Timed Recidivism. In search for critical periods to supplement interventions.

(1)

TIMED RECIDIVISM

IN SEARCH FOR CRITICAL PERIODS TO

SUPPLEMENT INTERVENTIONS

(2)

TIMED RECIDIVISM

IN SEARCH FOR CRITICAL PERIODS TO

SUPPLEMENT INTERVENTIONS

TAMÁS HODOZSÁN

Hodozsán T, Timed Recidivism. In search for critical periods to supplement interventions. Degree project in Criminology 30 Credits. Malmö University: Faculty of Health and Society, Department of Criminology, 2020.

ABSTRACT

Assessing risk had always been the key focus when it comes to recidivism. Using risk assessment instruments, it is possible to predict the outcome of recidivism dichotomously. These measures, however, can only predict between 70-80 percent of validity, and they specify only levels of risk (low-medium-high), but not time. Therefore, the aim of this study is to define time of recidivism to supplement risk assessment with a possible new actuarial approach and fill out gaps in the existing literature. To do so a systematic literature review was

(4)

INTRODUCTION

Studying post release performance of incarcerated individuals has always been a great interest for policymakers and practitioners within the field of criminology (Ostermann, 2015). The main goal of correctional systems (prison, social

services) and policymakers, to reorient prisoners into a life without crime

(Skardhamar and Telle, 2012). People released from any correctional system may face barriers that inhibit them from successfully obtain that life. (Stansfield and Williams, 2014)). There are numerous articles that target factors both static, thus unchangeable, that refer to events in the past and dynamic, that refer to current changeable characteristics; both on individual and structural level, that has a connection to recidivism (Cuevas et al., 2019, Andrews et al., 2006, Ropes Berry et al., 2020) . Many of them had come to the conclusion that a large percentage recidivate in a relatively short time, most of them highlighted the importance of the first year. (Stansfield and Williams, 2014, Langan and Levin, 2002). While these studies found support of risk factors and recidivism, unfortunately, there is lack of research related to the exact timing of recidivism. Which in fact would improve the understanding of time of relapse; and therefore, would result in better interventions. However, this faces a major problem, as there is no current method that focuses time explicitly, thus highlights the need for a new actuarial approach. Such approach would create categories (intervals) by the frequency recidivism occurs, as defining exact time points (exact day) is impossible. In other words, these categories would serve as critical periods of recidivism. The speed of which reoffending occurs also suggests more targeted intervention and more intense supervision in the early stages of the “at-risk” period (Stansfield and Williams, 2014). Which makes time as a risk factor of recidivism. Consequently, prioritizing research that strictly targets the timing of failure (the exact time of relapse) can improve the understanding of how recidivism related to time, and how would it improve the practical part of correctional systems and policymaking.

Aim and research questions

The aim of the current study can be defined as “developing an intervention

technique that fits into the RNR model as a time frame, in order to make the interventions more intense at critical (the highest chance to relapse) periods” – The author. In order to do so, a Systematic Literature Review (SLR from now on)

was conducted, guided by the following questions:

1. How well the timing of recidivism can be predicted using the current available literature?

2. What factors affect specifically the time of recidivism? 3. How is it possible to measure and predict time of recidivism? 4. Are there any methodological challenges that affect the efficiency of

predictions?

5. Is it possible to create a new actuarial approach that only relies on time as risk factor?

(5)

PRIOR RESEARCH

In criminal justice research recidivism had been used dichotomously as an outcome measure (Cuevas et al., 2019).Yukhnenko et al. (2020) conducted a meta analytic review on risk factor for recidivism on non-custodial sentences and found that, the most common static risk factor domains are age, gender, ethnicity,

educational problems (not having a diploma or having high educational needs)

and criminal history. Regarding the most common dynamic risk factor domains,

substance misuse, low income, association with antisocial peers (antisocial

friends or lack of pro-social ones), mental health needs, and marital status (being single or divorced) found to be the most useful for prediction. Having offence types dichotomously grouped (general and violent), Eisenberg et al. (2019) identified 16 risk domains on forensic outpatient population. This subgroup differs from the inpatient one, as they have less severe mental conditions (i.e. less major mental disorders), on the other hand, disorders regarding substance abuse and impulsive behavior is much higher. Regarding general recidivism 13 out of 16 risk domains were positively associated with recidivism. For instance, antisocial pattern of behavior, race and ethnicity, criminal history. For violent recidivism Criminal History, Antisocial Pattern, Antisocial Attitudes, Criminal Friends (antisocial peers), Substance Abuse, Education/Employment, Family/Partner, Personal/Psychological Problems, Living Environment, and Leisure. However, many of these domains consisted both static (mostly history of.) and dynamic factors (current). After tracking 272111 prisoners following their release Langan and Levin (2002) found similar results regarding the offenders’ characteristics and other static factors such as the number of prior arrests was a good predictor of recidivism. So was the incarceration time. Offense type is also a good predictor, but only for the same offense.

