Research Proposal

Effectiveness of a prolonged incarceration and rehabilitation measure for high-frequency offenders N. Tollenaar &A. M. van der Laan & P. G. M. van der Heijden Published online: 8 May 2013 # Springer Science+Business Media Dordrecht 2013 Abstract ObjectivesTo estimate the incapacitation effect and the impact on post-release recidivism of a measure combining prolonged incarceration and rehabilitation, the ISD measure for high frequency offenders (HFOs) was compared to the standard practice of short-term imprisonment.

MethodsWe applied a quasi-experimental design with observational data to study the effects of ISD. The intervention group consisted of all HFOs released from ISD in the period 2004–2008. Two control groups were derived from the remaining population of HFOs who were released from a standard prison term. To form groups of controls, a combination of multiple imputation (MI) and propensity score matching (PSM) was used including a large number of covariates. In order to measure the incapacitation effect of ISD, the number of convictions and recorded offences in a criminal case of the controls were counted in the same period as their ISD counterfactuals were incarcerated. The impact on recidivism was measured by the prevalence and the frequency of reconvictions corrected for time at risk. Robustness of the results were checked by performing a combined PSM and difference-in-difference (DD) design.

ResultsThe estimate of the incapacitation effect was on average 5.7 criminal cases and 9.2 offences per ISD measure. On average 2.5 convictions and 4 recorded offences per year per HFO are prevented. The HFOs released from ISD showed 12 to 16 % lower recidivism rates than their control HFOs released from prison (Cohen’s J Exp Criminol (2014) 10:29–58 DOI 10.1007/s11292-013-9179-y N. Tollenaar (*) :A. M. van der Laan Research and Documentation Centre (WODC), Turfmarkt 131, 2511 DP Den Haag, Netherlands e-mail: [email protected] A. M. van der Laan e-mail: [email protected] P. G. M. van der Heijden Utrecht University, Utrecht, Netherlands e-mail: [email protected] P. G. M. van der Heijden University of Southampton, Southampton, UK h=0.3–0.4). The recidivists of the ISD group also showed a lower reconviction frequency than the control group recidivists (Cohen’sd= 0.2).

ConclusionsThe ISD measure seems to be effective in reducing recidivism and crime. The estimated incapacitation effect showed that a large portion of criminal cases and offences was prevented. DD analysis and sensitivity analyses confirmed the robustness of the PSM results. Due to the absence of actual treatment data, the effects found cannot be attributed separately to resocialization, imprisonment, or improve- ment of life circumstances.

KeywordsDouble difference.

Frequent offenders.

Incapacitation.

Recidivism.

Propensity score matching Introduction Internationally, there has long been interest in the most frequent and persistent offenders. In the Netherlands at the beginning of the twenty–first century, the Department of Justice developed a focus on the most frequent offenders. It was mainly fed by the idea that reducing reoffending specifically of frequent offenders would provide a significant reduction of the overall crime level, a finding long common in criminology (Wolfgang et al.1972). Judicial policy focused on the most active offenders in the group of frequentoffenders, namely the so-called high- frequency offenders (HFOs). These were simply defined as those offenders who were arrested by the police for committing a crime 11 or more times in the previous 5 years. This group made up 30 % of the yearly population of suspects, starting as teenages, had a mean age of 30 years, and on average had 40 previous convictions.

It is plausible that a large overlap exists between the Dutch HFOs and high-rate chronic offenders found in the international life-course research. In various criminal career studies that used latent trajectory modelling (Nagin and Land1993), 5 % of a cohort of offenders was found to be of some type of high-rate chronic offender. This type could clearly be distinguished from the remaining offender groups by a prolongued period of frequent offending in their life-course (Blokland et al.2005; Piquero and Blumstein2007; Piquero et al.2010; Bersani et al.2009). They are designated as‘high- level chronics’(Nagin1999),‘high-level persisters’(Blokland et al.2005), high-rate chronics (Sampson and Laub2003), or high-rate offenders (Piquero et al.2010). Their corresponding conviction frequencies (Lambdas) were found to be between 1.5 and 3 convictions per year (Blokland et al.2005;Nagin1999; Piquero and Blumstein2007).

These rates, however, vary with age. Nagin (1999) found the average conviction rate amongst the chronic group to decline from age 18 to age 32 from 1.5 to .34, whereas Sampson and Laub (2003) mentioned that their high-rate chronic group showed a peak in their late 30s/early 40s and declined thereafter.

Piquero et al. (2010) also discovered a group of short-term high-frequency (STHF) offenders in their data. It was characterized by a high peak at adolescence of 1.5 convictions that quickly declined to a low rate of .2 convictions in the mid-20s and going to zero at the beginning of their 30s. This STHF offender group was found to be indistinguishable from the long-term low-rate offenders in their high early child- hood risk profile, illustrating the difficulty of early identification.

30N. Tollenaar et al. Recent Dutch life-course studies also mention high-rate chronic offenders (Blokland et al.2005; Bersani et al.2009). Bersani and collegues, for example, found that 4.2 % of their offender cohort consisted of chronic offenders. These offenders showed an average conviction frequency of 1.8 per year through their 20s and 30s, followed by a decline in their late 30s. Dutch HFOs also have a high risk profile. They were found to have serious problems in multiple areas of functioning (Jacobs and Essers2003). These offenders were typically characterized by addiction to drugsand/or alcohol, being jobless, and being low educated. More than half of the group had housing and financial problems and a third suffered from psychological or psychiatric problems (Tollenaar et al.2007). Others found that HFOs were arrested on a very regular basis on (minor) acquisitive crimes and nuisance crimes (see, e.g., Versteegh et al.2003). The criminal justice system disposed of these crimes mainly by a short prison term, ranging from a few days to 3 months. Upon release, HFOs tended to commit crimes as frequently as before their prison sentence.

To remediate this frequent cycle of crime and persecution, the Dutch Ministry of Justice regulated a severe sanction, the ISD measure (Inrichting voor Stelselmatige Daders,or Institution for habitual offenders), by law in late 2004 1[Ministerie van Justitie (Ministry of Justice)2003]. By this law, a HFO can be sentenced to a maximum sentence of 2 years even for a relatively minor offence. The conditions are that, in the preceding 5 years, the offender must have had three or more convictions sanctioned by a community sentence order, a prison term, or a measure of restraint. In practice, only suspects with 11 or more police contacts in 5 years were eligible for placement in the ISD, i.e. they had to conform to the HFO definition. Next, for those offenders who are motivated, treatment and rehabilitation programs are available. Interventions for addiction or other behavioral problems are offered which should lead to committing fewer crimes after the detention.

The main goals of the ISD are reducing crime byincapacitation and reducing post-release recidivism by re-socialization.

There are three aspects that distinguish the ISD measure from a standard sanction for HFOs. First, there is certainty of sanctioning. The public prosecutors’office is provided by the police with a list of all persons having 11 or more police contacts in the last 5 years. An offender who is on this list is told that, after being caught for an offence, he or she is likely to be sentenced with an ISD measure for the next offence. Second, there is a prolonged incarceration length. The incarceration length is much longer (up to 2 years) than the sanction for comparable types of crime (several weeks). And, third, as a consequence, there is more time and opportunity to actually rehabilitate the offender. Moreover, treatment programs should be offered to motivated offenders. So, apart from a longer incarceration period, the ISD measure also contains quasi-compulsory treatment for addicted HFOs. These three aspects could have a series of different effects with regard to crime reduction. The first one includes a general deterrent effect, the second a special deterrent effect and incapacitation effects, whereas the third one could lead to improve- ment of life circumstances and abate addiction problems and result into re-socialization.

In the Netherlands, only one study has been conducted on the effectiveness of the ISD measure and its predecessor the SOV on crime (Vollaard2012). In this study, the 1The problem of addicted offenders in the Netherlands was countered earlier with the SOV measure (Strafrechtelijke opvang verslaafdenor Rehabilitation of Drug-Addicted Offenders Act). This predecessor of the ISD had stricter inclusion criteria and was only applied on a small scale. Two hundred slots were created for addicted male habitual offenders without serious psychiatric disorders. The SOV was merged into the ISD at the end of 2004.

Effectiveness of a prolonged incarceration 31 effects of both the SOV and the ISD were lumped together. In a natural experiment, Vollaard showed a decrease of police-reported specific crime types, i.e. burglary and breaking into a car, after introduction of the measures. He used the fact that both measures were not implemented in all regions simultaneously. The study, however, was on the aggregate level. Therefore, general prevention effects, incapacitation effects, and recidivism reduction could not be differentiated. Moreover, the effect onindividualre-offending of all crime types of HFOs remains unknown.

The present study aims to estimate the relative effect of the ISD measure on the individual level compared to the‘treatment as usual’, being a short prison term. The focus is on both the incapacitation effect of the measure and the recidivism after release. In the following sections, we will therefore describe the findings from recent empirical studies on incapacitation effects and specific deterrent effects of (prolonged) incarceration.

Because a large part of the HFO group is addicted to heroin and/or cocaine, we will also discuss recent research on quasi-compulsory treatment of drug offenders and drug courts.

Incapacitation effects of prolonged incarceration of HFOs Most existing research concerning the effect of incarceration focus on the recidivism afterwards and do not differentiate incapacitation effects. Those that do provide an estimate of the latter are, however, very dependent on the estimators used, the investi- gated population, the type of crime, and the source used (Spelman2000; Piquero and Blumstein2007). Two approaches are possible for estimating an incarceration effect: a top–down (macro) or a bottom–up (micro) approach (Spelman2000). In most bottom– up approaches, the estimate is either based on the number of crimes committed in the period previous to incarceration or on a sample from the general offender population that had not been incarcerated. In the first case, it is unclear whether the assumption that the annual offending frequency would have been the same if the offender would not have been incarcerated. In both cases, the estimates suffer from bias due to stochastic selectivity, because those incarcerated are likely to be more frequent offenders than those not incarcerated (see, e.g., Piquero and Blumstein2007).

