the concentration of pollu-

tants. For robustness we follow this approach in investigating whether there is a CKC

in the US. Thus we include an additional cubic term of income. The idea is to allow

the data to reveal in a flexible way potential non-linearities. In particular, a third-degree

polynomial has the advantage over the quadratic specification that it is not symmetric,

hence the cubic function may rise faster than it declines or vice-versa. This approach

was first used to investigate the EKC by Grossman and Krueger (1995), Holtsz-Eakin

and Selden (1995) and Cole et al. (1997) among others. Our preferred specification thus

estimates:

Crimeit = αi + γt + ρ1Incomeit + ρ2Income2it + ρ3Income

3it + θXit + ηit (2)

Even in the case of the flexible third degree polynomial of income the parametric

approach may be arbitrarily restrictive. Thus, if the parametric assumptions are wrong

the results may be compromised. Thus, as a further robustness exercise we use nonpara-

metric methods to estimate the relationship between income and crime. By doing so we

relax any parametric assumption imposed on the data generating process and let the data

determine the appropriate model. We estimate a generic nonparametric equation of the

form:

Crimeit = f(Incomeit) + µit (3)

where f(Incomeit) is an unknown function. To estimate this equation we use Kernel-

weighted local polynomial smoothing. We do so for each state separately. As in the case

4Following the seminal contribution of Grossman and Krueger (1993) several studies have foundthat some pollutants follow an inverse U-shaped pattern relative to countries’ incomes (see for exampleGaleotti et al., 2006 and 2009; and Andreoni and Levinson, 2001). However, others have rejected theexistence of an EKC (Stern, 2004 and Poudel et a. , 2009). Dasgupta et al. (2002) and Dinda (2004)provided comprehensive reviews. The existence or not of an EKC ultimately depends, according to Lieb(2002) on the pollutant analyzed, the sample of countries studied and the empirical method used (eithercross-section or time series as in the case of the KC).

5

of the fixed effects parametric panel model, this accounts for the large heterogeneity that

exists across states, for instance in terms of per capita income which constitutes the main

variable in our analysis.

4 Results

4.1 Parametric results

Table 2 reports the results from estimating equation 2 (odd columns) and equation 3 (even

columns) on three different aggregations of crime. Columns 1 and 2 look at the effect

of income on the total crime rate, that aggregates the entire set of property and violent

crimes as reported in Panel A of Table 1. Columns 3 and 4 focus on the aggregation of

the property crimes only (burglary, larceny and car theft), and columns 5 and 6 on the

aggregate violent crime rate (composed by murder, assault, rape and robbery). Individual

crime categories are analyzed in Table 4.

All columns include state and year fixed effects. In Panel A, we add no further

controls, whereas Panel B includes as well the entire set of state-level time-varying controls

described in Panel B of Table 1. In Panel A, in most cases the coefficient of income is

positive and significant and that of income squared is negative and significative. This

combination is consistent with an inverted U-shaped relationship between income and

crime. The only exception is in the last column, that looks at the effect of income on the

rate of violent crime using a cubic polynomial of income. In that case, none of the terms

in the polynomial is statistically significant. Results for Panel B are very similar (both in

terms of the sign and size of coefficients for the polynomial terms). The only difference

worth highlighting is in the last column, where no term was significant without controls

yet with controls there is a positive linear term (significant at the 90% confidence level).

The main messages from this table can therefore be summarized as follows. First, there

is a robust CKC for total crime and for property crime. Second, since the results are very

similar with and without the set of controls often highlighted in theories of crime, the CKC

is not be driven by a correlation between income and such characteristics. Notably, since

inequality is included in the set of controls, the comparison between Panel A and Panel

B implies that the CKC is not a mere reflection of the traditional KC where income and

inequality have an inverse-U relationship. And finally, while the CKC is also apparent in

some specifications for violent crime, this relationship is less robust. In particular, in the

most convincing cubic specification allowing for a more flexible non-symmetric U-shape

we find that there is no CKC for violent crimes (regardless of whether state controls are

include or not).

Table 2 also reports in the last line the implied income for the crime turning point

in each of our specifications. Since results with and without controls are similar, we

6

report these only for the more demanding specifications with state controls. Imposing a

quadratic polynomial affects the estimated turning point. While in quadratic polynomials

the income threshold at which crime starts to fall is around 35 to 37 thousand dollars (of

2009) per capita, once a cubic term is included this falls to 20 to 24 thousand dollars per

capita. Hence, imposing symmetry does seem to be restrictive.

Table 3 looks more closely at the possibility that the traditional KC is driving our

results. While this is unlikely given our comparison of Panel A and B in Table 2, we

can also run our basic specifications, equations 2 and 3, with income inequality as the

dependent variable. We run this both with state controls (columns 1 and 2) and without

the controls (columns 3 and 4). Clearly, regardless of whether the quadratic or cubic

specification is used (in the odd and even columns respectively), there is simply no tradi-

tional KC in our sample period. Hence, we confirm that the CKC in our US sample is a

phenomenon that extends beyond Simon Kusnetz’ famous income-inequality relationship.

Table 4 looks at the individual components of property crime (columns 1 to 6) and of

violent crime (columns 7 to 14). As in Table 2, the odd columns estimate the quadratic

income regression (equation 2) and the even columns the cubic one (equation 3). Each

one of the three crime categories that compose the property crime rate features the

inverted-U relationship with income. The only exception is the rate of car theft in the

cubic polynomial specification. As for the categories comprising the violent crime rate,

the only one for which there seems to be a CKC is robbery. This lies in line with the

observation from Table 2 that the CKC is less robust for violent crime. The weaker

CKC for violent crime in Table 2 is driven solely by the behavior of robbery, which in

turn is closely related with property appropriation albeit its violent nature. In short, the

parametric results provide robust evidence of the existence of a CKC for property-related

crimes.

These results are consistent with the idea that property crimes are more likely to

depend on economic motivations than violent crimes. This is indeed the essence of the

benchmark economic model of crime of Becker (1968) and Ehrlich (1973).

However, our conclusion that there is a CKC in the US may depend on parametric

assumptions regarding the functional form of the true relationship between income and

crime. To deal with this concern, several scholars have adopted semi-parametric and

nonparametric techniques, which do not specify a functional form a priori (e.g. Bertinelli

and Stobl, 2005; Frazer, 2006 and Galeotti et al., 2006). The advantages of nonparametric

empirical analysis are summarized by DiNardo and Tobias (2001). We now explore the

robustness of our conclusions regarding the existence of CKC in the last four decades in

the US using a more flexible and unconstrained nonparametric specification.

7

4.2 Nonparametric results

We now show graphically the nonparametric Kernel estimates for each state over the

support of the state-specific real per capita income during the sample period. This allows

us to account for the large heterogeneity in states’ income. Overall, we observe a robust

inverted U-shaped relationship between income and total crime (Figure 1), property crime

(Figure 2) and also violent crime (Figure 3), with the exception of very few states.5 There,

crime rates either fail to revert to lower level as income grows and remain fluctuating near

the maximum level, or keep rising monotonically with income.6

Just as we ruled out in Table 3 the existence of a classical KC using our parametric

approach, we can also examine this with our nonparametric approach. Figure 4 shows

that during our sample period there is no evidence for a state level KC. The relationship

between inequality (measured in the vertical axis using the income Gini coefficient) and

income (measured in the horizontal axis) is monotonically increasing. This is true for all

states. Thus, again we are confident in claiming that the CKC in the US is a phenomenon

that is not explained by the behavior of income inequality.

