Disentangling Promotive and Buffering Protection: Exploring the Interface Between Risk and Protective Factors in Recidivism of Adult Convicted Males

Abstract

The quality of risk assessment instruments has improved greatly during the last 40 years. While assessing protective factors has become common practice, with some instruments now devoted entirely to such assessments, little is known about the effect of risk and protective factors on recidivism. The present study investigates the effects (promotive or buffering protective) of protective factors captured by the LS/CMI for a sample of 18,031 convicted adult males under the supervision of provincial services in Canada. Effects of protective factors and possible interactions between risk and protective factors were investigated using moderation analyses. Results indicate that protective factors can be both promotive and buffering protective for risk and that the benefits of protective factors are related to the risk to which people are exposed. Patterns of protective effects appear to differ for general and violent recidivism. Theoretical and clinical implications are discussed.

Keywords

protective factors risk assessment recidivism moderation analysis LS/CMI

Risk assessment gathers and organizes information to assess the risk that an individual will re-exhibit violent (or criminal) behavior and to identify the best available options to prevent such recidivism (Hart, 2009). Numerous methods and strategies to assess risk are available. While the primary goal of early structured methods was to quantify the level of risk, newer methods, such as structured professional judgment tools or actuarial assessments of risk and need based on the assessment of dynamic risk factors, are also aimed at assessing intervention needs and guiding rehabilitative practices. First-generation actuarial instruments were focused on predicting recidivism, while newer methods are increasingly aimed at understanding and focusing intervention on the factors potentially causing people to reoffend. Currently available tools cover a wide variety of risk domains, ranging from the generic, such as substance abuse or impulsivity, to risks specific to certain types of offenses, such as dynamic risk factors related to sexual recidivism (Hanson & Morton-Bourgon, 2005). While the use of structured risk and needs assessments is now common practice among professionals in westernized countries, many of them also rely on a determination of the protective factors when specifying risk levels, designing intervention plans, and promoting change.

Protective Factors: Definitions and Assessment

Protective factors can be defined as the personal, social, and institutional resources that foster competence and promote successful development, decreasing the likelihood of engagement in problematic behavior (Deković, 1999). It is now common practice to divide protective factors into two broad categories according to their effect on recidivism (Lösel & Farrington, 2012). Promotive factors have a direct inverse effect on the likelihood of recidivism, regardless of the risk factors to which the individual is exposed. For example, exposure to prosocial friends is potentially beneficial for most people. Buffering protective factors act indirectly on the probability of reoffending through mitigation of the negative impact of risk factors (Cording & Beggs Christofferson, 2017). They have a moderating effect in that they alter the direction or strength of the relationship between a predictor (risk) and an outcome (Hayes, 2017), in this case, recidivism: the effect of one variable depends on the presence and strength of another, creating a conditional relationship or a buffering protective effect (Lösel & Farrington, 2012). Buffering protective factors could, for example, affect the likelihood of recidivism only among people with certain characteristics or a certain risk level or profile. For example, the benefits associated with structured prosocial activities could be greater for people with fewer prosocial contacts, such as those who live alone or are unemployed.

Only a few tools formally consider protective factors. The most frequently referenced include the Structured Assessment of Violence Risk in Youth (SAVRY; Borum et al., 2002), the Short-Term Assessment of Risk and Treatability (START; Webster et al., 2004), the Inventory of Offender Risk, Needs, and Strength (IORNS; Miller, 2006), the Dynamic Risk Assessment for Offender Re-entry (DRAOR; Serin et al., 2007), and, more recently, the Service Planning Instrument (SPIN; Jones et al., 2015). However, the instrument that has generated the most interest is unquestionably the Structured Assessment of Protective Factors (SAPROF; de Vogel et al., 2009). SAPROF is a structured guide for professional judgment that makes it possible to evaluate 17 protective factors, two of which are static, while 15 are dynamic.

While instruments formally devoted to measuring protective factors are not the norm, different assessment tools draw on protective factors without explicitly referring to them as such. This is the case, for example, with instruments in the LS (Level of Service) family (Andrews et al., 2004), which make use of a number of protective factors, coding them as inverse risk factors (prosocial friends, productive integration at work, prosocial leisure activities, familial relationships, etc.). In the coding scheme of such instruments, the absence of a protective factor is considered a risk and increases the total risk score.

In clinical practice, the assessment of protective factors has been based on experience and intuition. Previous research in this area, while recognizing the importance of protective factors in planning interventions or encouraging change, has failed to identify which protective factors are the most effective in preventing recidivism. As the effects of certain protective factors are hypothesized to be conditional in nature (de Vries Robbé & de Vogel, 2009), it is reasonable to assume that most main-effect statistical techniques that do not consider buffering effects (such as correlative studies or meta-analytic investigations) may have underestimated or even missed the moderating effects of certain protective factors. And, as the majority of instruments currently used are based on different meta-analyses, it is also reasonable to assume that only the direct effects of protective (promotive) factors were recorded. Variables governed by more complex effects, such as moderating effects, would thus be systematically misrepresented.

Few previous studies have focused specifically on the interface between risk and protective factors on recidivism in adult individuals convicted of a crime (Cording & Beggs Christofferson, 2017). Studies have addressed different factors that protect people from adopting a delinquent lifestyle or committing a violent offense, whether as a juvenile (Jolliffe et al., 2016; Jones et al., 2007; Longshore et al., 2005) or as an adult (Bouffard & Laub, 2004; Horney et al., 1995; O’Connor & Perryclear, 2002; Uggen, 2000). Other research has tested the incremental validity of an instrument for the assessment of protective factors (SAPROF) compared with an instrument for risk assessment (Historical, Clinical, Risk Management-20 [HCR-20]; de Vries Robbé et al., 2011, 2015), or the incremental validity of strength scales over risk scales in predicting inpatient aggression (Desmarais et al., 2012). However, a better examination of the relationships between the various concepts in relation to recidivism is needed (Polaschek, 2016).

This Study

Despite the interest in this area, to the best of our knowledge, no study has addressed the interface between risk and protection in the prediction of recidivism in adults. This absence can be explained, at least in part, by the fact that studies of such relationships require large sample sizes to obtain stable and reproducible results (McClelland & Judd, 1993; Rosnow & Rosenthal, 1989). Furthermore, studies investigating the relevance of protective factors on recidivism have largely focused either on retrospective estimates or file analyses or have been conducted on small samples of people who were often assessed in psychiatric hospitals (de Vries Robbé et al., 2015). Finally, while many have noted the necessity of considering the nature of the effect of protective factors (whether promotive or buffering protective), none have empirically tested these effects on recidivism with adult individuals convicted of a crime. The aim of the current study is therefore to investigate the nature of the effects (promotive or buffering protective) of protective factors on general and violent recidivism in convicted adult males according to Summative Risk Scale and risk domains.

Method

Participants

Participants included 18,031 adult males under the supervision of the Ministry of Community Safety and Correctional Services (MCSCS) in Ontario, Canada. In Canada, correctional jurisdiction is decided by sentence length—in general, people who have been sentenced to less than 2 years are supervised in the provincial correctional system, while those serving 2 years or more are supervised in the federal system. People in our study were all in the provincial system. Adult males convicted of a crime during the period from 1998 to 2011 were included in our assessment. Mean age at release, or at the time of assessment in the community, was 33.9 (SD = 10.6) years. The majority were Caucasian (68.8%), while 12.5% were Aboriginal, Black (9.3%), or other or unknown (9.4%). While the majority of people (56.7%) were still incarcerated when the LS/CMI was administered, 39.3% were on probation and 4.0% had been released on a conditional sentence.