Both potential issues can be prevented by providing a counterfactual. The counterfactual approach was followed by Apel and Sweeten (2010): they applied a quasi-experimental design using PSM (propensity score matching), thus providing a between-person counter- factual. They estimated the incapacitation effect to be 6.1–14.1 self-reported offences per year prison in youth (13–18 years old) and 4.9–8.4 offences per year in the age range of 18– 24. Similarly, Owens (2009) estimated the incapacitation effect on incarceration 23- to 25- year-olds in a natural experiment by using difference-in-difference analysis. She found that the offenders under a more lenient regime, a sentence guideline resulting in an average 1 year less sentence length, had on average a yearly 2.8 arrests that would have been prevented under the stringent regime. As a matter of course, these incapacitation effects in the general population of juveniles and young adults will plainly not generalize to the (high-)frequency adult offenders.

Wermink et al. (2012) extended the age range to 12–50 years. Apart from the between-person counterfactual, they also provided within-person counterfactuals on conviction data of first-time imprisonment. For this low risk group, the incapacitation effect was estimated to be between .17 and .21 convictions per year. These estimate are somewhat downwardly biased because they included suspended prison sentences 32N. Tollenaar et al. in the control group, while they did not have actual prison stay data. An unknown portion of these is also incapacitated because of breaking conditions.

The results of these studies may not generalize to the high-frequency offenders population for a number of reasons. First, HFOs are in a later stage of their criminal career than the offenders in the studies discussed above. The latter may not have been not yet be on their peak lambda, and thus give a false impression of incapacitation effects later in their career. Secondly, the offense frequency distribution of imprisoned offenders is highly skewed (see also Piquero and Blumstein2007). This implies that estimates of the incapacitation effect in specific populations might not generalize to other offender populations. This holds especially in our case, where the offenders are actuallyselectedon their actual offending frequency in the previous 5 years. In other words, they have already retrospectively proven to be a high frequency offender.

Specific deterrent effects of incarceration The bottom–up studies on specific deterrent effect of incarceration can be classified in in two types: (1) the relative effect of an incarceration with respect to other sanctions, like a community sentence, and (2) the‘dose-response’effect of incarceration, i.e.

longer prison sentence versus a shorter prison sentence. Empirical results of both lines of research are presented below.

The relative effect of incarceration versus other sanctions and specific deterrence The literature on the effect of incarceration on post-release recidivism is incon- clusive (Nagin et al.2009;Spelman2000) and mainly considers general offender populations. In their extensive overview of the literature, Nagin et al. (2009) concluded that most studies with regard to the effects of incarceration were non- experimental and that (quasi-)experimental studies hardly exist. So, drawing clear conclusions about the effectiveness of incarceration was difficult. More recently, however, a few (quasi-)experimental studies on the effect of incarceration have been conducted. Wermink et al. (2010) compared community service (CS) to imprisonment using PSM. They found CS to have lower recidivism rates than imprisonment. This study was, however, limited to the first-time imprisonment of offenders aged 18–50. Therefore, the results do not apply to HFOs as they have typically already had a long sequence of imprisonments. Bales and Piquero (2012) also used a quasi-experimental design to study the effects of prison as opposed to community sanction on recidivism rates, within 3 years after release.

Their study compared three methods: exact matching, PSM, and regression-based models. They matched on and controlled for demographic and criminal career covariates. Prison sanctions were found to have a criminogenic effect compared to CS.

Both the Wermink et al. and the Bales et al. studies suffer from the fact that the propensity scores of the experimental group at the extreme ends were left out, due to insufficient common support. This leads to potential problems with generalizability because the resulting subset is likely not representative of the total group. As Cook et al. (2008) noted, the experimental group should be as intact as possible to lead to results comparable to a randomized controlled trial.

Effectiveness of a prolonged incarceration 33 Recently, Cid (2009) estimated the effect of imprisonment versus suspended sentence in Spain via regression correction. He found that suspended sentences lead to less re- incarceration than prison sentence. He compared offenders on a binary outcome with varying observation times using logistic regression. In order to estimate the effect of imprisonment over time, it would have been more appropriate to use survival analysis.

Most recently, both Loeffler (2013) and Nagin and Snodgrass (2013) estimated the effect of incarceration on re-arrest rates in natural experiments. In their research, they exploited the variation in imprisonment imposition rates among different judges within a specific county and used judge as an instrumental variable. Criminal cases were allocated randomly to judges in both studies. Neither of these studies found a statistically significant effect on re-arrest.

Taking the recent literature together, the evidence for a general effect for all offenders is contradictory. There might, however, be a differential effect for different types of offenders. In their review of differential deterrence research, Piquero et al.

(2011) concluded that there seems to be heterogeneity in the response to deterrents threats with regard to: (1) social bonding, (2) morality, (3) individual characteristics like impulsivity, discount rate and self-control, (4) emotional/pharmacological arous- al, (5) position in a social network, and (6) decision-making competence. Individual offenders differ in sanction threat perception, response to sanction threat, and re- sponse to punishment.

The applicability of the results in the aforementioned studies may be of limited value to the group of this study, as HFOs have already been proven not to be deterred by prison by their large criminal history. However, because they have mostly under- gone short prison terms, a prolonged prison term might have a distinguishable effect.

The dose response effect of incarceration length on recidivism Research on the effects of incarceration length on follow-up crime is scarce. Gendreau et al. (1999) found in their meta-analyses that longer prison sentences were correlated with higher post-release recidivism. The longer the follow-up time, the larger the differences in recidivism levels, so the effect seems to increase with time. A systematic review by Spelman (2000) was inconclusive about these effects. In their recent review of the literature on the effect of imprisonment on reoffending, Nagin et al. (2009)found generally inconclusive evidence with regard to the effect of prison term length on recidivism. Only one of two mentioned experimental studies showed significant effects.

The 17 non-experimental studies they included showed a mix of criminogenic effects, preventive effects, and non-significant effects of incarceration length. Some of these individual studies even found different results for different subgroups. A recently pub- lished study by Snodgrass et al. (2011) found no effect of sentence length on post-release recidivism. On the other hand, Meade et al. (2012) found a curvilinear relationship of time served with re-arrest: up to 2 years, the odds of re-arrest increased, whereas additional incarceration time decreased the odds.

Of specific relevance is a finding by DeJong (1997), who found that experienced offenders with few ties with society experienced a longer time to re-arrest given longer sentence length. This suggest a specific deterrent effect of longer sentences for this subgroup. This specific description, experienced with few ties to society, seems quite applicable to HFOs. 34N. Tollenaar et al. Quasi-compulsory treatment of drug-addicted offenders Substance abuse can be a causal factor in persistent offending, and might even prolong criminal careers that would otherwise have ended (Sampson and Laub 2003). Recently, a few international reviews have appeared concerning the effects of quasi-compulsory treatment ofaddicted offenders. In an international review study of the effects of quasi-compulsory treatment of addicts on addiction and crime, Stevens et al. (2005) found that positive effects can be expected, but that the designs of the studies do not yet allow conclusions on the effectiveness (see also Schaub et al.2010). Stevens and colleagues arrived at the tentative conclusion that quasi-compulsory treatment reduces addiction and crime as at least as well as voluntary treatment (Stevens et al.2005; Schaub et al.2010). In a systematic review, Mitchell et al. (2006) studied evaluations of the effects of quasi-compulsory treatment interventions for addicted detainees (incarceration- based drug treatment programs). They included only studies that compared an experimental with a control group. Quasi-compulsory interventions appear to be effective and to lead to lower recidivism rates among the treatment groups compared to control groups. The greatest reductions of recidivism and prevention of relapse in addiction were found in interventions that had a therapeutic environment (see also Lipsey2009). In their meta-analysis, Parhar et al. (2008) found that voluntary treatment had apositive effect on general recidivism regardless of whether this was in a custodial setting, whereas mandatory treat- ment did not have an effect.

In many countries—including the Netherlands—offenders addicted to heroin can participate in an opioid substitution treatment (OST) in prison. Larney et al. (2012) found that, as long as participants remained in OST after release from prison, the average risk of re-incarceration proved to be 20 % less for the duration of their OST.

In the United States, drug-addicted offenders may be sent to drug courts instead of the traditional justice system.Drug courts supply intensive treatment, supervision, testing for drug use, frequent court appearance for progress assess- ment, and rewards and punishments for meeting or not meeting obligations.

Rempel et al. (2012) found that drug courts, consisting of community-based treatment and intensive judicial oversight, significantly reduced self-reported crime.

In their study, Warner and Kramer (2009) compared offenders who completed a community-based treatment program for drug-dependent offenders (RIP/D&A, i.e.

Restrictive Intermediate Punishment/Drugs and Alcohol treatment) to a random comparable group traditionally sentenced before implementation of this program.

After 3 years, the treatment group had a risk of re-arrest that was 33 % lower than offenders undergoing probation, state incarceration, or county jail. However, of- fenders that dropped out of their treatment had an increased risk of re-arrest.

In the Netherlands, Koeter and Bakker (2007) conducted an effect study of the SOV, the predecessor of the ISD. The SOV was compared to three control groups: a regular detention and two quasi-compulsory treatments for addicted offenders. They found that, after a regression correction for initial differences in age, conviction history, addiction, psychosocial problems, and follow-up time, the SOV group had significantly lower self-reported and police-reported recidivism frequencies.

Effectiveness of a prolonged incarceration 35 Research questions To summarize, the incapacitation effect varies substantially across previous research.

Previous studies on the deterrent effect of incarceration found inconclusive results; however, some recent quasi-experimental studies show mainly criminogenic effects.

Little is known with regard to the effects of incarceration length on recidivism, but there seems to be ample evidence that a prolonged incarceration length for those with weak bonds to society seems to have a reducing effect onrecidivism. Finally, quasi-compulsory treatment for drug addicts seems to have a reducing effect on recidivism and relapse rates.

