Development, crime and punishment: accounting for the international differences in crime rates

Transcription

1 Journal of Development Economics 73 (2004) Development, crime and punishment: accounting for the international differences in crime rates Rodrigo R. Soares* Department of Economics, University of Maryland, 3105 Tydings Hall, College Park, MD 20742, USA Graduate School of Economics, Getúlio Vargas Foundation, Rio de Janeiro, Brazil Received 1 January 2000; accepted 1 December 2002 Abstract This paper analyzes the determinants of the heterogeneity in crime rates across countries, focusing on reporting rates and development. The behavior of the reporting rate is studied by comparing data from victimization surveys to official records. Reporting rates are strongly correlated with development: richer countries report a higher fraction of crimes. The positive relation between development and crime found in previous studies is shown to result from this correlation. Once the presence of the reporting error is accounted for, development does not affect crime. Reductions in inequality and increases in growth and education are associated with reductions in crime rates. D 2003 Elsevier B.V. All rights reserved. JEL classification: K42; O10; O17; O57; Z13 Keywords: Crime; Development; Reporting rate; Inequality; Victimization 1. Introduction Crime rates vary enormously across countries, and their variation in this dimension is orders of magnitude larger than their variation through time in any given country. For example, the number of homicides per 100,000 inhabitants, probably the most popular crime statistic, ranges from 17 to 0.6 for countries like, respectively, Mexico and Japan. At * Department of Economics, University of Maryland, 3105 Tydings Hall, College Park, MD 20742, USA. address: soares@econ.umd.edu (R.R. Soares) /$ - see front matter D 2003 Elsevier B.V. All rights reserved. doi: /j.jdeveco

2 156 R.R. Soares / Journal of Development Economics 73 (2004) the same time, changes in homicide rates within a country, even in considerably long periods of time, rarely go beyond 20%. 1 The possible explanations for these cross-country differences are many, ranging from distinct definitions of crimes and different reporting rates (percentage of the total number of crimes actually reported to the police), to real differences in the incidence of crimes, due to different culture, religion, level of economic development or natural conditions. The goal of this paper is to analyze the causes of the differences in crime rates across countries, paying special attention to reporting rates and development. The economic theory of crime offers a natural theoretical benchmark for such analysis. In its setup, criminals respond to economic incentives in the same way that legal workers do (Becker, 1968; Stigler, 1970). 2 Particularly, the attractiveness of the criminal activity is intimately related to variables that undergo significant changes during the process of economic development, such as income distribution, urbanization, per capita income and institutional development. The relation between development and crime, thus, seems to call naturally for an economic interpretation. In this paper, we look at how the changes usually associated with economic development affect crime rates. In trying to assess this question, we face the traditional problem of underreporting present in the international data. A new victim survey crosssection is used, together with an official records panel, to overcome this problem. The use of the two data sets allows us to analyze and control for the presence of the reporting error. Our results suggest that the positive link between crime and development usually cited in the criminology literature but regarded with suspicion by economists does not exist. Reporting rates of crimes are strongly related to development, mainly income per capita. Therefore, the positive correlation between crime and development sometimes reported is entirely caused by the use of official records. Development is not criminogenic. Once we devise and use a correction procedure that takes into account the reporting error, the evidence indicates that economic development seems to be unrelated to crime rates. Income inequality affects crime rates positively, while education and growth reduce crime. The paper begins with a discussion of the links between economic development and crime predicted by economic theory (Section 2). Section 3 presents the existing empirical evidence linking crime and development and discusses its problems. Section 4 describes the two data sets used here one panel based on official records and one cross-section based on victim survey data and uses them to illustrate and analyze the bias induced by the official data. An alternative approach, which accounts for the reporting error, is proposed and applied in Section 5. Section 6 concludes the paper. 1 After sustaining an impressive and unprecedent decline in the homicide rate in the decade following 1990, the U.S. reduced this statistic by only roughly 25% of its initial value. 2 In Stigler s words, [this type of criminal] seeks income, and for him the usual rules of occupational choice will hold (Stigler, 1970, p. 530). Although a considerable fraction of crimes does not constitute economic crimes (such as sexual and hate), most of the quantitative differences in international crime rates are due to economic crimes. Besides, these crimes are likely to be the most elastic ones, in terms of responses to policy measures and to changes in economic conditions.