Measures and Procedure

Participants were evaluated using the LS/CMI (the LS/CMI is the commercial name for the LSI-OR originally used by the MCSCS). The LS/CMI assesses static and dynamic factors linked to recidivism risk. The assessment is a revision and improvement of the LSI-R, which had 54 items. The LS/CMI has 43. Its first section is divided into eight major categories of criminogenic factors: Criminal History (eight items), Education/Employment (nine items), Family/Marital (four items), Leisure/Recreation (two items), Companions (four items), Alcohol and Drug Problems (eight items), Antisocial Attitude/Orientation (four items), and Antisocial Patterns (four items). The second section is used to record factors associated with any risks and needs, including factors specifically related to an increased risk of recidivism (i.e., anger management deficits) or that characterize the crimes committed by the person (i.e., a history of stalking behavior; Andrews et al., 2004). The majority of the LS/CMI items are coded “Yes” or “No” (0 = “No,” 1 = “Yes”), with certain items coded on a scale ranging from 0 to 3 (0 and 1 = Yes, 2 and 3 = No). The total of all items provides information about a person’s risk level, while subtotals indicate criminogenic factors. The LS/CMI was coded following clinical interviews and thorough readings of each case. All assessors were professionals who had completed 5.5 days of LS/CMI training.

Reorganizing the LS/CMI Items Into Risk and Protective Factors

The LS/CMI was developed as a tool for the evaluation of the risk of recidivism well before the growing interest in studying protective factors for recidivism, but protective factor markers were already integrated into it, although they were operationalized as the negative of a risk factor. In the original coding, the absence of a protective factor was considered a risk factor. Given this, items in the LS/CMI can be reorganized into three families of factors: unipolar risk factors (i.e., antisocial friends), unipolar protective factors (i.e., prosocial friends), and bipolar risk-protection factors (negative or positive attitudes toward conventional life). This last family includes cases where a factor can be considered a risk factor if it is problematic or a protective factor if it is beneficial. Risk factor scales were related to (a) Criminal History (eight items), (b) Antisocial Relations (three items), (c) Alcohol and Drug Problems (eight items), (d) Antisocial Attitudes (two items), and (e) Antisocial Personality (six items). A Summative Risk Scale was also created by summing all risk scales. Items measuring protective factors were (a) Grade 12 completed, (b) prosocial acquaintances (i.e., the person has acquaintances who are not involved in criminal activities and do not have criminal records), (c) prosocial friends (i.e., the person has close friends with whom he or she spends leisure time who are not involved in criminal activities and do not have criminal records), (d) satisfactory marital situation, (e) good parental relationships, (f) good relationships with relatives, (g) currently employed, (h) recent job stability, (i) prosocial organized activity, (j) good use of spare time, (k) negative attitudes toward crime, and (l) positive attitudes toward conventional life. A Summative Protective Scale (SPS) was also created by adding all 12 protective items.

Recidivism

Data on recidivism were compiled from the Offender Tracking and Information System (OTIS) database of the MCSCS in Ontario, Canada. Monitoring began when the individual was in the community either on parole or at the end of a sentence. Recidivism, assessed after a 2-year follow-up for each participant, was measured by new convictions (general recidivism) and new convictions for violent recidivisms (all violent offenses).

Data Analysis

Bivariate correlations were used to measure the association between risk scales, protective factors, and recidivism. To study the effect of protective factors on recidivism, we tested all protective factors to determine if they had promotive (direct) or buffering protective (moderated) effects on recidivism with binary logistic regression models using the Process macro 3.3 for SPSS version 25 (Hayes, 2017). Independent variables were standardized total risk and risk scales (Criminal History, Antisocial Relations, Alcohol and Drug Problems, Antisocial Attitudes, and Antisocial Personality). Moderator variables were the 12 binary dynamic protective factors as well as a standardized SPS. Dependent variables were general and violent recidivism after a 2-year follow-up. As the aim of this study was not to examine whether protective factors could contribute to risk assessment by adding variance over risk factors, focus was not on the possible increase of explained variance but on the interpretation of standardized odds ratio (OR). We looked instead at main and interactions effects and, because of the large number of analyses, all models were Bonferroni-corrected for controlling for a family-wise Type I error (p < .05/12 = .004).

Results

In total, 10,166 participants (56.4%) out of 18,031 had reoffended within the 2-year follow-up. Recidivism involved a violent offense for 15.6% and was related to a sexual offense for 0.5% and a nonviolent offense for 30.0%. Table 1 shows the descriptive statistics and the bivariate correlations of risk scales and protective factors with general and violent recidivism. With regard to protective factors, for all but six correlations—satisfactory marital situation for general recidivism, prosocial acquaintances, good parental relationships, relatives’ relationship satisfaction, prosocial organized activity, and negative attitudes toward crime for violent recidivism—there was a negative association at p < .001 between a protective factor and recidivism. Product moment correlations ranged from −.023 to −.163, all significant at p < .001.

Table 1:

Descriptive Statistics and Bivariate Correlations With General and Violent Recidivism

Risk scales	M (SD)	General recidivism (r_pb)	Violent recidivism (r_pb)
Criminal History (0–7)	4.66 (1.52)	.242***	.070***
Antisocial Relations (0–3)	2.25 (0.78)	.109***	.042***
Alcohol and Drug Problems (0-8)	4.58 (2.12)	.155***	.076***
Antisocial Attitudes (0–2)	0.97 (0.77)	.026***	.056***
Antisocial Personality (0–6)	2.52 (1.30)	.166***	.061***
Summative Risk Scale (0–26)	14.88 (4.13)	.247***	.102***
Summative Protective Scale (0–12)	4.77 (2.20)	−.194***	−.062***
Protective factors	N (%)	Phi	Phi
12th grade or more	5,023 (28.1)	−.029***	−.028***
Prosocial acquaintances	15,287 (89.2)	−.080***	−.017
Prosocial friends	12,904 (75.3)	−.120***	−.030***
Satisfactory marital situation	8,367 (48.8)	−.012	−.099***
Good parental relationships	4,955 (28.8)	−.052***	−.011
Relatives’ satisfactory relationship	8,785 (51.2)	−.048***	−.016
Currently employed	6,229 (35.0)	−.126***	−.026***
Recent job stability	7,443 (41.6)	−.163***	−.041***
Prosocial organized activity	1,004 (5.8)	−.023***	−.004
Good use of spare time	4,208 (24.4)	−.129***	−.026***
Negative attitudes toward crime	4,946 (29.3)	−.091***	−.002
Positive attitudes toward conventional life	6,923 (40.9)	−.119***	−.029***

Note. r_pb = point-biserial correlation coefficient.

***

p < .001.

Testing the Effect of Protective Factors as a Function of Summative Risk Scale

The following section looks at whether protective factors affect general and violent recidivism through promotive (direct) or buffering (moderating) effects (or both) on risk factors. Table 2 shows the results of the moderation analysis with binary logistic regression models testing the effects (direct-promotive or moderated-buffering protective effect) of SPS in relation to risk scales and Summative Risk Scale for general and violent recidivism. All models were statistically significant at p < .001 and the Nagelkerke pseudo-R² coefficients ranged from .009 to .10. Coefficients presented are standardized OR coefficients. Results suggest that the SPS had mainly a promotive effect on risk for general and violent recidivism. The higher the SPS, the lower the recidivism rate after controlling for risk scales. For general recidivism, significant buffering effects were observed for Criminal History and Alcohol and Drug Problems. More specifically, a significant interaction term for SPS × Criminal History indicated that the effect of Criminal History on general recidivism was decreased as the SPS increased. In this case, standardized ORs for low-, medium-, and high-level protection persons were 1.628, 1.547, and 1.470, respectively. With regard to Alcohol and Drug Problems, ORs decreased from 1.314 for low-level protection persons to 1.258 for medium-level protection to 1.204 for high-level protection. Regarding violent recidivism, relationships were smaller, as indicated by smaller Nagelkerke pseudo-R² coefficients. The effect of the SPS was mostly promotive, except for Antisocial Relations. The SPS appeared to buffer the effect of Antisocial Relations, lowering the ORs from 1.187 for low-level protection, 1.105 for medium-level protection, and a nonsignificant 1.028 for high-level protection. Finally, when tested against the Summative Risk Scale, the SPS was nonsignificant for violent recidivism.

