Inter-American Development Bank

Inter-American Development Bank Regional Policy Dialogue Poverty Reduction and Social Protection Network Studies on Poverty and Social Protection

Crime Victimization and Income Distribution Rafael Di Tella HBS Sebastian Galiani UdeSA and Ernesto Schargrodsky* UTDT

May 2, 2002

Abstract Crime levels have risen significantly in Argentina during recent years. In this study, we analyze the relationship between crime victimization and income distribution. Our main question is whether the rich or the poor have been the main victims of this crime rise. For home robberies, we found that the poor have suffered the main crime increases. For street robberies, both groups show similar augments in victimization. The findings are consistent with additional evidence showing that the rich are able to protect their houses through pecuniary security devices better than the poor. We find no differences in reporting rates and access to justice by social groups. JEL: K42 Keywords: Crime victimization, income distribution, private security.

*

Rafael Di Tella, Harvard Business School, Boston, MA 02163, US, [email protected], Tel: 617-4955048. Sebastian Galiani, Universidad de San Andres, Vito Dumas 284, (B1644BID) Victoria, Provincia de Buenos Aires, Argentina, Tel: (54-11) 4746-2608, [email protected]. Ernesto Schargrodsky, Universidad Torcuato Di Tella, Miñones 2177, (1428) Capital Federal, Argentina, [email protected], Tel: 54-11-47840080. This research is carried out with the support of the IDB Poverty Reduction and Social Protection Network. Matias Cattaneo provided excellent research assistance.

1. Introduction During the 1990s, official statistics show that the number of reported property crimes increased in Argentina by 85%, from 404,465 in 1990 to 753,727 in 2000, while the number of total crimes doubled from 560,240 in 1990 to 1,129,900 in 2000.1 This increase has been particularly severe in the second part of the decade. Unfortunately, Argentina has become a much more insecure country than it used to be. Not long ago, Buenos Aires was known as one of the safest Latin American cities. The rest of the country was even safer. Crime has now risen sharply by any measure. Figure 1 ilustrates the evolution of reported property crimes, total reported crimes, total criminal sentences, and unemployment since 1990.

Figure1 Crime and Unemployment Evolution, 1990-2001. 20

30,000 Property Crimes (in hundreds)

18

Total Reported Crimes (in hundreds)

25,000

Total Criminal Sentences

16 14

20,000

12 10

15,000

8 10,000

6

Unemployment Rate

Unemployment

4

5,000

2 0

0 1990

1991

1992

1993

1994

1995

1996

1997

1998

1999

2000

2001

The deterioration in the social conditions worsened since the country entered in an endless recession in 1998. Unemployment, which was at 6% at the beginning of the 1990s, reached 17.4% in 2001. It is expected to peak around 24% in the next household

1

Due to underreporting, these crime figures highly underestimate the problem. According to a recent victimization survey, 41% of the surveyed people declare to have suffered a crime during 1999, but only 29% of those crimes were reported to the police (Ministerio de Justicia, 2000). Fajnzylber, Lederman, and Loayza (1998) report a significant increase in crime throughout the world, including Latin American countries.

1

survey of May of 2002. Although official crime measures are not yet available for 2001, the number of police officers killed on duty in the Buenos Aires Metropolitan Area has set a sad record in that year. After the assassination of 55 police officers during 2000, 82 officers were killed in action last year. The figure has already reached 27 during the first three months of 2002 (Clarin, April 10, 2002). Crime is one of the main concerns of the population according to a recent opinion poll (Clarin, April 29, 2001).2 More than 30% of the people interviewed in our survey report that at least one member of his or her household has been robbed during 2001. In Ministerio de Justicia (2000), 35.5% of the households report to have suffered a crime during 1999. Another recent survey (Catterberg & Asoc, 2001) shows that 23.7% of the interviewed people declare to have suffered a crime during the past 12 months, and 5.1% during the last month. The figures reach 40.4% and 13.6%, respectively, when the question refers to all the members of the household.

The increase in crime rates is a pretty well documented fact. The purpose of our study is to analyze which social groups have been the main victims of this significant increase in crime. In order to address this question empirically, we run a survey on current and past victimization rates and income levels. For home robberies, we found that the poor have suffered the main crime increases. During the first half of the decade, high-income households had suffered a significantly higher victimization rate than low-income households. The difference has now turned non-significant. For street robberies, instead, both groups show similar augments in victimization. We find no significant differences in reporting rates and access to justice by social groups.

Our findings are consistent with additional evidence showing that the rich are able to protect their houses through pecuniary security devices better than the poor. The use of private security devices was highly unusual in Argentina a few years ago. The use of private security in home protection was restricted to a small number of very high-income 2

In that poll, 74.6% of the surveyed people declare to be concerned about unemployment, 66.3% concerned about crime, and 42.9% concerned about corruption, as the three main population worries. Opinion polls in 17 Latin American countries describe violence as “the region’s main social and economic problem” (Londoño and Guerrero, 1999).

2

people. Private neighborhoods with gated access (barrios privados) were largely nonexistent. With low levels of private protection and a relatively even distribution of public police protection, criminals tended to target high-income booties. However, this reality changed drastically when unemployment and crime started to rise. As the likelihood of becoming the victim of a crime raised, high and middle-income citizens increased the consumption of private security devices. Many high and middle-income neighborhoods have now private security protection and gated access. Homes and cars are now protected by a variety of security devices. Our findings show that high-income groups are more likely than low-income groups to use private security, alarms, insurance, and armored doors (instead, low-income families are more likely to have a watchdog for security purposes). However, the ability to use protection devices against street robberies seems to be limited.

If high and middle-income citizens responded to the crime increase by protecting their homes, low-income houses might have suffered the main burden of the crime increase, suggesting the presence of significant negative externalities associated with the use of home security devices (Di Tella and Schargrodsky, 2001). Without a similar ability to hire street protection devices, both rich and poor seem to have suffered a similar increase in street victimization rates.

We consider that it is important to analyze whether the poor are the main victims of crime and its increasing trend in Latin America. The social costs of crime in the region amount to 14.2% of GDP according to Londoño and Guerrero (1999)’s estimates. In addition to the direct costs of crime and the value of stolen goods, crime affects the access to capital markets (Miller, 2001), reduces the productivity of businesses, induces underinvestment, lowers the accumulation of social and human capital, and affects labor force participation (Fajnzylber, Lederman, and Loayza, 2000a; and Andalon-Lopez and Lopez-Calva, 2001). If the poor are the main victims of crime, and crime has a deleterious effect on social and economic development, then crime may be playing a significant role in the reproduction of poverty in the region.

3

In the rest of the paper, Section 2 describes the previous literature and its relationship to the empirical exercise that we conduct. Section 3 presents a model that ilustrates the effect of increases in victimization by social groups. Section 4 describes our survey, while the results are presented in Section 5. The last section discusses policy implications and concludes.

2. Literature Review The relationship between the protection of property rights and income distribution has received considerable attention. However, this attention has mostly focused on analyzing the effect of income distribution on crime levels. In one of the first papers in the crime literature, Ehrlich (1973) pointed out that both theory and empirical evidence suggest a positive correlation between income inequality and crimes against property. Since then, there has been a series of papers that have dealt with this question, either directly or indirectly, using a variety of empirical strategies.

A natural idea is to exploit cross-country variations in income inequality. Most work on crime, however, has been reluctant to make international comparisons. Cross-country studies of crime have important data problems because official crime data is often not comparable across countries. There are differences in underreporting and in the definitions of crime in different countries. Fajnzylber, Lederman and Loayza (2000b) is one of the first attempts at explaining crime patterns across countries. To solve the data limitations outlined above they focus on the rate of homicides (per 100,000 people) as the proxy for violent crimes, arguing that this type of crime suffers the least from underreporting and idiosyncratic classification. The authors use a sample of 45 developed and less developed countries for the period 1965-1995. The main conclusion of the paper is that income inequality, measured by the Gini index, has a robust, significant and positive effect on the incidence of violent crimes. The econometric methodology is a GMM IV estimator that controls for unobserved country-specific effects and the possible endogeneity of the explanatory variable. The strong correlation between income inequality, measured by the Gini index, and crime survives after controlling for relative

4

poverty, ethnic polarization, unequal distribution of police protection, and access to the judicial system.

An interesting aspect of this study is the connection between income inequality and economic opportunities in the fight against crime. Violent crime rates decrease when economic growth improves. However, the crime-reducing impact of the GDP growth rate is weaker when income inequality is larger. Interestingly, relative poverty does not seem to affect the crime rate once income inequality is included in the regression. It does, however, have an indirect impact. The positive effect of inequality falls as the poor become richer. It is also worth noting that crime exhibits a considerable degree of persistence measured by a large and significant coefficient on the lagged dependent variable.

Andalon-Lopez and Lopez-Calva (2001), and Fajnzylber, Lederman, and Loayza (2000a) also analyze the relationship between crime, income inequality, and polarization. Raphael and Winter-Ebmer (2001) summarize the results of 68 studies on unemployment and crime. The literature finds a strong positive relationship between the unemployment rate and, on the one hand, income inequality (e.g. Blinder and Esaki (1978)) and, on the other, crime rates (Ehrlich (1973), Freeman (1991), Tauchen, Witte and Griesinger (1994), inter alia -but also see Papps and Winkelman (1996)-).