In this study, we will investigate the incapacitation effect and the effect on recidivism of the ISD measure on the individual level. Firstly, we estimate the incapacitation effect of the ISD-measure with respect to sanctioning as usual, being mainly short prison terms. Second, we aimed to evaluate the effects of the ISD measure on reconviction after release from the institution, relative to the treatment as usual. Our research questions were therefore:

1. What is the incapacitation effect of the ISD measure executed between 2004 and 2008 relative to standard short-term imprisonments for HFOs?

2. Is the ISD measure executed between 2004 and 2008 effective in terms of reducing recidivism of HFOs compared to standard prison sanctions?

Methods This study used a quasi-experimental research design with observational data. The recidivism ofallHFOs released from ISD between 2004 and 2008 was compared to the recidivism in two control groups of very active offenders released from a standard prison sanction. Additionally, we estimated the size of the incapacitation effect of the measure compared to a standard prison sanction. In order minimize selection bias due to observational data, we applied propensity score matching (PSM) on a large set of available covariates that were related to the outcomes.

Outcome measures In order to estimate the incapacitation effect and the post-release recidivism, we used conviction data from the criminal justice system 2. We also counted the number of recorded offences. A conviction was defined as a valid disposal by the court or the public prosecutor. We included motoring offences like leaving the scene of accident and driving under influence of alcohol or drugs. Cantonal court cases (i.e. mis- demeanors) were not counted. Furthermore, acquittals, technical judgments, and technical dismissals were removed. The offences as measured in conviction data are an underestimate of the actual crime (Farrington2013; Farrington et al.2007; Koeter2004). However, the estimates can serve as a lower bound of the actual number of offences among HFOs. 2We also used arrest data of the HFOs. The resulting analyses yielded approximately the same effect sizes as those reported in this article and are available upon request by the first author.

36N. Tollenaar et al. Estimation of the incapacitation effect The incapacitation effect was defined as the number of convictions and recorded offences in convictions that have been prevented by incarcerating HFOs with an ISD measure. This effect was estimated by establishing the reconvictions of the counter- factuals who were released in the exact same period as were the ISD-subjects. For this purpose, we only used thesimultaneouscontrols (see later), because these observa- tions lay closest in time to the ISD observations. This between-subjects strategy was also followed by Sweeten and Apel (2007) and Wermink et al. (2012).

Recidivism Reconvictions were used as an indicator of recidivism. We evaluated the recidivism both in terms of prevalence and frequency per year free. Recidivism prevalence is the probability of a first conviction within a certain follow-up time after release from penitentiary. The duration is defined as the amount of time between the date of release and the minimum date of committing an offense of the offenses in the criminal case.

Recidivism frequency is the number of reconvictions for any crime type. The recidivism frequency was corrected for duration of follow-up detentions and defined as the number of convictions per year free.

Analytic strategy To adjust for a priori differences between the ISD group and both control groups, a propensity score matching (PSM; Rosenbaum and Rubin1983) was performed using 20 covariates for each control group. We used covariates that were related to crime of frequent offenders, including demographics, socio-economic variables, criminal ca- reer data, and (problems in) functioning in different areas of life.

In order to estimate the incapacitation effect of ISD, we used descriptive analyses to count the conviction frequencies of the counterfactuals. In order to estimate the recidivism prevalence rates, we conducted survival analysis. The recidivism preva- lence was calculated using product-limit estimation (Kaplan and Meier1958).

Differences in survival were tested with log-rate, Wilcoxon, and Tarone–Ware tests.

Differences in conviction frequencies were tested withttests. We used Cohen’s h and Cohen’sd(Cohen1988) to portray the effect sizes of ISD. To further test the robustness of the results with respect to the method used, we also performed a combination of PSM and difference-in-differences (DD) analysis (Ashenfelter1978).

Because the source data contained a substantial amount of missingness, all anal- yses were performed within a multiple imputation (MI) framework. We generated multiple imputed datasets. The analyses and resulting statistics were combined using the rules of Rubin (1987). To test the sensitivity of the results to the missing data methods, we additionally performed complete case analysis and applied the indicator method for missing values (see“Sensitivity of results to methods chosen”), and In the following sections we will succinctly describe the data, the definition of the groups, the variables used, the procedure for imputing missing data, the matching by pro- pensity scores, and the combination of imputation and PSM.

Effectiveness of a prolonged incarceration 37 Dataset and operalizations The data were obtained by linking police, public prosecutor’s office, probation and imprisonment data on the individual level. This was done for a subset of all suspects who were identified in the police registration as a HFO.

Linked data We used four national level data sources:

1.Police data. In order to identify the HFOs from the set of all suspects, we extracted data from the recognitionservice system (i.e.herkenningsdienstsysteem,HKS).Thisisa file on a national level of all local police databases containing all arrests leading to a police report of all crime suspects of age 12 years and older per year. The data are on the individual level. It includes data on ethnicity, geographical region, and crime types; 2.Conviction data. These data were extracted from the Dutch Offender’s Index. This is an anonymized index of the General Documentation Files (GDF). It provides a complete chronological overview of all criminal cases in which a person is convicted for a criminal offence. Convictions are registered for persons from ages 12 and older.

3.Probation data. In order to receive information on the functioning of the HFOs in different areas of life, we used probation data extracted from their client service system (clientvolgsysteem, CVS). This information is used in pre-sentence reports, probation plans with the client, and execution of community sentence orders. It contains information on the functioning of the offender with regard to different areas; 4.Prison data. The prison data were extracted from repositories of the Dutch prison system (Tenuitvoerlegginsprogramma gevangeniswezen, TULP-GW). These data contain the exact dates of entry into and release from prison from 1996. These data made it possible to correct recidivism frequencies for time at risk and to estimate the incapacitation effect. Unfortunately, we did not have data on stays in a hospital or psychiatric care, in which someone is also limited in committing crimes.

ISD group and the controls The total dataset was constructed as follows. In the period 2003–2008, the annual cohorts of suspects having 11 or more antecedents in a 5-year period was established, the complete list of HFOs during that period. An additionalrequirement was set for the prison control groups: an offender had to be at least once recognized as an HFO in the 4 years previous to release. From this dataset, the ISD group and both control groups were formed.

TheISD groupconsisted ofallHFOs released from an ISD measure from 2004 until December 2008. This group contained 558 persons. Four HFOs were dropped because of extreme pre-test crime frequencies due to too limited time at risk, leaving 554 persons in the treatment group. The majority of this group was released in 2007 and 2008. The average consecutive stay in prison was 834 days.

Thefirst control groupconsisted of all HFOs who were released from prison before the introduction of the ISD measure. Thehistorical control groupwas selected from HFOs released from prison between January 2003 and September 2004. This historical control group contained 4,092 offenders. Their average stay in a prison was 38N. Tollenaar et al. 108 days. This historical control group was used for ruling out selection effects by judges imposing ISD measures, because the ISD-measure had not yet been implemented. By using this group as a comparison, selection effects due to judicial decisions were minimized.

Thesecond control groupis a group of HFOs released from prison in 2007 and 2008, termed thesimultaneous control group. This control group was selected from all HFOs who were released from detention in 2007 and 2008 and contained 6,652 observations having a mean average incarceration length of 102 days. As this group was released in approximately thesame period as the ISD group, effects that, for instance, may have arisen from a selective law enforcement aimed at HFOs could be minimized.

Covariates To match the ISD group to both groups of controls, we used 20 covariates which can be classified into five groups: demographic characteristics, criminal career features, socio-economic status variables, characteristics of the conviction leading to incarcer- ation or ISD, and functioning in diverse life areas. Demographic characteristics were gender, age, and ethnicity, country of birth, and number of inhabitants of a residential municipality. It is well known that men are more likely than women to commit crime (Steffensmeier and Allan1996). Another well-known relationship is the age–crime curve (Farrington1986). The frequency to commit crime increased rapidly from 12 years to the start of the young adult life, and then decreases (the age–crime curve; see, for instance, Blokland2005for this curve in a Dutch population). This means that age effects can be expected.

We had two socio-economic status variables, namely work status and highest attended education. Not having a job is related to committing crime (Van der Geest2011). Little or no education increases the likelihood of committing crimes (Lochner2004).

Our criminal career features included age at the time of the first criminal case, the number of previous convictions, and the criminal case density. The latter is defined the number of criminal cases in the criminal career per unit time, uncorrected for time in detention. As a proxy of the severity of a criminal career, we used the average maximum potential sanction according to Dutch criminallaw for a specific crime type. This covariate is defined as the maximum possible penalty in the criminal case, expressed as the number of days of prison terms. This variable is averaged over all cases in the criminal career.

Also, characteristics of the case that led to the conviction to an ISD measure or a standard sanction, the index case were included. These were age at the time of the offence and the judicial district where the case was handled. The latter requires some explanation. The ISD cell capacity in the Netherlands was allocated proportionally to the local number of HFOs in the municipalities. In the districts of large cities, there was more capacity than in other cities. This means that offenders in the large cities were more likely to have ISD imposed on them. Additionally, local prosecution priorities may also have led to differences in recidivism for which we want to control.

The matching covariates also involved problems in different life areas. HFOs on whom an ISD measure was imposed showed to have significant problems in several areas of functioning (Goderie and Lünnemann2008) which may affect the impact of Effectiveness of a prolonged incarceration 39 the measure. Therefore, it is important to match individuals in the control group with similar problems. The probation CVS data provides data with regard to differ- ent areas of functioning. These are physical, psychological, addiction, relation- ship, housing, and financial matters. Problems in these life areas potentially maintain criminal behavior.

Finally, we also matched on whether a HFO has had a SOV measure in the past. It is possible that, having undergone a SOV measure, lowered the recidivism frequency (Koeter2004).