3 R.R. Soares / Journal of Development Economics 73 (2004) Economic development and the economics of crime Economic theories of crime relate the likelihood that an individual engages in criminal activities to the costs and benefits of these activities, when compared to legal occupations. At the aggregate level, the more prevalent the conditions which make crime attractive, the higher the crime rates. Ehrlich (1973) was the first to explicitly address this question. He constructed a model of participation on illegitimate activities, where individuals decided on the allocation of time across non-market, legal and illegal activities. Although much theoretical work has been done since then, most of his results still reflect the basic view of the profession. Taking the average income level as a measure of economic development, his model predicts that development has an indeterminate effect on crime rates. The direction of such an effect depends primarily on how risk aversion changes with income. Other interesting features of his model are the positive effect of inequality, and the negative effect of the probability of apprehension, on crime. The ambiguous effect of economic development obtained by Ehrlich may indeed be the reason why so little attention has been paid to the relationship between crime and development in economic literature. No significant work has been done in this area. However, the idea that the basic relation between income and crime should work through changes in the degree of risk aversion does not sound very intuitively appealing. In relation to the other variables, inequality is certainly the factor that received the most attention. For instance, Chiu and Madden (1998) have recently developed a model that analyzes in detail the determinants of burglary rates, concentrating on the discussion of the types of inequality increases that tend to increase crime. Bourguignon (1999), from an empirical perspective, investigated the sizeable economic costs that inequitable economic development may generate, via increases in crime and violence. The probability of apprehension has also been analyzed. Glaeser and Sacerdote (1996), for example, focused on the link between urbanization and crime, via an effect of the population density on the probability of apprehension. They also point other reasons one should expect population density to affect crime rates, such as higher pecuniary return to crime, social interactions and development of tastes. Other variables related to development that may be important in determining recorded crime rates are education and institutional development. Education may change criminal behavior via shifts on preferences or reporting behavior. Institutional development can help increase the confidence of the people in the system, thus increasing reporting rates, and make the record keeping inside the government more efficient, so that information is not lost once it enters the system. Although all these variables are thought to play some role in the determination of crime rates, their relative importance is still an empirical question. Besides, the effect of development itself which is theoretically indeterminate still remains the subject of debate. In this respect, most of the empirical studies available were not done by economists, and they are far from arriving at a consensus in terms of the effects of the different variables on crime. Section 3 presents a quick review of the existing empirical literature and discusses its main problems.

4 158 R.R. Soares / Journal of Development Economics 73 (2004) Previous evidence The empirical evidence on the determinants of cross-regional differences in crime rates is mainly concentrated on the effects of inequality and development (income level or some measure of poverty). As our primary interest here is also related to these two variables, we are going to center this quick review of the literature and much of the subsequent discussion on them (reviews of the criminology literature are also presented in Patterson (1991) and Fowles and Merva (1996)). Tables 1 and 2 summarize the results of several studies that tried to analyze the effects of, respectively, inequality and development on crime rates. These tables do not intend to be comprehensive reviews of all the evidence available, nor detailed descriptions of the techniques and strategies adopted in the various papers. Instead, their goal is to give a broad view of the general results obtained, and of how criminologists themselves see the present stage of this debate (in this direction, see Patterson, 1991 or Fowles and Merva, 1975). The statistical approaches used in the different studies are as diverse as they could possibly be, and, for this reason, we simply report the units and dimension of analysis, the types of crimes analyzed and the final conclusion as the authors themselves present it (or as their numbers would suggest). For the inequality case, 11 studies used cross-sections, 3 used panel data and 2 used time series; 13 used U.S. data (neighborhoods, cities, SMSA s, counties or national data) and only 3 used international data; the Gini coefficient was the choice of inequality measure in virtually all the cases (14). In the development studies, 16 cases used cross-sections, 6 used panel data and 1 used time series; 15 used U.S. data (neighborhoods, cities, SMSA s, counties or national data) and 8 used international data; the measure of development was income per capita in 4 cases, incidence of poverty (according to some income level or poverty line) in 15 cases and other measures (energy consumption, diversification of industry, etc.) in the rest. Specific details are presented in Tables 1 and 2. The major part of the evidence regards within United States studies, with the units changing from neighborhoods and cities to counties and metropolitan areas. As can be seen in Table 1, the results on inequality in this case vary between positive and nonsignificant from crime to crime and from study to study, leaving no clearly identifiable pattern. In relation to development, Table 2 shows that the U.S. studies most often indicate a negative effect of income level (or positive effect of poverty level) on crime rates, although non-significant and even positive results are sometimes present. Overall, it seems fair to say that the U.S. evidence suggests a negative effect of income levels on crime rates and, not very convincingly, a positive effect of inequality. The international evidence, surprisingly, suggests a conclusion strikingly different from this one. While the few inequality studies, as in the U.S. case, leave no clear answer, the evidence on development seems to be overwhelming: virtually, all the international evidence suggests that development and crime rates are positively and significantly correlated. This is certainly the most consistent of all the results that can be read from Tables 1 and 2. The only exception is the case of homicides. Although maybe surprising for economists, this result seems to be almost a stylized fact for criminologists and sociologists used to the international comparisons of crime rates.