A shorter literature, however, has analyzed the different question of who are the main victims of crime. From the basic model of crime of Becker (1968), one would expect that the rich are more attractive targets for the poor.3 On the other hand, if the rich can protect themselves so as to avoid victimization it is possible that any increase in crime falls on the disadvantaged. Levitt (1999) argues that the poor suffer disproportionately more from property crime today than what they did twenty years ago, possibly because of the

3

Becker’s work started an enormous literature in economics that uses that rational choice paradigm (see Witte (1980), McCormick and Tollison (1984), Ehrlich and Brower (1987), Andreoni (1991), Freeman (1996), Glaeser, Sacerdote and Scheinkman (1996), Levitt (1997), inter alia).

5

increased reliance on theft-prevention devices by higher income groups. He also indicates that, in stark contrast to property crime, homicide appears to have become more dispersed across income groups, at least based on neighborhood-level data for Chicago. Moreover, the political economy of crime protection suggests that more intense public police deterrence will occur in rich neighborhoods. Income distribution may affect public provision as richer households may be able to supplement votes with campaign contributions.

An important class of papers in this topic involves so-called crime victimization studies. Gaviria and Pages (2002) provide an example of such a study for Latin America. They show that property crime in Latin America affects mostly rich and middle class households living in larger cities. They also find that households living in cities with rapid population growth are more likely to be victimized than households living in cities with stable population. The main data set used is the Latinobarometer. This is a public opinion survey restricted to urban populations that covers 17 Latin American countries and more than 50,000 households between 1996 and 1998. The key survey question used in this paper is “Have you or any member of your family been assaulted, robbed or victimized in any way during the past twelve months?” Demographic characteristics of the respondent and the head of the household contained in the Latinobarometer are used in the analysis as control variables. The authors assume that all the reported victimization rates correspond to property crime. They construct an index of long-term economic status with data about the ownership of appliances and durable goods and housing characteristics that the survey contains (because household income or wealth is missing from this data set). Victimization surveys from Colombia, El Salvador and Peru are also used to complement the analysis and give a deaggregated picture of type of crime in those countries. Gaviria and Velez (2001) perform a similar study for Colombia. These studies do not analyze chages in victimization distribution under fluctuations in crime levels.

An important issue on victimization differences across social groups is that income levels affect households’ ability to protect themselves against crime. Garcette (2001) presents a

6

model to study how crime victimization is distributed across the poor and the rich. The main result of the paper is that crime victimization inequality increases in the income of the pivotal voter who sets the level of public protection expenditures. Empirical studies have to recognize that private security devices may reduce the level of crime in some areas, but displaced it to other areas. Thus, private security protection may induce significant negative externalities into neighboring areas.4 Di Tella and Schargrodsky (2001) find negative externalities in neighboring areas generated by observable police presence. The complementary set of questions is asked by Ayres and Levitt (1998), Lott (1998), and Duggan (2001). These papers study the effect of the introduction of unobservable protection devices (Lojack and concealed handguns) with potentially positive externalities. These market failures may generate suboptimal provision of private security by the free market.

3. A Simple Model Our hypothesis is that, as the crime increase led high-income citizens to protect themselves through private security devices, low-income households suffered the main burden of the increase in victimization. We can extend Becker (1968) to develop a model with these theoretical predictions. The Becker model is built around the choice of a potential criminal who has the opportunity of being honest or committing a criminal act.

In our case, the criminal has the choice of committing two potential illegal activities: stealing in neighborhood i or stealing in neighborhood j. If the agent decides to be honest, he will take home with certainty his wage w obtained in legal activities. Choosing malfeasance implies playing a lottery: with some probability (1-θ) the agent will escape detection and take home the booty b. With the complementary probability θ the agent will be caught and suffer the (negative) penalty f. Without loss of generality, let’s think that bi > b j , i.e. neighborhood i is the rich neighborhood with more valuable booties. Moreover, each potential criminal h has two idiosyncratic shocks ε ih and ε hj that affect 4

For a survey of the criminology literature on displacement the reader is referred to Cornish and

7

their returns from stealing in neighborhood i or j, respectively. These shocks may represent particular characteristics of the criminals that affect their ability to act in each of these neighborhoods in spite of the different booty values. Furthermore, assume that the shocks ε are distributed in the population following the joint density function g (ε i , ε j ) . Thus, criminal h will be honest if:

{

}

w > Max (1 − θ i )bi + θ i f i − ε ih , (1 − θ j )b j + θ j f j − ε hj ,

or will steal in neighborhood i if:

{

(1 − θ i )bi + θ i f i − ε ih > Max w, (1 − θ j )b j + θ j f j − ε hj

}

It is easy to observe that for any given set of the parameters, the cut-off values ε i* and ε *j would give the frecuency of crime in each neighborhood, Gi (θ i , bi , f i ,θ j , b j , f j , w) and G j (θ i , bi , f i ,θ j , b j , f j , w) .

Moreover, let’s call B at the value of the stolen goods for each neighborhood, where typically B > b, i.e. the owner values the lost property more than the thief. Let’s also consider that the rich neighborhood can install a security technology at a fixed cost K.5 The technology increases the probability of capturing a thief committing a criminal act in neighborhood i from θ i to θ i . Alternatively, the technology could increase penalties in neighborhood i (in absolute values) from f i to f i .6

Clarke (1987) and Hesseling (1994). 5 In reality, private security protection can also be hired in low-income neighborhoods. We are simply assuming here that the fixed cost K is high enough as to being unaffordable for the lowincome community j or that the value of the expected loss in the poor neighborhood is lower than the security costs (i.e., K > B j G j (.) ). The argument just needs that private protection is a normal good. Although legal penalties may be similar in all the neighborhoods, for a criminal the probability of getting shot while commiting a crime or of being actually prosecuted may differ across neighborhoods.

6

8

This simple model can be utilized to understand what happened in Argentina during the 1990s. At a high salary w obtained from legal activities, crime was low. Moreover, at those low crime levels, the fixed cost K disincentives the use of private security devices: Bi Gi (θ i , bi , f i ,θ j , b j , f j , w) < Bi Gi (θ i , bi , f i ,θ j , b j , f j , w) + K

as the expected loss without private protection (the value of the loss times the probability of suffering it) are lower than the expected costs under private protection. In this case, at similar levels of protection in rich and poor areas, crime should concentrate in highincome neighborhoods with higher booties bi. However, as the increase in unemployment lowers the legal salary w, the costs from the crime increase may exceed the fixed costs of private security protection for neighborhood i, making convenient for the rich to pay the protection cost K in order to reduce their victimization rates. Bi Gi (θ i , bi , f i ,θ j , b j , f j , w) ≥ Bi Gi (θ i , bi , f i ,θ j , b j , f j , w) + K

At that point, rich-neighborhoods buy private security increasing the expected punishment for the criminals that intend to act in those areas. For many criminals, the augment in punishment more than compensates the difference in booty values. Thus, the increase in protection and penalties in neighborhood i may induce an increase in crime in low-income neighborhoods j. Crime rises as a result of the increase in unemployment, but it may concentrate in poor neighborhoods as high-income neighborhoods hire private security devices that induced negative externalities over poor neighborhoods.7 As people usually describe it, crime may have become in Argentina a war of poor vs. poor, rather than a war of poor vs. rich.

7

We treat here the decision to install a private security device as a unified decision by the rich community (or as if all the rich households are similar). This is just a simplification to isolate our argument from problems of collective decisionmaking. Of course, negative externalities within the rich neighborhoods could also be present.

9

Is this theoretical description appropriate for Argentina? The main purpose of this study is to establish empirically the changes in the distribution of crime victimization by socioeconomic level experienced in Argentina during the decade. To address this question we run a victimization survey asking people of every social level the number of crimes they have suffered in the last ten years. We also use the survey to investigate differences in the consumption of private protection by income level.

4. Data Description

A household survey specially designed is the main source of information for this study. The target population of the study is the population of the Buenos Aires Metropolitan Area. We interviewed 200 households in the City of Buenos Aires and 200 households in the suburban Great Buenos Aires. The survey was conducted by the opinion poll company Catterberg & Asoc. during February of 2002 through telephone interviews. Due to the generalized fear of suffering a crime, it would have been virtually impossible to perform a personalized survey, as people do not allow interviewers to access their houses these days. This problem would have been particularly severe for interviewing highincome households, as access is precluded in gated neighborhoods and buildings with private security guards. The other alternative to phone interviews would have been street interviews, but again it would have been difficult to find high-income citizens and to convince them of answering a sensitive survey. The main drawback of the phone interviews is that our study could be missing the very low-income population, who may lack home telephone lines. However, the penetration of telephone lines in the area under analysis is very high.8 Moreover, we designed a four stage stratified random sampling of households following quotas by neighborhood, gender, education level, and age, in order to obtain full representation. Importantly, we increase the precision of our estimates using weighted regressions in the econometric analysis. The weights replicate the distribution of the population universe (known from census data) by these characteristics, as explained in Appendix 1.