In recent years, examining the time of failure has become more relevant. Langan and Levin (2002) pointed out that the level of risk of recidivism is the highest in the first months after releasing and it decreases over time until it reaches the onset risk level of a non-offender (Kurlychek et al., 2006). This pattern can be seen especially in adults as Durose et al. (2014) indicates in a 3-year period, 2/3 of adult convicts recidivate in the first 3-year. Numerous risk factors were identified that has a connection to either the likelihood of recidivism or the time of recidivism (Andrews et al., 2011). For instance, Type of sanction, individuals sentenced to prison are more likely to relapse into crime then on probation (Spohn and Holleran, 2002)). Incarceration time (Huebner and Berg, 2011). Gender differences can also be found regarding recidivism, (Benda, 2005)). There are also youth specific factors such as, higher levels of childhood

(6)

THEORETICAL FRAMEWORK

Assessing risk of recidivism is has a great concern to the criminal justice system, judicial and correctional services, and to the society as well (Sjöstedt and Långström, 2001). Recidivism in a nutshell can be defined as returning to crime after a prior offense. More formally, Maltz (1984) defines recidivism as to return to criminality after the execution of a formal sanction (e.g.: reconvictions, violations of sentences, arrests for new crimes). As it can be seen, recidivism comprises two elements, that is (1) the commission of an offense, (2) by an individual already committed at least one other offense (Harris et al., 2011). Risk factors (or predictors) can be defined as any factor that is correlated with crime. In other words, statistically associated with crime involvement. The theoretical basis used to develop risk assessment instruments is the risk–need–responsivity (RNR) model (Campbell et al., 2018). Based on cognitive social learning, general

personality models and scholars the model identified criminogenic risk factors both on micro and individual level that increase risk for recidivism (Andrews and Dowden, 2007).

1. Risk: This principle stands for matching the intensity of the program regarding the risk assessed by prior RAIs. (Risk assessment instruments), consequently, minimal intervention for low-risk offenders, and increased intervention for high-risk offenders.

2. Need: The need principle stands for targeting criminogenic needs of offenders, or those that are functionally related to crime, in other words criminal behavior.

3. Responsibility: Responsibility stands for matching the style and mode of the intervention to the offenders learning style and abilities.

The RNR model is a guideline for offender assessment and treatment and had been broadly discussed in the past years. It “underlies most of the risk-need

assessment instruments” (Andrews et al., 2011, p. 735) and assumes that

adherence to the particles, that is risk need and responsivity will result in reduction of recidivism rates (Grieger and Hosser, 2014). In other words “The

goal of this model is to determine the risk an offender poses for recidivating, target dynamic criminogenic needs of each offender, and deliver treatment in a way that an offender will be responsive and successful” (Campbell et al., 2018, p. 527).

These risk factors can be divided into two parts:

1. Big-Four Risk Factors: That strongly predict recidivism (1) History of Antisocial Behavior, (2) Antisocial Personality Pattern (3) Antisocial Cognition (4) Antisocial associates. 2. Moderate Four Risk Factors: That have moderate impact on

recidivism (5) family and marital circumstances, (6)

school/work, (7) leisure/work (8) substance abuse. (Andrews & Bonta, 2010a)

(7)

By looking at the risk factors of the RNR (especially the big four) it is notable that many of them are connected to the early stages of the developmental period of an individual and elements regarding adulthood. The developmental period is a key window for establishing self-control - as the learning process takes place in

both the primary and secondary socialization - via effective parenting. Regarding

social learning theory, „Behavior is acquired or conditioned by the effects,

outcomes or consequences it has on the person’s environment.” (Winfree Jr et al.,

1994, p. 149). Incomplete socialization or ineffective parenting (deficient attachment, discipline and monitoring) causes lack of self-control, which is – according to the theory – is a predictor behind criminality (Gottfredson and Hirschi, 1990). Secondary socialization refers to the phase of the individual when encountered by peers, role-models, beliefs outside family. At this point bonds to the society starts to shape (social bond theory), if these bonds are weak or broken, delinquent acts happen (Hirschi, 2002). In other words, self-control is part of (altogether with morality) crime propensity. According to Haar and Wikström (2010) criminal actions are outcomes of the perception-choice process (cognitive evaluation through a moral filter) of the interaction between criminogenic setting (temptations, provocations) and the person’s crime propensity (self-control, morality), which is the definition of SAT (Situational Action Theory). Targeting causes rather than predictors, as they can only forecast criminality, according to Wikström and Treiber (2017) is an effective way to reduce criminality. In other words, reducing crime propensity, will lead to less influence of the settings on people. On the other hand, focusing on risk factors as they can forecast criminality is also a great way to reduce crime. If this forecast can predict efficiently,

(8)

METHODS

The chosen method for the aim and research questions was a systematic

literature review. According to (Fink, 2019) a “stand-alone” literature review must be systematic, explicit comprehensive, and reproducible. Systematically following a methodological approach, explicitly explains the procedure of how it was

conducted, comprehensively including all relevant material, as a result of

reproducible research. The purpose of such studies is to identify evaluate and synthetize on already existing research written by researchers, practitioners, and

scholars. Research regarding failure of time had become a primary target in the academic world, as previous literature showed, that the relapse rate is the highest in the first year, especially in the early months. Therefore, conducting a

systematic literature review on the topic of time seems appropriate. The search was carried out by using “MAU Libsearch”. This search enables to use multiple databases at the same time (including open source, but peer-reviewed material as well). This multiple sided approach allowed the author to get a wide range of material on this particularly sensitive topic.

Procedure

The current SLR was carried out on the 15th of April 2020. Three different models (as the first indicated) were used to maximize the quantity and the value of the search. The search terms for each study slightly differed in order to maximize the effect and due to the high level of contamination. Due to the use of various statistical methods, previous research increased the author’s interest to include only quantitative studies. Furthermore, as the purpose is to review the current knowledge on the topic, it seems appropriate to review articles only from the past 20 years.

The first model used the following parameters “recidivism AND timing or time

AND criminology”. Which resulted in (N=14) studies. Due to the lack of studies a

slight change was necessary in the searching process as was stated above. This change was based on the result of this model. In other words, the first cycle served as a pretest in order to define the search string of the second cycle of the SLR.

Note: Criminology was added due to the contamination as studies in biomedicine also occurred.

The second model used the following modified parameters “recidivism AND

survival analyzes AND criminology”. This changed increased the number of

studies (N=183), but a personal notice indicated a second slight change in the keywords regarding the search, therefore a third model was added to once again maximize the results.