5 R.R. Soares / Journal of Development Economics 73 (2004) Table 1 Summary of the evidence on the effect of inequality a on crime rates Study Unit/dimension of analysis Type of crime Conclusions Eberts and Schwirian SMSA s/cross-section Total crime (official data) (1968) Danziger and U.S. national data/ Burglary (official data) Wheeler (1975) time series Assault Robbery Danziger and SMSA s/cross-section Burglary (official data) Wheeler (1975) Assault Robbery Jacobs (1981) SMSA s/cross-section Burglary (official data) Grand larceny Robbery Blau and Blau (1982) SMSA s/cross-section Murder (official data) Rape Robbery Assault Messner (1982) SMSA s/cross-section Murder (official data) Carrol and Jackson U.S. cities/cross-section Burglary (official data) Positve effect (1983) Robbery Positve effect Crime against the person Positve effect Williams (1984) SMSA s/cross-section Homicide (official data) Bailey (1984) U.S. cities/cross-section Murder (official data) Stack (1984) Countries/cross-section Property crime (official data) Patterson (1991) U.S. neighborhoods/ Burglary (victim surv. data) cross-section Violent crime Fowles and SMSA s/panel Aggravated assault (off. data) Merva (1996) Murder Motor vehicle theft Larceny/theft Robbery Burglary Rape Allen (1996) U.S. national data/ Robbery (official data) time series Burglary Vehicle theft Fanjzylber et al. Countries/panel Homicide (official data) (1998) Robbery Kelly (2000) U.S. counties/ Violent crime (official data) cross-section Property crime Assault Robbery Murder Rape Burglary Larceny Car crime Fanjzylber et al. (2000) Countries/panel Homicide (official data) Robbery a Gini coefficient in 14 of the cases; various different measures in Eberts and Schwirian (1968) and Fowles and Merva (1996).

6 160 R.R. Soares / Journal of Development Economics 73 (2004) Table 2 Summary of the evidence on the effect of development a on crime rates Study Unit/dimension of analysis Type of crime Conclusions Wolf (1971) Countries/panel Total crime (official data) Larceny Murder Wellford (1974) Countries/cross-section Homicide (official data) Sex offence Major larceny Minor larceny Fraud Counterfeit Drug Total crime Harries (1976) U.S. cities/cross-section Robbery Aggravated assault Burglary Auto theft McDonald (1976) Countries/cross-section Juvenile crime (official data) Theft Property Total crime Murder Krohn and Countries/cross-section Homicide (official data) Wellford (1977) Property crime Total crime Krohn (1978) Countries/cross-section Homicide (official data) Property crime Total crime Decker (1980) U.S. cities/cross-section Violent crime (official data) Property crime Viol. crime (vict. sur. data) Property crime Stack (1984) Countries/cross-section Property crime (official data) Watts and Watts (1981) U.S. cities/cross-section Major crimes (official data) Blau and Blau (1982) SMSA s/cross-section Murder (official data) Rape Robbery Assault Crutchfield et al. (1982) SMSA s/cross-section Robbery (official data) Assault Burglary Messner (1982) SMSA s/cross-section Murder (official data) Sampson and U.S. neighborhoods/ Theft viol. crim. Castellano (1982) panel (vict. sur. data) Messner (1983) SMSA s/cross-section Homicide (official data)

7 R.R. Soares / Journal of Development Economics 73 (2004) Table 2 (continued) Study Unit/dimension of analysis Type of crime Conclusions Loftinm and Parker U.S. cities/cross-section Total crime (1985) Family crime Other primary crime Homicide Messner and Manhattan Homicide (official data) Tardiff (1986) neighborhoods/ cross-section Sampson (1986) U.S. neighborhoods/ Theft (victim survey data) panel Personal crime Patterson (1991) U.S. neighborhoods/ Burglary (victim surv. data) cross-section Violent crime Fowles and SMSA s/panel Aggrav. assault (off. data) Merva (1996) Murder Motor vehicle theft Larceny/theft Robbery Burglary Rape Allen (1996) U.S. national data/ Robbery (official data) time series Burglary Vehicle theft Fanjzylber et al. Countries/panel Homicide (official data) (1998) Robbery Kelly (2000) U.S. counties/ Violent crime (official data) cross-section Property crime Assault Robbery Murder Rape Burglary Larceny Car crime Fanjzylber et al. Countries/panel Homicide (official data) (2000) Robbery a Income per capita or (inverse of) poverty index in 21 cases; various different measures in Wolf (1971) and Krohn (1978). Burnham (1990, p. 44), for example, in trying to set an agenda for the contemporary study of crime and development, argues that evidence as exists seems to suggest that development is indeed probably criminogenic. Along the same lines, Stack (1984, p. 236), when trying to select control variables to include together with a measure of inequality in his regression, decides to include the level of economic development, a factor found to be related positively to property crime rates in the previous cross-national research. Other papers cited in Table 2 also present arguments in this direction, together with intellectual roundabouts that try to rationalize these results. Nevertheless, these results may have an explanation far more simple than the industrialization induced social disintegration usually suggested in the sociological