Handling missing data The probation data suffered from missing data; up to 39 % of the observations were missing minimally one value. These missings were mainly concentrated in the probation data (i.e. life circumstances). The group with missing data on these characteristics differed significantly from the group without missing data. If we were to discard all cases with missing data, 39 % of the data would not be used, possibly leading to a selective sample of those HFOs that received ISD. Instead, we decided to apply multiple imputation (MI) by using the regression switching approach of van Buuren et al. (1999), which works as follows. In an initialization step, all missing values were replaced by random numbers, while the position of the missing data were retained in memory. Then, iteratively, for each independent variable, a regression analysis was performed by regressing iton the remaining covariates. From the resulting equation, P(X 1|X2,X 3,…X k), values for X 1were simulated and imputed. Then, X 2was regressed on the remaining predictors, now containing new values for X 1. This process over all covariates was repeatedntimes and resulted in a draw from the multivariate distribution of all Xs. For our analyses, five imputations were generated.

Matching by propensity scores In order to provide a counterfactual for the ISD group, individuals were matched using PSM. Instead of matching on the actual characteristics, individuals were matched on the probability to be allocated to the ISD group. Under the assumption of conditional independence, the treatment effect is simply the difference between the two groups on the outcome. Conditional independence implies that, after matching, the groups do not differ on unmeasured characteristics related to the outcome.

We applied nearest-neighbor matching without replacement and without a caliper.

If there were multiple matches, a random case was selected as the candidate. After matching, we asserted whether the matching had succeeded by comparing the groups on their covariates. To detect significant differences,ttests were applied at an alpha level of 0.05.

Combining MI and PSM There are many ways to combine MI methods and PSM methods (Hill et al.2004).

Hill and colleagues found in their Monte Carlo study that the combination of 40N. Tollenaar et al. separately imputing and matching the data had the least bias and variance for the estimate of the treatment effect. In our case, this combined method was performed as follows:

1. Generate 5 imputed datasets by means of switching regression (van Buuren et al.

1999) 2. Estimate a propensity score model on each dataset and match each ISD subject to a control subject within these datasets; 3. Calculate covariate balance statistics and intervention outcome on each imputa- tion sample; 4. Combine the five effect estimates and balance tests using the rules of Rubin (1987).

The outline of our procedure is depicted in Fig.1.

Difference-in-differences PSM requires the strong assumption of conditional independence. Therefore, residual confounding remains one of the possible threats to the validity of the estimate of the treatment effect. In orderto test the robustness of the recidivism results and relax this assumption, we also performed a combination of PSM and double difference or difference-in-differences analysis (DD; Ashenfelter1978).

In this method, complete elimination of bias is not assumed, but instead one assumes that the slopes of the pre-treatment and post-treatment effects are the same, the so-called parallelism assumption. Pre-treatment measures are manda- tory in DD, because these are compared with post-matching measures.

Therefore, the effect could not be estimated on recidivism prevalence but only Fig. 1Schematic representation of the steps involved in combining MI and PSM Effectiveness of a prolonged incarceration 41 on recidivism frequency. As opposed to the full PSM in the previous analysis, this pre-treatment covariate hadto be excluded from the matching.

The analyses were performed in Stata v.10.1. The PSM was done using the module psmatch2v,3.0.0 (Leuven and Sianesi2003). The MI was applied using the module mvis(Royston2004).

Results Propensity score matching Two propensity score models were fitted to the data: one to combine the ISD group with the historical control group and one to combine with the simulta- neous control group (see Appendix Table4). To check the common support condition, the distribution of the log of the propensity scores for the two models wereplottedinFig.2 3. It showed complete overlap between the ISD group and two control groups. This ensured that a reasonable match could be found for each ISD subject.

Background characteristics before and after matching To test whether the PSM succeeded well, we performed covariate imbalance tests. For these tests, differences between ISD and controls were calculated before and after matching (see Table1). We will first describe the background characteristics of the ISD-group (column‘ISD group’), and then describe the pre-match differences to both control group (columns‘pre-matching’). Finally, we will review the balances of the ISD with the control groups after matching (columns‘post-matching’).

Description of the ISD group The majority of the subjects in the ISD group was male and the subjects were on average nearly 40 years old when their ISD measure was imposed (first column of Table1). More than half of the ISD subjects was born in the Netherlands. More than half of the ISD subjects was not indigenous. Surinamese and Moroccans were the largest ethnic minority groups. For four out of ten HFOs in the ISD group, the highest education attained was primary education. Nearly nine in ten ISD subjects were unemployed or disabled.

The ISD subjects had an extensive career criminal. Their first conviction for a criminal offence was on average at a relatively young age. The ISD group had an average of more than 60 criminal convictions of a crime that had a mean penalty length in their career of over 4 years (1,557 days). A small part of the ISD subjects have undergone a SOV measure in the past. Clearly, the most frequent offenders were prioritized in the imposition of the measure in practice.

ISD subjects appeared to suffer in various domains of functioning. Over 80 % of the ISD subjects was addicted, more than half had housing problems and almost half 3The natural log was taken to enhance the visibility of the right tail of the distribution.

42N. Tollenaar et al. had financial problems. Four out of ten ISD subjects had relationship problems or psychological problems.

Pre-matching difference between ISD and controls Before matching, the background characteristics of the ISD group and both control groups clearly differed (see Table1). HFOs in the ISD group were on average older, were more often male, and more often lived in one of the four largest (G4) cities than HFOs in the control groups. In addition, they had more extensive criminal careers: they started earlier, had more convictions on their criminal record, and the average sentence length of the criminal cases was higher. Also, relatively more ISD subjects had had a SOV measure in the past. The ISD group also showed more problems in several areas of functioning (addiction, physical, psychological, and housing) than the control groups. These results showed clearly that the HFOs who were sanctioned with an ISD were far more risk-prone than the HFOs that had a standard prison sentence imposed. Matching was obviously required.

Matching balance After the matching, only one statistical significant difference was found between the ISD and the controls (see the last column in Table1). The ISD group only differed significantly from the simultaneous control group on the average number of criminal cases per year free preceding incarceration. We concluded that after matching the groups were comparable.

Incapacitation effect In order to estimate the size of the incapacitation effect of the ISD measure, the convictions and recorded offences of the HFOs were counted in the matched simul- taneous control group in the period the ISD HFOs were incarcerated. Because the subjects in this control group can be seen as the counterfactuals of the ISD subjects in terms of background characteristics, and the data span the same time period, this yields the incapacitation effect of the ISD measure.

Of the 554 HFOs in the simultaneous control group, 37 (7 %) did not come into contact with the judiciary in the period of their ISD counterparts. Those who were convicted during that period had 3.211 reconvictions in total (M = 5.7 convictions; SD = 4.4). These cases included 5.097 recorded offences (M = 9.2 offences; SD = 12.2). The types of recorded 0 1000 2000 3000 4000 5000 Frequency -20 -15 -10 -5 0 0 1000 2000 Frequency 3000 4000 -20 -15 -10 -5 0 log(Propensity score) ISD Historical control group log(Propensity score) ISD Simultaneous control group Fig. 2Common support of propensity scores of ISD and control groups Effectiveness of a prolonged incarceration 43 Table 1Pre- and post-matching background characteristics of ISD and control groups Pre-matching Post-matching ISD-group (n= 554)Historical controls 2003–2004 (n= 4,092)Simultaneous controls 2007–2008 (n= 6,652)Historical controls 2003–2004 (n= 554)Simultaneous controls 2007–2008 (n= 554) Demografic characteristics (in %) Male 94.0 92.8 94.6 94.2 93.7 Age 39.4 34.5**** 34.6**** 39.7 39.8 Country of birth (OBJD; in %) Netherlands 58.5 60.1 62.9* 59.8 57.5 Morocco 10.1 9.2 8.1 10.3 10.4 Neth. Antilles and Aruba 7.6 7.2 7.6 7.8 7.8 Surinam 14.8 9.7*** 7.8**** 14.3 16.2 Turkey 1.4 1.7 1.7 1.3 1.2 Other Western 2.9 5.0** 4.8* 2.6 2.6 Other non-Western 4.7 7.1** 6.9* 3.9 4.2 Ethnicity (HKS; in %) Netherlands 47.8 47.0 46.3 48.8 46.2 Morocco 12.5 12.7 14.2 12.2 12.4 Neth. Antilles and Aruba 7.9 7.8 8.4 8.0 8.2 Surinam 16.4 11.6** 10.9*** 16.1 17.9 Turkey 2.5 3.0 3.5 2.3 2.5 Other Western 7.0 8.5 7.9 7.0 6.8 Other non-Western 5.8 9.4*** 8.8** 5.6 6.1 Size of municipality (HKS; in %) <10.000 0.0 0.3 0.4 0.0 0.0 10.000–50.000 7.6 13.5**** 16.1**** 6.7 6.5 50.000–100.000 10.3 15.4*** 16.6**** 10.0 10.9 100.000–250.000 27.1 29.5 28.9 26.7 28.1 >250.000 inhabitants (G4) 53.6 39.5**** 35.7**** 55.4 53.5 Outside of the Netherlands 1.4 1.9 2.3 1.2 1.0 Education (CVS; in %) Primary or no education 21.0 21.9 21.4 21.5 21.9 Lower secondary without certificat 42.5 38.9 38.3 42.7 42.0 Lower secondary 17.0 18.7 17.7 16.4 17.5 Medium to higher secondary 8.2 10.6* 12.1** 8.8 7.9 Onbekend 11.3 10.0 10.4 10.7 10.6 Labor (CVS; in %) (partially) Unemployed/disabled 88.6 84.2** 79.8**** 89.4 89.7 Casual employment 6.0 8.1* 10.6*** 5.7 5.4 Employed 0.3 1.0 1.8** 0.2 0.2 Other 5.2 6.7 7.8* 4.8 4.7 44N. Tollenaar et al. Table 1(continued) Pre-matching Post-matching ISD-group (n= 554)Historical controls 2003–2004 (n= 4,092)Simultaneous controls 2007–2008 (n= 6,652)Historical controls 2003–2004 (n= 554)Simultaneous controls 2007–2008 (n= 554) Criminal career characteristics (M) Mean age at first conviction 18.2 18.8** 18.5 18.6 18.2 Mean previous convictions 61.3 38.0**** 31.8**** 61.0 60.9 Mean conviction density 1.6 1.4**** 1.3**** 1.6 1.6 Mean maximum penalty previous cases1,557.2 1,526.8* 1,452.6**** 1,547.5 1,567.4 Mean previous convictions before incarceration a 8.0 4.6**** 3.1**** 7.5 6.6* Had SOV (in %) 5.2 2.5** 2.1** 5.1 4.9 Court district (in %) Den Bosch 7.0 5.6 5.6 6.9 7.3 Breda 2.9 5.1** 5.4** 3.0 3.1 Maastricht 3.1 3.9 3.8 3.2 4.0 Roermond 1.1 1.2 1.5 1.0 1.0 Arnhem 2.0 7.1**** 6.8**** 1.8 1.7 Zutphen 3.8 2.3* 2.8 3.6 3.9 Zwolle-Lelystad 1.6 3.4** 3.8*** 1.8 1.4 Almelo 2.2 3.0 3.2 1.7 2.0 Den Haag 14.4 14.6 13.7 14.3 13.6 Rotterdam 16.8 13.4* 13.6 17.2 16.5 Dordrecht 1.6 2.8* 3.2 ** 1.8 1.6 Middelburg 0.9 1.2 1.5 1.0 0.8 Amsterdam 29.1 16.0**** 14.4**** 29.6 30.7 Alkmaar 1.8 2.2 2.1 1.6 1.7 Haarlem 3.2 4.0 3.5 3.6 3.3 Utrecht 6.9 8.5 7.5 6.7 6.6 Leeuwarden 0.4 2.6**** 3.2**** 0.1 0.0 Groningen 1.3 3.3**** 4.4**** 1.1 0.8 Problem areas (CVS; in %) Physical 23.4 18.8** 18.2** 24.5 23.1 Psychological/psychiatrical 43.3 38.2* 40.0 44.2 43.0 Addiction 81.6 75.0*** 68.0**** 82.6 83.0 Relations 40.0 38.0 39.9 39.4 39.3 Housing 55.3 45.8**** 43.8 56.5 57.7 Finances 48.8 45.1 46.5 48.2 49.7 *p< 0.05; **p< 0.01; ***p< 0.001; ****p< 0.0001 aEstimated Effectiveness of a prolonged incarceration 45 offences are depicted in Table2. The majority of offences were theft (41.6 %). More than 60 % within this category concerned shoplifting. Breaking into a house or car comprised 9.8 % of the offences, whereas 10.7 % concerned vandalism and public order crimes.