The third model used the following parameters ““recidivism AND time of

failure AND criminology”, which resulted in (N=67) studies.

Note II: If overlap was detected, the study got placed into the first group it matched in.

(9)

Overall (N=264) studies were included in the abstract review in order to define the inclusion and the exclusion categories. Studies that were not dealing with recidivism in the first hand were excluded from the SLR. So were studies, that used any other approach than quantitative. Thirdly, regarding the publishing dates, studies that were published before 2000. January 11. were also excluded from the study. Lastly, and yet most importantly, studies that did not deal with time in some way were excluded. In other words, if the study did not have a connection to the time of recidivism were excluded from the study, therefore the inclusion criteria can be defined as:

1. The article must deal with recidivism

2. The article should deal with time in some way 3. The article must be a quantitative study

4. The article must have been published in the past 20 years

Note III: Due to the high level of quantitative connection, basic terms must be defined before interpreting the results.

Figure 1. Flow chart of data selection.

1_{Publishing date: The main reasons behind this inclusion criteria are to avoid studies from the}

past that have been either revised or inclined. Secondly, as for an actuarial approach it is always better to work with the newest studies and articles.

Model I

"recidivism AND timing or time AND

criminology” Model II "recidivism AND survival analyzes AND criminology” Model III “recidivism AND

time of failure AND criminology” Overall studies N=264 Included N=14 Excluded N=250

(10)

Basic statistical terms

Receiver Operating Characteristics (ROC) analysis is a statistical technique to

determine the predictive accuracy between two (a.k.a. dichotomous) variables using correlations and identifying positive-, negative-, false positive- and false negative- (Bennell and Jones, 2005)) predictions (cases). In other words (Swets, 1988) a Roc curve is a graphical plot, a measure of discrimination accuracy, that determines whether predictions are depended more than by chance. This plot enables a direct visual comparison of two or more tests on common set of scales at cut points of prediction instruments (Stansfield and Williams, 2014).

Odds Ratio (OR)

The odds ratio is a number that indicates the probability of the outcome of a dichotomous variable. If the odds ratio Exp(B) (in statistical programs Exp(B) refers to the odds ratio) increases in a model (and of course the other variables remain the same) that means that variable has a higher chance of predicting the outcome of the dichotomous variable. The odds ratio can be greater than zero if the B (the number that indicates the predictive power within a model) is positive. By using this approach, the probability of the outcome of the dichotomous variable can be calculated by using the formula:

Note: The following example was made by using the variables in spss

P= e

constant+or-B value of the variable x the value of the variable

1+econstant+or-B value of the variable x the value of the variable

Note II: The equation above has only 1 variable, but multiple can be added, but all must be significant.

(11)

Kaplan-Meier Survival Analyzes (KM):

Survival analysis (or event-history analysis), was used originally when studying mortality by estimating probabilities of death rates over time (life-course). Survival analysis is a technique that - if used sporadically – can be applied to criminology to estimate the probability of offending and recidivism in order to find time (which is also known as survival time or failure time in survival analysis) in recidivism. (Stansfield and Williams, 2014)). Survival time is an interval of a starting point (in this case release from prison or a program) and an endpoint (in this particular case, recidivism). In order to conduct a survival analysis, one must define a follow-up period. In this follow-up period hazard rate and survival rate are calculated every other time, when one recidivism happened. Must be noted that survival analysis describes the survival according to only one factor under investigation. There are studies in which the main goal is not to give a survival curve, but to examine how much survival depends on different factors. In such cases, the Cox regression calculation, also known as the Cox proportional hazard model, is most commonly used. This assumes that the so-called hazard function (essentially instantaneous risk) can be given as a product of a factor that depends only on the follow-up time and the exponential functions of the

explanatory variables.

Cox proportional hazard:

The purpose of a cox proportional hazard is evaluation simultaneously of effects on several different factors on survival. Consequently, it makes examinations of how specified factors have an influence on an event (in this case, recidivism) at a certain (particular) point in time possible. These factors are the covariates (the predictor variables) in survival analysis (Cox, 1972).

Hazard Ratio (HR):

The hazard ratio can be used to express effects in studies comparing treatments when statistics which describe time-to-event or survival analyses are used. Applied to the field of criminology, hazard rate is the risk of recidivism at a specific time (Hill et al., 2008). In other words, if a hazard ratio above is 1 indicates positive association of the covariate with the event probability, which means negative association with the length of survival (Cox, 1972).

Comparison:

(12)

Ethical considerations

Ethical consideration must always be taken into consideration when conducting a research, that is no matter if quantitative, qualitative or in this case a Systematic Literature Review. These considerations serve as protection during the research for both the participants and the conductor of the study according to the 6 ethical principles. Regarding the current study, due to the chosen method (systematic literature review) these principles can be neglected. The reason of this neglection lies in the availability of the studies. Once a study is available in one of the online databases, has been already accepted, that makes it ethically approved by officials.

RESULTS

Fourteen studies were included in the final model (see Table 1) as they were the only ones that met with the inclusion criteria. It must be noted that, samples of the studies differ regarding the population. The samples varied in number (from N=58 to N=192556) and in the individuals they were focusing on. For instance, first-time incarceration, violent offenders, murderers and youth offenders. Moreover, some studies focused only on a small portion of the population for example: immigrants and first-born natives; or used a unique and costly approach such as, neuroimaging data. Secondly, they also differ in the used methods

(different risk assessment tools) and statistical analyzation processes (AUC-ROC, Bivariate and Multivariate regression analyzes, Multinomial regression analysis).