Table 2:

Moderation Models for Predicting General and Violent Recidivism Over 2 Years Based on Risk Scales

Variables	General recidivism	Violent recidivism
Criminal History	1.547***	1.175***
Summative Protective Scale	0.749***	0.886***
SPS × Criminal History	0.950*	0.989
Low protection	1.628***
Medium protection	1.547***
High protection	1.470***
Nagelkerke’s pseudo-R²	.100	.012
χ²	1393.76***	124.40***
Antisocial Relations	1.205***	1.105***
Summative Protective Scale	0.680***	0.864***
SPS × Antisocial Relations	0.988	0.931**
Low protection (–1SD)		1.187***
Medium protection (M)		1.105***
High protection (+1SD)		1.028
Nagelkerke’s pseudo-R²	.061	.010
χ²	83.82***	104.45***
Alcohol and Drug Problems	1.258***	1.193***
Summative Protective Scale	0.709***	0.888***
SPS × Alcohol and Drug Problems	0.957**	0.997
Low protection (–1SD)	1.314***
Medium protection (M)	1.258***
High protection (+1SD)	1.204***
Nagelkerke’s pseudo-R²	.065	.013
χ²	901.07***	137.69***
Antisocial Attitudes	0.998	1.129***
Summative Protective Scale	0.667***	0.867***
SPS × Antisocial Attitudes	0.974	0.963
Low protection (–1SD)
Medium protection (M)
High protection (+1SD)
Nagelkerke’s pseudo-R²	.053	.011
χ²	678.45***	109.63***
Antisocial Personality	1.241***	1.105***
Summative Protective Scale	0.730***	0.890***
SPS × Antisocial Personality	0.986	0.953
Low protection (–1SD)
Medium protection (M)
High protection (+1SD)
Nagelkerke’s pseudo-R²	.062	.009
χ²	849.03***	10.46***
Summative Risk Scale	1.539***	1.287***
Summative Protective Scale	0.794***	0.950
SPS × Antisocial Personality	0.992	0.980
Low protection (–1SD)
Medium protection (M)
High protection (+1SD)
Nagelkerke’s pseudo-R²	.094	.019
χ²	131.20***	197.02***

Note. SPS = Summative Protective Scale.

p < .05. **p < .01. ***p < .001.

Table 3 shows the results of individual protective factors on general recidivism and violent recidivism while controlling for Summative Risk Scale. The effects (OR) for those protected and not protected are presented only when the interaction terms were significant, indicative of a moderation effect. A reduction in the OR value for those protected is indicative of a reduction in the effects of the Summative Risk Scale (buffering effect). Only three protective factors influenced general recidivism when the Summative Risk Scale was considered. Promotive effects can be observed for the variables good parental relationship (OR = 0.890, p = .012) and currently employed (OR = 0.707, p < .001). Buffering effects of recent job stability on general recidivism were also detected, indicated by a significant interaction with the Summative Risk Scale (OR = 0.906, p = .049). Recent job stability reduced the effect of the Summative Risk Scale, lowering the OR from 1.659 to 1.502. No significant effect of protective factors could be detected at p < .05 for violent recidivism.

Table 3:

Moderation Models for Predicting 2 Years General and Violent Recidivism With Summative Risk Scale

Variables	12th grade or more	Prosocial acquaintances	Prosocial friends	Satisfactory marital situation	Good parental relationships	Good relationships with relatives	Currently employed	Recent job stability	Prosocial organized activity	Good use of spare time	Negative attitudes toward crime	Positive attitudes toward conventional life
General recidivism
Summative Risk Scale	1.706***	1.684***	1.769***	1.740***	1.740***	1.775***	1.638***	1.659***	1.694***	1.643***	1.729***	1.700***
Protective factor	0.993	0.871	0.837	1.020	0.890*	0.966	0.707**	0.661***	0.902	0.716	0.845	0.804
PF × Summative Risk Scale	0.971	1.005	0.925	0.970	0.929	0.927	1.006	0.906*	0.940	0.953	0.906	0.940
Unprotected								1.659***
Protected								1.502***
Nagelkerke’s pseudo-R²	.081	.083	.086	.086	.087	.086	.090	.092	.081	.086	.087	.089
χ²	1120.55***	1094.59***	1133.37***	1128.99***	1149.18***	1136.25***	1230.29***	1278.97***	1075.51***	1139.96***	1128.63***	1155.54***
Violent recidivism
Summative Risk Scale	1.334***	1.344***	1.366***	1.309***	1.333***	1.321***	1.350***	1.398***	1.324***	1.338***	1.383***	1.352***
Protective factor	0.903	1.098	1.063	0.972	0.952	0.602	1.016	1.032	1.014	1.015	1.200	1.020
PF × Summative Risk Scale	0.939	1.005	0.991	1.019	0.962	0.960	0.938	0.903	1.029	0.971	0.949	0.975
Unprotected
Protected
Nagelkerke’s pseudo-R²	.018	.019	.019	.032	.018	.019	.017	.018	.018	.018	.020	.019
χ²	190.25***	190.91***	194.92***	330.87***	187.05***	192.87***	183.09***	190.03***	181.80***	184.45***	203.61***	190.79***

Note. PF = protective factor.

p < .01. ***p < .001.

Testing the Effect of Protective Factors as a Function of Specific Risk Scales

The following section investigates the effect of different protective factors on specific risk scales in relation to general (Table 4) and violent (Table 5) recidivism. The tables can be analyzed by looking at the different effects of protective factors on recidivism for specific risk scales. All models were statistically significant at p < .01 and the Nagelkerke pseudo-R² coefficients ranged from .002 to .095.

Table 4:

Moderation Models for Predicting General Recidivism Over 2 Years Based on Risk Scales