8

83% for the City of Buenos Aires and 68% for the rest of the metropolitan area as of April 2001.

10

We collected information on victimization events, whether or not those events were reported to the police, whether or not any action was pursued in the judicial system, why crime events were not reported to the police and why legal actions were not pursued, information about the behavioral responses to crime, consumption of private protection, and income and demographic household information. Appendix 2 presents the exact version of our survey (in Spanish), and the number of households providing each response to the questions.

Although the survey is cross-sectional, we asked households to report retrospective information for the entire decade. However, retrospective information is always subject to recall bias. Thus, we designed the survey exploiting several techniques specially developed to minimize this nuisance. First, we restricted the information set to major crime events: armed robberies and forcible entry into their homes. The restriction to major events significantly reduces typical recall bias of retrospection, which is mainly associated to microscopic events (see Aday, 1996, and Reuband, 1994). Moreover, for our main results we just concentrate on whether the household has been victim of a crime during a period of time, but not on the number of times this has occurred. We should expect that recall bias has a larger effect on the latter, than on the former. Additionally, we used bounded recall procedures to reduce underreporting of crime events as a result of telescoping events from past periods. Although this procedure was designed for panel surveys, Sudman, Finn and Lannom (1984) consider its use for a cross-sectional survey. Subsequent research has confirmed that asking people twice about the same events, first in connection with a given reference period and then in connection with the period of interest, is very effective in reducing underreporting or, in general, recall bias (Aday, 1996). More importantly, our main question is which group has been mostly affected by the increase in crime levels. As this question refers to the changes in victimization rates rather than the levels, it will not be affected if recall bias affects both groups (rich and poor) similarly.

When we asked households retrospective information, we preferred to sacrify precision in the exact year of an event, but win confidence by considering longer time periods. Thus,

11

we defined three periods: 1990-1994 (the first part of the decade with low unemployment and strong growth), 1995-2000 (the period after the Tequila crisis with significant unemployment and a declining economy), and the final year of 2001 (right before the default of the external debt, the end of Convertibility, and the current unemployment peak).

A first look at the results in Table 1 shows that 10% of the households interviewed in our survey suffered a home robbery (forcible entry into their house) during 2001. This percentage was 9.8% for the whole period 1995-2000, and 9.1% for 1990-1994. 45% of these crimes were reported to the police in 2001, but the figure was larger in the previous years (48.7% for 1995-2000, and 80% for 1990-1994). For robberies outside the home, 31.8% of the interviews show that at least one member of the household has been robbed during 2001. This percentage was 27.8% for 1995-2000, and 9.4% for 1990-1994. The reporting rate of this type of crime was lower than for home robberies, but similarly decreasing over time (37.8% for 2001, 50.5% for 1995-2000, and 54.3% for 1990-1994).

Table 1 Home Robbery %Victimization %Reporting 1990-1994 9.1 80.6 1995-2000 9.8 48.7 2001 10.0 45.0

Street Robbery %Victimization %Reporting 9.4 54.3 27.8 50.5 31.8 37.8

These figures suggest that there was a significant increase in victimization rates (remember that the question refers to periods of different length), and a decrease in reporting rates. Consistently, Table 2 shows a growing feeling of insecurity in the population.

Table 2 Perceived Insecurity In your neighborhood, would you say that insecurity with respect to one decade ago has increased a lot, some, a little, has not changed at all, or has decreased?

12

%

Increased a lot Increased some Increased a little No change Decreased No answer

31.8 28.5 12.8 19.8 1.0 6.3

A final data issue is that the measure of income levels through surveys is always a delicate matter. Many people decline to reveal their income in a survey. When they answer, it is well known that the rich tend to underreport their income, while the poor may overdeclare. The best strategy to address this problem is to ask questions on education level, ocupation, and availability of cars, appliances (PC, air conditioner, and authomatic washing machine), and credit cards, in order to infer income levels from these variables. In the absence of income data, the opinion poll company, following Asociación Argentina de Marketing (1998), provided us with an index of income level that collapses all the indicators of household education, ocupation, and wealth into a continuous variable. Appendix 3 shows the details of this methodology. An important clarification is that this methodology has not been developed ad-hoc for this study. The index of income level has been provided to us by the poll company from standard survey practices. Using this index, we can now proceed to answer our research questions.

5. Empirical Results 5.1. The Evolution of Crime Victimization by Income Group

Our main question is which social group has been most affected by the significant increase in crime suffered in Argentina during the 1990s. Thus, we want to compare the change in crime levels for the high-income group relative to the change in crime levels for the low-income group. We took the median of the income level index to classify the households in the high and low-income level category. Note that this comparison will not be affected by recall bias as long as this bias is uncorrelated with income levels.9 9

In the first part of the survey we asked the number of times a member of the household had been robbed during the period 1990-94. To control for the consistency of the responses, at the very end of the questionnaire we asked the number of times a member of the household had been robbed during 1990-92, and during 1993-94. 92% of the interviewed people responded consistently (i.e.,

13

To analyze which social group has been most affected by the crime increase, we start by calculating whether the change in victimization rates was different across groups. Table 3 compares the differences in means to obtain a difference-in-differences estimator of the variation in home robberies by income levels. For the period 1990-1994, high-income households suffered a significantly higher home victimization rate than low-income families. After that period, low-income households suffered a significant increase in victimization likelihood, while high-income families show a non-significant decline. The cross-sectional difference becomes insignificant in those subsequent periods. Thus, the victimization rate of the low-income households caught up to the high-income rate during the decade. Importantly, the difference-in-differences tests show that the change in the victimization rate of the low-income group is significantly different from the change for the high-income households.

Table 3 [2001]-[95-00] [95-00]-[90-94] [2001]-[90-94] Diff T Diff t Diff t High 0.12 0.08 0.08 0.00 (0.01) -0.04 (1.32) -0.04 (1.34) Low 0.06 0.11 0.12 0.01 (0.18) 0.06** (1.97) 0.06** (2.14) Diff 0.07** -0.03 -0.03 -0.01 -0.10** -0.10** High-Low t (2.34) (0.96) (1.15) (0.14) (2.31) (2.44) Notes: * Significant at 10%; ** significant at 5%; *** significant at 1%. Home Robbery

90-94

95-00

2001

Table 4 performs a similar analysis for robberies suffered by household members outside the house. For the three periods, high-income households suffered a higher victimization rate than low-income families. The cross-sectional difference seems significant in more recent years. Moreover, both groups have suffered a significant increase in crime levels. The difference-in-differences tests, however, are never significant, suggesting that the evolution of victimization rates has not differed across groups.

Table 4 the sum of the responses to these two final questions equaled the previous response for the whole period). The correlation between our income level index and this consistency measure is very low

14

[2001]-[95-00] [95-00]-[90-94] [2001]-[90-94] Diff t Diff t Diff t High 0.11 0.33 0.37 0.04 (0.69) 0.22*** (5.61) 0.26*** (6.34) Low 0.08 0.22 0.27 0.05 (1.06) 0.14*** (3.98) 0.19*** (5.04) 0.03 0.12*** 0.10** -0.01 0.08 0.07 High-Low Diff t (1.05) (2.60) (2.22) (0.21) (1.57) (1.29) Notes: * Significant at 10%; ** significant at 5%; *** significant at 1%. Street Robbery

90-94

95-00

2001

We now want to explore whether the difference-in-differences results from Tables 3 and 4 are robust to the inclusion of pertinent control variables. We want to analyze the change in the likelihood of suffering a crime for different social groups after controlling for personal characteristics, neighborhood characteristics, and type of home (apartment or house). We also include random-effects to control for potential correlation in the victimization suffered by the same family over time. To identify which social group has been most affected by the crime increase, we first estimate the following Logit model:

Prob(zit = 1) = F (α H ,1990−94 H i I1990−94 +α H ,1995−2000 H i I1995−2000 +α H ,2001 H i I 2001 + +α L,1990−94 Li I1990−94 +α L,1995−2000 Li I1995−2000 +α L,2001 Li I 2001 + β x it + λi + µi ) where zit=1 if household i has been victimized during period t, Hi and Li are dummy variables indicating whether the family belongs to the high or low income group, I1990-94 I1994-2000, and I2001 are dummy variables for each period, xit is a vector of household covariates (that may or may not vary over time), λi is a neighborhood fixed effect, and µi is a household random effect. We are not interested in the α’s, which capture the level of the victimization likelihood for each group in each period. The focus of our analysis are neither the differences (α A,T1 −α A,T 2 ) , that capture the change for one group between two different periods T1 and T2, nor (α A,T 1 −α B,T 1 ) , that captures the cross-sectional difference between both groups at the same period. Instead, we want to test the null

(-0.12).

15

hypothesis of similar change for both groups. Defining ∆ A,T1−T 2 = α A,T1 −α A,T 2 , we test whether ∆ H ,T 1−T 2 = ∆ L,T1−T 2 .

Table 5 presents this difference-in-differences estimates for the home robberies. In column A, our estimates are obtained from a pooled Logit where the only control is whether the household lives in a building apartment or in a house. The next column incorporates household random effects in a panel Logit regression.10 The characteristics of the household head (education level, gender, age, and age squared) are incorporated in column C, while neighborhood dummies are added in the last column. The results are extremely similar for the four specifications. The increase in crime victimization for the low-income group is higher than the change experienced by the high-income group. The difference is significant at the 1% level.