As in the parametric specifications, the existence of a CKC is more evident in the

case of property crime than for violent crime. However, the contrast is not as sharp as in

the parametric results: Violent crimes also follow an inverted U-shaped relationship with

income in several states (Figure 3). This is confirmed by the equivalent nonparametric

estimates of the seven disaggregated outcomes, reported in Figures 5 to 11. Outcomes

related to property appropriation (including robbery, as discussed in the previous sub-

section) feature and inverted U-shaped relationship with income in most states, which is

consistent with the CKC stylized fact put forward by this paper. In addition, outcomes

related with violent crime (beyond robbery) also display such pattern in various states.

This is confirmed by figure 12, which shows the nonparametric estimates for the aggre-

gate violent crimes excluding robbery. Even in such case that focuses on the essentially

violent (an non appropriative) crimes, the inverted U relationship survives for at least

half of the states.

While the nonparametric results attenuate the differences across crime types identified

in the parametric analysis, the fact that the results are still much stronger and preva-

lent across states for property crime is likely to be related to the different motivation

in committing a crime. Indeed, property crime is more likely to depend on economic

motivations and rational opportunity cost calculations than violent crime. Thus, when

considering the relationship between crimes and income one should distinguish between

different types of crime.

5Each state sub-figure includes the 95 percent confidence interval (grey-shaded area).6Examples of this are Arkansas and West Virginia in the case of total crime in Figure 1 (top left

sub-figure and right-most sub-figure of the sixth line, respectively), and Alaska, Arkansas, Delaware,Hawaii, Kansas and others in the case of violent crime in Figure 3.

8

5 Conclusions

This paper documents for the first time the existence of a Crime Kuznets Curve at the

state level in the US since the 1970s. As real income per capita grows crime rates first

grow and then fall, effectively describing an inverted U-shaped relationship of the type

suggested by Simon Kuznets in 1955 for the relationship between income and inequality.

We however rule out the possibility that the CKC is explained by the evolution of

income inequality. Moreover our results are robust to controlling for a set of determinants

of crime identified by the previous literature, as well as for state-level and year fixed

effects. We document the existence of a CKC using both a flexible parametric specification

that includes a third-degree polynomial of income in the set of explanatory variables and

a nonparametric approach.

Consistent with the idea that income increases the opportunity cost of engaging in

appropriative illegal behavior, the CKC is present for crimes related to property but is

less robust for violent crimes. The only exception is robbery that in spite of its violent

nature is often motivated by the appropriation of someone else’s valuables. However,

a simple version of the “opportunity-cost theory” of crime is unable to account for the

patterns presented in this paper. Indeed, crime should fall monotonically with income if

the opportunity cost of crime is its main driver.

It is therefore necessary to develop and test new theories that can account for the

documented relationship. We conclude by suggesting some possible avenues for future

research, though of course many other interesting possibilities are open.

One hypothesis is that the provision of certain public goods with the potential to

reduce crime only increases significantly after communities have attained a sufficiently

high level of average income. These could be public goods affecting crime directly, like

police expenditures or investments in judicial efficiency, or indirectly, like schooling and

certain types of public infrastructure and amenities. A demand-side mechanism could

create this relationship if households exhibit the type of non-homothetic preferences often

used in problems involving subsistence levels of consumption and models of structural

change (e.g. Stone-Geary preferences). If public goods that are key drivers of crime

are outside the bundle of “subsistence” goods, then we would expect a strong demand

for them only after some minimal income has been attained. Conceivably, supply-side

mechanisms with the same spirit could also explain these patterns, if communities are

only able to provide certain public goods once a certain level of development has been

achieved.

Other hypotheses could emphasize the opposite direction of causality. For instance,

suppose that in early stages of development the types of activities that can enrich societies

are not too threatened by environments with relatively high incidence of crime. Perhaps

activities like resource extraction or industrial development intensive in physical capital

9

could survive or even thrive in spite of high rates of property and violent crime. However,

as the most productive activities in the technological frontier start demanding high levels

of human capital, it may be especially important to have an environment that attracts

highly-qualified individuals willing to live in these communities. In this hypothesis, only

when communities are able to diminish crime rates they may increase their income beyond

a certain threshold. The upward portion of the CKC could then be driven by some of

the theoretical mechanisms currently emphasized in the literature (like the fact that

wealthier communities are more attractive criminal targets) yet in the downward portion

the decrease in crime rates causes the rise in income.

As noted in the introduction, an interesting aspect of our results that may help guide

the search for the causes of the CKC is the apparent difference between property and

violent crimes. For instance, the first hypothesis would be consistent with our results

only to the extent that the public goods that are provided after certain level of income

have a stronger impact on property than on violent crimes. The second hypothesis seems,

prima facie, harder to reconcile with the presence of a CKC in property and not so much

in violent crimes. Indeed, highly-qualified individuals are likely to be just as unwilling to

live in places with high property crime rates as in places with high violent crime rates.

However, it is also clear that violent crime rates are much smaller, and therefore that

large decreases in crime rates are mainly associated with falls in property crimes. In any

case, these are all key questions that must be tackled to uncover the reasons behind the

CKC established in these paper.7

Our findings are also relevant for policy as they suggest that violent conflict cannot

be tackled solely by the trickle-down forces of economic growth and both more effective

policing as well as preventive strategies are needed. However, the scope of this paper

is limited in that it only describes a robust empirical pattern. As noted, this pattern

deserves more attention from both theorists and applied social scientists. Future efforts

in these two fronts are likely to converge in a better characterization of the micro drivers

of the macro patterns presented in this paper. Moreover, similar analyses testing the

existence of a CKC need to be done for other developed and developing countries. This

will examine the external validity of the stylized facts that we describe for the US.

We predict that the coming agenda of studying the CKC both theoretically and em-

pirically will be as active and vibrant as is the current debate on similarly policy relevant

issues like the EKC.

7We also emphasize that in some specifications the CKC is also apparent for violent crimes (namely,in our quadratic parametric exercises and in our nonparametric exercises for a number of states). Hence,while we are quite confident about the existence of a CKC, the idea that it is present mostly for thosetypes of crime more close related with economic appropriation is highly suggestive but not entirelyconclusive.

10

References

Aebi, M. (2004). “Crime Trends in Western Europe from 1990 to 2000”. European

Journal on Criminal Policy and Research, 10 (2-3), 163-186.

Allen, R. (1996). “Socioeconomic conditions and property crime”. American Journal of

Economics and Sociology , 55(3), 293-308.

Andreoni, J., & Levinson, A. (2001, May). “The simple analytics of the environmental

Kuznets curve”. Journal of Public Economics , 80 (2), 269-286.

Becker, G. (1968). “Crime and Punishment: An Economic Approach”. Journal of

Political Economy , 76 , 169.

Bertinelli, L., & Strobl, E. (2005, September.). “The Environmental Kuznets Curve

semi-parametrically revisited”. Economics Letters, Elsevier , 88(3), 350-357.

Buonanno, P., Drago, F., & Galbiati, R. (2014, February). “How much should we trust

crime statistics?A comparison between UE and US”. LIEPP Working Paper(19).

Buonanno, P., Drago, F., Galbiati, R., & Zanella, G. (2011). “Crime in Europe and in

the US: Dissecting the Reversal of misfortunes. Economic Policy , 26(67), 347-385.

Chiu, W., & Madden, P. (1998). “Burglary and income inequality”. Journal of Public

Economics , 69 , 123-141.

Cole, M., Rayner, A., & Bates, J. (1997, November). “The environmental Kuznets curve:

an empirical analysis”. Environment and Development Economics , 2 (04), 401-416.

Dasgupta, S., Laplante, B., Wang, H., & Wheeler, D. (2002, Winter). “Confronting the

Environmental Kuznets Curve”. Journal of Economic Perspectives , 16 (1), 147-168.