Variables	12th grade or more	Prosocial acquaintances	Prosocial friends	Satisfactory marital situation	Good parental relationships	Good relationships with relatives	Currently employed	Recent job stability	Prosocial organized activity	Good use of spare time	Negative attitudes toward crime	Positive attitudes toward conventional life
Criminal History	1.650***	1.727***	1.713***	1.753***	1.729***	1.746***	1.661***	1.681***	1.661***	1.669***	1.700***	1.729***
Protective factor	0.969	0.727***	0.685***	0.938	0.846***	0.902*	0.648***	0.607***	0.886	0.616***	0.828***	0.737***
PF × Criminal History	1.015	0.941	0.931	0.925	0.901**	0.927	0.942	0.875***	0.973	0.888*	0.910*	0.878**
Unprotected					1.729***			1.681***		1.669***	1.700***	1.729***
Protected					1.557***			1.470***		1.482***	1.547***	1.518***
Nagelkerke’s pseudo-R²	.077	.082	.086	.083	.084	.083	.090	.095	.077	.089	.083	.088
χ²	1061.91***	1081.28***	1139.59***	1087.74***	1108.65***	1094.82***	1231.98***	1309.97***	1014.09***	182.79***	1070.60***	1143.71***
Antisocial Relations	1.235***	1.283***	1.315***	1.312***	1.276***	1.287***	1.251***	1.244***	1.249***	1.237***	1.295***	1.285***
Protective factor	0.936	0.628***	0.602***	0.917	0.778***	0.823***	0.607***	0.539***	0.847	0.570***	0.705***	0.641***
PF × Antisocial Relations	1.028	0.979	0.928	0.910*	0.956	0.951	0.934	0.916*	0.942	0.941	0.877**	0.918*
Unprotected				1.312***				1.244***			1.295***	1.285***
Protected				1.194***				1.139***			1.136***	1.179***
Nagelkerke’s pseudo-R²	.017	.023	.032	.018	.021	.020	.034	.045	.016	.034	.026	.033
χ²	222.92***	295.57***	414.40***	231.53***	273.79***	260.80***	452.63***	612.43***	208.51***	43961***	331.24***	425.31***
Alcohol and Drug Problems	1.394***	1.330***	1.436***	1.389***	1.416***	1.440***	1.344***	1.366***	1.381***	1.344***	1.388***	1.373***
Protective factor	0.886**	0.668**	0.656***	1.035	0.818***	0.875	0.638***	0.556***	0.830	0.608***	0.716***	0.678***
PF × Alcohol and Drug Problems	0.936	1.014	0.902**	0.991	0.903*	0.903**	0.980	0.896**	0.920	0.932	0.937	0.938
Unprotected			1.436***		1.416***	1.440***		1.366***
Protected			1.295***		1.279***	1.301***		1.224***
Nagelkerke’s pseudo-R²	.033	.036	.044	.033	.036	.035	.047	.057	.033	.046	.041	.045
χ²	44344***	467.37***	567.16***	427.71***	474.50***	452.07***	628.67***	781.78***	429.00***	594.09***	522.46***	574.37***
Antisocial Attitudes	1.067***	1.127	1.115**	1.063**	1.045	1.100***	1.054**	1.060**	1.056**	1.050**	1.110***	1.090***
Protective factor	0.876***	0.601***	0.563***	0.955	0.792***	0.817***	0.584***	0.502***	0.815*	0.540***	0.673***	0.616***
PF × Antisocial Attitudes	0.954	0.904	0.886*	0.972	1.016	0.904*	0.975	0.942	0.915	0.913	0.819***	0.893**
Unprotected			1.115**			1.100***					1.110***	1.090***
Protected			0.988			0.995					0.909*	0.973
Nagelkerke’s pseudo-R²	.002	.010	.022	.001	.005	.005	.022	.038	.002	.023	.014	.020
χ²	27.63***	118.49***	266.18***	13.24**	55.60***	61.19***	281.33***	486.09***	20.07***	280.90***	179.40***	256.15***
Antisocial Personality	1.449***	1.288***	1.407***	1.473***	1.429***	1.436***	1.365***	1.323***	1.414***	1.361***	1.432***	1.387***
Protective factor	0.968	0.710***	0.711***	0.943	0.910*	0.921*	.643***	0.591***	0.864	0.638***	0.767***	0.719***
PF × Antisocial Personality	0.907*	1.083	0.950	0.936	0.962	0.976	1.018	0.983	0.940	0.949	0.882**	0.954
Unprotected	1.449***										1.432***
Protected	1.314***										1.264***
Nagelkerke’s pseudo-R²	.038	.040	.045	.040	.040	.040	.051	.056	.038	.048	.043	.047
χ²	514.11***	517.13***	582.12***	521.60***	523.58***	525.45***	690.68***	766.05***	488.30***	620.72***	551.95***	600.69***

Note. PF = protective factor.

p < .05. **p < .01. ***p < .001.

Table 5:

Moderation Models for Predicting Violent Recidivism Over 2 Years Based on Risk Scales

Variables	12th grade or more	Prosocial acquaintances	Prosocial friends	Satisfactory marital situation	Good parental relationships	Good relationships with relatives	Currently employed	Recent job stability	Prosocial organized activity	Good use of spare time	Negative attitudes toward crime	Positive attitudes toward conventional life
Criminal History	1.226***	1.243*	1.212***	1.263***	1.269***	1.281***	1.209***	1.200***	1.221***	1.208***	1.235***	1.226***
Protective factor	0.872*	0.955	0.903	0.578***	0.975	0.966	0.906	0.863**	0.982	0.902	1.127	0.929
PF × Criminal History	0.955	0.976	0.997	0.953	0.898	0.929	1.007	1.000	1.028	1.022	1.012	0.985
Unprotected
Protected
Nagelkerke’s pseudo-R²	.009	.009	.009	.027	.010	.010	.009	.010	.009	.009	.010	.010
χ²	98.77***	87.04***	91.31***	272.83***	102.38***	103.11***	94.79***	102.11***	88.39***	92.47***	94.70***	94.65***
Antisocial Relations	1.131***	1.195	1.242***	1.185***	1.173***	1.175***	1.158***	1.153***	1.128***	1.160***	1.187***	1.168***
Protective factor	0.863*	0.920	0.890	0.572***	0.945	0.928	0.882*	0.823***	0.978	0.865*	1.049	0.878*
PF × Antisocial Relations	0.955	0.957	0.902	0.909	0.853**	0.912	0.920	0.916	1.024	0.894	0.890	0.922
Unprotected					1.173***
Protected					1.001
Nagelkerke’s pseudo-R²	.004	.004	.005	.021	.004	.004	.004	.006	.003	.005	.004	.005
χ²	42.36***	40.35***	53.73***	215.06***	42.48***	38.66***	45.28***	57.34***	33.07***	46.85***	41.61***	47.97***
Alcohol and Drug Problems	1.254***	1.131	1.187**	1.212***	1.245***	1.269***	1.216***	1.247***	1.239***	1.234***	1.241***	1.254***
Protective factor	0.853**	0.915	0.905	0.606***	0.965	0.969	0.920	0.862**	0.963	0.932	1.074	0.927
PF × Alcohol and Drug Problems	0.929	1.108	1.055	0.975	0.983	0.949	1.020	0.939	0.985	1.006	1.022	0.959
Unprotected
Protected
Nagelkerke’s pseudo-R²	.011	.011	.011	.024	.010	.010	.010	.011	.010	.011	.011	.011
χ²	117.09***	105.33***	109.23***	24352***	105.42***	105.41***	103.67***	119.02***	102.12***	107.54***	106.57***	108.37***
Antisocial Attitudes	1.153***	1.251***	1.196***	1.127***	1.140***	1.204***	1.197***	1.202***	1.173***	1.194***	1.223***	1.152***
Protective factor	0.833**	0.951	0.870	0.592***	0.939	0.935	0.877*	0.811***	0.928	0.862**	1.027	0.858***
PF × Antisocial Attitudes	1.025	0.919	0.959	0.989	1.063	0.919	0.900	0.902	0.869	0.846**	0.823***	1.031
Unprotected										1.194***	1.223***
Protected										1.010	1.007
Nagelkerke’s pseudo-R²	.007	.006	.007	.020	.005	.006	.007	.009	.006	.007	.007	.007
χ²	65.28***	54.29***	64.88***	192.19***	51.43***	55.55***	65.40***	85.06***	53.49***	66.79***	70.69***	67.05***
Antisocial Personality	1.185***	1.215*	1.265***	1.189***	1.186***	1.227***	1.149***	1.143***	1.168***	1.186***	1.245***	1.202***
Protective factor	0.880**	1.007	0.988	0.579***	0.999	0.981	0.906	0.867**	0.984	0.901	1.096	0.960
PF × Antisocial Personality	0.931	0.971	0.904	0.961	0.947	0.904	1.049	1.026	1.077	0.893	0.858**	0.928
Unprotected											1.245***
Protected											1.069
Nagelkerke’s pseudo-R²	.007	.007	.007	.023	.006	.007	.007	.007	.006	.007	.008	.007
χ²	68.40***	67.30***	73.59***	231.65***	60.72***	67.70***	68.76***	75.94***	61.12***	66.96***	81.40***	72.23***

Note. PF = protective factor.

p < .01. ***p < .001.