Table 5 Dependent Variable: Home Robbery Dummy H95

(A)

(B)

(C)

(D)

-0.47** (2.00) -0.56 (1.47) -0.59 (1.52) -1.13*** (2.90) -0.13 (0.43) -0.01 (0.03)

-0.60 (1.52) -0.62 (1.57) -1.18*** (2.88) -0.15 (0.44) -0.01 (0.04) -1.98*** (6.64)

-0.61 (1.53) -0.63 (1.58) -0.91* (1.81) 0.13 (0.30) 0.26 (0.60) -0.67 (0.33)

-0.61 (1.52) -0.63 (1.57) -0.92* (1.83) 0.12 (0.27) 0.25 (0.58) -0.41 (0.20)

χ2 Null Hypothesis: ∆H,95-90 = ∆L,95-90 ∆H,01-90 = ∆L,01-90 ∆H,01-95 = ∆L,01-95

8.67*** 10.44*** 0.08

8.97*** 10.79*** 0.09

9.05*** 10.87*** 0.09

9.05*** 10.86*** 0.09

Individual Controls Zone Controls Random Effects

No No No

No No Yes

Yes No Yes

Yes Yes Yes

H01 L90 L95 L01 Constant

10

Note that the whole set of α’s cannot be identified in a fixed effects regression.

16

Observations 1195 1193 1193 1193 Notes: Weighted Logit model. All the regressions include a Building Apartment dummy. Absolute value of z-statistics in parentheses. * Significant at 10%; ** significant at 5%; *** significant at 1%.

In Table 6, we perform the same exercise for the robberies suffered by the household members outside the home. Again, we run four different specifications. We never find significant differences in the changes in crime victimization suffered by low and high income households. These results suggest that the evolution of crime victimization does not differ across groups for street robberies.

Table 6 Dependent Variable: Street Robbery Dummy H95 H01 L90 L95 L01 Constant χ2 Null Hypothesis: ∆H,95-90-∆L,95-90 ∆H,01-90-∆L,01-90 ∆H,01-95-∆L,01-95

(A)

(B)

(C)

(D)

1.41*** (4.47) 1.62*** (5.17) -0.39 (1.06) 0.91*** (2.96) 1.26*** (4.16) -2.14*** (8.01)

1.52*** (4.56) 1.76*** (5.28) -0.42 (1.10) 0.96*** (2.89) 1.35*** (4.10) -2.35*** (7.89)

1.52*** (4.56) 1.75*** (5.29) 0.05 (0.12) 1.45*** (3.75) 1.84*** (4.78) -7.07*** (4.00)

1.52*** (4.56) 1.75*** (5.28) 0.01 (0.03) 1.41*** (3.65) 1.80*** (4.68) -7.41*** (4.18)

0.07 0.00 0.19

0.09 0.00 0.20

0.07 0.01 0.21

0.07 0.01 0.22

Individual Controls No No Yes Yes Zone Controls No No No Yes Random Effects No Yes Yes Yes Observations 1182 1181 1181 1181 Notes: Weighted Logit model. Absolute value of z-statistics in parentheses. * Significant at 10%; ** significant at 5%; *** significant at 1%.

In these two tables, we have consider the dependent variable as a binary indicator that takes the value of 1 when household i has suffered a crime in period t, and 0 otherwise. This binary definition has the advantage of minimizing recall biases (it is easier to

17

remember whether a household has suffered a robbery, rather than the exact number of times it has happened), but at the cost of losing information by treating similarly at households that have suffered a different number of crimes. As a robustness checks, we now reproduce our exercise in a linear random-effects regression. In this specification, we also correct for the different duration of the time periods by defining the dependent variable qit as the average number of robberies per year that household i has suffered during period t:

q it = α H ,1990−94 H i I1990−94 +α H ,1995−2000 H i I1995−2000 +α H ,2001 H i I 2001 + +α L,1990−94 Li I1990−94 +α L,1995−2000 Li I1995−2000 +α L,2001 Li I 2001 + β x it + λi + µi +ν it where νit is the error term. Again, we are interested in testing the null hypothesis of similar change for both groups between two periods: ∆ H ,T 1−T 2 = ∆ L,T1−T 2 .

Table 7 presents these difference-in-differences estimates for the changes in the number of home robberies under different specifications. In the three cases, we find that the increase in crime victimization of the low-income group is higher than the change experienced by the high-income group. The difference is statistically significant at levels below 5% level. The only difference with the previous result is that, after correcting for the different period duration by taking the yearly average number of robberies, the difference-in-differences estimates are now significant for the changes from 2001 with respect to the other two periods, while in Table 5 the change became significant from 2001 and 1995-2000 relative to the first period.

Table 7 Dependent Variable: Number of Home Robberies H95 H01

(A)

(B)

(C)

-0.02 (0.23) 0.06 (0.65)

-0.02 (0.23) 0.06 (0.64)

-0.02 (0.24) 0.06 (0.64)

18

L90 L95 L01 Constant χ2 Null Hypothesis: ∆H,95-90 = ∆L,95-90 ∆H,01-90 = ∆L,01-90 ∆H,01-95 = ∆L,01-95

-0.02 (0.24) 0.00 (0.01) 0.34*** (3.84) 0.02 (0.25)

-0.07 (0.65) -0.05 (0.45) 0.29*** (2.66) 0.17 (0.38)

-0.09 (0.84) -0.07 (0.64) 0.27** (2.49) 0.05 (0.11)

0.12 5.72** 4.16**

0.12 5.72** 4.17**

0.13 5.87** 4.26**

Individual Controls No Yes Yes Zone Controls No No Yes Observations 1192 1192 1192 Notes: Weighted random effects model. All the regressions include a Building Apartment dummy. Absolute value of z-statistics in parentheses. * Significant at 10%; ** significant at 5%; *** significant at 1%.

In Table 8, we perform the same exercise for the number of robberies suffered by the household members outside the home. As it happened previously with our binary dependent variable specification, we find no significant differences in the changes in street robberies suffered by low and high-income households.

Table 8 Dependent Variable: Number of Street Robberies H95 H01 L90 L95 L01 Constant χ2 Null Hypothesis ∆H,95-90-∆L,95-90

(A)

(B)

(C)

0.08 (1.04) 0.63*** (8.29) -0.04 (0.49) 0.05 (0.68) 0.46*** (6.45) 0.06 (1.08)

0.08 (1.04) 0.63*** (8.30) 0.02 (0.21) 0.10 (1.18) 0.51*** (5.86) -0.31 (0.88)

0.08 (1.04) 0.63*** (8.22) 0.01 (0.14) 0.10 (1.12) 0.51*** (5.83) -0.37 (1.06)

0.00

0.00

0.00

19

∆H,01-90-∆L,01-90 ∆H,01-95-∆L,01-95

1.95 2.14

1.93 2.15

1.90 2.11

Individual Controls No Yes Yes Zone Controls No No Yes Observations 1181 1181 1181 Notes: Weighted random effects model. Absolute value of z-statistics in parentheses. * Significant at 10%; ** significant at 5%; *** significant at 1%.

5.2. The Use of Security Protection Devices

Our previous results show that the poor have suffered the main burden of the increase in home robberies; while both groups have suffered similar increases in victimization outside the house. These findings could be consistent with differences in the ability that people have to protect themselves against different types of crimes. We asked in our survey on the use of private security devices.11

Table 9 Security Device Home Device Watchdog Insurance Alarm Door Armored / Window Bar Private Security No Home Alone Street Device Darkness No Jewel No Documents Places Handgun

% 33.0 27.8 20.2 37.4 19.7 38.8 61.3 42.4 35.8 40.8 19.6

Table 9 shows the frequencies in the use of security devices in our sample. There are a number of security devices that households can use to protect their houses. Close to 40% of our households have armored doors or window bars, and more than 30% have a

20

watchdog for security purposes. Almost 40% of the families declare to leave permanently one member of the household at home for protection. 20% of the households have alarm devices, private security or handguns. Moreover, about 30% have acquired home insurance, which does not avoid the robbery but covers the losses.

We investigate the differential use of security devices by income level by running the following cross-sectional Logit model:

Prob(sdi = 1) = F ( α L Li + β x i + λi )

where sdi=1 if household i utilizes the security device under analysis, Li is a dummy variables indicating whether the family belongs to the low-income group, xi is a vector of household covariates, and λi is a neighborhood fixed effect. The significance of the coefficient α L will tell us whether low-income households show a different likelihood than high-income households to use the security device.

The results, presented in Table 10, show that low-income families appear more likely to own a watchdog for security protection. Instead, high-income households tend to use more private security, alarms, armored doors and window bars. They are also more likely to acquire home insurance. Thus, high-income people seem to use market based crime protection mechanisms, rather than non-pecuniary devices (Becker, 1965). However, both groups are similarly likely to leave a person permanently at home to protect the house. The groups show no difference in handgun property.