The analyzation process therefore consisted a controlled search, first and foremost to see if specific (exact) times of relapses are detectable (via survival analysis). Secondly, for factors that has a relation to this time in other words, either accelerates or slows (Hazard Ratios). Thirdly, factors that has increased likelihood ratios (Odds Ratios) regarding recidivism are also mentioned. Lastly, some additional not expected findings were also detected and mentioned.

For example, Cuevas et al. (2019) conducted multiple dichotomous and multinomial regression analysis respectively on a sample of 2523 participants with the purpose of how the central eight risk factors are related to the timing of recidivism. In order to do so, different time intervals were created (0-30 days; 30-90 etc.) to see which factors (or the combination of them) predict more efficiently regarding these stamps. As it was mentioned, many articles used a unique sample, to fill up a hole in the literature. These samples were only focusing on either minorities or a specific offense. For instance, (Stansfield and Williams, 2014) conducted a Cox proportional hazard (survival analysis) on a sample of 29317 perpetrators, regarding only domestic violence by using a RAI. While (Ramos and Wenger, 2020) mainly focused on incarcerated immigrants and native-born

individuals and how they will possibly relapse after release from prison by using mainly linear-regression models, thus odds ratios by focusing on risk factors that’s connection to recidivism had already been found. Moreover, targeting

developmental perspective Ozkan (2016) continuous survival analysis was conducted on a developmental perspective to investigate how developmental factors such as psychosocial maturity (temperance, responsibility and perspective)

(13)

As it can be seen, these studies are related to time somehow, thus got included in the final model however, some studies should have been excluded due to some specific approach. For example,(Kurlychek and Kempinen, 2006) also focused on time of recidivism and risk factors, but their mainly focus was on testing the applicability of a new policy, and how that policy will affect the time of relapse.

(14)

Table 1. Included articles with sample sizes and the statistical methods they used. (Continues).

Article Author Date Statistical Approach Sample

Dynamic risk factors and timing of recidivism for youth in residential placement

Cuevas; Wolff; Baglivio. 2019 Cox proportional hazard (survival analysis) Multinomial logistic regression

2523 Post-release Employment and Recidivism in

Norway

Skardhamar & Telle 2012 Cox proportional hazard (survival analysis) 7476 How do Former Inmates Perform in the

Community? A Survival Analysis of Rearrests, Reconvictions, and Technical Parole Violations

Ostermann 2015 Cox proportional hazard (survival analysis) 12187

Reoffending among serious juvenile offenders: A developmental perspective

Ozkan 2016 Cox proportional hazard (survival analysis) 1354 Predicting Family Violence recidivism using the

DVSI-R: Integrating survival analysis and Perpetrator characteristics

Stansfield &Williams 2014 ROC-AUC

Cox proportional hazard (survival analysis)

29317

Which Risk Factors are Really Predictive?

An analysis of Andrews’ and Bonta’s “central eight” Risk Factors for Recidivism in German youth correctional Facility inmates

Grieger &Hosser 2014 ROC-AUC

589

Time-Free effects in Predicting recidivism Using Both Fixed and variable Follow-Up Periods

Flores; Holsinger; Lowenkamp & Cohen

2017 ROC-AUC

(15)

Immigration and Recidivism: What Is the Link? Ramos & Wenger 2020 ROC-AUC

192556 The Intersectional Effects of Race and Gender on

Time to Reincarceration

Berry et al. 2020 ROC-AUC

21462 Effortful control, negative emotionality, and

juvenile recidivism: an empirical test of DeLisi and Vaughn’s temperament-based theory of antisocial behavior

Baglivio et al. 2016 Cox proportional hazard (survival analysis) 27713

Prediction of recidivism in a long-term follow-up of forensic psychiatric patients: Incremental effects of neuroimaging data

Delfin et al 2019 AUC-ROC Neuroimaging data

44

Low self-control and parole failure: An assessment of risk from a theoretical perspective

Langton 2006 Logistic regression 4146

Juvenile homicide offenders: A 35-year-follow-up study

Heide 2019 Logistic regression 59

Back on the Swagger

Institutional Release and Recidivism Timing Among Gang Affiliates

Caudill 2010 Cox proportional hazard (survival analysis) Royston-Parmar Regression model.

(16)

FINDINGS

Regarding the analyzation process, findings of the 14 (with a total sample of N= 329018) included studies are presented in 3 different ways. (1) Findings on

Time, that refers to exact timeframes of recidivism. (2) Findings on factors that affect the speed of relapse, that refers to factors, that have effect on the speed of

recidivism (hazard ratios). (3) Findings on the likelihood of recidivism, that refers to factors, that increase or decrease the likelihood of recidivism.

Findings on Time

The most useful way to find key time-patterns is via survival analysis, as it estimates the probability of recidivism, regarding time. In all of the 14 days, only 4 provided any kind of exact timeframes in recidivism. Although, all of the came to the conclusion that the first year is a key window for recidivism (Skardhamar and Telle, 2012, Grieger and Hosser, 2014), because the relapse rates were the highest, but when it came to the identification of shorter frames the studies differed.

For instance, Caudill (2010) in a 5-year long follow-up study found that, participants are at low risk to recidivism right after release then it peaks between the interval of 3-6 months, before it decreases again after 12 months. Ostermann (2015)also highlighted the first 6 months as the most challenging time for

reintegrating an individual. Similar results were found by Stansfield and Williams (2014). Within their study, the first 200 days represented this key window for reoffending. However, when they specifically analyzed domestic violence as a new family violence offense, the highest relapse rate was located in the first 100 days. To be more specific, there was a 5% relapse rate in the first 63 days. Dividing these results into different set of times can result in a brighter picture. The same method was used by Cuevas et al. (2019). Within their study, different time intervals (1. <30 days; 2 30-90 days; 3. 90-180 days) were created and participants (who did reoffend) were sorted into these groups by a regression model. In this multinomial regression analysis, the results indicated that when measures by risk, youth are more 1,567 time more likely to be rearrested in the first month and 1.261 times between the 31-90 days interval. Speaking of risk, in many cases risk is measured via RAIs, that are based on ROC-AUC measures. While the previously mentioned studies highlight the importance of the first six month after release, a study specifically on the application of a RAI with regards to time conducted by Flores et al. (2017) found contradiction in the applicability of these measures, as their predictive validity remains stable just after the 7th month (AUC=0.74). On the other hand, (Delfin et al., 2019) increased the predictive power of an instrument by studying cerebral blood flow (on forensic patients), a unique approach using neuroimaging data.

