The effect of police on crime, disorder and victim precaution. Evidence from a Dutch victimization survey

The effect of police on crime, disorder and victim precaution. Evidence from a Dutch victimization survey

International Review of Law and Economics 29 (2009) 336–348 Contents lists available at ScienceDirect International Review of Law and Economics The...

734KB Sizes 0 Downloads 2 Views

Recommend Documents

No documents
International Review of Law and Economics 29 (2009) 336–348

Contents lists available at ScienceDirect

International Review of Law and Economics

The effect of police on crime, disorder and victim precaution. Evidence from a Dutch victimization survey Ben Vollaard a , Pierre Koning b,∗ a b

CentER, University of Tilburg, Netherlands CPB Netherlands Bureau for Economic Policy Analysis, Netherlands

a r t i c l e JEL classification: K4 C23 Keywords: Police Crime Public disorder Victim precaution

i n f o

a b s t r a c t Using individual data from a large-scale Dutch crime victimization survey, we are able to expand the analysis of the effect of police on crime to crimes types that do not easily find their way into police statistics, and to public disorder and victim precaution. To address heterogeneity and simultaneity in the relation between police and crime, we model the police funding formula – used to distribute police resources across municipalities – to identify the endogenous variation in police levels. We use the remaining variation in police levels to identify the effect of police. We find significantly negative effects of higher police levels on property and violent crime, public disorder, and victim precaution. The effect on victim precaution is a hitherto largely ignored benefit of higher police levels not reflected in lower rates of crime and public disorder. © 2009 Elsevier Inc. All rights reserved.

1. Introduction The renewed interest in testing the theory of deterrence has resulted in a number of recent studies into the effect of police on crime. Creative research designs have been used to break through the simultaneity between police and crime levels and to address omitted variable bias. Di Tella and Schargrodsky (2004) and Klick and Tabarrok (2005) use shocks to police presence related to terrorist attacks and terrorist alerts to isolate causal direction. Corman and Mocan (2000, 2005) use high frequency data to escape the (slow) adjustment in allocation of police resources to crime rates. Levitt (2002) and Lin (2009) use instrumental variables to identify changes in police levels that are not related to changes in crime rates. These recent studies consistently find a negative effect of police on crime, with Di Tella and Schargrodsky (2004) and Klick and Tabarrok (2005) providing the clearest evidence that the estimated effects are the result of deterrence rather than incapacitation. Given the reliance on police statistics as source of crime data in the literature, the current evidence on the deterrent effect of police is mostly limited to crimes that are relatively well-reported by the public and well-recorded by the police, such as domestic burglary and car theft. Evidence on the effect of police on imperfectly measured violent crime is scarce, with only Levitt (2002)

∗ Corresponding author. E-mail addresses: [email protected] (B. Vollaard), [email protected] (P. Koning). 0144-8188/$ – see front matter © 2009 Elsevier Inc. All rights reserved. doi:10.1016/j.irle.2009.03.003

reporting an estimate that is of borderline statistical significance. Currently, no evidence on the effect of police on public disorder is available (see Weisburd & Eck, 2004 for a review of the literature). We are able to expand the scope of the analysis by using data from a large-scale Dutch victimization survey. Next to the traditional crime categories that also find their way into police recorded crime statistics, the victimization survey provides data on various types of public disorder such as nuisance from drug users, graffiti and littering. The survey also provides more reliable data on crimes that tend to be poorly reported and recorded, including violent threats. As a further extension of the study into how police affect incentives, we analyze how victim precaution is affected by greater levels of police protection. A lower level of victim precaution is a hitherto largely ignored gain of higher police levels that is not reflected in lower levels of crime and disorder.1 The victimization survey provides a rich source of individual data, allowing us to estimate the effect of police on crime at the level

1 Philipson and Posner (1996) argue that a decline in victim precaution due to better police protection partially offsets the effect of police on crime. An increase in the level of public protection, as by hiring more police, will cause the crime rate to fall and thus will lower the demand for private prevention – which will cause the crime rate to rise again, partially undoing the effect of the increase in public protection. Because of same-year simultaneity between crime and preventative behavior, we are not able to test whether lower levels of victim precaution in turn lead to higher levels of crime.

B. Vollaard, P. Koning / International Review of Law and Economics 29 (2009) 336–348

of the individual rather than at the level of city districts (Di Tella & Schargrodsky, 2004; Klick & Tabarrok, 2005), cities (Corman & Mocan, 2005; Levitt, 2002) or states (Lin, 2009). The survey includes a wide range of background characteristics of individuals (both victims and non-victims), including education, ethnicity, employment and type of residence. In contrast, when using data on offenders, individual characteristics can only be collected through painstaking efforts related to combining police statistics with sources such as school records and draft registration records. Because of the costs related to such data collection, the resulting data tend to be limited to one locality, such as the Philadelphia cohort of young men studied by Tauchen, Dryden Witte, and Griesinger (1994). As we will show in the analysis, individual characteristics are particularly important for explaining violent crime. To address endogeneity in the relation between police and crime in non-experimental data, we make use of the fact that the distribution of police resources across municipalities is based on a formula. The police funding formula includes predictors of local police workload such as housing density and length of roadways. Given the time needed to hire and train police personnel and the practice of smoothing year-to-year changes in local police resources, actual police levels differ from police levels prescribed by the funding formula. We use the difference between actual and prescribed police levels as source of exogenous variation in police levels. We find significantly negative effects of higher police levels on property and violent crime, public disorder, and victim precaution. The estimated effects on victimization of crime and also public disorder are similar in size to the effects on recorded crime found in previous studies. The estimated elasticity of police with respect to crime and public disorder varies between −0.2 and −0.5. In addition, we find people to move around more freely in response to greater police levels, with an estimated elasticity of −0.2. The rest of the paper is organized as follows. Section 2 describes our data. In Section 3, we discuss the empirical strategy to address unobserved heterogeneity, simultaneity and measurement error. Section 4 presents the estimation results. Section 5 concludes. 2. Data For our analysis, we use the Dutch Victimization Survey (PMB). The PMB is a repeated cross-section telephone survey that is unique in its sampling size. Whereas the US National Crime Victimization Survey covers 1 out of 2000 of the population above 12 years of age, the PMB covers some 1 out of 200 of the population above 15 years of age. Given the relatively infrequent nature of victimization of crime – with the average risk of victimization varying between 1 and 11 percent across crimes – a large sample size is essential for obtaining reliable estimates of the effect of police on victimization of crime. The PMB contains detailed information on victimization of crime, experience of public disorder and victim precaution. For every survey wave, respondents have been selected at random