Dills, A. K., Miron, J. A., & Summers, G. (2008). “What do economists know about

crime?”. National Bureau of Economic Research, Working Paper 13759 .

DiNardo, J., & Tobias, J. L. (2001, Fall). “Nonparametric Density and Regression

Estimation”. Journal of Economic Perspectives , 15 (4), 11-28.

Dinda, S. (2004, August). “Environmental Kuznets Curve Hypothesis: A Survey”.

Ecological Economics , 49 (4), 431-455.

Donohue, J. J., & Levitt, S. (2001). “The Impact Of Legalized Abortion On Crime”.

The Quarterly Journal of Economics , 116 (2), 379-420.

Durlauf, S., Navarro, S., & Rivers, D. (2010). “Understanding Aggregate Crime Regres-

sions.”. Journal of Econometrics, forthcoming..

Ehrlich, I. (1973, May-June). “Participation in Illegitimate Activities: A Theoretical and

Empirical Investigation”. Journal of Political Economy , 81 (3), 521-65.

Fougere, D., Kramarz, F., & Pouget, J. (2009). “Youth Unemployment and Crime in

France”. Journal of the European Economic Association, 7(5), 909-938.

Frazer, G. (2006). “Inequality and Development Across and Within Countries”. World

Development , 34 (9), 14591481.

Galeotti, M., Lanza, A., & Pauli, F. (2006). “Reassessing the environmental Kuznets

11

curve for CO2 emissions: A robustness exercise”. Ecological Economics , 57 (1), 152

- 163.

Galeotti, M., Manera, M., & Lanza, A. (2009, April). “On the Robustness of Robust-

ness Checks of the Environmental Kuznets Curve Hypothesis”. Environmental and

Resource Economics , 42 (4), 551-574.

Gallup, J. (2012, September). “Is There a Kuznets Curve?”. (Portland State University)

Glaeser, E., & Sacerdote, B. (1999, December). “Why Is There more Crime in Cities?”.

Journal of Political Economy, University of Chicago Press , 107(S6), S225-S258.

Goldberger, A., & Rosenfeld, R. (2009). “Understanding Crime Trends”. Washington,

DC: National Academies Press..

Gould, E., Weinberg, B., & Mustard, D. (2002). “Crime and Local Labor market

Opportunities in the United States: 1979-1997”. The Review of Economics and

Statistics , 87(3), 411- 422.

Grossman, G., & Krueger, A. (1995). “Economic growth and the Environment”. The

Quarterly Journal of Economics , 110 (2), 353-377.

Holtz-Eakin, D., & Selden, T. (1995). “Stoking the fires? CO2 emissions and economic

growth”. Journal of Public Economics , 57 (1), 85 - 101.

Kanbur, R. (2012). “Does Kuznets Still matter?” (Working Papers No. 128794). Retrieved

from http://EconPapers.repec.org/RePEc:ags:cudawp:128794

Kuznet, S. (1955). “Economic growth and Income Inequality”. The American Economic

Review , 45 (1), 1-28.

Levitt, S. (2004, Winter). “Understanding Why Crime Fell in the 1990s: Four Factors

that Explain the Decline and Six that Do Not”. Journal of Economic Perspectives ,

18 (1), 163-190.

Lieb, C. (2003). The environmental kuznets curve : A survey of the empirical evidence

and of possible causes. University of Heidelberg, Department of Economic.

Lin, M.-J. (2008). “Does Unemployment Increase Crime?: Evidence from U.S. Data

19742000”. Journal of Human Resources , 43 (2), 413-436.

Oster, A., & Agell, J. (2007). “Crime and Unemployment in Turbulent Times”. Journal

of the European Economic Association, 5 , 752775.

Poudel, B., Paudel, K., & Bhattarai, K. (2009). “Searching for an Environmental Kuznets

Curve in Carbon Dioxide Pollutant in Latin American Countries”. Journal of

Agricultural and Applied Economics , 41 (01), 13-27.

Raphael, S., & Winter-Ember, R. (2001, April). “Identifying the Effect of Unemployment

on Crime”. Journal of Law and Economics , 44 (1), 259-83.

Stern, D. (2004). “The Rise and Fall of the Environmental Kuznets Curve ”. World

Development , 32 (8), 1419 - 1439.

The Economist. (2013, July). “Falling crime: Where have all the burglars

gone?”. Retrieved from Availablefrom:http://www.economist.com/news/

12

http://EconPapers.repec.org/RePEc:ags:cudawp:128794

Availablefrom:http://www.economist.com/news/briefing/21582041-rich-world-seeing-less-and-less-crime-even-face-high-unemployment-and-economic


briefing/21582041-rich-world-seeing-less-and-less-crime-even-face

-high-unemployment-and-economic

13




Table 1: Descriptive statistics

Mean Std. Dev. Min. Max.

Panel A: Dependent variablesa

Property crime rate 4,029.474 1,207.784 1,269.464 7,996.011

Violent crime rate 430.349 223.443 38.08 1,244.326

Burglary rate 1,005.14 425.61 1.087 2,906.741

Larceny rate 2,645.335 757.274 243.673 5,106.13

Car theft rate 376.229 206.021 70.138 1,571.088

Murder rate 6.32 3.685 0.157 20.349

Assault rate 262.617 143.039 25.12 785.723

Rape rate 32.357 13.72 4.16 102.184

Robbery rate 129.005 95.417 6.396 684.006

Panel B: Independent variables

Real income per capitab 28.553 8.231 12.123 58.097

Gini coefficient 0.542 0.056 0.41 0.709

Employment ratec 54.762 6.668 36.72 77.055

Population densityd 169.082 235.975 0.547 1,189.316

Share males 0-15 years old 11.529 1.51 8.403 17.236











Notes: The number of observations in all cases is 2,100 (50 states × 42 years). a Crime

rates computed in terms of 100,000 inhabitants. b Real income per capita measures in

thousands of dollars using 2009 as the base year. c The employment rate is computed

using the employment figures of the US Bureau of Labor Statistics as the percentage

of employed individuals over the state population. d Population density is computed as

number of inhabitants per squared mile.

14

Table 2: Aggregate crime categories

Total crime Property crime Violent crime

(1) (2) (3) (4) (5) (6)

Panel A: Without state controls

Income 385.9*** 1,028*** 351.7*** 987.5*** 33.98** 40.42

(80.45) (148.0) (74.76) (140.1) (12.83) (26.11)

Income squared -5.377*** -25.97*** -4.929*** -25.32*** -0.445*** -0.651

(1.059) (3.926) (0.978) (3.814) (0.154) (0.697)

Income cubed 0.209*** 0.207*** 0.00209

(0.0353) (0.0349) (0.00671)

Constant -1,450 -7,366*** -1,263 -7,120*** -184.5 -243.8

(1,058) (1,555) (985.1) (1,450) (173.9) (283.2)

R-squared 0.678 0.715 0.680 0.722 0.491 0.492

Panel B: With state controls

Income 371.2*** 1,104*** 341.7*** 1,056*** 29.44** 48.54*

(102.3) (151.9) (97.76) (147.0) (12.61) (24.66)

Income squared -5.090*** -28.10*** -4.671*** -27.09*** -0.416** -1.015

(1.310) (4.007) (1.214) (3.911) (0.170) (0.663)

Income cubed 0.229*** 0.223*** 0.00596

(0.0364) (0.0361) (0.00645)

Constant -9,035** -16,829*** -7,926* -15,520*** -1,130*** -1,333***

(4,101) (3,995) (4,025) (3,918) (408.4) (452.7)

R-squared 0.703 0.743 0.701 0.745 0.579 0.581

Year FE X X X X X X

State FE X X X X X X

Implied income turning pointa 36.46 19.64 36.58 19.49 35.38 23.91

Notes: The number of observations in all cases is 2,100 (50 states × 42 years). Standard errors clustered at the state level

in parentheses. Controls include state and year fixed effects as well as income inequality, population density, employment

opportunities and the state-level male population for age brackets 0-15, 15-19, 20-24, 25-29, 30-34, 35-39, 40-44, 45-49,

50-54, 55-59 and 60-64. a Implied income turning point, in thousand of real USD of 2009, computed for the regressions

with state level controls (Panel B). *** is significant at the 1% level, ** is significant at the 5% level, * is significant at

the 10% level.