(17)

Findings on variables that accelerates recidivism

Multivariate analyzes are needed in order to calculate hazard ratios, a number that indicates the risk of recidivism at a specific time. In most of the cases this specific time is defined as “more quickly” or “slower”. In this paragraph, factors that accelerates recidivism are discussed.

First of all, many studies came to the conclusion that if the perpetrator is male and belong to an ethnic minority group, are at higher risk to be rearrested

faster.(Caudill, 2010, Ropes Berry et al., 2020, Cuevas et al., 2019, Ramos and Wenger, 2020, Baglivio et al., 2016a). Examining ethnic minorities specifically, black men are more likely to recidivate more quickly. Furthermore, when comparing immigrants and first-born natives, the hazard of reoffending in the three-year-period was nearly 17% lower for immigrants as opposed to

nonimmigrants (HR: 0.831). In other words, immigrants reoffend with a lower rate than native born. (Ramos and Wenger, 2020).

Secondly, from a developmental perspective, 2 components of the

psychological maturity (temperance, responsibility and perspective) are crucial predictors of time of both violent and income-related offences. All of the components had a larger impact on income-related offences than violent ones. Moreover, behavioral problems during childhood was not significant, while exposure to violence had a minor effect on the timing of income-related offences. Lastly, the involvement of delinquent peers has the largest effect on recidivism regarding time (Exp(B)=3.43; HR=2.15) (Ozkan, 2016). Same results were found by Langton (2006) as the results showed that low self-control had no significant effect on months to parole failure, moreover the hazard rate only decreased 0.6% in case of “higher” self-control. Furthermore, according to (Baglivio et al., 2016a) more impulsive youth with beliefs of out of control, antisocial behavior

reoffended faster. Secondly, juveniles without positive adult relationships, angry and having more hostile interpretation of others’ actions would reoffend faster. Most importantly, both low EC and high NE were related to a quicker time to re-offend (HR=0.876; 1.131 respectively) even after demographics (age/race) and individual risk factors (substance abuse, antisocial peers, age at first offense, etc.) were controlled.

Interestingly, first time offenders showed a 24% reduction in the hazard rate (HR=0.76), which means first time offenders experience some kind of deterrent effect (Langton, 2006). Such deterrent effect was concluded by (Skardhamar and Telle, 2012). Within their study regarding time of recidivism, the results indicated that, the recidivism rate is much lower for those who obtained a job post-release

(18)

Risk factors, such as parental status (HR=1.69) and crime type (HR=1.31) had a moderate effect on time of recidivism (Ramos and Wenger, 2020). Regarding such factors, in the study made by Cuevas et al. (2019). Within their study 2 patterns that made youth recidivate more quickly were identified (1) Youth measured by more risk across domains, were more likely to fail early, which highlights the importance of multisystemic interventions involving the individual, peers, school, and family (Borduin et al., 1990); and (2) measures of criminal history were unimportant in determining specific time intervals at which youth reoffended.

Findings on likelihood of recidivism

The third approach was to search for factors that increased or increases the likelihood of recidivism. In this paragraph odds ratios of factors that has a connection to recidivism are mentioned.

First of all, regarding cox-regression models, if the individual is male (Exp(B)=1,783), black or Hispanic has increased likelihood for rearrests

(Exp(B)=1,408 and 1,268 respectively (Cuevas et al., 2019, Ramos and Wenger, 2020, Ropes Berry et al., 2020, Baglivio et al., 2016a). Moreover, increased likelihood was shown for more prior felonies (Exp(B)=1,172). On the other hand, somehow unexpectedly, age of onset and age at release had decreased odds ratios (Exp(B)=0,855 and 0,919 respectively) (Cuevas et al., 2019). Almost the same result was shown by Caudill (2010) regarding prior adjudications (Exp(B)=1,15). Prior prison Commitment turned out to be another risk factor (ExpB=1.445) (Ramos and Wenger, 2020). Conversely, these odds are lower if the ex-inmate is older (Exp(B)=0.960), has a high school diploma (Exp(B)=0.845), married (Exp(B)=0.928), employed (Exp(B)=0.816), or has a current conviction (Ramos and Wenger, 2020). Returning to the old neighborhood also increases the chance of relapse (Heide, 2019).

Secondly, from the other perspective, that is time served in the correctional institute, participation in treatment programs increasing the chance of survival (Exp(B)=13.5 and 12.25 respectively). Similar effects were concluded by (Ostermann, 2015)), those who were released conditionally (parole supervision) were more unlikely to relapse when compared with unconditional release.

Methodological Issues regarding the articles

When reviewing all the articles, the author encountered some methodological issues, that must be mentioned.

First of all, as it was already mentioned, (Flores et al., 2017) found that the predictive validity remained stable only after 7 months. Which in fact questions the applicability of the RAIs. Must be noted, that they used one specific

instrument, so this observation, cannot be applied to other instruments, however, without proof, all risk assessment instruments may be affected.

(19)

Thirdly, in one case, the author noticed misinterpreted data regarding Odds ratios, which questions the creditability of the study.