337

from the total population over 15 years of age. Per police force area (and sometimes smaller areas), the interviewers used stratified sampling. A minimum of 1000 respondents were interviewed in each of the 25 police force areas. Response rates are relatively high, around 50 percent for the first two waves, and around 70 percent for the later waves when respondents also received a letter introducing the survey before the actual telephone interview. Analysis of age and motives of people refusing to participate in the survey does not indicate a systematic non-response bias (PMB, 2005). All victimization is observed at the individual level, only some property crimes are measured at the household level (bicycle theft, burglary, car theft, theft from car) as well as one form of victim precaution (not allowing children to go out). Respondents are interviewed in the first ten weeks in the year a survey is held, with the victimization reflecting the twelve months preceding the interview date. This means that a substantial part of the observed victimization occurs in the year preceding the interview year. For this reason, we use crime and public disorder data from the survey in year t for observations in year t − 1. The victimization survey does not only provide data on public disorder and victim precaution, but also provides measures of property and violent crime that are not affected by changes in reporting behavior of the public and recording behavior of the police. Changes in the way crime finds its way into police recorded crime may result in measurement error in this source of data (MacDonald, 2002). The Netherlands is no exception. Whereas Dutch victimization data and police statistics for property crime show somewhat similar trends, the trends for violent crime are rather different (Fig. 1). Research into violent crime statistics by Wittebrood and Junger (2002) shows that police reports are made for an increasing number of notified crimes and more police reports are finding their way into official records. Thus the discrepancy between victimization data and police statistics, the so-called ‘dark number’, is becoming smaller. The diverging trend between recorded violent crime and victimization of violent crime can be seen in many countries, including the US (Rand & Rennison, 2002). In addition to the survey data, data on police resources were obtained from the Dutch Interior Department. Historical series of police levels are only available at the regional level. For all 25 police force areas, growth in police personnel outstripped growth in population in this period. The total number of police personnel in full time equivalents increased from 39,462 in 1996 to 47,342 in 2004. We obtained data for some of the funding formula variables from the Netherlands Bureau of Statistics. We have pooled PMB data for the years 1996–2004. During this period, the PMB survey was conducted every other year. Thus our sample includes five waves. Table 1 summarizes the key variables we use for the empirical analysis. Background characteristics include age, gender, education level, ethnicity, housing type, household size, and income status (employed, student, housewife, or else).

Fig. 1. Trends in victimization data and police statistics, property and violent crime (1996 = 100). Source: Dutch Victimization Survey, Netherlands Bureau of Statistics.

338

B. Vollaard, P. Koning / International Review of Law and Economics 29 (2009) 336–348

Table 1 Sample statistics (1996–2004).

3. Research design

Deterrence measure Police personnel per 100,000 population

Mean

Standard deviation

273

103

Victimization of crime Bicycle theft (per household) Theft from car (per household) Theft of car (per household) Burglary (per household) Threat with violence Assault Robbery with violence

0.11 0.07 0.01 0.06 0.05 0.01 0.00

0.31 0.25 0.09 0.23 0.22 0.09 0.06

Frequent experience of public disorder Littering Graffiti Nuisance from youth Harassment (in public spaces) Public intoxication Vandalism Nuisance from drug users

0.29 0.14 0.12 0.03 0.08 0.19 0.07

0.46 0.35 0.33 0.18 0.27 0.39 0.25

0.09 0.17

0.29 0.38

0.17

0.38

0.54 47.8 0.24

0.50 17.4 0.43

0.05 0.04 0.57 0.20 0.22 0.15 0.57 0.11 0.03 0.08

0.22 0.04 0.50 0.40 0.42 0.23 0.50 0.03 0.01 0.06

Frequently taken forms of victim precaution Drive or walk round to avoid unsafe places Leaving valuable properties at home to prevent theft Not allowing children to go out because of safety reasons (per household) Individual and municipality characteristics Male Age Primary school or basic vocational training only Student Immigrant Employed Housewife Single household Childrena Terraced house Average moving mobilityb Average number of shopsc Average length of roadways (km)

We model the binary response variable y representing the event of becoming victim of crime, experiencing public disorder or taking precautionary measures y as follows: Pr(yijt = 1|˛, X) = ˛j + X ijt ˇ +  ln pjt−1

Notes: (a) Share of children in total household size. (b) Number of persons moved per 1000 population: number of persons moved is equal to the people moving within a municipality plus the sum of half of the people moving into the municipality and half of the people moving out of the municipality. (c) Number of commercial services establishments.

We estimate the effect of police on victimization of crime and experience of public disorder at the level of individual respondents, taking into account municipality-specific effects. In the pooled sample, we have 644 municipalities, with 599 individual observations on average. Table 2 shows the distribution of individual observations over municipalities. For the vast majority of municipalities, we have sufficient individual observations to obtain reliable municipality-specific effects. Table 2 Distribution of respondents across municipalities (1996–2004). Number of individual respondents per municipality <50 50–100 100–200 200–300 300–400 400–500 500–1000 >1000 Total number

Percent of municipalities

3.1. Heterogeneity across municipalities

Percent of respondents

11 13 19 13 6 8 19 11

1 2 5 5 4 6 23 55

644

385,543

Note: As a result of amalgamations, the number of municipalities declined from 625 in 1996 to 483 in 2004. We coded a subset of all amalgamations to limit the number of municipalities with a very low number of observations.

(1)

In this Linear Probability Model (LPM), ˛j indicates an effect on y that is specific for municipality j (with j = 1,. . . J). Xijt is a matrix representing the characteristics of individual i (with i = 1,. . . I) living in locality j at time t (t = 1,. . . T), with ˇ as a column vector describing the effects of X. ln pjt−1 represents the logarithmic value of the number of police personnel per capita in municipality j at time t − 1, and  is a parameter describing the effect of ln p. For ease of exposition, we initially assume that the number of police personnel per capita is observed at the level of municipalities. At the end of this section, we discuss the level of the analysis in more detail. As is well known in the literature, using the linear probability model as a convenient approximation to the underlying response probability may not be appropriate at extreme values of X, where predicted values may be outside the unit interval. However, since the main purpose of our analysis is to estimate the partial effect of X on the response probability, averaged across the distribution of X, the presence of predicted values outside the unit interval is not a major concern (see also Wooldridge, 2002). The choice for the linear probability model is further substantiated by the large sample properties of the data that we use, with the number of observations varying between 310,000 and 370,000.2 We include individual characteristics Xijt to prevent estimation bias through (observable) factors that affect both police and victimization levels (we also include household size and household composition for crimes measured at the household level). Individuals with a higher probability of victimization may also enjoy greater police protection. Ignoring the correlation between police and victimization levels would bias the estimated effect of police on crime towards zero. Victimization is not determined by individual characteristics alone. As we will show in Section 4, experience of disorder like vandalism and public intoxication is mostly determined by the characteristics of the municipality in which someone lives rather than by his or her individual characteristics like age and sex. Similarly, the number of police experienced at the individual level is affected by the type of community someone lives in. This co-determination of victimization and police numbers at the aggregated level calls for including locality-specific factors in the analysis. In the Netherlands, a funding formula is used to distribute police resources across municipalities. The funding formula includes municipality-specific characteristics such as length of roadways and number of population. Thus municipality-specific characteristics that are related to both police and victimization levels are all observable. By including the variables from the formula that are related to crime rates in the estimation equation, we are able to control for the fact that municipalities with persistently higher crime rates receive higher levels of police staffing.3 By explicitly modeling

2 Note that we can also exploit the large sample properties of our sample to justify the t-tests we apply in the following sections. In particular, we assume that, given large the sample size, the (average) effect the standard errors are independently and identically distributed. 3 The police budget has been distributed by means of a formula since the 1960s. Back then, the formula only included number of population. In 1985, number of population per house was added. After a thorough revision in 1989, the funding formula included number of population below 20 years of age, number of immigrants, number of unemployed, number of houses, number of personnel in service sector,

B. Vollaard, P. Koning / International Review of Law and Economics 29 (2009) 336–348

budget policy, we aim to eliminate correlation between ln p and ˛, the municipality-specific effect. As we will show later in this section, not all variation in police levels is explained by the crime-related variables of the funding formula. Redistribution of police personnel across municipalities takes long to crystallize out, both in the case of uneven growth in new personnel and in the case of shifts in existing resources between municipalities. New personnel needs to be hired and trained and transfers of personnel are slow. We use the remaining variation in police levels across municipalities and over time as source of exogenous variation. We include seven variables from successive revisions of the funding formula in our estimation equation (all per population): number of population below 20 years of age (we also include the other age categories), number of immigrants, number of shops, number of houses (we specify three types of housing), length of roadways, number of moves and the product of housing density and number of houses. Because of the focus of the survey questionnaire, we include the number of employed rather than the number of unemployed. Statistics on the number of cafe personnel and parking places are no longer available. To allow for the possibility that variation in police levels – after controlling for funding formula variables – may still be crimeinduced, we include additional municipality characteristics by averaging each individual control variable at the level of the municipality, including level of education, sex and two variables denoting whether the respondent is a student or house wife. We also include the average police level in a municipality. Year fixed effects are included to focus on differences between municipalities rather than capturing the correlation between national trends in police and victimization levels. Including municipality averages is in line with the modified random effects model developed by Mundlak (1978). Within this approach, the correlation between the municipality effect and the time varying observables is specified as a linear function of municipality averages: Pr(yijt = 1) = ajt + X ijt ˇ +  ln pjt−1

(2)

with the auxiliary regression: ajt = ı1 X.j + ı2 ln pj. + j.