15

Table 3: Is there a classical Kuznets Curve in our sample?

(1) (2) (3) (4)

Income -0.00197 -0.00329 0.000783 -0.00148

(0.00160) (0.00416) (0.00182) (0.00383)

Income squared 2.47e-05 6.69e-05 5.45e-06 7.63e-05

(2.24e-05) (0.000114) (2.27e-05) (0.000100)

Income cubed -4.28e-07 -7.06e-07

(1.08e-06) (9.48e-07)

Constant 0.484*** 0.496*** 0.308*** 0.331***

(0.0203) (0.0421) (0.0987) (0.0961)

R-squared 0.892 0.892 0.914 0.914

Year FE X X X X

State FE X X X X

Controls X X

Notes: The number of observations in all cases is 2,100 (50 states × 42 years).

Standard errors clustered at the state level in parentheses. Controls include

state and year fixed effects as well as population density, employment opportu-

nities and the state-level male population for age brackets 0-15, 15-19, 20-24,

25-29, 30-34, 35-39, 40-44, 45-49, 50-54, 55-59 and 60-64. *** is significant at

the 1% level, ** is significant at the 5% level, * is significant at the 10% level.

16

Table

4:Breakdown

bycrim

ety

pe

Pro

per

tycr

ime

Vio

lent

crim

e

Burg

lary

Lar

ceny

Car

thef

tM

urd

erA

ssau

ltR

ape

Rob

ber

y

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

(13)

(14)

Inco

me

61.9

8**

304.

3***

216.

9***

681.

4***

63.5

8***

72.2

2**

-0.1

03-0

.297

7.52

710

.50

0.23

81.

642

21.7

4***

36.9

5***

(30.

38)

(58.

93)

(61.

29)

(97.

16)

(17.

08)

(28.

77)

(0.2

00)

(0.3

14)

(8.6

94)

(19.

47)

(1.1

48)

(2.0

91)

(6.8

95)

(11.

46)

Inco

me

squar

ed-1

.016

***

-8.6

19**

*-2

.752

***

-17.

33**

*-0

.897

***

-1.1

680.

0017

20.

0078

3-0

.114

-0.2

07-0

.006

93-0

.051

0-0

.297

***

-0.7

74**

(0.3

74)

(1.6

77)

(0.7

42)

(2.6

33)

(0.2

25)

(0.8

80)

(0.0

0242

)(0

.007

89)

(0.0

966)

(0.5

20)

(0.0

122)

(0.0

508)

(0.1

07)

(0.2

90)

Inco

me

cub

ed0.

0757

***

0.14

5***

0.00

270

-6.0

8e-0

50.

0009

280.

0004

390.

0047

5*

(0.0

161)

(0.0

248)

(0.0

0926

)(7

.73e

-05)

(0.0

0477

)(0

.000

468)

(0.0

0281

)

Con

stan

t-1

,631

-4,2

07**

*-6

,589

**-1

1,52

6***

194.

010

2.2

-4.1

69-2

.100

-707

.8**

-739

.4**

7.14

6-7

.783

-414

.8**

-576

.5**

*

(1,4

32)

(1,3

46)

(2,9

85)

(3,0

41)

(748

.2)

(847

.3)

(7.1

96)

(7.5

15)

(302

.6)

(349

.2)

(36.

01)

(38.

60)

(183

.5)

(199

.5)

R-s

quar

ed0.

741

0.77

50.

668

0.71

40.

445

0.44

50.

575

0.57

50.

555

0.55

50.

594

0.59

50.

443

0.45

0

Yea

rF

EX

XX

XX

XX

XX

XX

XX

X

Sta

teF

EX

XX

XX

XX

XX

XX

XX

X

Con

trol

sX

XX

XX

XX

XX

XX

XX

X

Implied

inco

me

turn

ing

poi

nt

30.5

017

.65

39.4

119

.66

35.4

430

.92

––

––

––

36.5

023

.87

Not

es:

The

num

ber

ofob

serv

atio

ns

inal

lca

ses

is2,

100

(50

stat

es×

42ye

ars)

.Sta

ndard

erro

rscl

ust

ered

at

the

state

level

inpare

nth

eses

.C

ontr

ols

incl

ude

state

and

year

fixed

effec

tsas

wel

las

inco

me

ineq

uality

,

pop

ula

tion

den

sity

,em

plo

ym

ent

opp

ortu

nit

ies

and

the

stat

e-le

vel

mal

ep

opula

tion

for

age

bra

cket

s0-

15,

15-

19,

20-2

4,

25-

29,

30-3

4,

35-

39,

40-4

4,

45-

49,

50-5

4,

55-

59

and

60-6

4.

***

issi

gnifi

cant

at

the

1%

leve

l,

**is

sign

ifica

nt

atth

e5%

leve

l,*

issi

gnifi

cant

atth

e10

%le

vel.

17

Figure 1: Non parametric estimates for total crime20

0030

0040

0050

00

15 20 25 30 35kernel = epanechnikov, degree = 3, bandwidth = 3.32, pwidth = 4.98

Alabama

2000

3000

4000

5000

6000


Alaska

40005

00060

00700

08000


Arizona

2000

3000

4000

5000

10 15 20 25 30 35kernel = epanechnikov, degree = 3, bandwidth = 2.69, pwidth = 4.04