Lastly, regarding regressions models, some studies included many factors at once, but must be known that they can be overfitted. When this happens, coefficients may overestimate the effect.

DISCUSSION

In this systematic literature review 14 articles were analyzed in order to find patterns for the exact time of recidivism. The importance of this approach is to get a brighter picture on recidivism, when time comes to question. Secondly,

numerous risk factors were also identified and reviewed which both strengthened and weakened previous findings.

First and foremost, most of the findings were consistent with previous

literature, which pointed out that there are numerous factors regarding recidivism such as gender, race, residential placement, age at first offense, and prior felonies (Langan and Levin, 2002, Ropes Berry et al., 2020, Cuevas et al., 2019,

Ostermann, 2015), that are always present and can be relied on when conducting risk assessment, but that is not always the case. For instance, many studies used a unique approach in order to identify specific factors, that can be accounted for recidivism such as, gang affiliation (Caudill, 2010), being first generation native

born (Ramos and Wenger, 2020), mixed race and gender (Ropes Berry et al.,

2020), negative emotionality and effortful control (Baglivio et al., 2016a), speaking of control, - a classic - self-control (Langton, 2006) post-release

employment (Skardhamar and Telle, 2012) or cluster B personality disorder

(Delfin et al., 2019). On the other hand, certain factors can increase the likelihood of desistance, such as – in some cases – first time offending (Langton, 2006), post-release employment (Skardhamar and Telle, 2012).

Secondly, these factors always predict better if they are used together (Grieger and Hosser, 2014, Flores et al., 2017), that singlehandedly, which raises the conclusion that, one single factor cannot ever be the reason for recidivism. However, when conducting regression analyzes (which are the most used

statistical approaches regarding recidivism) one must have control of the included variables since, regression models could be overfitted, which can lead to

coefficients that overestimate the effect. In other words, a particular predictor can have a bigger effect on a phenomenon that it really has and also can infiltrate

(20)

However, as Flores et al. (2017) indicated AUC-ROC values based on a 7-month (or longer) follow-up period may suffice in giving researchers and practitioners the information they need, but before that, the predictive validity sometime not reliable. In other words, the predictive validity of an instrument remains stable just after the 7th month, indicating the need for either multiple risk assessment sessions at different time-intervals or the foundation of a new

approach.

Speaking of time, the main purpose of this SLR, was to look for different (exact) time patterns for recidivism, meaning different intervals as critical periods where the frequency of recidivism is increased. The results of multiple survival analysis showed that most of these periods can be found in the first year especially within the first half. Ropes Berry et al. (2020) found that those who will recidivate do so in the first half of the year and more than half do it by the end of it.

According to Stansfield and Williams (2014), the first 200 days is a key window for reoffending (regarding specifically domestic violence), and within that period even the first 100 (more precisely in the first 61 days) had the highest relapse rate. Similar results were shown by Cuevas et al. (2019) when predict the time of recidivism. Global risk score (generated by the RNR risk factors) increased the likelihood of being rearrested both, but more in the first month (30 days) and little less between the second and third month (31-90 days). Adding gang affiliation to the equation, (Caudill, 2010) found that, the relapse rate is low after the release, and peak between 3-6 months, which contradicts with the importance of the first month but raises it for the first 6. It must be noted that these results must be interpreted carefully, as they can vary regarding offense type used materials and risk factors that had been measured. Although, most of the recidivism happens in the first year (Langan & Lewin, 2002 (Langan and Levin, 2002, Cuevas et al., 2019, Stansfield and Williams, 2014, Ropes Berry et al., 2020, Caudill, 2010, Olver et al., 2012) that does not mean that the second year is unimportant. Many studies showed of the second and the third year (Skardhamar & Telle

(Skardhamar and Telle, 2012, Stansfield and Williams, 2014, Caudill, 2010, Langan and Levin, 2002). Consequently, interventions prison services,

policymakers and social workers should target these periods specifically in order to maximize the deterrent effect of intervention programs, meaning the

increasement of the number of sessions with each individual.

The significance of the first half year contradicts with the usefulness of the risk assessment tools as their predictive validity (mentioned above) remains stable only after the 7th month thus indicating the need of either the (also mentioned) risk assessment sessions at multiple times (especially within this period) or the

creating a new actuarial approach that only relies on time as critical periods to supplement risk assessment, where not risk assesses time, but time assesses risk.

Furthermore, these periods could also be grouped by offense types, they can accelerate the time of recidivism (Ropes Berry et al., 2020). Regarding this acceleration, there are also numerous factors that play a significant role. For instance, (Cuevas et al., 2019) identified two patterns that can be accused of such acceleration, first highlights the importance of dynamic risk factors such as

(21)

developmental factors such as psychological maturity that are accounted for accelerated recidivism, especially income-related and general offenses. These findings showed the importance of developmental factors regarding the time of recidivism; therefore, one must take these factors into consideration when timing is in question.

Notable, that exact comparison of the odds ratios and hazard ratios must be avoided due to the difference between their formulas (the denominator when

calculating odds ratios is equal to the number of non-recidivists, while in case of hazard ratios, it equals the number of recidivists divided by the total sample size)

and interpretation (Flores et al., 2017). In other words, they are not the same when it comes to interpreting results, but as it was mentioned before, the over-fitness of the data must be avoided as well.

The results overall indicated the need to focus on the first 180 days of release, especially between the 3-5 month of interval. Regarding this timeframe, the intensity of programs should be increased. Secondly, despite the fact that

participation in an intervention program targets dynamic factors, the importance of static factors when it comes to prediction cannot be neglected, as they might be better predictors, thus indicating the actuarial part of the assessment.