(3)

where X.j represent characteristics of municipality j, and ı1 describes the effect of these characteristics on ˛. ln pj. is the average value of police levels per municipality. j. represents the remaining municipality effect. We assume this variable to be independent and identically distributed. By adding average values of X and ln p as a set of controls for unobserved heterogeneity, we disentangle the ‘within’ from the ‘between’ estimators of both coefficients. Thus the coefficient estimates are identified from variation of X and ln p, holding the averages constant.4

number of personnel in cafes, and the number of parking places. In 1996, the year our analysis starts, the formula was revised again. The current set of variables include number of population, number of immigrants, moving mobility, number of houses, number of shops, length of roadways and the product of housing density and the number of houses. 4 Modifying the random effects model along these lines is related to the empirical literature on cluster or peer effects (Manski, 1993). In this literature, adding group averages has been used to estimate the importance of cluster effects—that is, the effect of the behavior of other people in a reference group on the behavior of an individual. Within the context of our model, however, the interpretation of cluster effects is different. The coefficients for individual characteristics reflect the effect of an individual’s behavior on the risk of victimization. The coefficients for the municipality averages reflect both cluster and sorting effects. Cluster effects occur if reference group behavior affects the individual risk of victimization. Think of vandalism indiscriminately affecting people living in a certain municipality. Sorting effects stem from individuals with similar characteristics concentrating into municipalities

339

Given the size of the data set, we expect the linear probability model to provide consistent estimates of the partial effects for the average values of X and ln p, provided that estimation techniques are robust to heteroscedasticy. The linear probability model allows us to avoid the computationally complex estimation of municipality specific in a non-linear setting. 3.2. Simultaneity For our estimates of police effectiveness to be consistent, changes in police personnel should not be correlated with local trends in crime. If they are, the estimated effect of police on crime and public disorder will be biased. Simultaneity is not likely to be a strong source of estimation bias. The funding formula focuses on overall crime, whereas we relate changes in police levels to trends in specific types of crime, public disorder, and victim precaution. There is quite some variation in trends between different crime categories, with some types of property crime going down and violent crime going up. In addition, it takes at least two years to hire and train new personnel. Therefore, budget decisions have a lagged effect on police levels, making it more difficult to respond to differences in crime trends (rare shifts of existing resources between municipalities were also slow to materialize). Finally, none of the funding formula variables has been updated every year; some variables have not been updated at all during 1996–2004. Regular updating of all variables would have resulted in reoccurring major redistributions of police resources across municipalities, which was seen as undesirable. Clearly, infrequent updating of variables impairs a policy response to differing local crime trends. Simultaneity may not be strong, but cannot assumed to be completely absent. To control for changes in the distribution of police resources that are related to local crime trends, we allow funding formula variables that were updated during the period of our analysis as well as the other municipality characteristics to vary over time. Thus, X.j in the auxiliary regression varies from year to year and becomes X.jt. We take the average value for the number of shops, moving mobility and length of roadways, since these funding formula variables have not been updated. Table 3 shows that the variables from the funding formula, the other municipality characteristics and the national trend together explain some three quarters of the variation in police levels. We assume the remaining variation to be exogenous to the relation between police and crime. An important related assumption is that the police does not change its prioritization of resources targeted at combating specific types of crime and public disorder. To illustrate the estimation approach, Fig. 2 in the appendix shows scatter plots for all 14 types of crime and public disorder with the deviation from police per capita levels prescribed by the funding formula on the horizontal axis and the risk of victimization of crime on the vertical axis. To make the crime rate in all areas comparable, the risk of victimization is adjusted for the average risk of victimization and the national trend. Generally, the scatter plots show that the greater the level of police per capita, the lower the risk of victimization of crime and disorder. 3.3. Victim precaution In analyzing the effect of police on victim precaution, we focus on three forms of precaution that people can easily alter according to changes in the conditions in their locality: drive or walk round to

with an a priori higher risk of victimization. In Section 4, we estimate the relative importance of individual versus municipality effects for several types of crime and public disorder.

340

B. Vollaard, P. Koning / International Review of Law and Economics 29 (2009) 336–348

Table 3 The police funding formula as predictor of police levels. Demographic structure Age < 20 20 < Age < 30 30 < Age < 50 50 < Age < 60 Housing type Terraced house Detached house Apartment Immigrants Employed Housing density * houses Average number of shops Average number of moves Average length of roadways Number of observations R2

0.13 −0.94** −0.11 −0.11

(0.47) (0.40) (0.37) (0.35)

0.57* 0.41 1.31***

(0.29) (0.27) (0.32)

1.54*** 0.12 0.14*** −0.61 −2.84*** 0.67**

(0.44) (0.38) (0.03) (2.27) (0.93) (0.31)

370,437 0.74

Notes: Results for year fixed effects are not reported. Standard errors are adjusted for correlation within police regions. Standard errors between parentheses. * Statistically significant at the 10-percent level. ** Statistically significant at the 5-percent level. *** Statistically significant at the 1-percent level.

municipalities, then – using similar arguments as in the case of police personnel – consistency is still achieved. This is not an argument to estimate the model using police force areas as reference groups for all independent variables. Given the number of independent variables in our model, the number of police force areas is very limited, casting doubts on the efficiency of the estimated parameters of the averaged variables.7 This is also likely to affect the efficiency of the estimated effect of police on crime. Thus, observing police personnel at the level of police force areas rather than municipalities does not imply switching to police force areas as cluster groups for the other independent variables as well. 4. Estimation results In this section we present the estimation results. We start by presenting the estimation results of the modified random effects approach as defined by Eqs. (2) and (3): the ‘preferred model’. In order to assess the robustness of our findings, we re-estimate our model using various other specifications. Then, we go into the explanatory power of control variables at the individual versus the municipality level. 4.1. Effect of police on crime and public disorder

avoid unsafe places, leaving valuable properties at home to prevent theft, and not allowing children to go out because of safety reasons. These are also measures people decide for themselves, as against measures like additional hinges and locks on doors and windows that local building associations may decide to install or the police might advise about.5 Just like the one-year lag between the effect of police on crime, we assume a one-year lag between higher police presence and changes in victim precaution. We assume that when making a decision on whether to avoid a certain street for instance, potential victims treat public expenditures on crime control as exogenous. Since our estimation approach addresses simultaneity between police and crime, we also deal with simultaneity between police and victim precaution related to crime levels. 3.4. Measurement error We use municipalities as the relevant geographical or reference unit in our analysis. Thus far, we have abstracted from the fact that police personnel per capita is measured with error—that is, we observe police personnel per capita for 25 police force areas, instead of 644 municipalities. Suppose we define the natural logarithm of police personnel per capita as the sum of a police force area component and a municipality-specific component. Furthermore, we assume the municipality-specific component to have an expected value of zero and to be independent and identically distributed. Then, in order to obtain consistent estimates, the municipalityspecific component should not be correlated with the regional component. This assumption is common for various applications of stratified sampling, where units in the sample are represented with different frequencies than they are in the population.6 Measurement error may also arise with respect to the averages of the other independent variables in our model. Suppose that these variables would be averaged across police force areas instead of

5 In the case the police advice citizens on precautionary measures, more police could lead to more victim precaution. We focus on forms of victim precaution that are most likely to be individual decisions affected by the degree of police protection rather than by advice from the police. 6 This contrasts to situations where the classical errors in variables (CEV) assumption applies. In that case, measurement errors are (fully) correlated with the observed variable, causing estimates to be biased to zero (attenuation bias).