Table 10 Dependent Variable: Low Income

Watchdog (A) (B) 0.65*** 0.65** (2.85) (2.57)

Insurance Alarm (C) (D) (E) (F) -1.69*** -1.68*** -1.12*** -1.22*** (6.37) (5.88) (4.11) (4.10)

11

To identify which private security devices are normally used we followed previous surveys (Clarin, June 9, 1998; Pagina 12, September 6, 1998; Gallup, August 29, 1999; Ministerio de Justicia, 2000; Catterberg & Asoc, 2001).

21

Individual and Zone Controls Observations

No

Yes

No

Yes

No

Yes

396

396

393

393

394

394

Table 10 (Cont.) Dependent Variable: Low Income

Armored Door / Window Bar (G) (H) -0.59*** -0.58** (2.61) (2.30)

Private Security

No Home Alone

(I) -0.47* (1.73)

(K) -0.02 (0.10)

(J) -0.67** (2.21)

(L) -0.13 (0.55)

Handgun (M) 0.06 (0.24)

(N) 0.07 (0.24)

Individual and No Yes No Yes No Yes No Yes Zone Controls Observations 397 397 395 395 396 396 394 394 Notes: Weighted Logit model. All the regressions include a Building Apartment dummy. The constant is not reported. Absolute value of z-statistics in parentheses. * Significant at 10%; ** significant at 5%; *** significant at 1%.

Perhaps, there is little that people can do to avoid becoming the victim of a robbery outside their houses.12 More than 60% of the interviewed people declare to avoid being outside at night, and 40% have changed their itineraries or transportation means. Again, 20% of the households have a handgun. Moreover, around 40% avoid carrying documents, credit cards, or jewelry in order to prevent the losses. However, according to our interviews there is not much else that people can do to avoid street robberies.13 Moreover, the superiority of high-income families to protect themselves against this type of crime seems weaker. In Table 11 we find that rich groups avoid carrying valuable objects. Low-income families are more likely to avoid dark areas, but a similar pattern is not observed for changes in itineraries or transportation means. As reported before, both groups show no differences in handgun property.

Table 11 No Documents 12

No Jewel

Of the 127 households that declare that one of its members has been robbed outside the house during 2001, those robberies took place in the street (92), in a car or public transportation (19), at work (3), in a shop or restaurant (11), or at a bank or ATM (2). 13 Very rich people can, of course, resort to bodyguards and armored cars, but these is extremely expensive and infrequent in Argentina.

22

Dependent Variable: Dependent Low Income Variable:

No Documents


No Jewel

-0.52** (A) (2.40)

-0.40* (B) (1.69)

-0.54** (C) (2.56)

-0.46** (D) (2.00)

No

Yes

No

Yes

399

399

395

395

Table 11 (cont.) Darkness Low Income

(E) 0.28 (1.33)

Places

(F) 0.44* (1.85)

(G) -0.22 (1.04)

Handgun (H) -0.04 (0.15)

(I) 0.19 (0.75)

(J) 0.06 (0.22)

Individual and No Yes No Yes No Yes Zone Controls Observations 396 396 397 397 394 394 Notes: Weighted Logit model. The constant is not reported. Absolute value of z-statistics in parentheses. * Significant at 10%; ** significant at 5%; *** significant at 1%.

Thus, there seem to be strong differences in the home protection devices implemented by different social groups, while cross-sectional differences in street protection devices appear weaker. It is important to note the high observability of home protection devices used by the households in our sample. It is expected that the use of observable devices by a household leads criminals to target another property, inducing negative externalities. Instead, the diffusion of unobservable security devices may generate positive externalities by increasing the expected probability for a criminal of getting caught when facing any target. This is the focus of a significant literature in economics and criminology (recent papers include Ayres and Levitt (1998) who focus on unobservable anti-theft devices produced by the Lojack company, Lott and Mustard (1997) and Duggan (2001) who focus on the right-to-carry concealed handguns laws, and Di Tella and Schargrodsky (2001) who study the effect of observable police presence). Thus, we not only find that rich-households are more likely to use pecuniary security devices (with the exception of watchdogs), but the observable nature of these devices suggest that they may shift crime from the areas that have observable protection to those that do not or, in our set-up, from rich to poor areas.

23

5.3. Access to Justice and Public Police Protection

If private provision is relevant for crime protection, income inequality may explain why poor people are underprotected against crime. The poor may lack resources to hire and buy protection devices (Anderson, 1999). Without resources to “buy” security, poverty leads to underprotection. However, we may fear that some additional factors worsen the situation. Poor citizens may suffer a weak definition of their property rights. Without legal proof of their property rights (or without proper personal identification documents),14 the poor cannot resort to the police and the judiciary. In addition, public officials frequently have biases against the poor in the provision of their services (World Bank, 2000). Moreover, in many occasions, the poor suffer harassment from the police and other public officers. The poor usually “fear” rather than trust the police. Denounces of police corruption and of police abuses against the poorest frequently appear in the media.15 In addition, the complexity of the legal system, educational handicaps, and the costs associated with judiciary processes (court fees, legal aid, time and transportation costs) may act as exclusionary barriers to the access to justice by the poor (World Bank, 2000). Furthermore, if the consumption of private protection by the rich induces negative externalities, it should be optimal that public police forces follow crime by moving to low-income areas. We also exploit our survey to analyze whether there are differences in the treatment and protection that the poor receive from the police and the judicial system relative to the rich.

Table 12 presents the survey responses on public police protection. We ask at which frequency police walks or drives in front of people’s houses. At first glance, police protection in high-income areas seems to be slightly better. 14

This problem may be particularly relevant for illegal immigrants. For example, Bolivian workers are frequent targets of robberies in Northern Great Buenos Aires (La Nacion, June 24, 2000). Policemen are suspects of several of these attacks (Clarin, June 29, 2000). 15 CELS, Centro de Estudios Legales y Sociales (www.cels.org.ar), and COFAVI, Comision de Familiares de Victimas Indefensas de la Violencia Social (Policial-Judicial-Institucional) (www.derechos.org/cofavi/), are two Argentine human rights organizations endorsing these denounces and providing free legal aid.

24

Table 12

Total number of households How often does the police usually patrol your house? Every day Twice or three times a week Once a week At least once a month Less than once a month Never No answer

Income Level Total Low High 196 204 400

91 22 13 12 3 30 25

103 23 8 9 7 27 27

194 45 21 21 10 57 52

We then perform an Ordered Logit of this frequency on income levels in Table 13. With no controls in Column A and with household controls in Column B, police protection seems to be similar across social groups. However, we can still wonder whether this is optimal. If high-income groups protect themselves through the hiring of private security guards, it might be socially optimal that police forces follow crime moving to lowincome areas. This does not seem to be the case. In the last column of the table we include the use of private security as a control in order to analyze whether private protection reduces the frequency of public police protection. The security variable results insignificant. It does not seem to be the case that public police is being redeployed to compensate for negative externalities generated by private security.

Table 13 Dependent Variable: Frequency of police patrolling Low Income

(A)

(B)

(C)

0.26 (1.23)

0.01 (0.05)

0.03 (0.12) -0.04 (0.14)

No

Yes

Yes

359

359

359

Private Security


25

Notes: Weighted Ordered Logit model. The constant is not reported. Absolute value of z-statistics in parentheses. * Significant at 10%; ** significant at 5%; *** significant at 1%.

We also analyze whether income level affects the likelihood of reporting a criminal act. The Logit regressions presented in Table 14 show that both social groups seem similarly likely to report a crime to the police. The income level dummies are never significant.

Table 14 Dependent Variable: Low Income

Home Robbery Reporting (A) (B) -0.42 -0.66 (0.61) (0.75)

Street Robbery Reporting (C) (D) 0.42 0.50 (1.15) (1.22)

Individual and Zone No Yes No Yes Controls Observations 43 41 130 130 Notes: Weighted Logit model. The constant is not reported. Absolute value of z-statistics in parentheses. * Significant at 10%; ** significant at 5%; *** significant at 1%.

Finally, Table 15 describes the frequency of crime reporting by social group, the result of the reporting, and the reasons for no reporting. Receiving no answers from the police and the judiciary is the most frequent response after reporting. The low value of the loss, skepticism about the investigation, and reporting costs seem to be the main causes of the lack of reporting. The results look very similar for both social groups. Perhaps, skepticism seem higher in high-income groups, and reporting costs appear more relevant for low-income groups, as reasons for no reporting.

Table 15 Home Street Robbery Robbery Income Level Low High Low High Total Number of Crimes 23 17 52 75 Have your reported the crime to the police or to the justice? Yes What happened? 10 8 22 26 I never had an answer. 5 5 15 18 The report got stuck in the 1 2 3 1

26

Total Low High 75 92

32 20 4

34 23 3

No

justice. I was called to declare but nothing happened after all. There was an investigation, but no results. There was an investigation with some results. There was an investigation and the case was solved. Other reasons / No answer. Why? Because the loss was worthless / There was no loss. Because of the reporting costs / The loss of hours of work. Because the police would not have done anything. Because the justice would not have done anything. Because there was no evidence to do anything. Other reasons / No answer.