Certain protective factors had promotive effects on the risk for general recidivism, while others had buffering effects. One particularly notable example of a promotive effect was being employed, which had a direct effect on general recidivism over all risk scales. It had a promotive effect on general recidivism after controlling for Criminal History (OR = 0.648, p < .001), Antisocial Relations (OR = 0.607, p < .001), Alcohol and Drug Problems (OR = 0.638, p < .001), Antisocial Attitudes (OR = 0.584, p < .001), and Antisocial Personality (OR = 0.643, p < .001). Prosocial acquaintances had promotive effects after controlling for Criminal History (OR = 0.727, p < .01), Antisocial Relations (OR = 0.628, p < .001), Alcohol and Drug Problems (OR = 0.668, p < .001), Antisocial Attitudes (OR = 0.601, p < .001), and Antisocial Personality (OR = 0.710, p < .001). Finally, good use of spare time showed promotive effects after controlling for Antisocial Relations (OR = 0.570, p < .001), Alcohol and Drug Problems (OR = 0.608, p < .001), Antisocial Attitudes (OR = 0.540, p < .001), and Antisocial Personality (OR = 0.638, p < .001).

Numerous protective effects for general recidivism were buffering protective. This was true for recent job stability, which moderated the effect of Criminal History (reducing the OR from 1.681 to 1.482), Antisocial Relations (from 1.244 to 1.139), and Alcohol and Drug Problems (from 1.366 to 1.224). It was also the case for the attitude variables that buffered Criminal History, Antisocial Relations, and Antisocial Attitudes, although this effect may be due partially to the bipolar nature of the attitude variables. In only a few cases did the moderating effect eliminate the effect of risk factors. This was true for prosocial friends, which entirely buffered the effect on general recidivism of Antisocial Attitudes, reducing the OR from 1.115 to a nonsignificant 0.988. Good relationship with relatives also had that type of effect on Antisocial Attitudes (OR from 1.100 to 0.995), as did positive attitudes toward conventional life (OR from 1.090 to 0.973). Except for prosocial acquaintances and prosocial organized activity, all protective factors had some buffering effects on different risk factors, reducing, but not eliminating, their effect. Examples of buffering effects of protective factors on Antisocial Attitudes are presented in Figure 1.

Figure 1:

Examples of Buffering Effect of Protective Factors on Antisocial Attitudes in General and Violent Recidivism

Table 5 shows the results of all moderation analyses conducted on all combinations of specific risk scales and protective factors for violent recidivism. All models were statistically significant at p < .001 and the Nagelkerke pseudo-R² coefficients ranged from .003 to .024. The promotive effect of satisfactory marital situation was observable in all models predicting violent recidivism, as indicated by significant OR coefficients in models with Criminal History (OR = 0.578, p < .001), Antisocial Relations (OR = 0.572, p < .001), Alcohol and Drug Problems (OR = 0.606, p < .001), Antisocial Attitudes (OR = 0.592, p < .001), and Antisocial Personality (OR = 0.579, p < .001). To a lesser degree, this was also the case for recent job stability and having completed 12th grade or higher.

Certain protective factors showed buffering effects on risk factors for violent recidivism. Good parental relationships had buffering protective effects on Antisocial Relations, reducing its effects from an OR of 1.173 to a nonsignificant 1.001. Good use of spare time had a similar effect on Antisocial Attitudes (OR from 1.194 to a nonsignificant 1.01), as did negative attitudes toward crime on Antisocial Attitudes (OR from 1.223 to a nonsignificant 1.007) and Antisocial Personality (OR from 1.245 to a nonsignificant 1.069).

Discussion

While systematic research dealing with the effect of protective factors on recidivism in adults convicted of a crime is relatively recent, a knowledge base of variables that affect the development of delinquency is beginning to be developed (Farrington et al., 2016; Jolliffe et al., 2016; Kim et al., 2016). While early instruments for risk assessment were used during the 1980s, a tool specifically dedicated to the assessment of protective factors was not designed until 2007 (de Vogel et al., 2009). Even today, despite clear interest in these questions, there are few empirical studies that measure the effects of protective factors on recidivism among convicted adults. The present study is intended to contribute to our understanding of these factors by focusing on their effect on recidivism for a large sample of adult males, using the LS/CMI, a fourth-generation actuarial instrument that addresses certain protective factors.

One general conclusion is that the nature of the effect, whether promotive or buffering protective, is relatively complex. When associated with Summative Risk Scale, only being employed had a promotive effect. Regarding violent recidivism, the selected protective factors showed largely nonsignificant effects over the risk scales considered. When the Summative Risk Scale was divided into its constituents, a plethora of more complex relationship emerged. It appears that no single factor acts as a talisman that protects all risk profiles. With only a few exceptions, and in accord with common wisdom, protective factors useful for reducing recidivism are related to the risk profile of the individual.

The Protective Effects of Employment

Among the protective factors that emerged in this study were the promotive effect of employment and the buffering protective effect of recent job stability on general recidivism. Results showed that being employed was good for almost everyone, while stable employment reduced the effects of risk factors on general recidivism. The benefits of work on recidivism have been studied previously (Bucken & Zajac, 2009; Duwe, 2013, 2015a, 2015b; Sampson & Laub, 2005; Uggen, 2000) and several researchers have found that success in finding legal employment is one of the best predictors of successful release (Visher et al., 2005). Being employed full-time and holding a job for a certain period of time had positive effects and, for nonviolent recidivists, these effects were relatively independent of risk factors.

The mechanisms through which employment reduces recidivism need further clarification but it is easy to speculate about the many benefits of employment. For example, stable work provides a certain pattern to peoples’ lives and a relatively stable schedule allows the individual to settle into a job’s rhythm. Employment also provides an opportunity for new prosocial contacts, making it possible to develop new friendships and diminish time spent with delinquent peers (Warr, 1998; Wright & Cullen, 2004). These new relationships act as an informal social control and can lead to changes in antisocial values that are difficult to reconcile with a more organized lifestyle (Sampson & Laub, 2001). Being legally employed over a certain period can also affect perceptions of legal work—an individual with a full-time job may experience cognitive dissonance if his or her previous mind-set makes it difficult to justify investing so much time in work and it may be easier to modify one’s beliefs than give up the employment. Work also makes it possible for people to discover new competencies and develop a new identity as an employee (Bushway & Paternoster, 2017) as well as providing an opportunity to build a positive self-image, something that may not have been possible previously.

Good Friends and Acquaintances

Having a prosocial network has been shown to be protective for general recidivism. Our results show that relationships with prosocial people had a promotive effect, particularly for sample participants with alcohol and drug problems and antisocial relations and personality. The positive effect of positive relationships may have several sources. First, such relationships provide connections to potential prosocial activities and opportunities. Individuals sharing time with prosocial friends may be exposed to activities that lead them to refrain from recidivating. Second, such relationships also seem to affect thought patterns: having prosocial friends had important buffering protective effects, partially buffering the effects of alcohol and drug problems and entirely buffering the predictive effects of antisocial attitudes.