Table 4 presents the estimation results based on the preferred model.8 When analyzing property crime, we find the police to be effective in reducing bicycle theft and theft from cars. The effect of police on victimization of burglary and car theft is not statistically significant. The result for car theft is most likely due to low victimization rates (1 percent of households) and correspondingly high standard errors. In contrast to police statistics, a victimization survey provides less precise figures for relatively rare types of crimes. Apparently, the police are not very successful in bringing down burglary, a result that stands in contrast to other studies that find a statistically significant negative effect of police on burglary of around −0.30 (Marvell & Moody, 1996; Corman & Mocan, 2005; Klick & Tabarrok, 2005). We estimate a one percent increase in police personnel to result in a decrease in threat with violence of 0.4 percentage point. We also find a negative effect on assault and robbery with violence, but the effect is not statistically significant. Victimization of both types of crimes occurs even less frequently than victimization of car theft, which results in high standard errors. We select seven measures of public disorder included in the survey that are available for each wave: littering, graffiti, vandalism, harassment, youth nuisance, public intoxication and drug nuisance. We find the effect of police to be negative and statistically significant for littering, harassment, youth nuisance, public intoxication and drug nuisance. The size of the effect on public disorder is similar to the estimated effect on crime. We find no significant effect of police on graffiti and vandalism. Given the observed rates of change in police levels, the size of the estimated effect of police on crime and disorder is substantial. Our estimates imply that the 30 percent increase in police per capita in the Netherlands over the period 1996–2004 resulted in a decrease in crime and disorder by some 10 percent. Although the growth in police levels was relatively strong over this period, a similar rate of expansion is not uncommon. The Netherlands experienced comparable growth in police levels in the years preceding

7 See Wooldridge (2002), who discusses the importance of a high number of clusters to be able to apply panel data methods. 8 The elasticity is defined as the estimated coefficient for ln p (as in the appendix) divided by the average value of that specific type of crime or public disorder (as in the summary statistics).

B. Vollaard, P. Koning / International Review of Law and Economics 29 (2009) 336–348

341

Table 4 Effect of police on crime and public disorder—implied elasticities. Property crime Burglary Car theft Theft from car Bicycle theft

Violent crime −0.14 0.17 −0.31** −0.48***

(0.14) (0.41) (0.14) (0.10)

Public disorder

Threat with violence Assault Robbery with violence

−0.38** −0.32 −0.37

(0.16) (0.40) (0.59)

Littering Graffiti Vandalism Harassment Youth nuisance Public intoxication Drug nuisance

−0.16*** −0.07 −0.11 −0.47** −0.42*** −0.20* −0.36***

(0.06) (0.09) (0.08) (0.20) (0.10) (0.12) (0.13)

Notes: Estimation results for all other variables are included in the appendix. Standard errors are between parentheses. * Statistically significant at the 10-percent level. ** Statistically significant at the 5-percent level. *** Statistically significant at the 1-percent level.

Table 5 Point estimates for effect of police on property and violent crime.

This study Marvell and Moody (1996) Levitt (2002) Di Tella and Schargrodsky (2004) Klick and Tabarrok (2005) Corman and Mocan (2005) Lin (2009)

Unit of analysis

Property crime

Car theft and theft from cars

Violent crime

Dutch municipalities Major US cities Major US cities Buenos Aires neighborhoods Washington D.C. city districts New York City US states

−0.35***

−0.28** −0.85**c

−0.31**b

a

−0.50**

−2.18**

−0.44* −0.33***c,d −0.86** −0.56**c −4.14***c

Notes: (a) Includes burglary, car theft, theft from car and bicycle theft. (b) Includes threat with violence, assault and robbery with violence. (c) Excludes theft from cars. (d) Deterrence effects only. * Statistically significant at the 10-percent level. ** Statistically significant at the 5-percent level. *** Statistically significant at the 1-percent level.

the analysis, 1992–1996, and during the period 1970–1985. At the individual level, the increase in police levels imply a small but not negligible decrease in the probability of victimization. For instance, the average probability of becoming victimized by theft from car decreased from once every 13 years in 1996 to once every 14 years in 2004 as a result of the expansion in police personnel. 4.2. Comparison with other studies Table 5 shows that our estimated elasticities for property and violent crime are in line with recent studies based on police statistics, with the exception of Lin (2009) whose estimates are much higher than any estimates previously reported (we only include estimates statistically significant at the 10-percent level or lower). For reasons of comparability, we have lumped together several types of property crimes and violent crimes. Although country-specific characteristics may affect the size of the estimated effect, the results do not suggest that the use of a victimization survey as alternative source of crime data greatly affects the estimated effect of police on crime. If anything, our point estimates are somewhat lower. Thus, our results confirm the existing evidence using a different source of crime data and a different estimation method. 4.3. Effect of police on victim precaution Table 6 provides the estimated effect of police on victim precaution. As expected, more police leads to lower victim precaution. A Table 6 Effect of police on victim precaution, estimated elasticities. Drive or walk round to avoid unsafe places Leaving valuable properties at home to prevent theft Not allowing children to go out because of safety reasons

−0.25**

(0.11)

−0.16**

(0.08)

−0.08

(0.15)

Notes: Standard error between parentheses. * Statistically significant at the 10percent level. ** Statistically significant at the 5-percent level. *** Statistically significant at the 1-percent level.

one percent increase in police levels leads to a 0.2 percent decrease in people who frequently drive or walk round to avoid unsafe places and a similar decrease in people who frequently leave valuable properties at home to prevent theft. The effect on parents’ behavior towards their children is not statistically significant. Thus, we provide evidence that there is an additional benefit of police on victim precaution that is not reflected in lower levels of crime and public disorder. 4.4. Robustness To assess the importance of controlling for the policy response to different crime (trends) and to validate the robustness of our results, we re-estimate the model using various other specifications. We vary the set of control variables, the method of estimation, and the circumstances under which the police work. Table 7 provides an overview of the results—with the preferred model as our benchmark. In the second specification, we exclude municipality characteristics that are part of the funding formula. As discussed in Section 3, in addition to individual background characteristics, we include the funding formula variables to control for simultaneity in the relation between police and crime at the level of the municipality. The estimation results show that not explicitly controlling for the relation between budget policy and victimization rates produces similar estimates for property and violent crime, while resulting in overestimation of the effect of police on public disorder. As we discuss in the below, municipality characteristics explain most of the variation in public disorder, whereas individual characteristics are of more importance for explaining variation in property and violent crime. Consequently, leaving out the funding formula variables biases the results for public disorder more than the results for crime. Given the upward bias in specification (ii), the allocation of police resources tends to be at a disadvantage to municipalities with severe and persistent disorder problems. We re-estimate the model using fixed effects estimation in the third specification.9 In previous studies into the effect of police