8 162 R.R. Soares / Journal of Development Economics 73 (2004) literature. One major statistical problem is systematically overlooked in the crossnational studies discussed here: the non-randomness of the reporting error. Official data is known to greatly underestimate actual crime rates, and this can constitute a serious problem if the degree of underestimation is correlated with the characteristics of the country. If this is really the case, the evidence cited above cannot be seriously taken into account until one is able to determine the degree of bias introduced by the reporting error. Some of the previous studies acknowledged this problem and its potential severity, and tried to concentrate the analysis on crimes thought to be less subject to it (Fanjzylber et al., 1998, 2000 center their discussions in homicides), while most papers simply ignored it (Krohn and Wellford, 1977; Krohn, 1978; Stack, 1984). In Section 4, we use a new cross-country data set, based on victim survey data, and a traditionally used data set, based on official records, to analyze the characteristics of the reporting error and the kind of bias that the use of official data may introduce. 4. The international data on crime rates 4.1. The data There are few sources of data on crime rates for different countries and, until very recently, all the information available was based on official records. In the end of the 1980s and beginning of the 1990s, a new data set based on victimization surveys was compiled by a group of different institutions. This data set constitutes today the International Crime Victim Survey (ICVS), a survey conducted by a group of international research institutes under the coordination of the United Nations Interregional Crime and Justice Research Institute (UNICRI). It contains data for selected countries, irregularly distributed over the years 1989, 1992 and/or 1996/1997. This data set has the obvious advantage of being free of the reporting problem typical of the official data, but it has the drawback of, up to now, being useful only as a crosssection. 3 Other data sets available, thus, still keep their relevance because they allow an exploration of the panel feature of the crime phenomenon. It is, therefore, important to understand the behavior of the reporting problem in the official data, and whether it is still possible to use this kind of information in any meaningful way. To address this question, we use the United Nations Survey of Crime Trends and Operations of Criminal Justice Systems (UNCS). This is a data set created by the United Nations with information related to several crime and justice related variables, based on official records. Several countries and years are irregularly covered in the period between 1971 and We concentrate our analysis on the three types of crimes that can be compared across the victim survey (ICVS) and the official records survey (UNCS). The definitions of these 3 As mentioned, the panel feature cannot be explored because the time span is still very short and observations are very irregularly distributed across countries and time periods.

9 R.R. Soares / Journal of Development Economics 73 (2004) crimes are presented below, together with the way in which the UNCS data were made compatible to the ICVS. Thefts: thefts of bicycle, motorcycle, other personal thefts, pick-pocketing and car crimes on the ICVS. Thefts and major thefts on the UNCS. Burglaries: burglaries and attempts on the ICVS. Burglaries on the UNCS. Contact crimes: robberies, sexual incidents and/or threats/assaults on the ICVS. Major assaults, assaults, rapes and robberies on the UNCS. Although there is considerable heterogeneity within some of these categories, we believe that heterogeneity across groups is much larger, such that thefts, burglaries and contact crimes actually do represent very distinct types of crimes. Thefts here are property crimes that generally do not involve direct contact between victim and perpetrator, and do not involve invasion of a house or building. Burglaries are very narrowly defined, and have a perfect match between the two data sets. Contact crimes, on the other hand, include all sorts of crimes that involve some form of physical violence or threat. It is certainly the most heterogeneous group of the three, since it includes not only property crimes, such as robbery, but sexual crimes as well. Table 3 presents some descriptive statistics for the two data sets (countries included in at least part of the sample are listed in Appendix A). The numbers are extremely different. Comparing the cross-country averages from the ICVS with the ones from the UNCS (based on a within country average from 1989 to the last year available), we have the following numbers: according to the official records, 2.1% for thefts, 0.7% for burglaries and 0.3% for contact crimes; according to the victim survey, 25.1% for thefts, 6.7% for burglaries and 7.7% for contact crimes. Although the magnitude may be surprising, the underestimation present on the official data was already expected. It does not constitute a problem in itself if it is not correlated with the countries characteristics. Table 3 Descriptive statistics Official data Victim survey data Theft Burglary Contact Theft Burglary Contact Mean S.D Max Min No. of obs Correlations Theft Burglary Contact Obs.: Data is number of crimes as a percentage of population. Official data is taken from the UNCS data set and victim survey data from the ICVS. For comparability between the two data sets, statistics for the official data are calculated from country averages, from 1989 to the last year available. ICVS data are averages for all the surveys in which the country was included (1989, 1992 and/or 1996/1997).