Speaking in counterfactual terms, 7 % of the HFOs in ISD would not have had a new conviction within 2 years if they had been sanctioned with a standard sanction.

On average, an estimated 5.7 convictions and 9.2 recorded offences have been prevented by a single incarceration in the ISD. To make these estimates more comparable to estimates from other studies on the incapacitation effect, we rescaled these to numbers per year. The incapacitation effects then were 2.5 (SD = 1.9) convictions and 4.0 (SD = 3.4) recorded offenses per year of incapacitation.

Post-release recidivism prevalence The ISD group showed a significantly lower recidivism prevalence than both control groups. The failure curves of recidivism prevalence are depicted in Fig.3. All three tests were significant at the .001 level. Two years after release we found that the ISD group showed an estimated 16 % less recidivism prevalence compared to the historical control group (72 and 88 % recidivated, respectively). After 4 years, the difference increased to 19 % (75 and 94 % recidivated, respectively). In terms of an effect size (Cohen’s h), the differences in prevalence between both groups increased from‘small’(Cohen’sh=.42) 2 years after release to‘moderate’(Cohen’sh= .55) 4 years after release. In the domain of criminology, however, effect sizes this large are rare, so these effects can be considered quite substantial (see, e.g., Lipsey2000).

The differences with the simultaneous group were smaller (Fig.3b). Two years after release, the probability of recidivating was 0.84 for the HFOs who had a standard short-term imprisonment, an estimated 12 % more recidivists than among those HFOs released from ISD (Cohen’sh= .29).

Post-release recidivism frequency The ISD group also differed significantly from both control groups in terms of frequency of reconvictions and recorded offences per year free. Table3shows the means and standard deviations of the frequency of reconviction and recorded Table 2 Type of recorded offence in prevented convictions a aEstimated on the simultaneous control group Type of recorded offence Percentage Breaking into house or car 9.8 Theft 41.6 Other property crime 0.2 Assault and battery 5.5 Vandalism/public order 10.7 Drugs 5.0 Weapons 0.5 Traffic 3.6 Miscellaneous 23.1 46N. Tollenaar et al. offences in units per year free. All tests showed statistical significant differences between the ISD group and its controls. The effect sizes are small following statistical effect size conventions (Cohen’sd< 0.3), but are large when it concern effects for crime interventions.

The differences between ISD and controls were also calculated for the subset of those HFOs who did recidivate. By doing this, we could establish whether ex-ISD subjects who still recidivated tended to‘calm down’compared to their controls. These tests also showed significant results. This finding indicates that the ISD measure seems to reduce recidivism frequency of at least a part of its releasees.

Difference-in-differences on recidivism frequency To test the robustness of the recidivism frequency findings, a combined PSM and DD analysis was performed on both control groups. In Fig.4, the actual progress is shown from pre- to post-treatment reconviction frequencies for the ISD group (dotted lines) and its controls (straight lines). Due to the PSM, the controls can be seen as a counterfactual to the HFO in the ISD group. So if those HFOs would not be sanctioned with an ISD measure, a DD analysis assumes that their progress in reconviction frequency would be the same as their controls. This expected progress is shown in the dotted lines. The difference between the actual ISD slope of reconviction frequency and the expectation is an estimate of the treatment effect of the ISD group versus the two control groups.

Figure4ashows pre- and post-test means on the reconviction frequency of the ISD and historical controls and the expected frequency based on the slope of the historical controls. Figure4bvisualizes the same information for the simultaneous control group. The mean difference in the expected reconviction frequency of the ISD group and the actual post-release frequency in the historical group is 3.3. Relative to the simultaneous control group, the estimate of the ISD effect is somewhat larger. The difference in the expected and observed post-release reconviction frequency of the ISD group is 5.8. Using this alternative method, we arrived at the same conclusion:

compared to standard short-term imprisonment the ISD measure reduced the amount of recidivism among HFOs after release. 0 .25 .5 .75 1 Recidivism probability 0 500 1000 1500 Time in days ISD Historical control group 95% C.I. 95% C.I. 0 .25 .5 .75 1 Recidivism probability 0 500 1000 1500 Time in days ISD Simultaneous control group 95% C.I. 95% C.I. Fig. 3Recidivism prevalence of ISD versus the historical and simultaneous controls. The curves for the control groups are combined from the imputed datasets. TheFstatistics are obtained by combining the five X 2statistics from the five imputations Effectiveness of a prolonged incarceration 47 Table 3Imputed means and standard deviations of ISD and control groups Recidivism and recorded offencesISD Historical control groupSimultaneous control group(A) vs. (B) (A) vs. (C) (A) (B) (C) Mean (SD) Mean (SD) Mean (SD) Total group Reconvictions per year free 3.4 (6.1) 5.9 (15.4) 6.4 (13.8)tvalue (df= 12.09) = 2.66*d= 0.23tvalue (df= 32.56) = 3.31**d= 0.26 Recorded offences per year free 5.2 (10.2) 10.3 (27.7) 10.8 (27.6)tvalue (df= 10.97) = 3.09**d= 0.27tvalue (df= 25.68) = 3.62***d= 0.27 Recidivists Reconvictions per year free 4.7 (6.7) 6.6 (16.2) 7.4 (14.5)tvalue (df= 3.58) = 3.81*d= 0.17tvalue (df= 17.74) = 4.25 *** d= 0.21 Recorded offences per year free 7.1 (11.3) 11.5 (29.3) 12.3 (29.0)tvalue (df= 6.12) = 4.10**d= 0.23tvalue (df= 12.75) = 4.49***d= 0.23 Thetstatistics are combined from fivetstatistics from the five imputations *p< 0.05; **p< 0.01; ***p< 0.001 48N. Tollenaar et al. Sensitivity of results to methods chosen In order to test the robustness of our results with regard to the imputations, we also generated the results using the following methods:

&Complete case analysis: In this analysis, no imputation was used and nearest- neighbor matching PSM was performed on listwise deleted datasets.

&The indicator method. In this method, for each variable with missing data, an indicator iscreated(1=missing,0=nonmissing).The missing value in the actual variable is replaced by a zero. Although this method yields biased regression estimates, it is often suitable for estimation of the treatment effect (see van Buuren2012).

&Kernel matching. In this analysis, imputation was used, but the matching proce- dure was replaced by Kernel matching with a bandwidth of 0.03. In this method, a match is constructed for each subject in the ISD using a weighted average over multiple persons in the comparison group (Heckman et al.1998a; Heckman et al.

1997,1998b) instead of actual one-by-one matching.

These analyses proved to yield the same results as the analyses shown. The results therefore seem to be robust to imputation and matching methods chosen.