9 The fixed effects model is based on a linear probability model as well. The advantage of using a linear probability specification rather than a binary model is that fixed

342

B. Vollaard, P. Koning / International Review of Law and Economics 29 (2009) 336–348

Table 7 Effect of police on crime and public disorder—elasticities for various model specifications. (i) Preferred model

(ii) Excl. funding formula variables

(iii) Fixed effects estimation

(iv) Highly urbanized areas only

Property crime Burglary Car theft Theft from car Bicycle theft

−0.14 0.17 −0.31** −0.48***

(0.14) (0.41) (0.14) (0.10)

−0.19 −0.07 −0.26** −0.55***

(0.14) (0.38) (0.13) (0.10)

0.05 −0.16 −0.24** 0.26*

(0.19) (0.57) (0.19) (0.14)

−0.18 0.13 −0.37*** −0.51***

(0.14) (0.41) (0.14) (0.10)

Violent crime Threat with violence Assault Robbery with violence

−0.38** −0.32 −0.37

(0.16) (0.40) (0.59)

−0.41** −0.34 −0.35

(0.15) (0.38) (0.77)

−0.42** −0.48 −0.49

(0.21) (0.52) (0.79)

−0.39** −0.34 −0.48

(0.16) (0.40) (0.60)

Public disorder Littering Graffiti Vandalism Harassment Youth nuisance Public intoxication Drug nuisance

−0.16*** −0.07 −0.11 −0.47** −0.42*** −0.20* −0.36***

(0.06) (0.09) (0.08) (0.20) (0.10) (0.12) (0.13)

−0.30*** −0.23*** −0.27*** −0.75*** −0.46*** −0.22*** −0.57***

(0.05) (0.08) (0.08) (0.18) (0.09) (0.12) (0.12)

−0.20*** −0.32*** −0.07 1.26*** −0.61*** 0.08 0.74***

(0.07) (0.12) (0.10) (0.26) (0.13) (0.16) (0.17)

−0.14*** −0.09 −0.16** −0.48** −0.44*** −0.15 −0.31**

(0.06) (0.09) (0.08) (0.20) (0.10) (0.13) (0.13)

Notes: Standard errors are between parentheses. * Statistically significant at the 10-percent level. ** Statistically significant at the 5-percent level. *** Statistically significant at the 1-percent level.

on crime, fixed effects estimation is the default option to address unobserved heterogeneity. In the fixed effects models, differences between municipalities are specified as an effect that is constant over time. A comparison between random effects and fixed effects estimation results shows that if we control for all possible municipality characteristics that are constant over time, but do not control for simultaneity, we are likely to underestimate the effect of police on crime. As we would expect, changes in police personnel per capita tend to be positively correlated with trends in crime. In the case of disorder, ignoring simultaneity tends to lead to underestimation as well, with graffiti and youth nuisance as the exceptions. Thus municipalities with relatively unfavorable trends in disorder tend to profit from resource decisions made during 1996–2004. This finding stands in contrast with specification (ii), where excluding all funding formula variables resulted in overestimation of the effect. Specification (iii) shows that this overestimation is not likely to be the result of simultaneity, but of time-constant characteristics of municipalities. In other words, municipalities with severe and persistent public disorder problems have not been allocated relatively high staffing levels. The use of non-experimental data allows us to study the robustness of our findings under different conditions. In specification (iv), we vary the degree of urbanization of a municipality in which the police are active. Intuitively, we expect the police to be most effective in fighting crime and public disorder in urban areas. After all, a police officer in a densely populated area can control more people than a police officer in a sparsely population area. To test the difference in effectiveness between urban and rural regions, we include an interaction term for police levels and average police levels in the four most urbanized police force areas in the preferred model.10 We assume that the impact of all other explanatory variables in our model is equal for regions with a high and low degree of urbanization. The last column of the table shows that the estimation results for urban regions are very similar to the results for all regions together in the first column. Thus effectiveness in fighting crime and public disorder does not differ greatly between urban-

effects estimation is subject to the incidental parameters problem (see Wooldridge, 2002). The estimation results are robust to the choice between the linear probability model, Probit and Logit. 10 The most urbanized police force areas include: Amsterdam-Amstelland, Rotterdam-Rijnmond, Haaglanden and Utrecht.

ized and less urbanized regions, which is in line with Kovandzic and Sloan (2002, p. 73). 4.5. Victimization: is it the individual or the municipality? Since we use individual victimization data, we are able to control for background characteristics both at the level of the individual and of the municipality. The estimation results for the control variables allow us to analyze whether victimization of crime and experience of public disorder is dominated by individual characteristics or by factors at the level of the municipality. Individual characteristics can reflect the extent to which people are willing to take risks for instance. But higher victimization rates may also be related to the municipality in which someone lives. In that case, people experience crime and public disorder regardless of their individual background characteristics. To see whether individual or municipality factors dominate, we decompose the explained variance in crime and public disorder rates into two parts: the proportion of the explained variance due to differences in background characteristics of individual respondents and the proportion due to differences in characteristics of municipalities (all estimates are based on the preferred model). Table 8 presents the results of the variance decomposition. Table 8 Variance decomposition for property crime, violent crime and public disorder. Explanatory power of individual effects

Explanatory power of municipality effects

Property crime Burglary Car theft Theft from car Bicycle theft

0.71 0.51 0.49 0.47 0.72

0.29 0.49 0.51 0.53 0.28

Violent crime Threat with violence Assault Robbery with violence

0.77 0.79 0.83 0.23

0.23 0.21 0.17 0.77

Public disorder Littering Graffiti Vandalism Youth nuisance Harassment Public intoxication Drug nuisance

0.29 0.26 0.29 0.42 0.48 0.26 0.57 0.29

0.71 0.74 0.71 0.58 0.52 0.74 0.43 0.71

B. Vollaard, P. Koning / International Review of Law and Economics 29 (2009) 336–348

Factors at the level of the municipality are particularly important when explaining experience of public disorder. This result makes intuitive sense: there is not much an individual can do about experiencing problems like littering, graffiti and drug nuisance. Public disorder is high in municipalities with a young, low educated, immigrant population outside the labor force living in apartment buildings (see tables in the appendix). Public disorder is simply part of living in municipalities with these characteristics. In the case of property crime and particularly violent crime, individual effects dominate, with robbery with violence as the exception. Especially assault, threat with violence and bicycle theft vary across individuals rather than across municipalities. Thus, compared to public disorder, these types of crime are not as likely to be a characteristic of a municipality. Victimization of property and violent crime decreases with age and education level, and is relatively high among females, immigrants, and people living in apartment buildings (Tables 9 and 10). These findings underline the importance of controlling for individual background characteristics, especially in the case of violent crime. 5. Conclusions As a result of a reliance on police recorded crime statistics, the literature on police effectiveness tends to provide evidence on crimes that are relatively well-reported and well-recorded only. By using data from a large-scale victimization survey, we are able to expand the analysis to crimes that are not well-covered in police statistics such as threats with violence, to several types of public disorder and to different forms of victim precaution. When estimating police effectiveness, the survey data allows us to control for both municipality effects and individual characteristics of both victims and non-victims. We model the crime-related variables that are part of the police funding formula to identify the endogenous variation in police levels, and use the remaining variation to estimate the effect of police on crime and public disorder. We find significantly negative effects of higher police levels on property and violent crime, public disorder, and victim precaution. The estimated effects on victimization of crime and also public disorder are similar in size to the effects on recorded crime found in previous studies. The estimated elasticity of police with respect to crime and public disorder varies between −0.2 and −0.5. In addi-