10 164 R.R. Soares / Journal of Development Economics 73 (2004) Table 4 Cross-section regressions Official data Victim survey data Thefts ln(gnp) (0.3191) (0.2443) (0.1610) (0.1459) (0.0625) (0.0374) Ratio (0.0502) (0.0624) (0.0622) (0.0279) (0.0120) (0.0095) Urb (0.0206) (0.0169) (0.0132) (0.0101) (0.0056) Educ (0.0356) (0.0241) (0.0175) (0.0061) Growth (0.0533) (0.0419) (0.0205) (0.0098) ln(pol) (0.1950) (0.0576) Chr (0.6939) (0.2657) Const (3.2632) (2.9546) (1.3275) (1.1502) (0.5418) R No. of obs (0.0037) (0.2657) Burglaries ln(gnp) (0.2559) (0.2818) (0.1837) (0.3068) (0.1623) Ratio (0.0426) (0.0596) (0.0600) (0.0452) (0.0205) Urb (0.0123) (0.0174) (0.0134) (0.0213) (0.0109) Educ (0.0457) (0.0331) (0.0279) (0.0138) Growth (0.0421) (0.0453) (0.0457) (0.0209) ln(pol) (0.0929) (0.1209) Chr (0.5213) (0.5111) Const (3.4295) (3.0923) (1.3400) (2.4265) (1.2129) R No. of obs (0.1023) (0.0155) (0.0074) (0.5786) Contact crimes ln(gnp) (0.2012) Ratio (0.0407) Urb (0.0132) (0.1769) (0.0624) (0.0139) (0.1661) (0.0543) (0.0139) (0.1154) (0.0196) (0.0061) (0.0768) (0.0163) (0.0060) (0.0816) (0.0110) (0.0058)

11 R.R. Soares / Journal of Development Economics 73 (2004) Table 4 (continued) Official data Victim survey data Contact crimes Educ (0.0249) (0.0200) (0.0181) (0.0086) Growth (0.0416) (0.0363) (0.0215) (0.0123) ln(pol) (0.1667) (0.0656) Chr (0.4576) (0.2875) Const (2.4509) (2.2378) (1.1328) (1.3657) (0.7221) (0.4471) R No. of obs Obs.: Numbers below the coefficients are robust standard errors (Huber/White/Sandwich estimator of the variance used). For comparability between the two data sets, statistics for the official data are calculated from UNCS country averages, from 1989 to the last year available. Victim survey data are averages for all the ICVS surveys in which the country was included (1989, 1992 and/or 1996/1997). Dependent variable is the log of the number of thefts, burglaries or contact crimes as percentage of the total population. Independent variables are ln of the GNP per capita; ratio between income or consumption per capita of the 20% richest and of the 20% poorest; percentage of population living in urban areas; ln of the number of policemen as a percentage of the population; gross primary enrollment rate; average growth rate of the GNP per capita in the period; and a dummy indicating whether at least 60% of the population is Christian. In terms of the cross-country differences in crime rates, it is interesting to notice that the official data seems to increase the dispersion of the cross-country distribution in relation to its mean: while the ratio of the standard deviation to the mean is between 1.08 and 1.25 for the three types of crimes in the official data, it is between 0.27 and 0.56 in the victimization data. This tends to increase the relative differences across countries in the official records in relation to the differences in the victimization data, suggesting that there is some noise added when we go from the victimization to the official data. Anyhow, even the victim survey shows that crime rates can be quite different across countries. Some places have from 4 to 20 times higher rates than others, illustrating the relevance of the problem that we want to analyze. We now present some evidence on the different conclusions that are obtained when each of these alternative data sets is used Cross-section analysis As an exploratory approach, we run, for the two data sets, cross-section regressions of the three types of crimes on our variables of interest and on a set of control variables. The UNCS data used is the average for each country of the crime rates between 1989 and the last year available. The basic specification of the regressions is: lnðcrime i Þ¼b 0 þ b 1 lnðgnp i Þþb 2 ratio i þ b 3 urb i þ b 4 educ i þ b 5 growth i þ b 6 lnðpol i Þþb 7 chr i þ e i ; ð1þ