Discussion The ISD measure is a severe penal measure meant to reduce crime among high frequent offenders (HFO) through incapacitation and preventing recidivism. Important aspects of the ISD are that there is certainty of sanctioning, it contains a prolonged incarceration up to 2 years, and for those HFOs who are motivated interventions and rehabilitation are available. Using a retrospective quasi-experimental design, we investigated the effective- ness of the ISD measure compared to that of a standard sanction for HFOs, mainly a short- term imprisonment in terms of reduction in (re)conviction and registered crime. The incapacitation effect of the ISD with respect to the standard sanction pattern after release was shown to be substantial. In addition, the reconviction rate of a group of HFOs released from ISD in the period 2004–2008 compared to two comparable control groups was found to be substantially lower. 012 3 4 5 6 7 8 9 10 Before incarceration After release Mean conviction frequency ISD Simultaneous controls Expectation ISD Treatment effect 0 1 2 3 4 5 6 78 9 Before incarceration After release Mean conviction frequency ISD Historical controls Expectation ISD Treatment effect ab Fig. 4Difference-in-difference estimates of the ISD group (dotted lines) versus the two control groups (straight lines) on average number of convictions per year free Effectiveness of a prolonged incarceration 49 Incapacitation effect Our results showed that, when HFOs areincapacitated in an ISD institution, a significant higher number of convictions and recorded offences is prevented com- pared with when they received a standard sanction. This study delivers empirical evidence with regard to chronic offenders that incapacitation has a reducing effect on crime (DeLisi and Piquero2011). Moreover, by using a counterfactual approach, we could estimate thesizeof the incapacitation effect. Taking into account the time that the matched counterparts of the ISD subjects in the simultaneous control group were included, we estimated an average of 5.7 criminal convictions that included 9.2 recorded offences.. This boils down to a preventive effect of on average 2.5 convic- tions and 4 recorded offenses per chronic offender per year. The majority of the recorded offences were related to theft (especially shoplifting). Other offences related to burglary from homes or cars, vandalism, and public order offences.

This is a large incapacitation effect when compared with estimates at the individual level in Wermink et al. (2012), that range from .17 to .21 convictions, but is quite similar to the 2.8 re-arrests per year per person found by Owens (2009). These studies do, however, pertain to groups that are not comparable tothe HFOs, namely first-time imprisoned 18– 50 year olds and imprisoned 23–25 year olds, respectively. The HFOs have been preselected by the police and public prosecutor on their actual arrest frequency, namely at least 11 police contacts in 5 years. This yields a minimal lambda of 2.2 police contacts per year before incarceration. Therefore, larger incapacitation effects can be expected.

Our estimate is probably a large underestimation of the number of actual committed crimes that have been prevented. It is well known that, in the judicial chain, the number of offences are filtered (Farrington2013; Piquero and Blumstein2007). We do not know exactly how large this underestimation is, but previous research showed that, as the number of actual offences increases, the likelihood that these facts are recorded in convictions decreases (Farrington et al.2007). In the Netherlands, Koeter (2004) found that, among a group of addicted criminals, the number of self-reported offences was 4–20 times higher than the offences known to the police, depending on the type of crime. The only way to further investigate this is to also conduct research based on data that do not rely on police or justice efforts, such as self-reports of offending among high-level chronic offenders.

Effect on post-release recidivism Our results showed a statistically significant small effect of the ISD measure on the post-release recidivism of its participants, when compared to standard short impris- onment for HFOs. This holds for the recidivism prevalence as well as the frequencies of reconvictions and recorded offences. Two years after discharge from the ISD, 72 % of the offenders were reconvicted compared to 84–88 % of the HFOs who were released from a standard short-term imprisonment. A part of these ISD releasees might actually have desisted, because instantaneous desistance was shown to exist in long follow-up data (Kurlycheck et al.2012).

These results are in line with those of an earlier study on the effects of the precursor of the ISD measure, the SOV measure (Koeter and Bakker2007).

The SOV measure, however, differed from the ISD with regard to the inclusion of offenders. In the SOV, only male addicted HFOs were included while offenders with 50N. Tollenaar et al. serious psychiatric symptoms were excluded. Our results seem to confirm the results found by DeJong (1997). The habitual experienced offender with weak bonds to conventional society is more susceptible to increased sentence length.

As we did not have individual level intervention or treatment data, we cannot discern treatment from deterrence effects. If rehabilitation had been the only effective ingredient of the measure, the results are somewhat surprising because an important condition of the What Works approach, namely program integrity (Lipsey2009), seems not to be met. An evaluation of the program integrity of the ISD (Goderie and Lünnemann2008) showed that, until 2008, the measure in practice was not implemented as it was intended to be. In particular, psychiatric care for offenders with psychiatric problems was hardly available, and ISD convicts having addiction problems remained untreated throughout the duration of the measure in detention.

The interventions and training that were offered during their stay in ISD did not seem to be aimed at addressing the core problems of its participants (Goderie and Lünnemann2008). Although we had no information about which inmates received which interventions or training during the ISD, it is unlikely that the situation in our research differed. An alternative explanation for the positive effects found for the ISD measure on recidivism of the participants could be the following. The majority of the HFOs in the 2004–2008 ISD group were shown to be a very problematic group with a risky lifestyle characterized by addiction, unemployment, mental, relational, and housing problems. It is conceivable that the prolonged removal of these offenders from their habitual environment and placement in prison, where at least basic care was provided, could have helped them to overcome some of their basic problems, at least for a limited period of time. Daily care and routine in prison for an extended time could have helped them to improve their physical and mental health. This may for some have had a lasting effect on recidivism after their detention, at least up to 4 years after discharge. Another alternative explanation could be that the prolonged and certain incarceration of specific chronic offenders motivates prison staff to invest more in depth in these offenders, as opposed to the standard short-term imprisonment where it is mostly certain that the offenders will be released from prison in a short period and prison staff effort will be in vain.

Context specific factors The ISD measure seems to deliver promising results in terms of incapacitation and recidivism reduction amongst a group of high-rate chronic offenders. We are not sure for several reasons if these results can be generalized to other countries and other judicial systems. In the Netherlands, even after a period of increasingly more punitive sanctioning causing the imprisonment rates to rise, the actual lengths of prison terms are still relatively low. It is conceivable that, in countries that impose lengthier imprisonments, a HFO group may not even show up because the offenders are unable to obtain convictions that frequently.

Countries can differ considerably on which offenses are counted as crimes (see, e.g., Aebi et al.2010). For example, drug possession and drug use are not indictable offenses as opposed to, for example, the U.S.A., so these will not count as crimes in the Netherlands. This will work through both in the definition of the group as in recidivism and the incapacitation effect. Traffic offenses like driving without a license, Effectiveness of a prolonged incarceration 51 however, are counted as crimes. So, both the size of the HFO group, as its recidivism is dependent on what is counted, and the absence of drug possession and use crimes will shrink both the group and the incapacitation effect. The amount of crime per capita can also vary substantially. In Europe, the Netherlands holds sixth place in 2007 on the number of criminal offenses (Aebi et al.2010: 37). Another source of variation across countries might be the varying amount of supervision and/or prison aftercare.

Supervision was shown to be negatively correlated with recidivism (Lund et al.2012), specifically with disordered offenders with substance abuse disorder.

Limitations Random assignment to the treatment conditions is absent in our study. So, we cannot completely rule out alternative explanations for the positive effects of ISD with regard to recidivism. However, potential selection bias was minimized as much as possible.

Furthermore, the research literature describe a number of conditions to prevent selection bias in quasi-experimental studies (Cook et al.2008; Glazerman et al.

2002). These are:

1. After matching, the intervention group should be as intact as possible; 2. Initial differences between intervention and control groups should be minimized; 3. There is matching on geographic proximity of the cases; 4. Matching should be on pre-intervention measures of the outcome.

All four conditions are covered in our analysis: (1) we used 99 % of those HFOs who were released from ISD in the period 2004–2008; (2) after matching, only one difference remained between the intervention and controls; (3) the matching covar- iates contained the municipality of residence and the court district that persecuted the HFOs; and (4) we matched on pre-detention conviction frequency per year free.

A fifth condition mentioned in the literature regarding PSM is that the number of covariates should be maximized (see, e.g., Apel and Sweeten2010). Bales and Piquero (2012) also found that their model with the maximum number of covariates showed the lowest difference in recidivism rates between prison and the CS group, suggesting bias reduction. In our study, we used 20 covariates covering different domains of functioning related to the outcome and control groups have been shown to be equal after matching.

On the other hand, some conditions canexacerbatethe bias and cause large differences between estimates based on quasi-experimental and experimental designs (Cook et al.2008):

1. Intervention and comparison groups are selected from different datasets or samples; 2. Thenof both groups is limited; 3. Matching only involves demographic variables; 4. There is a large heterogeneity between populations; 5. The assignment of cases to intervention and control groups is very complex and unclear.

Our study does not suffer from the first four conditions. The fifth condition could be a potential problem because the assignment of HFOs to ISD follows a multi-stage 52N. Tollenaar et al. process. First, each court district makes a list of the target HFOs, then offenders with the highest priority are selected in a local consultation between the police, public prosecutor, and probation. Finally, the judge considers whether the offender is eligible for placement in the ISD. To counter this potential assignment problem, we also assembled a historical control group that was released from prison before it was possible to impose the ISD. The drawback of the latter is that their time at risk falls in a different period from the ISD group, making it more susceptible to temporal differences in prosecution priorities.

Compared to both groups of controls, ISD showed a significant effect on reconviction rates. A part of the recidivism reduction may be explained because of offenders dying after drug treatment due to an overdose (see, e.g., Dumont et al.2012). We cannot exclude this alternative explanation because we did not have date of death outside the period of judicial affairs. Nevertheless, this does not explain why the recidivists of the ISD group have a lower reconviction frequency.

Finally, our study focused on officially recorded crime. Registered crime is only a lower bound of re-offending (see, e.g., Piquero and Blumstein2007; Farrington2013).

Conclusion We conclude that the ISD measure is more effective in terms of recidivism and has a substantially larger incapacitation effect than the imposition of a standard custodial sentences to high frequency offenders. The effectiveness of ISD, both in terms of incapacitation and recidivism, apply to the group of high frequency offenders in the period 2004–2008 who were released from an ISD measure. This group of offenders is characterized by a risky lifestyle with addiction, lack of housing, unemployment, and relational problems. It is plausible that within this extreme high-risk group, prolonged incarceration can bring improvements in their lifestyle, that continues after release, thus having a prolonged effect on reducing recidivism. It is, however, not expected that broadening the measure to groups of offenders with a less problematic background will show similar effects. Future research should seek to replicate these results and focus on finding which components of the ISD measure contribute to the effects found.