1

0

0

2

1

2

1

0

1

1

2

1

0

1

2

1

2

2

2 0 13

0 0 9

1 0 30

1 2 49

3 0 43

1 2 58

7

2

9

13

16

15

2

0

7

3

9

3

2

3

5

19

7

22

0

1

1

4

1

5

1 0

1 1

4 0

2 0

5 10

3 13

6. Policy Implications and Conclusions

Crime has risen significantly in Argentina during recent years, together with unemployment and the deterioration of social conditions. We analyze which social groups have been the main victims of this increase in crime. For home robberies, we found that the poor have suffered the main crime increases. During the first half of the decade, high-income households used to suffer a significantly higher home victimization rate than low-income households. The difference has now turned non-significant. For street robberies, instead, both groups show similar augments in victimization. Our findings are consistent with additional evidence showing that the rich have been able to protect their houses through pecuniary security devices better than the poor. Instead, the ability to use protection devices against street robberies seems limited.

Our results on the distributional impact of crime have important policy implications. A number of interventions are guided by their perceived impact on income distribution. Our estimates suggest that such interventions may need re-directing. If the increase in crime

27

tends to affect disproportionately low-income households, there might be appropriate to produce crime-corrected measures of income distribution before using these variables to target intervention. Moreover, our results could improve the cost-benefit evaluation of several crime reduction policies that are currently under discussion.

In particular, our results have implications for two areas of public policy. If high-income neighborhoods hire crime protection devices to protect themselves, it would be optimal that police forces follow crime moving to low-income areas. Our evidence on police patrolling does not suggest that a compensatory redeployment of public protection has been taking place.

Second, if there are significant negative externalities associated with the use of private security devices, some form of taxation or regulation of the crime protection industry may be necessary. Ayres and Levitt (1998) show the economic importance of private security expenditures in the US and how such private spending has outgrown public spending over the recent past. This is also true in many developing countries, including Argentina. The sector is largely unregulated, particularly outside the US. In most countries that we know of, citizens are free to hire visible private protection. If such activities induce negative externalities, these market failures may generate an overprovision of private security by the free market.

28

References: Aday, L. (1996), Designing and Conducting Health Surveys, Jossey-Bass. Andalón-López, Mabel and Luis F. López-Calva (2001), “Income Inequality, Polarization, and Violent Crime: Evidence from Mexico,” mimeo, El Colegio de México. Anderson, Michael (1999), “Access to Justice and Legal Process: Making Legal Institutions Responsive to Poor People in LDCs,” World Bank. Andreoni, James (1991) “Reasonable Doubt and the Optimal Magnitude of Fines: Should the Penalty Fit the Crime?”, RAND Journal of Economics, 22 (3), 385-95. Asociación Argentina de Marketing (1998), “Indice de Nivel Socio Económico Argentino”. Ayres, Ian and Steven Levitt (1998) “Measuring Positive Externalities from Unobservable Victim Precaution: An Empirical Analysis of Lojack”, Quarterly Journal of Economics, 113 (1), 43-77. Becker, Gary (1965) “A Theory of the Allocation of Time,” Economic Journal, 75, 493517. Becker, Gary (1968) “Crime and Punishment: An Economic Approach”, Journal of Political Economy, 76 (2), 169-217. Blinder, A. and Howard Esaki (1978) “Macroeconomic Activity and Income Distribution in the Postwar United States”, Review of Economics and Statistics, 60, 604-9. Catterberg & Asoc. (2001), “Encuesta sobre Inseguridad Ciudadana”, mimeo. Cornish, Derek and Ronald Clarke (1987) “Understanding Crime Displacement: An Application of Rational Choice Theory”, Criminology, 25, 933-47. Di Tella, Rafael and Ernesto Schargrodsky (2001) “Using a Terrorist Attack to Estimate the Effect of Police on Crime”, Working paper CREDPR, Stanford University N°90. Duggan, Mark (2001) “More Guns, More Crime”, Journal of Political Economy, 109, 1086-1114. Ehrlich, Isaac (1973) “Participation in Illegitimate Activities: A Theoretical and Empirical Investigation”, Journal of Political Economy, 81 (3), 521-65. Ehrlich, Isaac and George Brower (1987) “On the Issue of Causality in the Economic Model of Crime and Law Enforcement: Some Theoretical Considerations and Experimental Evidence”, American Economic Review, 77 (2), 99-106. Fajnzylber Pablo, Daniel Lederman, and Norman Loayza (1998), “Determinants of Crime Rates in Latin America and the World, An Empirical Assessment,” The World Bank. Fajnzylber Pablo, Daniel Lederman, and Norman Loayza (2000a), “Crime and Victimization: An Economic Perspective,” Economia (LACEA), 1(1), 219-302. Fajnzylber, Pablo, Daniel Lederman, and Norman Loayza (2000b) “What Causes Violent Crime?”, European Economic Review, forthcoming. Freeman, Richard (1996) “Why Do So Many Young American Men Commit Crimes and What Might We Do About It?”, Journal of Economic Perspectives, 10 (1), 25-42. Freeman, R. (1991) "Crime and the Employment of Disadvantageous Youths", NBER 3875. Gaviria, Alejandro and Carmen Pages (2002) “Patterns of Crime Victimization in Latin American Cities,” Journal of Development Economics, 67, 181-203.

29

Gaviria, Alejandro and Carlos Vélez (2001) “Who Bears the Burden of Crime in Colombia?”, mimeo. Garcette, Nicolas (2001) “Income Inequality and Crime Protection”, mimeo DELTA. Glaeser, Edward, Bruce Sacerdote, and Jose Scheinkman (1996) “Crime and Social Interactions”, Quarterly Journal of Economics, 111 (2), 507-48. Hesseling, Rene (1994), “Displacement: A Review of the Empirical Literature”, in Ronald Clarke, ed., Crime Prevention Studies, III, 197-230. Levitt, Steven (1997) “Using Electoral Cycles in Police Hiring to Estimate the Effect of Police on Crime”, American Economic Review, 87 (3), 270-90. Levitt, Steven (1999) “The Changing Relationship between Income and Crime Victimization”, FRBNY Economic Policy Review, 87-99. Londoño, Juan L. and Rodrigo Guerrero (1999), “Violencia en América Latina: Epidemiología y Costos,” InterAmerican Development Bank. Lott, John and David Mustard (1997) “Crime, Deterrence, and Right-to-Carry Concealed Handguns”, Journal of Legal Studies, 26 (1), 1-68. Lott, John (1998) More Guns, Less Crime, The University of Chicago Press: London. McCormick, Robert and Robert Tollison (1984) “Crime on the Court”, Journal of Political Economy, 92 (2), 223-35. Miller, Matthew (2001) “The Poor Man´s Capitalist”, New York Times Magazine, July 1, 2001. Ministerio de Justicia (2000) “Estudio de Victimización en Centros Urbanos de la Argentina”, mimeo. Papps, K. and R. Winkelmann (1996) “Unemployment and Crime: New Answers to an Old Question”, mimeo. Raphael, Steven and Rudolf Winter-Ebmer (2001) “Identifying the Effect of Unemployment on Crime”, Journal of Law and Economics 44(1), 259-283. Reuband, K. (1994), “Reconstructing Social Change Through Retrospective Questions: Methodological Problems and Prospects”, in Autobiographical Memory and the Validity of Retrospective Report, Schwartz, N. and S. Sudman (eds.), SpringerVerlag. Sudman, S., A. Finn and L. Lannon (1984), “The Use of Bounded Recall Procedures in Single Interviews”, Public Opinion Quarterly, 48, 520-24. Tauchen, Helen, Ann Dryden Witte, and Harriet Griesinger (1994) "Criminal Deterrence: Revisiting the Issue with a Birth Cohort", Review of Economics and Statistics, 76, 3, 399-412. Witte, Ann (1980) “Estimating the Economic Model of Crime with Individual data”, Quarterly Journal of Economics, 94 (1), 155-67. World Bank (2000), “Attacking Poverty,” World Development Report 2000/2001.

30

Appendix 1: Computation of Households Weights The survey interviews 200 households in the City of Buenos Aires and 200 households in the suburban area of the Province of Buenos Aires. However, it is clear that these quantities do not follow the true distribution of Population or Income Level (IL) of this regions.16 To replicate the true distribution the weight of our observations has to be adjusted. In our sample we have the following distribution of observations: City of Province of Income Level Buenos Aires Buenos Aires 57 20 High 87 92 Medium 56 88 Low Total 200 200 According to the 1991 Census, surveys and other marketing sources, the true Population distribution between the City of Buenos Aires and the suburban area of the Province of Buenos Aires is 32% and 68% respectively. Moreover, according to these sources the distribution of the Income Level index is as follows: City of Province of Income Level Buenos Aires Buenos Aires 21% 4% High 54% 38% Medium 25% 58% Low Total 100% 100% It is straightforward that the true distribution joint of these variables is: City of Province of Income Level Buenos Aires Buenos Aires 6.72% 2.72% High 17.28% 25.84% Medium 8.00% 39.44% Low Total 32% 68%

Total 9.44% 43.12% 47.44% 100%

Note that the total of each row reflects the distribution of the Income Level for both regions. Similarly, the total of each column shows the distribution of Population for both region. Finally, six weight are computed. The procedure is described in the following matrix of weights. City of Income Level Province of Buenos Aires Buenos Aires High Medium Low

16

6.72% × 400 = 0.47157895 57 17.28% × 400 = 0.79448276 87 8.00% × 400 = 0.57142857 56

2.72% × 400 = 0.544 20 25.84% × 400 = 1.12347826 92 39.44% × 400 = 1.79272727 88

Details on the calculation of the IL index are presented in Appendix 3.