The Benefits of Prosocial Romantic Involvement on Violent Recidivism

The results of this study also highlight the preventive benefits of a satisfactory relationship with a prosocial partner on violent recidivism (Bucken & Zajac, 2009; Horney et al., 1995; Laub & Sampson, 1993). The effect of such a relationship on violent acts can be partly explained by the fact that it promotes important life changes, whether through hobbies, activities, or the choice of friends (Horney et al., 1995). Altercations between strangers, in bars, for example, might be significantly reduced. Being part of a couple can lead an individual to restructure his or her life conditions (Laub & Sampson, 1993). Another possible explanation for such a result is that people may tend to engage in conflict with those close to them (Cusson, 2005). When relations with a partner are harmonious, chances of offenses against that person, at least within the relationship context, are significantly reduced.

Idle Hands Are the Work of the Devil: The Need to Interfere in Criminogenic Situations

Good use of spare time also appeared to be an important protective factor and should not be ignored (Hoge et al., 1996). Its effect was related to risk profile, as it was largely a promotive factor. An explanation for the importance of prosocial use of leisure time could be that it interferes with the development of criminogenic situations, understood as the moment when a person finds himself in a context in which a crime is possible but has not yet occurred (Bonta & Andrews, 2017; Parent, 2013; Wikström et al., 2012). While a motivated offender, an attractive target, and the absence of a capable guardian are sometimes considered convergent elements for the commission of a crime (Cohen & Felson, 1979), an idle and outgoing lifestyle could be considered fertile ground for criminogenic situations (Parent, 2013; Parent et al., 2016). A person with little investment in conventional activities, such as school or work, instead spending time hanging out or partying, may be more frequently exposed to opportunities for crime and have more time to identify potential targets (Cusson, 2005; Felson & Boba, 2010).

As emphasized by Wikström et al. (2012), some criminal acts may depend on context (identifying criminal opportunities without having sought them out), while others depend on propensity for crime (an active search for criminal opportunities by a persistent offender.) It is therefore not surprising that a person with antisocial attitudes or substance use problems who is kept busy with prosocial activities may have some protection from recidivism. Such activities help create schedules that cut into free time that would otherwise be used in the active search for criminal activities, as well as decreasing chance of criminal encounters, thus reducing the development of criminogenic situations.

Limitations, Implications, and Directions for Future Research

Although this study benefits from important methodological strengths (e.g., a very large sample, good interrater reliability and predictive validity, recidivism measured by new convictions, a thorough investigation of all combination of risk and protective factors), the data were not initially collected to investigate the effect of protective factors and certain limitations should be taken into account in future studies of the interaction between risk and protective factors. Some limitations are related to the representativeness of the sample: participants had received short sentences (less than 2 years) and patterns of protective factors may be different for people given longer sentences. As well, female or juvenile individuals convicted of crimes were not considered. Further investigation of the robustness of the detected relations is therefore warranted with other samples and other kinds of individuals involved with the criminal justice system. The criteria used to measure criminal persistence (i.e., recidivism) may have influenced the results as well: the use of official records meant that only official recidivism detected by the judicial system was considered and the dark figure of crime could not be taken into account. We did not investigate some kinds of recidivism such as sexual recidivism and recidivism in cases of domestic violence. Projects investigating self-reported crimes or arrest data might also generate different results.

Conducting moderation analyses on a very large sample has some drawbacks: in certain cases, we found very small effect sizes statistically significant. Because we performed numerous analyses, Bonferroni correction was also used to help identify the most crucial effects, but by applying such a correction we may have inflated Type II errors (Perneger, 1998).

There are also limitations related to the nature of the association between predictors, moderators, and recidivism. Certain interactions between risk and protective factors could be the result of more complex patterns than those considered here. This study tested the effects of protective factors and did not put much emphasis on SPSs, which were deemed too broad and vague. We found that aggregating the various items into scales that measured various constructs appeared to obscure relationships rather than clarifying them. More general profiles that considered all risk factors simultaneously or looked at more complex interaction patterns might have produced different results. Also, looking at meaningful patterns of interactions between protective and risk profiles (i.e., convergence of multiple risk factors in complex patterns or risk profiles) rather than risk factors alone might have produced different results.

Another potential limitation is related to what was measured. The LS/CMI, a powerful tool for structuring risk assessment and tailoring intervention (Olver et al., 2014), probably captures the essence of numerous protective constructs but it was initially operationalized for use with summative scales measuring risk. As well, only criminogenic factors were considered in the analyses and the selection of protective factors was restricted to factors commonly used to predict recidivism. Results showed that those protective factors were insufficient with regard to violent recidivism. It is possible that extending the range of variables and considering noncriminogenic factors, specific responsivity characteristics, or sociodemographic factors such as mental health issues, cultural factors, age, and sex might contribute to refining our understanding of the nature of the relationships between risk and protective factors.

Limitations may also have resulted from the operationalization of protective factors. In this study, binary protective items, with only a limited amount of variance, were tested as moderating factors on composite measures of risk domains. While clinically more informative than a general summative protection scale, results were probably influenced by the limited explanatory power they contained. Composite scales of protective domains, containing greater variance, might have helped uncover promotive or buffering effects. Regarding the issue of operationalization, protective factors are generally described as unipolar or bipolar. The choice of operationalization may well have repercussions for the analyses that can be performed and the conclusions that can be drawn about their effects on recidivism (Polaschek, 2017). In bipolar operationalizations, the construct has two poles, generally a low-level pole in which the domain constitutes a risk factor and a high-level pole where the domain constitutes a protective factor. The classic example of a bipolar operationalization is family relationships: at the lower end, conflictual and antisocial families act as a risk factor, while at the higher end of the continuum, stable and supportive families act as a protective factor. With that choice of operationalization, families are either a protective factor or a risk factor (or somewhere in between). They cannot act as a risk factor in certain aspects of a person’s life (i.e., through promoting the view that violence is sometimes needed to deal with problems) and as protective in others (i.e., in promoting work as important and encouraging the person to invest in his professional life). The same kind of problem arises when certain members of the family are prosocial influences and can be protective, while other members are antisocial influences. While internal protective factors generally measure traits and therefore can be more easily operationalized as bipolar constructs (a person can hardly have high impulsivity and good self-control at the same time), external protective factors are possibly better served by unipolar operationalizations in which protective factors are measured on a conceptual continuum ranging from the absence of the protective factor to a high degree of protection. The typical case of a unipolar operationalization is friends and relationships: people can have both prosocial and antisocial friends, so these relationships can simultaneously act as a risk and a protective factor. Choosing a bipolar operationalization for certain domains sometimes forces the assessor to speculate about a middle ground (a mean score) and to weigh the importance of all constituents in order to rate it. In the previous example, practitioners may be faced with trying to figure out what one’s close criminal friend and two relatively prosocial casual friends add up to, which can create interrater reliability issues. In the present study, both bipolar risk-protection items and unipolar protective items were used, depending on the initial measures of those items in the LS/CMI. The initial operationalization may therefore have obscured some relationships. Variance related to the bipolar protective factors could have been explained, at least partially, by the added variance from the negative pole of those items (Polaschek, 2016). This possibility highlights the pressing need to further clarify the definition and operationalization of both dynamic risk factors and dynamic protective factors. Such efforts will help disentangle their various effects and elucidate the underlying mechanisms by which they act to reduce recidivism and promote desistance (Cording & Beggs Christofferson, 2017; Polaschek, 2016; Serin et al., 2019; Thornton, 2015; Ward & Fortune, 2016). It will also prevent us from attempting to predict recidivism by analyzing new ways of operationalizing previously documented risk factors now named differently (Kroner et al., 2005). Finally, further refinement of the LS family or other actuarial instruments could also consider reorganizing items conceptually into subscales with regard to their effect (risk vs. protection) instead of reversing the coding of protective factors for summing risk purposes.