343

tion, we find people to move around more freely in response to greater police levels, with an estimated elasticity of −0.2. Lower levels of victim precaution are an additional benefit of higher police levels not reflected in a decline in victimization rates. Comparing estimates from different model specifications and different estimation techniques, we show the importance of controlling for simultaneity. We find trends in police levels to be correlated with changes in municipality-specific characteristics. This effect would not have been picked up in a fixed effects estimation approach, in which case the effect on crime and disorder would be underestimated. We find experience of public disorder mostly to be a characteristic of the municipality in which someone lives, with little variation across individuals in a municipality, whereas victimization of property crime and particularly violent crime varies across individuals rather than municipalities. These findings underline the importance of controlling for municipality-specific characteristics that are included in the police funding formula, especially in the case of public disorder. Leaving out municipality characteristics from the funding formula leads to overestimation of the effect of police on public disorder, suggesting that initial staffing levels were at a disadvantage to municipalities with severe and persistent public disorder problems. As a final note, finding a negative effect of expanding police levels on crime and disorder does not imply that expanding police levels is socially beneficial—even if the costs of doing so do not exceed the benefits, as suggested by Levitt (1997). The literature on what works in bringing down crime suggests that a one-off investment in improving police effectiveness may well be a more cost effective way of bringing down crime than perpetually higher costs to sustain higher police levels (see Weisburd & Eck, 2004 for an overview). Acknowledgements We thank the Dutch Interior Department and Justice Department for providing data from the Dutch Victimization Survey. We thank Jeannette Verbruggen and Ali Aouragh for providing excellent research assistance and Arie Kapteyn, David Weisburd and Gregory Ridgeway for useful comments on an earlier version.

344

B. Vollaard, P. Koning / International Review of Law and Economics 29 (2009) 336–348

Appendix A

Fig. 2. Deviation from level of police per capita prescribed by police funding formula vs. risk of victimization, by police force area and year (1996–2004). Note: The risk of victimization in a particular year is adjusted for the average risk of victimization in the police force area and the national trend.

B. Vollaard, P. Koning / International Review of Law and Economics 29 (2009) 336–348

345

Fig. 2. (Continued ).

Table 9 Effect of police on victimization of property crime—the modified random effects model. Burglary Ln (police)(t − 1)

−0.008

Individual characteristics Male −0.005*** Age < 20 0.039*** 20 < age < 30 0.028*** 30 < age < 50 0.021*** 50 < age < 60 0.018*** Education level 1 −0.021*** Education level 2 −0.015*** Education level 3 −0.011*** Education level 4 −0.004** Employed −0.001 Student −0.004 House wife −0.005*** Immigrant 0.004** Terraced house −0.009*** Detached house 0.010*** Apartment −0.016*** Municipality characteristics 0.004 Ln (police)(t − 1) Male 0.024***

Car theft

Theft from car

Bicycle theft

(0.008)

0.001

(0.004)

−0.021**

(0.010)

−0.053***

(0.011)

(0.001) (0.004) (0.002) (0.002) (0.001) (0.002) (0.002) (0.002) (0.002) (0.001) (0.003) (0.001) (0.002) (0.002) (0.002) (0.002)

−0.002*** 0.010*** 0.007*** 0.005*** 0.004*** 0.001 0.001 0.001 0.001 0.001 −0.002 0.000 0.006*** −0.003*** −0.002** 0.001

(0.000) (0.002) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.002) (0.001) (0.001) (0.001) (0.001) (0.001)

−0.006*** 0.060*** 0.078*** 0.049*** 0.033*** −0.020*** −0.017*** −0.008*** −0.005** 0.008*** −0.010*** −0.003** 0.017*** −0.009*** −0.005** 0.011***

(0.001) (0.004) (0.002) (0.002) (0.001) (0.002) (0.002) (0.003) (0.002) (0.001) (0.004) (0.001) (0.003) (0.002) (0.002) (0.003)

0.006*** 0.138*** 0.074*** 0.050*** 0.032*** −0.040*** −0.034*** −0.028*** −0.011*** 0.011*** 0.062*** −0.007*** 0.018*** −0.018*** −0.024*** 0.007***

(0.001) (0.006) (0.002) (0.002) (0.002) (0.002) (0.002) (0.003) (0.002) (0.002) (0.005) (0.002) (0.003) (0.002) (0.002) (0.003)

(0.008) (0.009)

−0.002 0.006

(0.003) (0.004)

0.020** 0.014

(0.010) (0.010)

0.014 −0.029**

(0.011) (0.012)

346

B. Vollaard, P. Koning / International Review of Law and Economics 29 (2009) 336–348

Table 9 (Continued ) Burglary Age < 20 20 < age < 30 30 < age < 50 50 < age < 60 Education level 1 Education level 2 Education level 3 Education level 4 Employed Student House wife Immigrant Terraced house Detached house Apartment Housing density Shops Moves Length of roads Number of obs.

Car theft

−0.031 0.002 −0.012 −0.011 0.030*** 0.006 −0.017 0.063*** −0.004*** 0.013** −0.036*** 0.000** 0.100*** 0.074*** 0.119** 0.009*** 0.176* −0.028*** −0.068***

(0.034) (0.015) (0.014) (0.014) (0.011) (0.010) (0.018) (0.013) (0.011) (0.031) (0.013) (0.019) (0.013) (0.014) (0.015) (0.002) (0.074) (0.004) (0.010)

368,638

0.004 0.001 0.015*** 0.001 0.006 −0.002 −0.018** 0.010* −0.005 −0.011 −0.004 0.007 0.013** 0.017*** 0.020*** 0.000 0.073** −0.006*** −0.019***

Theft from car (0.015) (0.006) (0.006) (0.006) (0.005) (0.004) (0.007) (0.006) (0.005) (0.013) (0.005) (0.008) (0.006) (0.006) (0.007) (0.001) (0.032) (0.002) (0.004)

307,115

−0.122*** −0.033** −0.011 −0.014 0.066*** 0.022* 0.027 0.098*** 0.026** 0.091*** 0.014 0.065*** 0.062*** 0.058*** 0.101*** 0.023*** 0.217*** −0.022*** −0.067*** 307,114

Bicycle theft −0.064 0.070*** 0.075*** 0.006 0.031** −0.033** −0.106*** 0.024 −0.024 0.217*** −0.047*** −0.136*** 0.125*** 0.078*** 0.054*** 0.050*** −0.153 0.012* −0.048***

(0.038) (0.017) (0.015) (0.016) (0.013) (0.012) (0.021) (0.015) (0.013) (0.035) (0.014) (0.023) (0.015) (0.016) (0.017) (0.003) (0.085) (0.005) (0.011)

(0.044) (0.020) (0.018) (0.018) (0.015) (0.014) (0.024) (0.018) (0.015) (0.041) (0.017) (0.026) (0.017) (0.019) (0.020) (0.003) (0.100) (0.007) (0.013)

340,501

Notes: Results for year fixed effects, household size and composition are not reported. Standard errors between parentheses. * Statistically significant at the 10-percent level. ** Statistically significant at the 5-percent level. *** Statistically significant at the 1-percent level.