12 166 R.R. Soares / Journal of Development Economics 73 (2004) where ln(crime) is the natural logarithm of the different measures of crime (as percentages of the population); ln(gnp) is the natural logarithm of the GNP per capita at constant 1995 US$; ratio is a measure of inequality, based on the ratio between the share of income or consumption of the 20% richest of the population to the share of income or consumption of the 20% poorest; 4 urb is the percentage of the population living in urban areas; educ is the gross primary enrollment rate; growth is the average GNP per capita growth in the period; ln(pol) is the natural logarithm of the percentage of policemen in the population; and chr is a dummy indicating whether the religious majority in the population is Christian. 5 Appendix B presents the description and sources of these variables. The first three right hand side variables together with economic growth are the development related variables that constitute our main interest. The education and religion variables are introduced as controls for possible taste shifts that may be correlated with economic development itself, and may affect both crime and reporting behavior. The police variable is a natural control for the crime prevention measures taken by the different countries. Three specifications of this equation are run for each type of crime. We begin by including all six variables, and then consecutively exclude the police and religion indicators, and the growth and education variables. Table 4 presents the results. The first three columns are related to the official data and the last three to the victim survey data. Robust standard errors are used in all cases. Some clearly identifiable differences arise when we compare the results from the two different data sets. 6 In the official data regressions, the effect of income (ln(gnp)) is always positive and statistically significant. For eight out of nine cases, the effect of inequality (ratio) is positive, but only the results for contact crimes are statistically significant. The 4 Atkinson and Brandolini (2001) raise doubts regarding the international comparability of the income distribution data collected by Deininger and Squire (1996). It is possible that our inequality variable based on the share of income or consumption of different groups of the population suffers from the same kind of problems they discuss. This is an issue that cannot be dealt with in this paper, but should be kept in mind when analyzing the results. Measurement error on the inequality variable alone would bias its coefficient towards zero and the coefficients on the other independent variables in unpredictable ways. 5 The coefficients have the following interpretation: for ln(gnp) and ln(pol), they are simply elasticities; for ratio, urb, educ and growth, the relative change on the dependent variable given a one unit change in the independent variable (respectively, a one time increase in the gnp of the 20% richest of the population in relation to the gnp of the 20% poorest, 1% more of the population living in an urban area, 1% more primary enrollment or 1% more economic growth); for the religion dummy, the relative increase in crime if the country has a majority of that religion. It is important to keep in mind that these are percentage and relative changes on the rates of crimes, not absolute changes in its level. 6 There is a well-known problem of endogeneity of the police variable here (see, for example, Levitt, 1997). As we did not find a good instrument for it, we chose to present the equations with and without the ln(pol) variable included on the right-hand side. As can be seen from the tables, there is almost no change on the qualitative results (in terms of significance and sign of the coefficients) as police is excluded from the regressions. Besides, the presence of the police variable in the UNCS data set is very irregular, so that its inclusion in the regression hugely reduces the number of observations. For these reasons, we ignore the coefficients on the police variable in the following discussion, since we do not have any particular interest on them. Just for the record, it shows up as positive and borderline significant in the regression for official data on thefts, and in both regressions for burglaries.

13 R.R. Soares / Journal of Development Economics 73 (2004) effect of urbanization (urb) has different signs in the different specifications, never being significant, and the same happens for education (educ) and growth (growth). For the victim survey data, in six out of nine cases the effect of income is negative (in three cases it is borderline significant, although only one case is significant at the 5% level). The effect of inequality is positive, again, in eight out of nine cases, and it is significant or borderline significant in all the cases. Urbanization has a positive effect in seven cases, being close to significant for all the burglaries regressions and for the shortest specification for both thefts and contact crimes. All the other variables are generally not significant and change signs in the different specifications. Given this general description, it seems fair to say that the two data sets describe very different pictures regarding the relationship between crime and development. The official records suggest that crime rates increase with income per capita, and seem to be positively affected by inequality, although the evidence regarding the latter is not very strong. On the other hand, the victim survey indicates that, if anything, crime rates seem to decrease with income per capita, although a more precise statement would be that these two variables are not very strongly related. Moreover, inequality has a strong correlation with crime rates in this case and there is some weak evidence that urbanization may also have a positive effect. If one believes that there is a reporting problem in the official data, and that this problem is less severe in the victim surveys, these different results are actually telling something about the nature of the reporting error, and the way in which it correlates with the independent variables. The cross-section regressions indicate that the reporting error is not random, and introduces systematic biases on the estimates obtained from official data. The extremely different conclusions obtained from the two data sets, particularly in respect to income, support the hypothetical relation between underreporting and economic development, and indicate that this relation is serious enough to call into question the results of the studies discussed in Section 3. In Section 4.3, we analyze explicitly the determinants and the characteristics of the reporting rate The determinants of the reporting rate Evidence from Section 4.3 stresses the importance of the underreporting of crimes in official data, and strongly suggests that it may be affected by variables related to development. 7 If we assume that the victim survey data represent the real crime rate or, at least, that their deviations from the real rate are not correlated with the exogenous variables, we can use the two different data sets to recover a cross-section of the reporting error. This cross-section can then be used to analyze the relation between the reporting error and the development-related variables. 7 The analysis of the differences between data from official records and from surveys is a recurrent subject in applied criminology research. References in this area include Kitsuse and Cicourel (1963), Skogan (1976), Cohen and Land (1984), Biderman and Lynch (1991), Figlio (1994), O Brien (1996), Levitt (1998) and many others. Although the topics covered in this literature are very diverse, the discussion is almost always centered on national data (where the problem is most likely less serious), and nobody addresses the same problem that we are trying to address here.