CONCLUSION

For the author’s best knowledge this is the first literature review that targets exact time patterns as a purpose. Within this paper a systematic literature review was conducted in order to find exact time patterns in recidivism. First and

foremost, the results indicated that most of the incarcerated individuals recidivate within the first year of the release. Moreover, the relapse rate was even higher in the first half of that year. To be more precise the end of first month then

somewhere between the second and third and at the end of the 6th are critical periods when it comes to identify the timing of recidivism. Consequently, interventions should be more intense in these periods in order to maximize the deterrent effect of such programs.

LIMITATIONS

(22)

FUTURE STUDIES

These critical periods are a good start to for future research in order to better understand the time of recidivism, therefore studies should prioritize them. Especially, because the results indicated, that risk assessment instruments have a decreased predictive validity in the early months. The risk that serves as the core of treatment programs. Combining these results, future studies either should take the advice from Flores et al. (2017) to reassess risks periodically or create a new actuarial approach where the standardized risk score is replaced by time. As an initiative, the author would like to take a suggestion for that approach.

The first approach consists a search in preexisting data in police records, to look for exact times. Notable, that those times, will be much precise as the exact date of relapse must be documented, which allows the researchers to calculate the exact survival time of each offender. At this point multiple timestamps can be created using frequencies of recidivism. As a result, a number of different groups can be created in which potential risk factors can be reviewed of how they are related to the time. Unfortunately, this approach requires too much time and workforce to get results.

The second approach is a randomized hierarchical cluster analysis (for 3-4-5-6 clusters) to see if patterns can be found. If yes, each cluster can be reviewed and explained of how potential factors were sorted for that specific cluster. If no, - that is randomized results – that cluster must be excluded.

Furthermore, such new approach can replace the dichotomous oppression, as Cuevas, Wolff and Baglivio (2019) indicated, these times frames should be grouped by offense types, not just dichotomously (violent and non-violent

(sometimes the term “general” is used), but rather using any kind of offense type, such as burglary, theft, sexual abuse etc., in order to get a brighter picture on the time of recidivism on each offense type.

(23)

REFERENCES

ALLISON, P. D. 2010. Survival analysis using SAS: a practical guide, Sas Institute.

ANDREWS, D. A., BONTA, J. & WORMITH, J. S. 2006. The recent past and near future of risk and/or need assessment. Crime & Delinquency, 52, 7-27.

ANDREWS, D. A., BONTA, J. & WORMITH, J. S. 2011. The risk-need-responsivity (RNR) model: Does adding the good lives model contribute to effective crime prevention? Criminal Justice and Behavior, 38, 735-755.

ANDREWS, D. A. & DOWDEN, C. 2007. The risk–need–responsivity model of assessment and human service in prevention and corrections: Crime-prevention jurisprudence. Canadian Journal of Criminology and Criminal

Justice, 49, 439-464.

BABYAK, M. A. 2004. What you see may not be what you get: a brief, nontechnical introduction to overfitting in regression-type models.

Psychosom Med, 66, 411-21.

BAGLIVIO, M. T., WOLFF, K. T., DELISI, M., VAUGHN, M. G. & PIQUERO, A. R. 2016a. Effortful control, negative emotionality, and juvenile

recidivism: An empirical test of DeLisi and Vaughn’s temperament-based theory of antisocial behavior. The Journal of Forensic Psychiatry &

Psychology, 27, 376-403.

BAGLIVIO, M. T., WOLFF, K. T., PIQUERO, A. R., BILCHIK, S., JACKOWSKI, K., GREENWALD, M. A. & EPPS, N. 2016b. Maltreatment, child welfare, and recidivism in a sample of deep-end crossover youth. Journal of youth and adolescence, 45, 625-654. BENDA, B. B. 2005. Gender differences in life-course theory of recidivism: A

survival analysis. International Journal of Offender Therapy and

Comparative Criminology, 49, 325-342.

BENNELL, C. & JONES, N. J. 2005. Between a ROC and a hard place: A method for linking serial burglaries by modus operandi. Journal of

Investigative Psychology and Offender Profiling, 2, 23-41.

BORDUIN, C. M., HENGGELER, S. W., BLASKE, D. M. & STEIN, R. J. 1990. Multisystemic Treatment of Adolescent Sexual Offenders. International

Journal of Offender Therapy and Comparative Criminology, 34, 105-113.

CAMPBELL, C., PAPP, J., BARNES, A., ONIFADE, E. & ANDERSON, V. 2018. Risk assessment and juvenile justice: An interaction between risk, race, and gender. Criminology & Public Policy, 17, 525-545.

(24)

DUROSE, M. R., COOPER, A. D. & SNYDER, H. N. 2014. Recidivism of

prisoners released in 30 states in 2005: Patterns from 2005 to 2010, US

Department of Justice, Office of Justice Programs, Bureau of Justice …. EISENBERG, M. J., VAN HORN, J. E., DEKKER, J. M., ASSINK, M., VAN

DER PUT, C. E., HENDRIKS, J. & STAMS, G. J. J. 2019. Static and dynamic predictors of general and violent criminal offense recidivism in the forensic outpatient population: A meta-analysis. Criminal justice and

behavior, 46, 732-750.

FINK, A. 2019. Conducting research literature reviews: From the internet to

paper, Sage publications.

FLORES, A. W., HOLSINGER, A. M., LOWENKAMP, C. T. & COHEN, T. H. 2017. Time-free effects in predicting recidivism using both fixed and variable follow-up periods: Do different methods produce different results.

Criminal justice and behavior, 44, 121-137.

GOTTFREDSON, M. R. & HIRSCHI, T. 1990. A general theory of crime, Stanford University Press.

GRIEGER, L. & HOSSER, D. 2014. Which risk factors are really predictive? An analysis of Andrews and Bonta’s “Central Eight” risk factors for

recidivism in German youth correctional facility inmates. Criminal justice

and behavior, 41, 613-634.