Arkansas

30004

00050

00600

070008

000


California

20003

00040

00500

060007

000


Colorado

2000

3000

4000

5000

6000


Connecticut

3000

4000

5000

6000

7000


Delaware

40005

00060

00700

080009

000


Florida

20003

00040

00500

060007

000


Georgia

3000

4000

5000

6000

7000


Hawaii

20002

50030

00350

040004

500


Idaho

3000

4000

5000

6000


Illinois

30003

50040

00450

05000


Indiana

20002

50030

00350

040004

500


Iowa

30003

50040

00450

05000


Kansas

2000

2500

3000

3500


Kentucky

3000

4000

5000

6000

7000


Louisiana

1000

2000

3000

4000

5000


Maine

3000

4000

5000

6000

20 30 40 50kernel = epanechnikov, degree = 3, bandwidth = 4.28, pwidth = 6.42

Maryland

30004

00050

00600

07000


Massachusetts

3000

4000

5000

6000

7000


Michigan

25003

00035

00400

04500


Minnesota

1000

2000

3000

4000

5000


Mississippi

3500

4000

4500

5000

5500


Missouri

25003

00035

00400

045005

000


Montana

20002

50030

00350

040004

500


Nebraska

40005

00060

00700

08000


Nevada

1000

2000

3000

4000

5000


New Hampshire

2000

3000

4000

5000

6000


New Jersey

4000

5000

6000

7000


New Mexico

20003

00040

00500

060007

000


New York

2000

3000

4000

5000

6000


North Carolina

1000

1500

2000

2500

3000


North Dakota

30003

50040

00450

05000


Ohio

2000

3000

4000

5000

6000


Oklahoma20

00300

040005

00060

00700

0


Oregon

2000

2500

3000

3500


Pennsylvania

2000

3000

4000

5000

6000


Rhode Island

20003

00040

00500

06000


South Carolina

1500

2000

2500

3000

3500


South Dakota

20003

00040

00500

06000


Tennessee

30004

00050

00600

07000


Texas

3000

4000

5000

6000


Utah

1000

2000

3000

4000

5000


Vermont

2500

3000

3500

4000

4500


Virginia

3000

4000

5000

6000

7000


Washington

1000

1500

2000

2500

3000


West Virginia

25003

00035

00400

04500


Wisconsin

25003

00035

00400

04500


Wyoming

18

Figure 2: Non parametric estimates for property crime10

0020

0030

0040

0050

00

15 20 25 30 35kernel = epanechnikov, degree = 3, bandwidth = 3, pwidth = 4.5

Alabama

20003

00040

00500

06000


Alaska

30004

00050

00600

070008

000


Arizona

2000

3000

4000

5000


Arkansas

30004

00050

00600

07000


California

20003

00040

00500

060007

000


Colorado

2000

3000

4000

5000

6000


Connecticut

3000

4000

5000

6000


Delaware

30004

00050

00600

070008

000


Florida

20003

00040

00500

06000


Georgia

30004

00050

00600

07000


Hawaii

20002

50030

00350

04000


Idaho

25003

00035

00400

045005

000


Illinois

2500

3000

3500

4000

4500


Indiana

20002

50030

00350

04000


Iowa

30003

50040

00450

05000


Kansas

2000

2500

3000


Kentucky

2000

3000

4000

5000

6000


Louisiana

15002

00025

00300

035004

000


Maine

30003

50040

00450

050005

500


Maryland

2000

3000

4000

5000

6000


Massachusetts

3000

4000

5000

6000


Michigan

2500

3000

3500

4000

4500


Minnesota

1000

2000

3000

4000


Mississippi

3000

3500

4000

4500

5000


Missouri

25003

00035

00400

04500


Montana

20002

50030

00350

04000


Nebraska

30004

00050

00600

07000


Nevada

1000

2000

3000

4000

5000


New Hampshire

2000

3000

4000

5000

6000


New Jersey

3000

4000

5000

6000


New Mexico

20003

00040

00500

06000


New York

20003

00040

00500

06000


North Carolina

1000

1500

2000

2500

3000


North Dakota

25003

00035

00400

04500


Ohio

2000

3000

4000

5000


Oklahoma20

00300

040005

00060

00


Oregon

1500

2000

2500

3000

3500


Pennsylvania

2000

3000

4000

5000

6000


Rhode Island

2000

3000

4000

5000


South Carolina

1500

2000

2500

3000


South Dakota

1000

2000

3000

4000

5000


Tennessee

3000

4000

5000

6000

7000


Texas

3000

4000

5000

6000


Utah

1000

2000

3000

4000

5000


Vermont

20002

50030

00350

04000


Virginia

3000

4000

5000

6000

7000


Washington

1000

1500

2000

2500


West Virginia

25003

00035

00400

04500

15 20 25 30 35 40kernel = epanechnikov, degree = 3, bandwidth = 3.33, pwidth = 5

Wisconsin

20002

50030

00350

040004

500


Wyoming

19

Figure 3: Non parametric estimates for violent crime20

040

060

080

0


Alabama

200

400

600

800


Alaska

3004

0050

0600

700


Arizona

2003

0040

0500

600


Arkansas

400

600

800

1000


California

3003

5040

0450

5005

50


Colorado

100

200

300

400

500


Connecticut

3004

0050

0600

7008

00


Delaware

400

600

8001

0001

200


Florida

200

400

600

800


Georgia

100

150

200

250

300


Hawaii

1001

5020

0250

300


Idaho

400

600

800

1000


Illinois

2003

0040

0500

600


Indiana

010

020

030

040

0


Iowa

100

200

300

400

500


Kansas

2002

5030

0350

4004

50


Kentucky

400

600

800

1000


Louisiana

5010

0150

2002

50


Maine

5006

0070

0800

900


Maryland

200

400

600

800


Massachusetts

500

600

700

800


Michigan

1502

0025

0300

350


Minnesota

100

200

300

400

500


Mississippi

3004

0050

0600

7008

00


Missouri

100

200

300

400


Montana

100

200

300

400

500


Nebraska

200

400

600

8001

000


Nevada

5010

015

020

0


New Hampshire

3004

0050

0600

700


New Jersey

200

400

600

8001

000


New Mexico

2004

0060

08001

00012

00


New York

300

400

500

600

700


North Carolina

010

020

030

0


North Dakota

2503

0035

0400

4505

00


Ohio

2003

0040

0500

600


Oklahoma20

030

040

050

060

0


Oregon

2002

5030

0350

4004

50


Pennsylvania

2002

5030

0350

400


Rhode Island

2004

0060

0800

1000


South Carolina

010

020

030

040

0


South Dakota

200

400

600

800

15 20 25 30 35kernel = epanechnikov, degree = 3, bandwidth = 3.34, pwidth = 5

Tennessee

3004

0050

0600

7008

00


Texas

1001

5020

0250

3003

50


Utah

5010

015

020

0


Vermont

200

250

300

350

400


Virginia

200

300

400

500


Washington

1001

5020

0250

3003

50


West Virginia

5010

0150

2002

5030

0


Wisconsin

010

020

030

040

0


Wyoming

20

Figure 4: Inequality and Income.4

5.5

.55

.6.6

5


Alabama

.3.4

.5.6

.7


Alaska

.4.4

5.5

.55

.6.6

5


Arizona

.45

.5.5

5.6

.65

.7


Arkansas

.45

.5.5

5.6

.65


California

.45

.5.5

5.6


Colorado

.45

.5.5

5.6

.65

.7


Connecticut

.45

.5.5

5.6


Delaware

.45

.5.5

5.6

.65

.7


Florida

.45

.5.5

5.6

.65


Georgia

.45

.5.5

5.6


Hawaii

.45

.5.5

5.6

.65


Idaho

.45

.5.5

5.6

.65


Illinois

.4.4

5.5

.55

.6.6

5


Indiana

.4.4

5.5

.55

.6


Iowa

.45

.5.5

5.6

.65


Kansas

.45

.5.5

5.6

.65


Kentucky

.45

.5.5

5.6

.65

.7


Louisiana

.4.4

5.5

.55

.6


Maine

.4.4

5.5

.55

.620 30 40 50kernel = epanechnikov, degree = 3, bandwidth = 3.85, pwidth = 5.78

Maryland

.45

.5.5

5.6

.65


Massachusetts

.4.4

5.5

.55

.6


Michigan

.45

.5.5

5.6


Minnesota

.45

.5.5

5.6

.65

.7


Mississippi

.45

.5.5

5.6

.65


Missouri

.4.5

.6.7


Montana

.45

.5.5

5.6


Nebraska

.4.4

5.5

.55

.6.6

5


Nevada

.4.4

5.5

.55

.6