14 168 R.R. Soares / Journal of Development Economics 73 (2004) We do this by constructing a measure of the reporting rate (fraction of crimes actually registered by the official records) and running the following regression: lnðrrate i Þ¼a 0 þ a 1 lnðgnp i Þþa 2 educ i þ a 3 urb i þ a 4 ratio i þ a 5 lnðpol i Þ þ a 6 chr i þ m i ; ð2þ where rrate is the reporting rate for the different crimes. This variable is constructed as the ratio of the crime rate obtained from official records (registered crimes) to the crime rate obtained from the victim survey ( real number of crimes). 8 The specification of this equation tries to capture the different factors that may play a role in determining reporting rates. Income per capita is the most commonly used indicator of overall development level, and it is highly correlated with several factors omitted from this equation, such as institutional development, degree of law enforcement, and corruptibility of the police. The educational variable tries to capture the population s degree of knowledge of individual rights, and also social transformations that may affect the reporting rates of certain types of crimes, such as sexual and violent ones. Level of urbanization and police presence are important determinants of the costs of access to police departments or law related offices, as well as of the ability of the information to navigate the system, from where the crime was first registered to the central office where statistics are computed. Finally, inequality may affect the level of social conflict inside a country, and religion affects, through culture, the habits of the population. These two factors may be important in explaining how citizens regard their country s institutions, and how much they respect and trust them. Table 5 shows the results of this regression with two sets of independent variables: all the variables included in the specification above and only ln(gnp). The numbers are overwhelming: the full specification explains 83% of the underreporting for thefts, 83% for burglaries and 65% for contact crimes. Indeed, despite the fact that a couple of variables, such as urb and ln(pol), are borderline significant in some cases, the variable ln(gnp) is by far the most important factor: it alone explains 68% of the underreporting for thefts, 60% for burglaries and 48% for contact crimes. Countries with higher income per capita have significantly and systematically higher reporting rates. Figs. 1 3 plot the ln(rrate) against the ln(gnp) for, respectively, thefts, burglaries and contact crimes. The close positive association between reporting rate and income is clear in all three graphs. The coefficient on income per capita here can be interpreted as an elasticity, which means that a 10% increase in gnp increases the reporting rate in 9% for thefts, 10% for burglaries and 6% for contact crimes. These elasticities imply that if a country like Colombia increases its gnp to the level of Netherlands, its reporting rate for thefts will increase from 0.7% to 9%, for burglaries from 0.4% to 6% and for contact crimes from 1.4% to 12%. These results also show where the frequently documented positive relation between development and crime comes from: it is a product of the positive correlation between 8 The victim survey rate of burglaries for Finland is slightly smaller than the rate calculated from the official data (UNCS). That gives a reporting rate bigger than 100%. Since this was the only case for which this happened, we think that it is probably due to some minor measurement error, and ignore it.

15 R.R. Soares / Journal of Development Economics 73 (2004) Table 5 Cross-section regressions for the reporting rate Thefts Burglaries Contact crimes ln(gnp) (0.2019) (0.1085) (0.3629) (0.1358) (0.1659) Educ (0.0282) (0.0541) (0.0213) Urb (0.0135) (0.0195) (0.0112) Ratio (0.0320) (0.0520) (0.0377) Chr (0.6349) (0.8151) (0.4655) ln(pol) (0.2080) (0.1785) (0.1479) Const (2.4798) (0.9960) (4.0679) (1.1771) (1.7210) R No. of obs (0.1011) (0.9120) Obs.: Numbers below the coefficients are robust standard errors (Huber/White/Sandwich estimator of the variance used). Dependent variable (reporting rate) is defined as the ln of the ratio of official crime rates (UNCS averages from 1989 to last year available) to victim survey crime rates (ICVS averages from 1989, 1992 and/or 1996/1997 surveys). Rates are number of crimes as a percentage of total population. Independent variables are ln of the GNP per capita; gross primary enrollment rate; percentage of population living in urban areas; ratio between income or consumption per capita of the 20% richest and of the 20% poorest; ln of the number of policemen as a percentage of the population; and a dummy indicating whether at least 60% of the population is Christian. development and reporting rates. In other words, developed countries report more crimes (as a percentage of the total number of crimes), and no evidence presented until now suggests that developed countries actually have higher crime rates. The conclusions from the studies cited in Section 3 were seriously harmed by the use of official data. In addition, this evidence sheds light on the distinct relation between development and homicide reported by these same studies. It is likely that the elasticity of the reporting rate in relation to development is much smaller for homicides than for other types of crimes. Death certificates have (almost) always to be filed, and when the cause is identified as homicide, the crime must be reported to the police. In this case, reporting does not depend directly on the willingness of citizens, and the record keeping has automatic mechanisms that work outside of the police and judicial structures. Another curious fact from Tables 1 and 2 that is perfectly consistent with the results from this section is the difference in conclusions when we compare U.S. and international studies. The fact that the within U.S. studies do not suggest any clear relation between crime and development, while the international studies do, should be expected, since reporting rates probably vary much more across countries than within countries. Nevertheless, we do not read these results as implying that income, per se, tends to increase the fraction of crimes reported to the police. As mentioned before, we think that the ln(gnp) variable is capturing effects related to development that cannot be precisely

16 170 R.R. Soares / Journal of Development Economics 73 (2004) Fig. 1. Relation between reporting rates and GNP for thefts.

17 R.R. Soares / Journal of Development Economics 73 (2004) Fig. 2. Relation between reporting rates and GNP for burglaries.

18 172 R.R. Soares / Journal of Development Economics 73 (2004) Fig. 3. Relation between reporting rates and GNP for contact crimes.