Conclusion

During the past 10 years, protective factors have become increasingly important in the assessment of risk for recidivism. Although there has been a great deal of speculation about their potential effects, there have been few empirical investigations of their relationship to recidivism. This study focused on testing the nature of the effects (promotive or buffering protective) of protective factors on general and violent recidivism in convicted adult males according to Summative Risk Scale and risk domains. Results suggest that the effects of protective factors on general and violent recidivism are complex and vary according to the risk profile of the individual. Further work on operationalization of protective factors and their measurement is needed to clarify various effects and further understand desistance mechanisms.

Footnotes

Authors’ Note:

The authors would like to thank Greg Wright, Michael Kirk, and Paula Davis from the Program Effectiveness Statistics and Applied Research Ministry of Community Safety and Correctional Services. The views expressed are those of the authors and do not necessarily reflect the views of the Ministry of Community Safety and Correctional Services of Ontario.

ORCID iDs

Jean-Pierre Guay

Massil Benbouriche

Jean-Pierre Guay, PhD, is a professor at the School of Criminology at the University of Montreal and a senior researcher at the International Centre for Comparative Criminology and at the Philippe-Pinel National Institute of Forensic Psychiatry. His research focuses on risk assessment, psychopathic traits, sadism, and predatory processes. He has also been leading a research program on the development of Autonomous Virtual Agents for clinical training purposes.

Geneviève Parent, PhD, has been an associate professor in the Department of Psychoeducation and Psychology, Université du Québec en Outaouais since 2017. Her research interests focus on risk assessment, evidence-based practices with juvenile and adult offenders, as well as sexual well-being of women.

Massil Benbouriche, PhD, is an associate professor in psychology and justice at the University of Lille (France) and an associate researcher at the Philippe-Pinel National Institute of Forensic Psychiatry (Canada). He holds a PhD in psychology (experimental social psychology) and a PhD in criminology. His research focuses on risk assessment and recidivism, as well as self-regulation in maladaptive behaviors.

References

Andrews

D. A.

Bonta

Wormith

S. J.

(2004). The Level of Service/Case Management Inventory (LS/CMI). Multi-Health Systems.

Bonta

Andrews

D. A.

(2017). The psychology of criminal conduct (6th ed.). https://doi.org/10.4324/9781315677187

Borum

Bartel

Forth

A. E.

(2002). Manual for the Structured Assessment of Violent Risk in Youth (SAVRY). University of South Florida.

Bouffard

L. A.

Laub

J. H.

(2004). Jail or the army: Does military service facilitate desistance from crime? In Maruna

Immarigeon

(Eds.), After crime and punishment: Pathways to offender reintegration (pp. 129–151). William Publishing.

Bucken

K. B.

Zajac

(2009). But some of them don’t come back (to prison!): Resource deprivation and thinking errors as determinants of parole success and failure. The Prison Journal, 89(3), 239–264. https://doi.org/10.1177/0032885509339504

Bushway

S. D.

Paternoster

(2017). Understanding desistance: Theory testing with formal empirical models. In MacDonald

(Ed.), Measuring crime and criminality (pp. 299–333). Routledge. https://doi.org/10.4324/9780203785997

Cohen

L. E.

Felson

(1979). Social change and crime rate trends: A routine activity approach. American Sociological Review, 44(4), 588–608. https://doi.org/10.2307/2094589

Cording

Beggs Christofferson

(2017). Theoretical and practical issues for the measurement of protective factors. Aggression and Violent Behavior, 32, 45–54. https://doi.org/10.1016/j.avb.2016.12.007

Cusson

(2005). La délinquance, une vie choisie: Entre plaisir et crime [Delinquency, a chosen life: Between pleasure and crime]. Hurtubise HMH.

10.

Deković

(1999). Risk and protective factors in the development of problem behavior during adolescence. Journal of Youth and Adolescence, 28(6), 667–685. https://doi.org/10.1023/A:1021635516758

11.

Desmarais

S. L.

Nicholls

T. L.

Wilson

C. M.

Brink

(2012). Using dynamic risk and protective factors to predict inpatient aggression: Reliability and validity of START assessments. Psychological Assessment, 24(3), 685–700. https://doi.org/10.1037/a0026668

12.

de Vogel

de Ruiter

Bouman

de Vries Robbé

. (2009). SAPROF. Guidelines for the assessment of protective factors for violence risk (English version). Forum Educatief.

13.

de Vries Robbé

de Vogel

. (2009). Assessing protective factors for violence risk. American Psychology-Law Society News, 29(3), 11–12.

14.

de Vries Robbé

de Vogel

de Spa

. (2011). Protective factors for violence risk in forensic psychiatric patients: A retrospective validation study of the SAPROF. International Journal of Forensic Mental Health, 10(3), 178–186. https://doi.org/10.1080/14999013.2011.600232

15.

de Vries Robbé

de Vogel

Koster

Bogaerts

. (2015). Assessing protective factors for sexually violent offending with the SAPROF. Sexual Abuse: A Journal of Research and Treatment, 27(1), 51–70. https://doi.org/10.1177/1079063214550168

16.

Duwe

(2013). What works with Minnesota prisoners: A summary of the effects of correctional programming on recidivism, employment, and cost avoidance. Minnesota Department of Corrections. https://mn.gov/doc/assets/What_Works_with_MN_Prisoners_July_2013_tcm1089-272845.pdf

17.

Duwe

(2015a). The benefits of keeping idle hands busy: The impact of a prisoner reentry employment program on post-release employment and offender recidivism. Crime & Delinquency, 61(4), 559–586. https://doi.org/10.1177/0011128711421653

18.

Duwe

(2015b). An outcome evaluation of a prison work release program estimating its effects on recidivism, employment, and cost avoidance. Criminal Justice Policy Review, 26(6), 531–554. https://doi.org/10.1177%2F0887403414524590

19.

Farrington

D. P.

Ttofi

M. M.

Piquero

A. R.

(2016). Risk, promotive, and protective factors in youth offending: Results from the Cambridge study in delinquent development. Journal of Criminal Justice, 45, 63–70. https://doi.org/10.1016/j.jcrimjus.2016.02.014

20.

Felson

Boba

(2010). Crime and everyday life (4th ed.). SAGE.

21.

Hanson

R. K.

Morton-Bourgon

K. E.

(2005). The characteristics of persistent sexual offenders: A meta-analysis of recidivism studies. Journal of Consulting and Clinical Psychology, 73(6), 1154–1163. https://doi.org/10.1037/0022-006X.73.6.1154

22.

Hart

S. D.

(2009). Evidence-based assessment of risk for sexual violence. Chapman Journal of Criminal Justice, 1(1), 143–165.

23.

Hayes

A. F.

(2017). Introduction to mediation, moderation, and conditional process analysis: A regression-based approach (2nd ed.). The Guilford Press.

24.

Hoge

R. D.

Andrews

D. A.

Leschied

A. W.

(1996). An investigation of risk and protective factors in a sample of youthful offenders. Journal of Child Psychology and Psychiatry, 37(4), 419–424. https://doi.org/10.1111/j.1469-7610.1996.tb01422.x

25.