31

Appendix 2: Survey Questionnaire (in Spanish) and Responses. Pregunta

Opciones

Sexo del entrevistado Masculino Femenino Podría decirme su edad? 36-45 46-55 56 y mas (volver a escribir la edad) Edad Podría decirme el máximo nivel educativo que Ud. alcanzó? Sin Estudios Primaria Incompleta Primaria Completa Secundaria Incompleta Secundaria Completa Terciaria Incompleta Terciaria Completa Universitaria Incompleta Universitaria Completa Postgrado En su barrio Ud. diría que la inseguridad con respecto a una década atrás: aumentó mucho, aumentó bastante, aumentó un poco, sigue Mucho igual, o disminuyó? Bastante Poco Sigue igual Disminuyó No sabe/No contesta En el año 2001 Ud. o algún miembro de su familia que vive en su mismo hogar sufrió un asalto fuera de su casa? Si No No sabe/No contesta

Cantidad de Respuestas 400 161 239 400 178 129 93 400 400 400 2 15 113 24 82 7 25 42 86 4 400 127 114 51 79 4 25 400 127 273 0

Cuántas veces (en total sumando a todos los miembros de su familia que viven en su hogar)? Número de veces

127 127

La última vez que Ud. o un miembro de su familia conviviente fue asaltado, fue un asalto a mano armada? Si No No sabe/No contesta

127 92 34 1

Y dónde ocurrió?

127 92 19 3 11 2 0 0

En la calle En un vehículo o medio de transporte En el trabajo En un comercio o restaurante En un banco o cajero automático Otro lugar:......... No sabe/No contesta Dónde? Descripción Ud. o su familiar resultó lastimado? Si No No sabe/No contesta Por favor, cuéntenos todas las cosas que le robaron a Ud. o al familiar asaltado?

127 13 114 0

Descripción

Denuncio/ciaron el hecho ante alguna autoridad competente?

127

32

Si No No sabe/No contesta Luego de haber hecho la denuncia, que ocurrió? Nunca tuve ninguna respuesta La denuncia quedó estancada en la justicia Me llamaron a declarar pero nunca paso mas nada Se investigó pero sin ningún resultado Se investigó y se obtuvieron algunos resultados Se investigó y se resolvió el caso No sabe/No contesta En general cuál fue la razón principal por la que no denunció este hecho (No leer las opciones. Indicar sólo una)?

48 79 0 48 33 4 2 2 3 2 2 79

La pérdida fue de escasa importancia/ no hubo pérdida Por el costo económico/ pérdida horas de trabajo/ costo de denuncia La policía no hubiera hecho nada La justicia no hubiera hecho nada La policía y la justicia no podrían haber hecho nada / por falta de pruebas Por miedo/ temor/ disgusto a relacionarse con la policía Por miedo / temor a represalias de los ladrones Porque no tenía seguro/ el seguro no cubría La policía desaconsejó / desalentó la denuncia Fue la policía Falta de documento de identidad/papeles en regla/domicilio/ Otras razones No sabe / No contesta

22 10 24 5 6 0 2 2 0 0 7 1 0

Ahora le voy a hacer preguntas referidas a hechos ocurridos desde 1990 hasta 1994 y desde 1995 hasta el 2000. Por favor tómese su Si tiempo para recordar. Entre 1995 y 2000 Ud. o algún miembro de No su familia que vive en su mismo hogar sufrió un asalto fuera de su No sabe / No contesta casa?

400 111 288 1

Cuántas veces (en total sumando a todos los miembros de su familia que viven en su hogar? Cantidad de veces

111 111

La última vez que Ud. o un miembro de su familia conviviente fue asaltado, fue un asalto a mano armada? Si No No sabe / No contesta

111 70 38 3

Y dónde ocurrió?

111 67 20 9 11 1 2 1

En la calle En un vehículo o medio de transporte En el trabajo En un comercio o restaurante En un banco o cajero automático Otro lugar:......... No sabe / No contesta Dónde? Descripción Ud. o su familiar resultó lastimado? Si No No sabe / No contesta Por favor, cuéntenos todas las cosas que le robaron a Ud. o al familiar asaltado?

Descripción

33

111 16 95 0

Denuncio/ciaron el hecho ante alguna autoridad competente? Si No No sabe / No contesta

111 55 54 2

Entre 1990 y 1994 Ud. o algún miembro de su familia que vive en su mismo hogar sufrió un asalto fuera de su casa? Por favor tómese Si su tiempo para recordar. No No sabe / No contesta

400 36 346 18

Cuántas veces (en total sumando a todos los miembros de su familia que viven en su hogar)? Cantidad de veces

36 36

La última vez que Ud. o un miembro de su familia conviviente fue asaltado, fue un asalto a mano armada? Si No No sabe / No contesta

36 23 13 0

Y dónde ocurrió? En la calle En un vehículo o medio de transporte En el trabajo En un comercio o restaurante En un banco o cajero automático Otro lugar:......... No sabe / No contesta

36 23 4 3 5 1 0 0

Descripción

0 0

Si No No sabe / No contesta

36 4 32 0

Dónde?

Ud. o su familiar resultó lastimado?

Por favor, cuéntenos todas las cosas que le robaron a Ud. o al familiar asaltado?

Descripción


36 19 16 1

En el año 2001, alguna vez han robado o intentado robar en su casa (habiendo o no habiendo gente)? Si No No sabe / No contesta

400 40 359 1

Cuántas veces? Cantidad de veces

40 40


40 6 32 2


40 0 40 0

La última vez fue un asalto a mano armada?



Descripción


40

34


18 22 0

Obtuvo/vieron algún resultado de la policía o de la justicia luego de haber hecho la denuncia? Nunca tuve ninguna respuesta La denuncia quedo estancada en la justicia Me llamaron a declarar pero nunca paso mas nada Se investigó pero sin ningún resultado Se investigó y se obtuvieron algunos resultados Se investigó y se resolvió el caso No sabe / No contesta

18 10 3

En general cuál fue la razón principal por la que no denunció este hecho (No leer las opciones. Indicar solo una)?

22 La pérdida fue de escasa importancia/ no hubo pérdida Por el costo económico/ pérdida horas de trabajo/ costo de denuncia La policía no hubiera hecho nada La justicia no hubiera hecho nada La policía y la justicia no podrían haber hecho nada / por falta de pruebas Por miedo/ temor/ disgusto a relacionarse con la policía Por miedo / temor a represalias de los ladrones Porque no tenía seguro/ el seguro no cubría La policía desaconsejó / desalentó la denuncia Fue la policía Falta de documento de identidad/papeles en regla/domicilio/ Otras razones No sabe / No contesta

1 1 1 2 0

9 2 5 1 2 1 0 0 0 0 1 0 1

Ahora le voy a hacer preguntas referidas a hechos ocurridos en el pasado. Por favor tómese su tiempo para recordar. Entre 1995 y Si 2000 alguna vez han robado o intentado robar en su casa (habiendo No o no habiendo gente)? No sabe / No contesta

400 39 360 1


39 39


39 10 24 5


39 2 37 0




Descripción


Entre 1990 y 1994 alguna vez han robado o intentado robar en su casa (habiendo o no habiendo gente)? Por favor tómese su tiempo para recordar.


39 19 20 0


400 36 359 5

35


36 36


36 4 29 3


36 2 34 0



Por favor, cuéntenos todas las cosas que le robaron a ud. o al familiar asaltado?

Descripción


36 29 7 0

Ahora voy a leerle una lista de cosas que la gente hace para evitar ser victima de algún delito, y quisiera que responda para cada caso Si si Ud lo hace. Para evitar ser victima de algún delito Ud. Sale No menos cuando esta oscuro? No sabe / No contesta

400 244 154 2

Desde cuándo? Desde el último año Desde hace 1 a 5 años Desde hace 6 a 10 años Desde hace más de 10 años No sabe / No contesta

244 45 128 20 48 3


400 167 227 6

Desde el último año Desde hace 1 a 5 años Desde hace 6 a 10 años Desde hace más de 10 años No sabe / No contesta

167 26 92 27 21 1


400 143 256 1


143 40 75 15 12 1


400 179 219 2

Desde el último año Desde hace 1 a 5 años Desde hace 6 a 10 años

179 14 70 25

Ud. dejó de llevar o cambio el tipo de reloj o de joyas que usaba?

Desde cuándo?

Ud. dejó de llevar documentos, tarjetas de crédito, o tarjetas para cajeros automáticos?

Desde cuándo?

Han puesto rejas en su casa?

Desde cuándo?

36

Desde hace más de 10 años No sabe / No contesta

68 2

Uds. no dejan su casa sola y algún miembro de su hogar debe estar permanentemente? Si No No sabe / No contesta

400 154 243 3

Desde cuándo? Desde el último año Desde hace 1 a 5 años Desde hace 6 a 10 años Desde hace más de 10 años No sabe / No contesta

154 22 47 18 66 1


400 162 235 3


162 39 105 12 6 0


400 131 266 3


131 9 51 32 38 1

Ud. dejó de caminar por ciertos lugares o cambio su medio de transporte habitual?