HAAR, D. H. & WIKSTRÖM, P. O. H. 2010. Crime propensity, criminogenic exposure and violent scenario responses: Testing situational action theory in regression and Rasch models. European Journal of Applied

Mathematics, 21, 307-323.

HARRIS, P. W., LOCKWOOD, B., MENGERS, L. & STOODLEY, B. 2011. Measuring recidivism in juvenile corrections. OJJDP.

HEIDE, K. M. 2019. Juvenile homicide offenders: A 35‐year‐follow‐up study.

Behavioral sciences & the law, 37, 493-511.

HIRSCHI, T. 2002. Causes of delinquency, Transaction publishers.

HUEBNER, B. M. & BERG, M. T. 2011. Examining the sources of variation in risk for recidivism. Justice Quarterly, 28, 146-173.

KURLYCHEK, M. & KEMPINEN, C. 2006. Beyond boot camp: The impact of aftercare on offender reentry. Criminology & Public Policy, 5, 363-388. KURLYCHEK, M. C., BRAME, R. & BUSHWAY, S. D. 2006. Scarlet letters

and recidivism: Does an old criminal record predict future offending?

Criminology & Public Policy, 5, 483-504.

LANGAN, P. A. & LEVIN, D. J. 2002. Recidivism of prisoners released in 1994.

Federal Sentencing Reporter, 15, 58-65.

LANGTON, L. 2006. Low self-control and parole failure: An assessment of risk from a theoretical perspective. Journal of Criminal Justice, 34, 469-478. MALTZ, M. D. 1984. Recidivism, Michael Maltz.

MENARD, S. 2010. Logistic Regression: From Introductory to Advanced

Concepts and Applications, SAGE Publications.

OLVER, M. E., STOCKDALE, K. C. & WONG, S. C. 2012. Short and long-term prediction of recidivism using the youth level of service/case management inventory in a sample of serious young offenders. Law Hum Behav, 36, 331-44.

OSTERMANN, M. 2015. How do former inmates perform in the community? A survival analysis of rearrests, reconvictions, and technical parole

(25)

RAMOS, J. & WENGER, M. R. 2020. Immigration and Recidivism: What Is the Link? Justice Quarterly, 37, 436-460.

ROPES BERRY, K., KENNEDY, S. C., LLOYD, M., VEEH, C. A. & TRIPODI, S. J. 2020. The Intersectional Effects of Race and Gender on Time to Reincarceration. Justice Quarterly, 37, 132-160.

SJÖSTEDT, G. & LÅNGSTRÖM, N. 2001. Actuarial assessment of sex offender recidivism risk: A cross-validation of the RRASOR and the Static-99 in Sweden. Law and Human Behavior, 25, 629-645.

SKARDHAMAR, T. & TELLE, K. 2012. Post-release employment and

recidivism in Norway. Journal of quantitative criminology, 28, 629-649. SPOHN, C. & HOLLERAN, D. 2002. The effect of imprisonment on recidivism

rates of felony offenders: A focus on drug offenders. Criminology, 40, 329-358.

STANSFIELD, R. & WILLIAMS, K. R. 2014. Predicting family violence

recidivism using the DVSI-R: Integrating survival analysis and perpetrator characteristics. Criminal Justice and Behavior, 41, 163-180.

STEINER, H., CAUFFMAN, E. & DUXBURY, E. 1999. Personality traits in juvenile delinquents: Relation to criminal behavior and recidivism.

Journal of the American Academy of Child & Adolescent Psychiatry, 38,

256-262.

SWETS, J. A. 1988. Measuring the accuracy of diagnostic systems. Science, 240, 1285-1293.

TAKÁCS, S. 2017. Bevezetés a matematikai statisztikába I, Budapest, Antarész kiadó.

WIKSTRÖM, P.-O. H. & TREIBER, K. 2017. Beyond Risk Factors: An Analytical Approach to Crime Prevention. In: TEASDALE, B. &

BRADLEY, M. S. (eds.) Preventing Crime and Violence. Cham: Springer International Publishing.

WINFREE JR, L. T., BÄCKSTRÖM, T. V. & MAYS, G. L. 1994. Social learning theory, self-reported delinquency, and youth gangs: A new twist on a general theory of crime and delinquency. Youth & Society, 26, 147-177. WOLFF, K. T. & BAGLIVIO, M. T. 2017. Adverse childhood experiences,

negative emotionality, and pathways to juvenile recidivism. Crime &

Delinquency, 63, 1495-1521.

WOLFF, K. T., BAGLIVIO, M. T. & PIQUERO, A. R. 2017. The relationship between adverse childhood experiences and recidivism in a sample of juvenile offenders in community-based treatment. International Journal of

Offender Therapy and Comparative Criminology, 61, 1210-1242.

WOLFF, K. T., BAGLIVIO, M. T., PIQUERO, A. R., VAUGHN, M. G. & DELISI, M. 2016. The triple crown of antisocial behavior: Effortful control, negative emotionality, and community disadvantage. Youth

Timed Recidivism. In search for critical periods to supplement interventions.

TIMED RECIDIVISM

IN SEARCH FOR CRITICAL PERIODS TO

SUPPLEMENT INTERVENTIONS

TABLE OF CONTENTS

TIMED RECIDIVISM

IN SEARCH FOR CRITICAL PERIODS TO

SUPPLEMENT INTERVENTIONS

TAMÁS HODOZSÁN

ABSTRACT

INTRODUCTION

PRIOR RESEARCH

THEORETICAL FRAMEWORK

METHODS

RESULTS

FINDINGS

DISCUSSION

CONCLUSION

LIMITATIONS

FUTURE STUDIES

REFERENCES