19 R.R. Soares / Journal of Development Economics 73 (2004) measured, such as institutional development and degree of law enforcement. To uncover the exact mechanism behind this relation is an important topic for future research, but the simple fact that this correlation exists is enough to bias the coefficients on development related variables estimated from traditional crime regressions. Despite this problem, if one wants to explore the time dimension of the behavior of crime rates, hope will still rest on official data sets, like the UNCS, since the International Crime Victim Survey is very recent. For this reason, in Section 5, we suggest a way of correcting the official records with information obtained from the victim survey crosssection, such that the reporting error is taken into account. The idea is to understand under what conditions, observing only one cross-section of reporting rates, we would be able to estimate the underreporting structure and, with that in hand, eliminate the bad variation from the official data. If these conditions are not very strict, they will allow the use of the panel data set from the UNCS, controlling for contamination from the reporting error. In the last two parts of the section, we apply the strategy to the UNCS data and discuss the results. 5. An alternative use of the official data 5.1. Econometric approach Suppose that crime rates are determined according to the following equation: Y * ¼ X h þ e; ð3þ where Y* stands for the logarithm of the crime rate of a specific type of crime, X is a vector of country s characteristics and e is an error term for which Cov(e,X)= 0. The only data observed on a panel basis is the official data, which is the true data plus a reporting error : Y ¼ Y * þ m: ð4þ The reporting error, as the evidence presented on the preceding sections suggests, is assumed to be correlated with the country s characteristics, such that Cov(m,X ) p 0. For the usual reasons, if we regress Y on X, we get a biased estimator of h, for which E(ĥAX)= h+(xvx) 1 XVE(mAX ) p h. If we could obtain an estimator of m such that E(mˆAX)=E(mAX), we could build the series ( Y mˆ), and regress it on X to obtain an unbiased estimate of h. The only hope in this direction lies with the cross-section observations available from the victim survey data (from the ICVS data set). The comparison of this data with the UNCS data (based on official records) allows us to build a vector m t of cross-section observations of the reporting error at a given point in time. If, additionally, the joint distribution of r and X is invariant across countries and time, this single cross-section will allow us to obtain all the relevant information regarding the correlation between m and X. Maintaining this assumption, and supposing that m and X are jointly normally distributed, we have that EðmAX Þ¼X c ð5þ

20 174 R.R. Soares / Journal of Development Economics 73 (2004) where c is the vector of coefficients of the linear regression of m on X (where X includes the unit vector). In this case, the projection of the cross-section vector r t on the corresponding matrix X t will, given our invariance assumption, give an unbiased estimate of c. We can then go on to construct mˆ = Xĉ for all the periods and countries covered by the official data, with E(mˆAX)=XE(ĉAX)=XE(ĉAX t )=Xc = E(mAX). With this estimate of m in hand, the official data Y can be corrected and an unbiased estimate of h can be obtained from the regression of ( Y mˆ) on X. The Appendix derivation proves that this procedure produces an unbiased estimator of h and, additionally, derives an unbiased estimator of its covariance matrix. In Section 5.2, we apply this strategy to our data set Estimation We apply the approach described in Section 5.1 to the UNCS data set, using the crosssection from the ICVS to construct the vector m t. The matrix X here has the same variables used in the right hand side of the cross-section equations (see Eq. (1)). Due to data limitations, the income inequality variable is country specific, in the sense that it changes from country to country, but remains constant for a given country through time. Deininger and Squire (1996) have noticed, after extensively documenting the methodology and availability of international data on inequality, that changes in the Gini coefficient of inequality tend to be small compared to changes in other economic variables (Deininger and Squire, 1996, p. 587). This reduces the concern in relation to this limitation of the data. The religion dummy, for obvious reasons, is also constant through time, while all other variables change with time and country. The characteristics matrix X can, thus, be divided into two subsets, V and F, where the typical vector of V is the time and country variant m it, and the typical vector of F is the country variant and time fixed f i. Estimates of pooled regressions and within and between decompositions are presented. In relation to ĉ (the estimated c) and the correction of Y discussed in Section 5.1, one main concern guides our approach. We want to eliminate the correlation between X and m from the official data, but we want c to be estimated with some precision, to avoid actual differences in crime rates to be also eliminated from the data. This constitutes a problem since the cross-section that we have available for m is a small sample, and some of the variables included in X are highly correlated with each other. For this reason, we decide to restrict the X s included in Eq. (5) only to those that show up significantly in the reporting rate regressions, and so, based on the evidence from Table 5, we end up using only the ln(gnp) to correct the official data. 9 The estimated c is then used to correct the observed Y in the way described in Section Inclusion of other variables, such as urb and educ, do not change the main conclusions. These results are available from the author upon request. 10 This approach limits the use of the correction procedure proposed as a tool for prediction. As income level is probably capturing other variables correlated with development, and the relation between income and these variables may not be stable through time, to use the estimated relation to forecast reporting rates in the long run may be misleading. This would correspond, in the econometric model discussed, to the relation between m and the observed X not being invariant through time.