Desde cuándo?

Para evitar ser víctima de algún delito, en su casa tienen un perro guardián?

Desde cuándo?

Uds. han asegurado los electrodomésticos o muebles que hay en su casa?

400 Si No No sabe / No contesta

110 286 4


110 14 46 23 25 2


400 80 317 3


80 14 35 19 10 2

Si

400 81

Desde cuándo?

Uds. colocaron una alarma contra robos?

Desde cuándo?

Participan Uds. de un plan de vigilancia entre los vecinos?

37

No No sabe / No contesta

314 5


81 16 42 8 14 1


400 119 279 2


119 11 57 29 22 0


400 78 318 4


78 6 48 9 13 2


400 77 316 7


77 0 14 7 55 1


400 20 379 1


20 5 8 5 2 0

Desde cuándo?

Colocaron Uds. en su casa una puerta blindada o cerraduras especiales?

Desde cuándo?

Contrata Ud. sólo o junto con sus vecinos o en su consorcio a un vigilante o empresa de seguridad?

Desde cuándo?

Para evitar ser víctima de algún delito, Ud. o alguien en su casa tiene un arma de fuego: revolver, escopeta, rifle?

Desde cuándo?

Con el propósito de evitar ser victima de algún delito, Ud. se ha mudado?

Desde cuándo?

En que barrio vivía antes? Nombre del barrio En que tipo de casa vivía antes? Vivienda en villa de emergencia Inquilinato / pensión Departamento en edificio (monoblock) Departamento tipo casa (prop. Horizontal)

38

20 1 1 4 4

Casa Barrio cerrado / country Otros No sabe / No contesta Su vivienda actual es: Vivienda en villa de emergencia Inquilinato / pensión Departamento en edificio (monoblock) Departamento tipo casa (prop. Horizontal) Casa Barrio cerrado / country Otros No sabe / No contesta Con cuánta frecuencia diría Ud. que pasa la policía frente a su casa, ya sea a pie o en auto? Todos los días Dos o tres veces por semana Una vez por semana Por lo menos una vez por mes Menos de una vez por mes Nunca No sabe / no contesta Necesito nuevamente hacerle preguntas referidas a hechos ocurridos en el pasado. Por favor tómese su tiempo para recordar. Número de veces Cuántas veces entre 1990 y 1992 inclusive Ud. o algún miembro de su familia que vive en su mismo hogar fue asaltado fuera de su casa? Cuántas veces entre 1993 y 1994 inclusive Ud. o algún miembro de su familia que vive en su mismo hogar fue asaltado fuera de su Número de veces casa?

9 0 0 1 400 1 3 155 32 206 1 2 0 400 194 45 21 21 10 57 52

400

400

Podría decirme la edad del jefe o sostén del hogar?

Puede decirme el máximo nivel educativo que alcanzo el jefe de hogar?

Número de años

400

Sin estudios Primaria incompleta Primaria completa Secundaria incompleta Secundaria completa Terciaria incompleta Terciaria completa Universitaria incompleta Universitaria completa Posgrado

400 1 16 109 23 104 5 33 28 75 6

Masculino Femenino

400 318 82

Trabaja Desocupado Subocupado (trabaja pocas hs.) Jubilado o Pensionado Rentista No sabe / No contesta

400 214 52 20 110 2 2

De qué sexo es el jefe de hogar?

En que situación se encuentra el jefe de familia? (Leer)

El jefe de familia trabaja (o trabajaba) en relación de dependencia o por cuenta propia (autónomo)?

400 Relación de dependencia Cuenta propia

39

248 152

Y a que se dedica (dedicaba)? Empleada doméstica Trabajador familiar Obrero no calificado Obrero calificado Empleado estado sin jerarquía Empleado privado sin jerarquía Jefe intermedio estado Jefe intermedio privado Gerente estado Gerente privado Alta dirección estado Alta dirección privado No sabe / No contesta

248 9 4 18 23 48 78 21 29 3 8 2 2 3

Changarín Trabajos no especializados Comerciante sin personal Técnico / artesano / trabajador especializado Profesional independiente Otros autónomos empleador de 1 a 5 empleados empleador de 1 a 20 empleados empleador de mas de 21 empleados rentista No sabe / No contesta

152 7 12 36 23 43 10 11 4 0 1 5

Y a que se dedica (dedicaba)?

Solo con fines estadísticos, podría decirme cuantos de los siguiente bienes poseen en su hogar: computadora, aire acondicionado, tarjeta Uno de crédito, lavarropas programable automático? Dos Tres Cuatro No sabe / No contesta

400 98 71 59 60 112

En su hogar tienen auto? (no como herramienta de trabajo, si la respuesta es afirmativa, preguntar cuántos)

400 No tiene auto

128 10 9

Chico mas de 10 años Chico entre 5 y 10 años Chico entre 2 y 5 años Chico menos de 2 años Mediano mas de 10 años Mediano entre 5 y 10 años Mediano entre 2 y 5 años Mediano menos de 2 años Grande mas de 10 años Grande entre 5 y 10 años Grande entre 2 y 5 años Grande menos de 2 años No sabe / No contesta

138 5 10 7 2 18 30 25 8 16 6 5 1 5


138 74 57 7

Desde el último año Desde hace 1 a 5 años

73 5 28

Podría decirme la antigüedad y el tamaño del auto principal?

Utilizan para el auto alguno de los siguientes mecanismos de protección o de seguridad? Candado, cadena, traba de volante

253

Uno Dos o mas No sabe / No contesta

Desde cuándo?

40

Desde hace 6 a 10 años Desde hace más de 10 años No sabe / No contesta

25 15 0


138 89 42 7


87 6 43 26 12 0


138 106 28 4


106 4 47 33 22 0


138 32 101 5


32 1 17 9 5 0

Capital Sur Capital Centro Capital Norte Capital Noroeste GBA Norte GBA Oeste GBA Sur

400 36 78 50 36 78 76 46

Alarma, corta corriente o corta nafta

Desde cuándo?

Seguro contra robo

Desde cuándo?

Aunque tiene garaje propio, estaciona en un garaje fuera de su casa

Desde cuándo?

Zona

41

Appendix 3: Computation of the Income Level Index17 The Income Level index (IL) assigns a point average for each household according to three variables. The index can take values between 4 and 100 points. The variables and the maximum values are summarized in the following table: MAXIMUM VARIABLE POSSIBLE VALUE - Education 32 - Occupation 40 - Wealth a. goods and services 14 b. automobile 14 Total 100 Assignment of Points for each variable 1. Educational level of the household head. The values vary from 0 to 32 points according to the following table: EDUCATIONAL LEVEL POINTS No studies 0 Primary School Incomplete 5 Primary School Complete 9 High School Incomplete 13 High School Complete 17 Vocational School Incomplete 19 University Incomplete 22 Vocational School Complete 27 University Complete 31 Postgraduate Studies 32 2. Occupation of the household head. The assigned points range from 4 to 40 according to the following table:

17

See Asociación Argentina de Marketing (1998).

42

NON-EMPLOYEE Do Not Work Asset Holder

POINTS

EMPLOYEE POINTS Domestic Employee 7 20 Family Worker without Fixed Income 13 Non-Qualified Operator 9 Qualified Operator 17 Self-Employed Day Laborer 4 Technician / Foreman 23 11 Other Non-Specialized job Low Hierarchy Employee Retailer without Employees 18 Public Sector 12 Technician/Specialized worker 24 Private Sector 17 Independent Professional 30 Middle Hierarchy Employee Other Self-Employed 17 Public Sector 19 Private Sector 24 High Hierarchy Employee Employer 1-5 employees 30 Public Sector 26 6-20 employees 36 Private Sector 30 21 or more employees 40 Top Hierarchy Employee Public Sector 28 Private Sector 37 Note: 2/3 of the points of his/her last occupation are assigned to unemployed, retired or pensioner household heads. 3. Wealth. a. Goods and services. It measures the household capacity of accumulation of goods and services. The points are assigned according to the following table. NUMBER OF THE FOLLOWING GOODS AND SERVICES OWNED: PC, AIR CONDITIONER, CREDIT CARD AND Points AUTOMATIC WASHING MACHINE 0 0 1 3 2 7 3 11 4 14 b. Automobile: the questions asked are concerned with (i) the number of automobiles owned, (ii) the branch, model and age of the first automobile, if applies, and (iii) the branch, model and age of the second automobile, if applies. Using this information, points are assigned separately for each car according to the following table. BRANCH/MODEL AGE 10 and more years Between 6 and 9 years Between 3 and 5 years Less than 2 years

INFERIOR

MEDIUM

SUPERIOR

1.5 3.5 5.5 6.5

2 6 7 8

2.75 6.5 8.5 9.5

Finally, the automobile point assignment must satisfy the following two rules: (1) if the final number of points is between 1 and 3, then zero is assigned to this category; and (2) if the sum of the points assigned for both cars together reaches 15 points or more, then 14 is assigned to this category.

43