New Hampshire

.45

.5.5

5.6

.65


New Jersey

.45

.5.5

5.6

.65

.7


New Mexico

.45

.5.5

5.6

.65

.7


New York

.45

.5.5

5.6

.65


North Carolina

.4.6


North Dakota

.4.4

5.5

.55

.6


Ohio

.4.5

.6.7


Oklahoma.4

5.5

.55

.6.6

5


Oregon

.4.4

5.5

.55

.6


Pennsylvania

.45

.5.5

5.6


Rhode Island

.45

.5.5

5.6

.65


South Carolina

.4.4

5.5

.55

.6.6

5


South Dakota

.45

.5.5

5.6

.65


Tennessee

.45

.5.5

5.6

.65


Texas

.45

.5.5

5.6


Utah

.45

.5.5

5.6


Vermont

.45

.5.5

5.6


Virginia

.45

.5.5

5.6


Washington

.4.4

5.5

.55

.6.6

5


West Virginia

.4.4

5.5

.55

.6


Wisconsin

.4.4

5.5

.55

.6.6

5


Wyoming

21

Figure 5: Non parametric estimates for burglary40

0600

80010

00120

01400


Alabama

050

010

0015

00


Alaska

5001

0001

5002

0002

500


Arizona

600

800

1000

1200


Arkansas

5001

0001

5002

0002

500


California

500

1000

1500

2000


Colorado

500

1000

1500

2000


Connecticut

6008

00100

012001

40016

00


Delaware

5001

0001

5002

0002

500


Florida

6008

00100

012001

40016

00


Georgia

500

1000

1500

2000


Hawaii

400

600

8001

0001

200


Idaho

6008

0010

00120

01400


Illinois

600

8001

0001

2001

400


Indiana

400

600

800

1000


Iowa

600

8001

0001

2001

400


Kansas

6007

0080

0900

1000


Kentucky

8001

0001

2001

4001

600


Louisiana

4006

0080

010001

20014

00


Maine

6008

00100

012001

40016


Maryland

500

1000

1500

2000


Massachusetts

500

1000

1500

2000


Michigan

400

600

8001

0001

200


Minnesota

050

010

0015

00


Mississippi

6008

00100

012001

40016

00


Missouri

200

400

600

8001

000


Montana

4005

0060

0700

800


Nebraska

5001

00015

00200

02500


Nevada

2004

0060

08001

00012

00


New Hampshire

500

1000

1500

2000


New Jersey

500

1000

1500

2000


New Mexico

050

0100

0150

0200

0


New York

6008

00100

012001

40016

00


North Carolina

200

300

400

500


North Dakota

600

8001

0001

2001

400


Ohio

6008

00100

012001

40016

00


Oklahoma50

010

0015

0020

00


Oregon

400

600

800

1000


Pennsylvania

500

1000

1500


Rhode Island

8001

00012

00140

01600


South Carolina

3004

0050

0600

700


South Dakota

6008

0010

00120

01400


Tennessee

500

1000

1500

2000


Texas

4006

0080

0100

01200


Utah

4006

0080

010001

20014

00


Vermont

4006

0080

0100

01200


Virginia

80010

00120

014001

60018

00


Washington

300

400

500

600

700


West Virginia

400

600

800

1000


Wisconsin

3004

0050

0600

7008

00


Wyoming

22

Figure 6: Non parametric estimates for larceny10

00150

020002

50030

00


Alabama

15002

00025

00300

03500


Alaska

2000

3000

4000

5000


Arizona

1000

1500

2000

2500

3000


Arkansas

15002

00025

00300

035004

000


California

1000

2000

3000

4000

5000


Colorado

15002

00025

00300

03500


Connecticut

15002

00025

00300

035004

000


Delaware

2000

3000

4000

5000


Florida

1000

2000

3000

4000


Georgia

20002

50030

00350

040004

500


Hawaii

1000

1500

2000

2500

3000


Idaho

1500

2000

2500

3000

3500


Illinois

1500

2000

2500

3000


Indiana

1500

2000

2500

3000


Iowa

1000

2000

3000

4000


Kansas

10001

20014

00160

018002

000


Kentucky

15002

00025

00300

035004

000

15 20 25 30 35 40kernel = epanechnikov, degree = 3, bandwidth = 3.34, pwidth = 5

Louisiana

1000

1500

2000

2500

3000


Maine

2000

2500

3000

3500


Maryland

1000

1500

2000

2500

3000


Massachusetts

2000

2500

3000

3500


Michigan

1500

2000

2500

3000


Minnesota

5001

00015

00200

02500


Mississippi

1500

2000

2500

3000


Missouri

15002

00025

00300

03500


Montana

10001

50020

00250

03000


Nebraska

1000

2000

3000

4000

5000


Nevada

50010

00150

020002

50030

00


New Hampshire

10001

50020

00250

03000


New Jersey

2000

2500

3000

3500

4000


New Mexico

15002

00025

00300

03500


New York

10001

50020

00250

030003

500


North Carolina

5001

0001

5002

0002

500


North Dakota

1500

2000

2500

3000


Ohio

10001

50020

00250

03000


Oklahoma10

0020

0030

0040

0050

00


Oregon

500

1000

1500

2000


Pennsylvania

1500

2000

2500

3000


Rhode Island

10001

50020

00250

030003

500


South Carolina

1000

1500

2000

2500


South Dakota

50010

00150

020002

50030

00


Tennessee

15002

00025

00300

035004

000


Texas

20002

50030

00350

040004

500


Utah

50010

00150

020002

50030

00


Vermont

1500

2000

2500

3000


Virginia

20002

50030

00350

040004

500


Washington

500

1000

1500

2000


West Virginia

1500

2000

2500

3000


Wisconsin

1500

2000

2500

3000

3500


Wyoming

23

Figure 7: Non parametric estimates for car theft15

0200

2503

0035

0


Alabama

020

040

060

080

0


Alaska

4006

0080

0100

01200


Arizona

1001

5020

0250

3003

50


Arkansas

400

600

800

1000


California

020

040

060

080

0


Colorado

2003

0040

0500

6007

00


Connecticut

2003

0040

0500

600


Delaware

200

400

600

8001

000


Florida

2003

0040

0500

6007

00


Georgia

2003

0040

0500

6007

00


Hawaii

5010

015

020

025

0


Idaho

2003

0040

0500

6007

00


Illinois

200

300

400

500


Indiana

100

150

200

250


Iowa

200

250

300

350


Kansas

100

200

300

400


Kentucky

2003

0040

0500

6007

00


Louisiana

5010

015

020

025

0


Maine

3004

0050

0600

700


Maryland

050

010

0015

00


Massachusetts

3004

0050

0600

7008

00


Michigan

100

200

300

400

20 30 40 50kernel = epanechnikov, degree = 3, bandwidth = 3.34, pwidth = 5

Minnesota

010

020

030

040

0


Mississippi

200

300

400

500

600


Missouri

150

200

250

300

350


Montana

1502

0025

0300

3504

00


Nebraska

400

600

8001

0001

200


Nevada

5010

0150

2002

5030

0


New Hampshire

2004

0060

0800

1000


New Jersey

200

300

400

500

600


New Mexico

020

0400

6008

00100

0


New York

1502

0025

0300

3504

00


North Carolina

5010

015

020

0


North Dakota

2003

0040

0500

600


Ohio

200

300

400

500

600


Oklahoma0

200

400

600

800


Oregon

100

200

300

400

500


Pennsylvania

200

400

600

8001

000


Rhode Island

2002

5030

0350

4004

50


South Carolina

5010

015

020

0


South Dakota

2003

0040

0500

6007

00


Tennessee

200

400

600

800


Texas

2002

5030

0350

4004

50


Utah

5010

0150

2002

5030

0


Vermont

1001

5020

0250

3003

50


Virginia

2003

0040

0500

6007

00


Washington

5010

015

020

025

0


West Virginia

1502

0025

0300

3504

00


Wisconsin

5010

0150

2002

5030

0


Wyoming

24

Figure 8: Non parametric estimates for murder6

810

1214

16


Alabama

05

1015


Alaska

67

89

1011


Arizona

46

810

12


Arkansas

46

810

1214


California

24

68

20 25 30 35 40 45kernel = epanechnikov, degree = 3, bandwidth = 5, pwidth = 7.5

Colorado

23

45

6


Connecticut

24

68

10


Delaware

510

15


Florida

510

1520


Georgia

02

46

8


Hawaii

12

34

56


Idaho

46

810

12


Illinois

34

56

78


Indiana

11.