Appendix Ta b l e 4Coefficients of propensity score models of the probability of receiving ISD: ISD versus historicalor simultaneous controls ISD vs. historical ISD vs. simultaneous ISD/ prisonβSEzvalueβSEzvalue Sex (male) 0.08 0.21 0.39 0.22 0.22 1.02 Age 0.15 0.01 11.16**** 0.12 0.01 9.16**** Effectiveness of a prolonged incarceration 53 Table 4(continued) ISD vs. historical ISD vs. simultaneous ISD/ prisonβSEzvalueβSEzvalue Country of birth (OBJD) Netherlands 0 0 Morocco 0.24 0.35 0.70 0.52 0.34 1.51 Neth. Antilles and Aruba 0.33 0.78 0.43 0.41 0.75 0.54 Surinam 0.37 0.40 0.92 0.31 0.38 0.81 Turkey 0.80 0.59 1.34 0.06 0.59 0.10 Other Western 0.46 0.34 1.35 0.78 0.35 2.19* Other non-Western 0.37 0.31 1.18 0.24 0.33 0.71 Ethnicity (HKS) Netherlands (reference) 0 0 Morocco 0.52 0.33 1.60 0.50 0.32 1.56 Neth. Antilles and Aruba 0.23 0.76 0.31 0.67 0.74 0.91 Surinam 0.07 0.38 0.17 0.51 0.36 1.41 Turkey 0.70 0.46 1.51 0.15 0.46 0.32 Other Western 0.00 0.23 0.01 0.07 0.25 0.28 Other non-Western 0.36 0.27 1.37 0.36 0.28 1.28 Size of municipality (HKS) <50.000 (reference) 0 0 50.000–100.000 0.06 0.23 0.24 0.10 0.35 0.27 100.000–250.000 0.50 0.20 2.46* 0.37 0.34 1.07 >250.000 inhabitants (G4) 0.32 0.21 1.54 0.52 0.35 1.50 Outside of the Netherlands 0.31 0.45 0.70 0.14 0.64 0.22 Education (CVS) Primary or no education (reference) 0 0 Lower secondary without certificat 0.22 0.13 1.67 0.11 0.13 0.84 Lower secondary 0.11 0.16 0.68 0.15 0.16 0.92 Medium to higher secondary 0.03 0.21 0.12 0.11 0.21 0.51 Unknown 0.04 0.19 0.22 0.05 0.19 0.29 Werk (CVS) 0.00 0.00 0.00 0.00 (partially) Unemployed/disabled (reference) 0 0 casual employment 0.02 0.22 0.10 0.09 0.21 0.44 Employed 0.82 1.10 0.74 0.83 1.04 0.80 Other 0.09 0.23 0.39 0.21 0.24 0.86 Criminal career characteristics Mean age at first conviction 0.14 0.02 8.00**** 0.11 0.02 6.91**** Mean previous convictions 0.02 0.00 3.94**** 0.02 0.00 6.00**** Mean conviction density 1.59 0.32 4.99**** 2.85 0.33 8.53**** Mean maximum penalty previous cases 0.00 0.00 7.66**** 0.00 0.00 9.37**** Mean previous convictions before incarceration a0.08 0.01 8.15**** 0.20 0.01 15.84**** Had SOV (%) 0.34 0.25 1.40 0.91 0.26 3.53 54N. Tollenaar et al. References Aebi, M. F., de Cavarlay, B. A., Barclay, G., Gruszcyńska, B., Harrendorf, S., Markku, H., et al. (2010).

European sourcebook of crime and criminal justice statistics–2010. The Hague: WODC.

Apel, R. J., & Sweeten, G. (2010). Propensity score matching in criminology and criminal justice. In A.

Piquero & D. Weisburd (Eds.),Handbook of quantitative criminology(pp. 543–562). New York:

Springer.

Ashenfelter, O. (1978). Estimating the effect of training programs on earnings.The Review of Economics and Statistics, 6(1), 47–57.

Table 4(continued) ISD vs. historical ISD vs. simultaneous ISD/ prisonβSEzvalueβSEzvalue Court district (in %) Den Bosch (reference) 0 0 Breda 1.03 0.33 3.15** 0.92 0.33 2.76** Maastricht 0.18 0.33 0.52 0.33 0.34 0.99 Roermond 0.19 0.46 0.41 0.25 0.46 0.55 Arnhem 1.56 0.35 4.43**** 1.80 0.38 4.76**** Zutphen 0.18 0.32 0.57 0.23 0.31 0.73 Zwolle-Lelystad 0.99 0.40 2.46* 1.22 0.41 2.96** Almelo 0.51 0.36 1.41 0.71 0.38 1.88 Den Haag 0.36 0.24 1.46 0.26 0.25 1.07 Rotterdam 0.04 0.25 0.18 0.19 0.25 0.76 Dordrecht 0.61 0.40 1.50 0.89 0.41 2.17* Middelburg 0.24 0.53 0.46 0.10 0.53 0.19 Amsterdam 0.35 0.24 1.48 0.35 0.24 1.44 Alkmaar 0.47 0.42 1.11 0.16 0.42 0.38 Haarlem 0.45 0.32 1.43 0.34 0.32 1.05 Utrecht 0.68 0.27 2.48* 0.48 0.28 1.74 Leeuwarden 2.01 0.70 2.87** 2.59 0.80 3.25** Groningen 1.73 0.45 3.84**** 2.33 0.51 4.53**** Problem areas (CVS) (yes/no) Physical 0.14 0.14 1.01 0.06 0.14 0.41 Psychological/psychiatrical 0.23 0.12 1.90 0.01 0.12 0.09 Addiction 0.03 0.14 0.22 0.24 0.14 1.77 Relations 0.03 0.13 0.21 0.00 0.13 0.00 Housing 0.27 0.13 2.04* 0.32 0.13 2.44* Finances 0.01 0.13 0.11 0.16 0.13 1.19 Constant 9.68 0.74 13.12 11.33 0.78 14.50 Coefficients and standard errors are combined over five imputations *p< 0.05; **p< 0.01; ***p< 0.001; ****p< 0.0001 Effectiveness of a prolonged incarceration 55 Bales, W. D., & Piquero, A. R. (2012). Assessing the impact of imprisonment on recidivism.Journal of Experimental Criminology, 8(1), 71–101.

Bersani, B. E., Nieuwbeerta, P., & Laub, J. H. (2009). Predicting trajectories of offending over the life course: findings from a Dutch conviction cohort.Journal of Research in Crime and Delinquency, 46, 468–494.

Blokland, A. (2005). Crime over the life span. Trajectories of criminal behavior in Dutch offenders.

Dissertation, University of Leiden Blokland, A. A. J., Nagin, D. S., & Nieuwbeerta, P. (2005). Life span offending trajectories of a Dutch conviction cohort.Criminology, 43, 919–954.

Cid, J. (2009). Is imprisonment Criminogenic? A comparative study of recidivism rates between prison and suspended prison sanctions.European Journal of Criminology, 6, 459–480.

Cohen, J. (1988).Statistical power analysis for the behavioral sciences. Hillsdale: Lawrence Erlbaum.

Cook, T. D., Shadish, W. R., & Wong, V. C. (2008). Three conditions under which experiments and observational studies produce comparable causal estimates: New findings from within-study compar- isons.Journal of Policy Analysis and Management, 27, 724–750.

DeJong, C. (1997). Survival analysis and specific deterrence: integrating theoretical and empirical models of recidivism.Criminology, 35, 561–575.

DeLisi, M., & Piquero, A. R. (2011). New frontiers in criminal careers research, 2000–2011: a state-of-the- art review.Journal of Criminal Justice, 39, 289–301.

Dumont, D. M., Brockmann, B., Dickman, S., Alexander, N. & Rich, J. D. (2012). Public health and the epidemic of incarceration.Annual Review of Public Health, 21(33), 325–339.

Farrington, D. P. (1986) Age and crime. In Tonry, M. and Morris, N. (Eds)Crime andJustice: An Annual Review of Research, vol 7. Chicago: University of Chicago Press (pp. 189-250).

Farrington, D. P. (2013). Self-reported versus official offending. In D. P. Farrington, A. R. Piquero, & W. G.

Jennings (Eds.),Offending from childhood to late middle age: Recent results from the cambridge study in delinquent development(pp. 27–37). New York: Springer.

Farrington, D. P., Jolliffe, D., Loeber, R., & Hornish, D. L. (2007). How many offences are really committed per juvenile court offender?Victims and Offenders, 2, 227–249.

Gendreau, P., Goggin, C., & Cullen, F. (1999).The effects of prison sentences on recidivism. A report to the Corrections Research and Development and Aboriginal Policy Branch, Solicitor General of Canada.

Ottawa, Ontario: Public Works & Government Services Canada Glazerman, S., Levy, D. M., & Myers, D. (2002).Nonexperimental replications of social experiments: A systematic review. NJ: Mathematica Policy Research.

Goderie, M., & Lünnemann, K. D. (2008).De Maatregel Inrichting voor Stelselmatige Daders. Utrecht:

Verwey-Jonker Instituut.

Heckman, J., Ichimura, H., Smith, J., & Todd, P. (1998a). Characterizing selection bias using experimental data.Econometrica, 66, 1017–1098.

Heckman, J. J., Ichimura, H., & Todd, P. E. (1997). Matching as an econometric evaluation estimator:

evidence from evaluating a job training programme.Review of Economic Studies, 64, 605–654.

Heckman, J. J., Ichimura, H., & Todd, P. (1998b). Matching as an econometric evaluation estimator.Review of Economic Studies, 65, 261–294.

Hill, J. L., Reiter, J. P., & Zanutto, E. L. (2004). A comparison of experimental and observational data analyses. In A. Gelman & X. L. Meng (Eds.),Applied Bayesian modeling and causal inference from incomplete data perspectives(pp. 51–60). New York: Wiley.