Table 10 Effect of police on victimization of violent crime—the modified random effects model. Threat with violence

Assault

Robbery with violence

Ln (police)(t − 1) Individual characteristics Male Age < 20 20 < age < 30 30 < age < 50 50 < age < 60 Education level 1 Education level 2 Education level 3 Education level 4 Employed Student House wife Immigrant Terraced house Detached house Apartment

−0.019**

(0.008)

−0.003

(0.003)

−0.001

(0.002)

−0.034*** 0.105*** 0.058*** 0.037*** 0.022*** −0.019*** −0.011*** −0.006*** 0.001 0.001 −0.006* 0.000 −0.014*** −0.009*** −0.011*** 0.004**

(0.001) (0.004) (0.002) (0.001) (0.001) (0.002) (0.002) (0.002) (0.002) (0.001) (0.004) (0.001) (0.002) (0.001) (0.002) (0.002)

−0.003*** 0.032*** 0.014*** 0.008*** 0.004*** 0.000 0.001 0.001 0.001** −0.001** −0.001 0.000 0.002*** −0.003*** −0.002*** 0.000

(0.000) (0.002) (0.001) (0.001) (0.000) (0.001) (0.001) (0.001) (0.001) (0.001) (0.002) (0.000) (0.001) (0.001) (0.001) (0.001)

0.000 0.005*** 0.001** 0.000 0.001 −0.001*** −0.001** −0.001 0.000 0.000 −0.001 0.001* 0.001* −0.002*** −0.002*** −0.001**

(0.000) (0.001) (0.000) (0.000) (0.000) (0.000) (0.000) (0.001) (0.000) (0.000) (0.001) (0.000) (0.001) (0.000) (0.001) (0.001)

Municipality characteristics Ln (police)(t − 1) Male Age < 20 20 < age < 30 30 < age < 50 50 < age < 60 Education level 1 Education level 2 Education level 3 Education level 4 Employed Student House wife Immigrant Terraced house Detached house Apartment Housing density Shops Moves Length of roadways

0.012 0.000 −0.027 0.026** 0.017 0.023* 0.012 0.010 0.019 0.035*** −0.024** 0.009 −0.036*** 0.074** 0.015 0.007 0.009 0.014*** 0.195*** −0.011*** −0.029***

(0.008) (0.009) (0.028) (0.013) (0.011) (0.012) (0.010) (0.009) (0.017) (0.012) (0.010) (0.028) (0.011) (0.018) (0.012) (0.013) (0.014) (0.002) (0.068) (0.004) (0.009)

0.001 −0.001 0.000 0.002 −0.007 0.002 0.000 0.000 0.013* 0.006 −0.001 −0.012 −0.003 0.010 0.009* 0.009 0.003 0.005*** −0.009 −0.001 −0.003

(0.003) (0.004) (0.012) (0.006) (0.005) (0.005) (0.004) (0.004) (0.007) (0.005) (0.004) (0.012) (0.005) (0.008) (0.005) (0.005) (0.006) (0.001) (0.029) (0.002) (0.004)

0.002 −0.001 0.003 0.008*** 0.007*** 0.008** 0.006*** 0.001 0.002 0.008*** −0.003 −0.003 −0.004 0.019*** 0.007*** 0.006** 0.011*** 0.001** 0.037** −0.002** 0.004

(0.002) (0.002) (0.007) (0.003) (0.003) (0.003) (0.002) (0.002) (0.004) (0.003) (0.002) (0.006) (0.003) (0.005) (0.003) (0.003) (0.003) (0.001) (0.018) (0.001) (0.002)

Number of observations

368,666

368,696

368,700

Notes: Results for year fixed effects are not reported. Standard errors are between parentheses. * Statistically significant at the 10-percent level. ** Statistically significant at the 5-percent level. *** Statistically significant at the 1-percent level.

B. Vollaard, P. Koning / International Review of Law and Economics 29 (2009) 336–348

347

Table 11 Effect of police on experience of public disorder (1)—the modified random effects model. Littering Ln (police)(t − 1) Individual characteristics Male Age < 20 20 < age < 30 30 < age < 50 50 < age < 60 Education level 1 Education level 2 Education level 3 Education level 4 Employed Student House wife Immigrant Terraced house Detached house Apartment Municipality characteristics Ln (police)(t − 1) Male Age < 20 20 < age < 30 30 < age < 50 50 < age < 60 Education level 1 Education level 2 Education level 3 Education level 4 Employed Student House wife Immigrant Terraced house Detached house Apartment Housing density Shops Moves Length of roadways Number of observations

Graffiti

Youth nuisance

−0.046***

(0.016)

−0.009

(0.013)

−0.051***

(0.012)

0.032*** 0.054*** −0.003 0.014*** 0.034*** −0.002 0.015*** 0.008** 0.019*** −0.027*** 0.014*** 0.000 −0.053*** −0.026*** −0.047*** 0.097***

(0.002) (0.006) (0.003) (0.003) (0.003) (0.003) (0.003) (0.004) (0.003) (0.002) (0.005) (0.002) (0.004) (0.003) (0.004) (0.004)

0.007*** 0.063*** −0.003 0.002 0.017*** −0.015 0.002 0.005* 0.010*** −0.006*** 0.009* 0.004*** −0.035*** −0.020*** −0.041*** 0.059***

(0.001) (0.005) (0.003) (0.002) (0.002) (0.002) (0.002) (0.003) (0.002) (0.002) (0.005) (0.002) (0.003) (0.002) (0.003) (0.003)

0.009*** 0.085*** 0.049*** 0.038*** 0.033*** 0.028*** 0.028*** 0.020*** 0.018*** −0.015*** −0.001 −0.007*** 0.012*** −0.015*** −0.033*** 0.043***

(0.001) (0.005) (0.002) (0.002) (0.002) (0.002) (0.002) (0.003) (0.002) (0.002) (0.004) (0.002) (0.003) (0.002) (0.003) (0.003)

−0.042*** −0.004 −0.016 0.226*** 0.160*** 0.032 0.181*** 0.113*** 0.094*** 0.070*** −0.017 0.134*** −0.084*** 0.458*** 0.014 −0.148*** 0.063** 0.074*** 0.129 −0.079*** −0.170***

(0.016) (0.018) (0.060) (0.028) (0.023) (0.027) (0.020) (0.019) (0.035) (0.025) (0.022) (0.058) (0.023) (0.038) (0.026) (0.028) (0.029) (0.004) (0.144) (0.009) (0.021)

−0.008 −0.013 −0.031 0.230*** 0.175*** 0.089*** 0.057*** 0.075*** 0.027 0.065*** −0.052*** 0.140*** −0.098*** 0.117*** 0.081*** −0.023 0.182*** 0.024*** 0.351*** −0.041*** −0.150***

(0.012) (0.013) (0.045) (0.020) (0.017) (0.019) (0.015) (0.015) (0.026) (0.019) (0.016) (0.044) (0.017) (0.028) (0.018) (0.019) (0.021) (0.107) (0.003) (0.005) (0.150)

0.012 0.008 0.014 0.054*** 0.121*** 0.084*** 0.175*** 0.102*** 0.038 0.113*** −0.018 −0.030 −0.077*** 0.085*** 0.125*** 0.076*** 0.143*** 0.015*** 0.570*** 0.028*** −0.177***

(0.012) (0.014) (0.045) (0.021) (0.017) (0.020) (0.015) (0.014) (0.025) (0.018) (0.016) (0.043) (0.017) (0.028) (0.019) (0.020) (0.021) (0.003) (0.107) (0.009) (0.015)

368,439

367,329

367,753

Notes: Results for year fixed effects are not reported. Standard errors are between parentheses. * Statistically significant at the 10-percent level. ** Statistically significant at the 5-percent level. *** Statistically significant at the 1-percent level.