52

2.5


Iowa

23

45

67


Kansas

46

810

12


Kentucky

510

1520


Louisiana

11.

52

2.5

33.

5


Maine

68

1012


Maryland

12

34

5


Massachusetts

68

1012

14


Michigan

12

34


Minnesota

68

1012

1416


Mississippi

67

89

1011


Missouri

23

45

6


Montana

12

34

5


Nebraska

68

1012

1416


Nevada

11.

52

2.5

33.

5


New Hampshire

34

56

7


New Jersey

46

810

12


New Mexico

46

810

1214


New York

68

1012

14


North Carolina

01

23

4


North Dakota

45

67

89


Ohio

46

810


Oklahoma2

34

56


Oregon

55.

56

6.5

7


Pennsylvania

12

34

5


Rhode Island

510

1520


South Carolina

12

34

5


South Dakota

68

1012

14


Tennessee

510

15


Texas

12

34


Utah

01

23

4


Vermont

46

810

12


Virginia

23

45

6


Washington

34

56

78


West Virginia

23

45


Wisconsin

02

46

8


Wyoming

25

Figure 9: Non parametric estimates for assault20

030

040

050

060

0


Alabama

1002

0030

0400

5006

00


Alaska

100

200

300

400

500


Arizona

100

200

300

400


Arkansas

2003

0040

0500

600


California

1502

0025

0300

3504

00


Colorado

5010

0150

2002

50


Connecticut

100

200

300

400

500


Delaware

200

400

600

800


Florida

100

200

300

400

500


Georgia

5010

015

020

0


Hawaii

5010

015

020

025

0


Idaho

1002

0030

0400

5006

00


Illinois

100

200

300

400


Indiana

5010

0150

2002

50


Iowa

1001

5020

0250

3003

50


Kansas

1001

5020

0250

3003

50


Kentucky

2003

0040

0500

6007

00


Louisiana

050

100

150

200


Maine

200

300

400

500


Maryland

1002

0030

0400

5006

00


Massachusetts

100

200

300

400

500


Michigan

5010

015

020

0


Minnesota

100

150

200

250

300


Mississippi

100

200

300

400

500


Missouri

5010

0150

2002

5030

0


Montana

1001

5020

0250

3003

50


Nebraska

1002

0030

0400

500


Nevada

2040

6080

1001

20


New Hampshire

1001

5020

0250

300


New Jersey

020

040

060

080

0


New Mexico

200

300

400

500


New York

2503

0035

0400

450


North Carolina

050

1001

5020

0250


North Dakota

1001

5020

0250

300


Ohio

100

200

300

400

500


Oklahoma10

0150

2002

5030

0350


Oregon

5010

015

020

025

0


Pennsylvania

100

150

200

250

300


Rhode Island

200

400

600

800


South Carolina

050

1001

5020

0250


South Dakota

1002

0030

0400

500


Tennessee

100

200

300

400

500


Texas

5010

015

020

025

0


Utah

2040

6080

1001

20


Vermont

1001

2014

0160

1802

00


Virginia

1001

5020

0250

300


Washington

5010

0150

2002

5030

0


West Virginia

050

100

150

200


Wisconsin

5010

0150

2002

50


Wyoming

26

Figure 10: Non parametric estimates for rape15

2025

3035


Alabama

2040

6080

100


Alaska

2025

3035

4045


Arizona

1020

3040

50


Arkansas

2030

4050

60


California

(file graphs/lpoly_rape_r.gph saved)> F format)(file /Users/paolinhoz/Dropbox/research/progetti_jfv/kuznets/estimates/graphs/lpoly_murder_r.pdf written in PD(file graphs/lpoly_murder_r.gph saved)> PDF format)(file /Users/paolinhoz/Dropbox/research/progetti_jfv/kuznets/estimates/graphs/lpoly_property_r.pdf written in (file graphs/lpoly_property_r.gph saved)> DF format)(file /Users/paolinhoz/Dropbox/research/progetti_jfv/kuznets/estimates/graphs/lpoly_violent_r.pdf written in P(file graphs/lpoly_violent_r.gph saved) 5. } 4. graph export ${graphs}`y'.pdf, replace. graph save ${graphs}`y', replace 3. > */ ${graphs}`y'_46.gph ${graphs}`y'_47.gph ${graphs}`y'_48.gph ${graphs}`y'_49.gph ${graphs}`y'_50.gph> */ ${graphs}`y'_41.gph ${graphs}`y'_42.gph ${graphs}`y'_43.gph ${graphs}`y'_44.gph ${graphs}`y'_45.gph /*> */ ${graphs}`y'_36.gph ${graphs}`y'_37.gph ${graphs}`y'_38.gph ${graphs}`y'_39.gph ${graphs}`y'_40.gph /*> */ ${graphs}`y'_31.gph ${graphs}`y'_32.gph ${graphs}`y'_33.gph ${graphs}`y'_34.gph ${graphs}`y'_35.gph /*> */ ${graphs}`y'_26.gph ${graphs}`y'_27.gph ${graphs}`y'_28.gph ${graphs}`y'_29.gph ${graphs}`y'_30.gph /*> */ ${graphs}`y'_21.gph ${graphs}`y'_22.gph ${graphs}`y'_23.gph ${graphs}`y'_24.gph ${graphs}`y'_25.gph /*> */ ${graphs}`y'_16.gph ${graphs}`y'_17.gph ${graphs}`y'_18.gph ${graphs}`y'_19.gph ${graphs}`y'_20.gph /*> */ ${graphs}`y'_11.gph ${graphs}`y'_12.gph ${graphs}`y'_13.gph ${graphs}`y'_14.gph ${graphs}`y'_15.gph /*> */ ${graphs}`y'_6.gph ${graphs}`y'_7.gph ${graphs}`y'_8.gph ${graphs}`y'_9.gph ${graphs}`y'_10.gph /*> /* 2. gr combine ${graphs}`y'_1.gph ${graphs}`y'_2.gph ${graphs}`y'_3.gph ${graphs}`y'_4.gph ${graphs}`y'_5.gph> oly_burglary_r lpoly_larceny_r lpoly_car_theft_r lpoly_index_r {. foreach y in lpoly_violent_r lpoly_property_r lpoly_murder_r lpoly_rape_r lpoly_robbery_r lpoly_assault_r lp. .

(file graphs/lpoly_index_r_50.gph saved)(file graphs/lpoly_index_r_49.gph saved)(file graphs/lpoly_index_r_48.gph saved)(file graphs/lpoly_index_r_47.gph saved)