Horney

Osgood

D. W.

Marshall

I. H.

(1995). Criminal careers in the short-term: Intra-individual variability in crime and its relation to local life circumstances. American Sociological Review, 60(5), 655–673. https://doi.org/10.2307/2096316

26.

Jolliffe

Farrington

D. P.

Loeber

Pardini

(2016). Protective factors for violence: Results from the Pittsburgh Youth Study. Journal of Criminal Justice, 45, 32–40. https://doi.org/10.1016/j.jcrimjus.2016.02.007

27.

Jones

N. J.

Brown

S. L.

Robinson

Frey

(2015). Incorporating strengths into quantitative assessments of criminal risk for adult offenders: The service planning instrument. Criminal Justice and Behavior, 42(3), 321–338 https://doi.org/10.1177/0093854814547041

28.

Jones

Cauffman

Piquero

A. R.

(2007). The influence of parental support among incarcerated adolescent offenders the moderating effects of self-control. Criminal Justice and Behavior, 34(2), 229–245. https://doi.org/10.1177/0093854806288710

29.

Kim

B. K.

Gilman

A. B.

Hill

K. G.

Hawkins

J. D.

(2016). Examining protective factors against violence among high-risk youth: Findings from the Seattle Social Development Project. Journal of Criminal Justice, 45, 19–25. https://doi.org/10.1016/j.jcrimjus.2016.02.015

30.

Kroner

D. G.

Mills

J. F.

Reddon

J. R.

(2005). A coffee can, factor analysis, and prediction of antisocial behavior: The structure of criminal risk. International Journal of Law and Psychiatry, 28(4), 360–374. https://doi:10.1016/j.ijlp.2004.01.011

31.

Laub

J. H.

Sampson

R. J.

(1993). Turning points in the life course: Why change matters to the study of crime. Criminology, 31(3), 301–325. https://doi.org/10.1111/j.1745-9125.1993.tb01132.x

32.

Longshore

Chang

Messina

(2005). Self-control and social bonds: A combined control perspective on juvenile offending. Journal of Quantitative Criminology, 21(4), 419–437. https://doi.org/10.1007/s10940-005-7359-2

33.

Lösel

Farrington

D. P.

(2012). Direct protective and buffering protective factors in the development of youth violence. American Journal of Preventive Medicine, 43(2), S8–S23. https://doi.org/10.1016/j.amepre.2012.04.029

34.

McClelland

G. H.

Judd

C. M.

(1993). Statistical difficulties of detecting interactions and moderator effects. Psychological Bulletin, 114(2), 376–390. https://doi.org/10.1037/0033-2909.114.2.376

35.

Miller

H. A.

(2006). A dynamic assessment of offender risk, needs, and strengths in a sample of pre-release general offenders. Behavioral Sciences & the Law, 24(6), 767–782. https://doi.org/10.1002/bsl.728

36.

O’Connor

T. P.

Perryclear

(2002). Prison religion in action and its influence on offender rehabilitation. In O’Connor

T. P.

Pallone

N. J.

(Eds.), Religion, the community, and the rehabilitation of criminal offenders (pp. 11–33). The Harworth Press.

37.

Olver

M. E.

Stockdale

K. C.

Wormith

J. S.

(2014). Thirty years of research on the level of service scales: A meta-analytic examination of predictive accuracy and sources of variability. Psychological Assessment, 26(1), 156–176. https://doi:10.1037/a0035080

38.

Parent

(2013). L’explication de la délinquance prolifique: L’importance de l’interaction entre le risque individuel et le risque contextuel [Studying the prolific offender: The importance of the interaction between individual and contextual risks]. Papyrus Repository. https://papyrus.bib.umontreal.ca/xmlui/bitstream/handle/1866/10795/Parent_Genevieve_2013_these.pdf?sequence=2&isAllowed=y

39.

Parent

Guay

J.-P.

Laurier

Fredette

(2016). Explaining the frequency and variety of crimes through the interaction of individual and contextual risk factors. Canadian Journal of Criminology and Criminal Justice, 58(4), 465–501. https://doi.org/10.3138/cjccj.2015E11

40.

Perneger

T. V.

(1998). What’s wrong with Bonferroni adjustments. The BMJ, 316, 1236–1238. https://doi.org/10.1136/bmj.316.7139.1236

41.

Polaschek

D. L. L.

(2016). Desistance and dynamic risk factors belong together. Psychology, Crime & Law, 22(1–2), 171–189. https://doi.org/10.1080/1068316X.2015.1114114

42.

Polaschek

D. L. L.

(2017). Protective factors, correctional treatment and desistance. Aggression and Violent Behavior, 32, 64–70. https://doi.org/10.1016/j.avb.2016.12.005

43.

Rosnow

R. L.

Rosenthal

(1989). Definition and interpretation of interaction effects. Psychological Bulletin, 105(1), 143–146. https://doi.org/10.1037/0033-2909.105.1.143

44.

Sampson

R. J.

Laub

J. H.

(2001). Crime and deviance in the life course. In Piquero

Mazarolle

(Eds.), Life-course criminology: Contemporary and classic readings (pp. 21–42). Wadsworth Thomson Learning.

45.

Sampson

R. J.

Laub

J. H.

(2005). A life-course view of the development of crime. The ANNALS of the American Academy of Political and Social Science, 602(1), 12–45. https://doi.org/10.1177%2F0002716205280075

46.

Serin

R. C.

Chadwick

Lloyd

C. D.

(2019). Integrating dynamic risk assessment into community supervision practice. In Polaschek

D. L. L.

Day

Hollin

C. R.

(Eds.), The Wiley international handbook of correctional psychology (pp. 725–743). Wiley-Blackwell.

47.

Serin

R. C.

Mailloux

D. L.

Wilson

N. J.

(2007). The Dynamic Risk Assessment for Offender Re-Entry (DRAOR) [Unpublished user manual].

48.

Thornton

(2015). Developing a theory of dynamic risk. Psychology, Crime & Law, 22(1–2), 138–150. https://doi.org/10.1080/1068316X.2015.1109092

49.

Uggen

(2000). Work as a turning point in the life course of criminals: A duration model of age, employment, and recidivism. American Sociological Review, 65(4), 529–546. https://doi.org/10.2307/2657381

50.

Visher

C. A.

Winterfield

Coggeshall

M. B.

(2005). Ex-offender employment programs and recidivism: A meta-analysis. Journal of Experimental Criminology, 1(3), 295–316. https://doi.org/10.1007/s11292-005-8127-x

51.

Ward

Fortune

C.-A.

(2016). From dynamic risk factors to causal processes: A methodological framework. Psychology, Crime & Law, 22(1–2), 190–202. https://doi.org/10.1080/1068316X.2015.1117080

52.

Warr

(1998). Life-course transitions and desistance from crime. Criminology, 36(2), 183–216. https://doi.org/10.1111/j.1745-9125.1998.tb01246.x

53.

Webster

C. D.

Martin

M. L.

Brink

Nicholls

T. L.

Middleton

(2004). Manual for the Short-Term Assessment of Risk and Treatability (START) (Version 1.0, Consultation ed.). Forensic Psychiatric Services Commission, St. Joseph’s Healthcare.

54.

Wikström

P. O. H.

Oberwittler

Treiber

Hardie

(2012). Breaking rules: The social and situational dynamics of young people’s urban crime. Oxford University Press.

55.

Wright

J. P.

Cullen

F. T.

(2004). Employment, peers, and life-course transitions. Justice Quarterly, 21(1), 183–205. https://doi.org/10.1080/07418820400095781