Table 12 Effect of police on public disorder (2)—the modified random effects model. Public intoxication

Vandalism

Ln (police)(t − 1) Individual characteristics Male Age < 20 20 < age < 30 30 < age < 50 50 < age < 60 Education level 1 Education level 2 Education level 3 Education level 4 Employed Student House wife Immigrant Terraced house Detached house Apartment

−0.016**

Harassment (0.007)

−0.025***

Drug nuisance (0.009)

−0.016*

(0.010)

−0.021

(0.015)

0.005*** 0.018*** 0.014*** 0.007*** 0.006*** 0.012*** 0.008*** 0.004*** 0.006*** −0.005*** 0.003 −0.003*** −0.001 −0.015*** −0.015*** 0.011***

(0.001) (0.003) (0.001) (0.001) (0.001) (0.001) (0.001) (0.002) (0.001) (0.001) (0.003) (0.001) (0.002) (0.001) (0.001) (0.002)

0.000 0.055*** 0.040*** 0.041*** 0.041*** 0.020*** 0.011*** 0.001 0.005*** −0.012*** −0.009*** −0.005*** 0.018*** −0.023*** −0.029*** 0.030***

(0.001) (0.004) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.001) (0.003) (0.001) (0.003) (0.002) (0.002) (0.002)

0.000 0.108*** 0.095*** 0.052*** 0.037*** −0.019*** −0.024*** −0.024*** −0.014*** −0.003** 0.005 −0.005*** 0.000 −0.041*** −0.040*** 0.022***

(0.001) (0.004) (0.002) (0.001) (0.001) (0.002) (0.002) (0.002) (0.002) (0.001) (0.004) (0.001) (0.002) (0.002) (0.002) (0.002)

0.011*** 0.070*** 0.011*** 0.023*** 0.046*** 0.056*** 0.057*** 0.038*** 0.036*** −0.006*** 0.004 −0.004*** −0.014*** 0.029*** −0.006** 0.041***

(0.001) (0.005) (0.003) (0.002) (0.002) (0.002) (0.003) (0.002) (0.002) (0.005) (0.002) (0.003) (0.003) (0.003) (0.003)

Municipality characteristics Ln (police)(t − 1) Male Age < 20 20 < age < 30

0.001 −0.011* −0.075*** 0.042***

(0.006) (0.006) (0.020) (0.010)

−0.023*** −0.046*** −0.144*** 0.236***

(0.009) (0.009) (0.032) (0.015)

−0.025*** −0.014 −0.113*** 0.206***

(0.009) (0.011) (0.037) (0.017)

−0.014 0.020 0.226*** 0.141***

(0.015) (0.017) (0.055) (0.026)

348

B. Vollaard, P. Koning / International Review of Law and Economics 29 (2009) 336–348

Table 12 (Continued ) Harassment 30 < age < 50 50 < age < 60 Education level 1 Education level 2 Education level 3 Education level 4 Employed Student House wife Immigrant Terraced house Detached house Apartment Housing density Shops Moves Length of roads Number of obs.

0.013* 0.030*** 0.064*** 0.035*** −0.015 0.063*** −0.018*** 0.068*** −0.002 0.063*** 0.057*** 0.047*** 0.078*** 0.021*** 0.135*** −0.022*** −0.016*** 357,943

Drug nuisance (0.008) (0.009) (0.007) (0.007) (0.012) (0.009) (0.007) (0.021) (0.008) (0.014) (0.008) (0.009) (0.010) (0.002) (0.049) (0.002) (0.006)

0.121*** 0.092*** 0.207*** 0.098*** −0.050*** 0.154*** −0.161*** 0.211*** −0.069*** 0.292*** 0.139*** 0.139*** 0.109*** 0.053*** 0.877*** −0.051*** −0.019* 362,362

Public intoxication (0.012) (0.014) (0.010) (0.010) (0.018) (0.013) (0.011) (0.031) (0.012) (0.022) (0.014) (0.014) (0.015) (0.003) (0.074) (0.004) (0.010)

0.043*** 0.030** 0.096*** 0.029*** −0.035* 0.050*** −0.048*** 0.148*** −0.004 −0.139*** 0.054*** 0.082*** −0.014 0.058*** 1.741*** 0.093*** −0.026** 364,609

Vandalism (0.014) (0.015) (0.012) (0.011) (0.020) (0.015) (0.013) (0.035) (0.014) (0.022) (0.015) (0.017) (0.017) (0.003) (0.094) (0.010) (0.012)

0.274*** 0.276*** 0.101*** 0.071*** 0.007 −0.023 −0.147*** −0.055 −0.159*** 0.392*** 0.158*** 0.027 0.287*** −0.059*** −0.394*** −0.059*** −0.452***

(0.021) (0.025) (0.018) (0.017) (0.032) (0.022) (0.020) (0.053) (0.022) (0.034) (0.023) (0.025) (0.026) (0.004) (0.127) (0.006) (0.018)

359,522

Notes: Results for year fixed effects are not reported. Standard errors are between parentheses. * Statistically significant at the 10-percent level. ** Statistically significant at the 5-percent level. *** Statistically significant at the 1-percent level.

References Corman, Hope, & Naci Mocan, H. (2000). A time-series analysis of crime, deterrence, and drug abuse in New York City. American Economic Review, 90(3), 584–604. Corman, Hope, & Naci Mocan, H. (2005). Carrots, sticks, and broken windows. Journal of Law and Economics, 48(1), 235–266. Di Tella, Rafael, & Schargrodsky, Ernesto. (2004). Do police reduce crime? Estimates using the allocation of police forces after a terrorist attack. American Economic Review, 94(1), 115–133. Klick, Jonathan, & Tabarrok, Alexander. (2005). Using terror alert levels to estimate the effect of police on crime. Journal of Law and Economics, 48(1), 267–279. Kovandzic, Tomislav V., & Sloan, John J. (2002). Police levels and crime rates revisited. A county level analysis from Florida (1980–1998). Journal of Criminal Justice, 30(1), 65–76. Levitt, Steven D. (1997). Using electoral cycles in police hiring to estimate the effect of police on crime. American Economic Review, 87(3), 270–290. Levitt, Steven D. (2002). Using electoral cycles in police hiring to estimate the effect of police on crime: Reply. American Economic Review, 92(4), 1244–1250. Lin, Ming-Jen (2009). More police, less crime: Evidence from state data. International Review of Law and Economics, 29(2), 73–80. MacDonald, Ziggy. (2002). Official crime statistics: Their use and interpretation. The Economic Journal, 112(477), F85–F106.

Manski, C. F. (1993). Identification of social endogenous effects: The reflection problem. Review of Economic Studies, 60(3), 531–542. Marvell, Thomas B., & Moody, Carlisle E. (1996). Specification problems, police levels, and crime rates. Criminology, 34(4), 609–646. Mundlak, Yair. (1978). On the pooling of time series and cross section data. Econometrica, 46(1), 69–85. Philipson, Tomas J., & Posner, Richard A. (1996). The economic epidemiology of crime. Journal of Law and Economics, 39(2), 405–433. PMB (Politiemonitor Bevolking). (2005). Landelijke rapportage Politiemonitor Bevolking 2005. Hilversum: Intomart/GfK. Rand, Michael R., & Rennison, Callie Marie. (2002). True crime stories? Accounting for differences in our national crime indicators. Chance, 15(1), 47–51. Tauchen, Helen, Dryden Witte, Ann, & Griesinger, Harriet. (1994). Criminal deterrence: Revisiting the issue with a birth cohort. Review of Economics and Statistics, 76(3), 399–412. Weisburd, David, & Eck, John E. (2004). What can police do to reduce crime, disorder and fear? The Annals of the American Academy of Political and Social Science, 59(3), 42–65. Wittebrood, Karin, & Junger, Marianne. (2002). Trends in violent crime: A comparison between police statistics and victimization surveys. Social Indicators Research, 59(2), 153–173. Wooldridge, J. M. (2002). Econometric analysis of cross section and panel data. Cambridge: The MIT Press.