Jacobs, M., & Essers, A. A. M. (2003). De misdaad van de straat. Op zoek naar achtergrondenvan veelplegers.Tijdschrift voor Criminologie, 45(2), 140–152.

Kaplan, E. L., & Meier, P. (1958). Nonparametric estimation from incomplete observations.Journal of American Statistic Association, 53, 457–481.

Koeter, M. W. J. (2004).Vroeghulp aan verslaafden II: Het effect van de Vroeghulp Interventie Aanpak op criminele recidive en verslavingsgedrag. Amsterdam: The Amsterdam Institute for Addiction Research.

Koeter, M. W. J., & Bakker, M. (2007).Effectevaluatie van de strafrechtelijke opvang verslaafden (SOV).

Den Haag: WODC.

Kurlycheck, M. C., Bushway, S., & Brame, R. (2012). Long-term crime desistance and recidivismpatterns - Evidence from the Essex County Convicted Felon Study.Criminology, 50, 71103.

Larney, S., Toson, B., Burns, L., & Dolan, K. (2012). Effect of prison-based opioid substitution treatment and post-release retention in treatment on risk of re-incarceration.Addiction, 107, 372–380. 56N. Tollenaar et al. Leuven, E., Sianesi, B. (2003). PSMATCH2: Stata module to perform full Mahalanobis and propensity score matching, common support graphing, and covariate imbalance testing.http://ideas.repec.org/c/ boc/bocode/s432001.html.

Lipsey, M. W. (2000). Statistical conclusion validity for intervention research: A significant (p< .05) problem. In L. Bickman (Ed.),Validity and social experimentation: Donald Campbell’s legacy(pp.

101–120). Thousand Oaks: Sage.

Lipsey, M. W. (2009). The primary factors that characterize effective interventions with juvenile offenders:

A meta-analytic overview. Victims and Offenders, 4, 124–147. In: M. Tonry (Eds.), Crime & Justice (pp. 359–505). Chicago: University of Chicago.

Lochner, L. (2004). Education, work and crime: a human capital approach.International Economic Review, 45(3), 811–843.

Loeffler, C. E. (2013). Does imprisonment alter the life-course? Evidence on crime and employment from a natural experiment.Criminology 51. doi:10.1111/1745-9125.12000.

Lund, C., Forsman, A., Anckarsater, H., & Nilsson, T. (2012). Early criminal recidivism among mentally disordered offenders.International Journal of Offender Therapy and Comparative Criminology, 56, 749–768.

Meade, B., Steiner, B., Makarios, M., & Travis, L. (2012). Estimating a Dose–response relationship between time served in prison and recidivism.Journal of Research in Crime and Delinquency.

doi:10.1177/0022427812458928.

Mitchell, O., Wilson, D. B., & MacKenzie, D. L. (2006). The effectiveness of incarceration-based drug treatment on criminal behavior. Campbell Systematic Reviews no. 2006:11. The Campbell Collaboration Nagin, D. S. (1999). Analyzing developmental trajectories: A semiparametric, group-based approach.

Psychological Methods, 4(2), 139–157.

Nagin, D. S., & Land, K. (1993).“Age, criminal careers, and population heterogeneity: Specification and estimation of a nonparametric mixed poisson model.”Criminology 31, 327–362.

Nagin, D. S. & Snodgrass, G. M. (2013). The effect of incarceration on reoffending: Evidence from a natural experiment in Pennsylvania.Journal of Quantitative Criminology. doi:10.1007/s10940-012- 9191-9.

Nagin, D. S., Cullen, F. T., & Jonson, C. L. (2009). Imprisonment and reoffending.Crime and Justice: A Review of Research, 38,115–200.

Owens, E. G. (2009). More time, less crime? Estimating the incapacitative effect of sentence enhancements.

Journal of Law and Economics, 52, 551–579.

Parhar, K. K., Wormith, J. S., Derkzen, D. M., & Beauregard, A. M. (2008). Offender coercion in treatment:

a meta-analysis of effectiveness.Criminal Justice and Behaviour, 35(9), 1109–1135.

Piquero, A., & Blumstein, A. (2007). Does incapacitation reduce crime?Journal of Quantitative Criminology, 23, 267–285.

Piquero, A. R., Sullivan, C. J., & Farrington, D. P. (2010). Assessing differences between short-term, high- rate offenders and long-term, low-rate offenders.Criminal Justice and Behavior, 37, 1309 –1329.

Piquero, A. R., Paternoster, R., Pogarsky, G., & Loughran, T. (2011). Elaborating the individual difference component in deterrence theory.Annual Review of Law and Social Science, 7(7), 335–360.

Rempel, M., Green, M., & Kralstein, D. (2012). The impact of adult drug courts on crime and incarceration:

findings from a multi-site quasi-experimental design.Journal of Experimental Criminology, 8,165–192.

Rosenbaum, P. R., & Rubin, D. B. (1983). The central role of the propensity score in observational studies for causal effects.Biometrika, 70,41–55.

Royston, P. (2004). Multiple imputation of missing values.The Stata Journal, 4(3), 227–241.

Rubin, D. B. (1987).Multiple imputation for nonresponse in surveys. New York: Wiley.

Sampson, R. J., & Laub, J. H. (2003). Life-course desisters? Trajectories of crime among delinquent boys followed to age 70.Criminology, 41, 555–592.

Snodgrass, G. M., Blokland, A. J., Haviland, A., Nieuwbeerta, P., & Nagin, D. S. (2011). Does the time cause the crime? An examination of the relationship between time served and reoffending in the Netherlands.Criminology, 49(4), 1149–1194.

Sweeten, G., & Apel, R. J. (2007). Incapacitation: revisiting an old question with a new method and new data.Journal of Quantitative Criminology, 23, 303–326.

Schaub, M., Stevens, A., Berto, D., Hunt, N., Kerschl, V., McSweeney, T., et al. (2010). Comparing outcomes of‘Voluntary’and‘Quasi-compulsory’treatment of substance dependence in Europe.

European Addict Research, 60,53–60.

Spelman, W. (2000). What recent studies do (and don’t) tell us about imprisonment and crime.Crime and Justice: A Review of Research, 27, 419–494. Effectiveness of a prolonged incarceration 57 Steffensmeier, D., & Allan, E. (1996). Gender and crime. Toward a gendered theory of female offending.In Annual Review of Sociology, 22, 459–487.

Stevens, A., Berto, D., Heckmann, W., Kerschl, V., Oeuvray, K., Van Ooyen-Houben, M. M. J., et al.

(2005). Quasi-compulsory treatment of drug dependent offenders: an international literature review.

Substance Use & Misuse, 40, 269–283.

Tollenaar, N., Meijer, R. F., Huijbregts, G. L. A. M., Blom, M., & el Harbachi, S. (2007).Monitor Veelplegers. Jeugdige en zeer actieve Veelplegers in Kaart gebracht. Den Haag: WODC.

van Buuren, S. (2012).Flexible imputation of missing data. Boca Raton: Chapman & Hall/CRC.

Van Buuren, S., Boshuizen, C., & Knook, L. (1999). Multiple imputation of missing blood pressure covariates in survival analysis.Statistics in Medicine, 18, 681–694.

Van der Geest, V. (2011). Working their way into adulthood. Delinquency and employment in high-risk boys to age 32. Dissertation, Free University Amsterdam van Justitie, M. (2003).Plaatsing in een Inrichting voor Stelselmatige Daders. Tweede Kamer stukken (vergaderjaar 2002–2003) TK 28 980 nr. 3. Den Haag: Sdu Uitgevers.

Versteegh, P., Janssen, J., & Bernasco, W. (2003). Beginners, doorstromers en veelplegers.Tijdschrift voor Criminologie, 45, 127–139.

Vollaard, B. A. (2012). Preventing crime through selective incapacitation.The Economic Journal.

doi:10.1111/j.1468-0297.2012.02522.x.

Warner, T. D., & Kramer, J. H. (2009). Closing the revolving door? Substance abuse treatment as an alternative to traditional sentencing for drug-dependent offenders.Criminal Justice and Behavior, 36, 89–109.

Wermink, H., Apel, R., Nieuwbeerta, P., & Blokland, A. J. (2012). The incapacitation effect of first-time imprisonment: a matched samples comparison.Journal of Quantitative Criminology. doi:10.1007/ s10940-012-9189-3.

Wermink, H., Blokland, A., Nieuwbeerta, P., Nagin, D., & Tollenaar, N. (2010). Comparing the effects of community service and short-term imprisonment on recidivism: a matched samples approach.Journal of Experimental Criminology, 6, 325–349.

Wolfgang, M. E., Figlio, R. M., & Sellin, T. (1972).Delinaquency in a BirthCohort. Chicago: University of Chicago Press Nikolaj Tollenaaris a researcher at the WODC (Research and Documentation Centre). He is project leader of the national frequent offender monitor. His research interests include recidivism prediction models, criminal careers of chronic offenders, and research methodology.

André M. van der LaanPhD is a Senior Researcher at the WODC (Research and Documentation Centre) of the Dutch ministry of Security and Justice. He studied developmental psychology at the University of Leiden and took his PhD at the University of Groningen on the topic of“Defiance and Delinquency”. His research interests are in developmental and life-course criminology, criminal careers of chronic offenders and (explanations of) trends in juvenile crime rates. Furthermore, he is interested in the effects of juvenile sanctions and juvenile experiences of judicial sanctions.

Peter G. M. van der Heijdenis Professor of Statistics for the Behavioral and Social Sciences at Utrecht University and Professor of Social Statistics at University of Southampton, GB. At Utrecht University he is head of the Department of Social Sciences, Methodology and Statistics. For the Ministry of Safety and Justice he chairs the steering committee on research into recidivism. He teaches courses on multivariate analysis and multilevel analysis. He carries out research in surveys on sensitive questions and in population size estimation.

58N. Tollenaar et al. R epro duce d w ith p erm is sio n o f th e c o pyrig ht o w ner. F urth er r e pro ductio n p ro hib ite d w ith out p erm is sio n.