3035

4045

50


Colorado

510

1520

2530


Connecticut

2040

6080


Delaware

1020

3040

5060


Florida

1020

3040

50


Georgia

1020

3040


Hawaii

1020

3040

50


Idaho

1520

2530

3540


Illinois

1520

2530

3540


Indiana

510

1520

2530


Iowa

1020

3040

50


Kansas

1520

2530

35


Kentucky

2025

3035

4045


Louisiana

010

2030

40


Maine

2025

3035

4045


Maryland

1015

2025

3035


Massachusetts

2040

6080


Michigan

1020

3040

5060


Minnesota

1020

3040

50


Mississippi

2025

3035

15 20 25 30 35 40kernel = epanechnikov, degree = 3, bandwidth = 3, pwidth = 4.5

Missouri

1020

3040

50


Montana

010

2030

40


Nebraska

020

4060

80


Nevada

010

2030

40


New Hampshire

1015

2025

3035


New Jersey

2030

4050

60


New Mexico

1015

2025

3035


New York

1015

2025

3035


North Carolina

010

2030

4050


North Dakota

1020

3040

50


Ohio

1020

3040

50


Oklahoma10

2030

4050


Oregon

1015

2025

30


Pennsylvania

010

2030

40


Rhode Island

1020

3040

50


South Carolina

020

4060

80


South Dakota

1020

3040

50


Tennessee

2030

4050


Texas

1020

3040

50


Utah

010

2030


Vermont

1520

2530


Virginia

1020

3040

5060


Washington

05

1015

2025


West Virginia

510

1520

25


Wisconsin

1020

3040


Wyoming

27

Figure 11: Non parametric estimates for robbery0

5010

015

020

0


Alabama

4060

8010

0120


Alaska

5010

015

020

0


Arizona

4060

8010

0120

140


Arkansas

1502

0025

0300

3504

00


California

(file graphs/lpoly_robbery_r.gph saved)> format)(file /Users/paolinhoz/Dropbox/research/progetti_jfv/kuznets/estimates/graphs/lpoly_rape_r.pdf written in PDF (file graphs/lpoly_rape_r.gph saved)> F format)(file /Users/paolinhoz/Dropbox/research/progetti_jfv/kuznets/estimates/graphs/lpoly_murder_r.pdf written in PD(file graphs/lpoly_murder_r.gph saved)> PDF format)(file /Users/paolinhoz/Dropbox/research/progetti_jfv/kuznets/estimates/graphs/lpoly_property_r.pdf written in (file graphs/lpoly_property_r.gph saved)> DF format)(file /Users/paolinhoz/Dropbox/research/progetti_jfv/kuznets/estimates/graphs/lpoly_violent_r.pdf written in P(file graphs/lpoly_violent_r.gph saved) 5. } 4. graph export ${graphs}`y'.pdf, replace. graph save ${graphs}`y', replace 3. > */ ${graphs}`y'_46.gph ${graphs}`y'_47.gph ${graphs}`y'_48.gph ${graphs}`y'_49.gph ${graphs}`y'_50.gph> */ ${graphs}`y'_41.gph ${graphs}`y'_42.gph ${graphs}`y'_43.gph ${graphs}`y'_44.gph ${graphs}`y'_45.gph /*> */ ${graphs}`y'_36.gph ${graphs}`y'_37.gph ${graphs}`y'_38.gph ${graphs}`y'_39.gph ${graphs}`y'_40.gph /*> */ ${graphs}`y'_31.gph ${graphs}`y'_32.gph ${graphs}`y'_33.gph ${graphs}`y'_34.gph ${graphs}`y'_35.gph /*> */ ${graphs}`y'_26.gph ${graphs}`y'_27.gph ${graphs}`y'_28.gph ${graphs}`y'_29.gph ${graphs}`y'_30.gph /*> */ ${graphs}`y'_21.gph ${graphs}`y'_22.gph ${graphs}`y'_23.gph ${graphs}`y'_24.gph ${graphs}`y'_25.gph /*> */ ${graphs}`y'_16.gph ${graphs}`y'_17.gph ${graphs}`y'_18.gph ${graphs}`y'_19.gph ${graphs}`y'_20.gph /*> */ ${graphs}`y'_11.gph ${graphs}`y'_12.gph ${graphs}`y'_13.gph ${graphs}`y'_14.gph ${graphs}`y'_15.gph /*> */ ${graphs}`y'_6.gph ${graphs}`y'_7.gph ${graphs}`y'_8.gph ${graphs}`y'_9.gph ${graphs}`y'_10.gph /*> /* 2. gr combine ${graphs}`y'_1.gph ${graphs}`y'_2.gph ${graphs}`y'_3.gph ${graphs}`y'_4.gph ${graphs}`y'_5.gph> oly_burglary_r lpoly_larceny_r lpoly_car_theft_r lpoly_index_r {. foreach y in lpoly_violent_r lpoly_property_r lpoly_murder_r lpoly_rape_r lpoly_robbery_r lpoly_assault_r lp. .

(file graphs/lpoly_index_r_50.gph saved)

5010

015

020

0


Colorado

5010

015

020

025

0


Connecticut

100

150

200

250


Delaware

100

200

300

400


Florida

5010

0150

2002

50


Georgia

050

100

150


Hawaii

1020

3040


Idaho

1502

0025

0300

3504

00


Illinois

8010

012

014

0


Indiana

2030

4050


Iowa

4060

8010

012

0


Kansas

6070

8090

100


Kentucky

100

150

200

250

300


Louisiana

010

2030

40


Maine

1502

0025

0300

3504


Maryland

100

150

200

250


Massachusetts

100

200

300

400


Michigan

6080

100

120


Minnesota

050

100

150


Mississippi

100

150

200

250


Missouri

1020

3040

50


Montana

2040

6080


Nebraska

010

020

030

040

0


Nevada

010

2030

40


New Hampshire

100

150

200

250

300


New Jersey

6080

1001

2014

0160


New Mexico

1002

0030

0400

5006

00


New York

050

100

150

200


North Carolina

05

1015

20


North Dakota

1201

4016

0180

2002

20


Ohio

4060

8010

0120


Oklahoma50

100

150

200


Oregon

1001

2014

0160

1802

00


Pennsylvania

6080

100

120

140


Rhode Island

050

100

150

200


South Carolina

1015

2025


South Dakota

5010

015

020

025

0


Tennessee

100

150

200

250


Texas

4050

6070

80


Utah

010

2030


Vermont

6080

100

120

140


Virginia

8010

012

014

0


Washington

2030

4050

60


West Virginia

2040

6080

1001

20


Wisconsin

010

2030

40


Wyoming

28

Figure 12: Non parametric estimates for robbery20

0300

4005

0060

0


Alabama

200

400

600

800


Alaska

200

300

400

500

600


Arizona

100

200

300

400

500


Arkansas

2003

0040

0500

6007

00


California

2002

5030

0350

4004

50


Colorado

1001

5020

0250

300


Connecticut

1002

0030

0400

5006

00


Delaware

200

400

600

800


Florida

200

300

400

500


Georgia

5010

015

020

0


Hawaii

5010

0150

2002

5030

0


Idaho

2003

0040

0500

6007

00


Illinois

100

200

300

400


Indiana

5010

0150

2002

5030

0


Iowa

100

200

300

400


Kansas

100

200

300

400


Kentucky

2003

0040

0500

6007

00


Louisiana

050

100

150

200


Maine

200

300

400

500

600


Maryland

1002

0030

0400

5006

00


Massachusetts

2003

0040

0500

600


Michigan

5010

015

020

025

0


Minnesota

1001

5020

0250

3003

50


Mississippi

200

300

400

500


Missouri

010

020

030

040

0


Montana

100

200

300

400


Nebraska

1002

0030

0400

5006

00


Nevada

4060

8010

0120

140


New Hampshire

1001

5020

0250

3003

50


New Jersey

200

400

600

800


New Mexico

2003

0040

0500

600


New York

2503

0035

0400

4505

00


North Carolina

010

020

030

0


North Dakota

1001

5020

0250

3003

50


Ohio

1002

0030

0400

500


Oklahoma10

020

030

040

0


Oregon

100

150

200

250

300


Pennsylvania

1001

5020

0250

300


Rhode Island

2004

0060

0800

1000


South Carolina

010

020

030

0


South Dakota

2003

0040

0500

600


Tennessee

200

300

400

500

600


Texas

5010

0150

2002

5030

0


Utah

4060

8010

0120

140


Vermont

100

150

200

250


Virginia

100

200

300

400


Washington

1001

5020

0250

300


West Virginia

5010

015

020

0


Wisconsin

5010

0150

2002

5030

0


Wyoming

29

Date post:	28-Dec-2021
Category:	Documents
Upload:	others
View:	2 times
Download:	0 times

THE CRIME KUZNETS CURVE

Documents