Change in Level of Service Inventory–Ontario Revised (LSI-OR) Risk Scores Over Time: An Examination of Overall Growth Curves and Subscale-Dependent Growth Curves

Abstract

The dynamic nature of risk to re-offend is an important issue in the management of offenders and has stimulated extensive research into dynamic risk factors that can alter an individual’s overall risk to re-offend if addressed. However, few studies have examined the relative importance of these dynamic risk factors, complicating the task of developing case management and treatment plans that will effect the most change. Using a large, high-risk sample and multi-wave data of a common risk assessment tool, the Level of Service Inventory–Ontario Revised (LSI-OR), the current study investigated the relationship among criminogenic risk factors and their role in influencing the overall risk score. Results indicated a diverse pattern of effects on the eight subscale scores, specifically suggesting that changes on Procriminal Attitude/Orientation, Criminal History, and Leisure/Recreation subscales resulted in a quicker rate of change to the overall risk score over time. These results suggest that some factors may be driving the change in overall risk and could potentially effect the most change if prioritized for intervention. Practical implications and implications for further research are discussed.

Keywords

risk assessment Level of Service Inventory–Ontario Revised latent growth curve modeling dynamic risk

Over the past two decades, the purpose of offender risk assessment has expanded from the mere prediction of recidivism to assistance in the overall management of the offender. This shift acknowledges the utility of identifying dynamic (i.e., changeable) risk factors for both prediction and treatment purposes, rather than focusing solely on predictive, static (i.e., unchangeable) risk factors (Andrews, Bonta, & Wormith, 2006; Cauldy, Durso, & Taxman, 2013). With what are typically referred to as third- or fourth-generation risk assessments, an attempt is made to identify one or more criminogenic (i.e., dynamic) need areas that, if altered, would be expected to result in a lower risk of recidivism than if the factor were not addressed (Andrews et al., 2006).

With medium- to high-risk offenders, who often have several criminogenic needs, the task of identifying the relative importance of these dynamic risk factors, and therefore which factor(s) should be first addressed, can be daunting, particularly given the paucity of empirical research exploring the relationship among dynamic risk factors. And yet, the importance of this task is tied closely to effective case management, which requires an identification of areas of need that should take prominence in treatment and will yield the greatest overall effect on the offender’s behavior (Douglas & Skeem, 2005). Using multi-wave data of a common risk assessment tool, the Level of Service Inventory–Ontario Revised (LSI-OR), the current study investigated the relationship among criminogenic risk factors and their role in influencing the overall risk score for offenders who remain in contact with the criminal justice system.

Static Versus Dynamic Risk Factors

Many of the most commonly used risk assessment measures consider both static and dynamic factors in their assessment of risk (e.g., Offender Intake Assessment [OIA] of Correctional Service Canada; Motiuk, 1997; Service Planning Instrument [SPIn]; Van Dieten & Robinson, 2007; Level of Service Inventory–Revised [LSI-R]; Andrews, Bonta, & Wormith, 1995). Static risk factors are historical, unchangeable factors considered not amenable to treatment, such as prior offenses, gender, or age at first offense. By contrast, dynamic risk factors are potentially changeable factors, such as antisocial attitudes, employment status, and current substance use, which can increase or decrease an individual’s probability of risk if altered. Although both sets of risk factors independently predict recidivism (Gendreau, Little, & Goggin, 1996; Zamble & Quinsey, 1997), several studies have found that the prediction of risk is more accurate when both static and dynamic risk factors are combined (e.g., Brown, St. Amand, & Zamble, 2009; Hanson & Harris, 2000). For example, Beech, Friendship, Erikson, and Hanson (2002) investigated the utility of assessing both dynamic and static risk factors with 53 sexual offenders. Results indicated that, when predicting sexual recidivism, predictive accuracy was significantly enhanced when changeable factors, such as pro-offending attitudes and socio-affective problems, were considered in addition to the Static-99, a sex offender risk assessment tool consisting solely of static factors.

As one of the few studies utilizing multi-wave data, N. J. Jones, Brown, and Zamble (2010) assessed 127 male offenders on community supervision at 1, 3, and 6 months post-release from custody and documented recidivism after 6.5 years. They found that predictive models that incorporated both static (e.g., criminal history) and dynamic (e.g., criminal attitude) risk factors predicted recidivism more accurately than those that simply considered static variables (area under the curve [AUC] = 0.86 compared with 0.81, respectively). In an assessment of the predictive validity of change in dynamic factors, N. J. Jones and colleagues (2010) also found that the re-assessment of dynamic risk factors at Times 2 and 3 yielded greater predictive accuracy than the assessment of dynamic factors at Time 1 (AUC = 0.79 compared with 0.70, respectively).¹ This finding is consistent with other forensic studies demonstrating that change in dynamic factors increases the predictive validity of risk assessment instruments (e.g., Motiuk, Bonta, & Andrews, 1990; Quinsey, Jones, Book, & Barr, 2006). Given the increased predictive accuracy of combining both static and dynamic factors and the additional information gained by assessing changeable factors, risk assessments have continued to incorporate both types of risk factors, such as the Level of Service–Ontario Revised. Despite the importance of identifying relevant dynamic and criminogenic need areas, few studies have examined the relationship among these factors.

LSI-OR

The LSI-OR was designed to provide an estimate of an offender’s risk of re-offending and to identify criminogenic needs using a combination of static and dynamic risk factors. Briefly, the offender is assessed on eight subscales, consisting of Criminal History, Education/Employment, Family/Marital, Leisure/Recreation, Companions, Procriminal Attitude/Orientation, Substance Abuse, and Antisocial Pattern, that generate a total score placing an offender into one of five risk levels (i.e., very low, low, medium, high, or very high). The total score for the LSI-OR and all LSI derivatives (e.g., LSI-R, Level of Service Inventory-Saskatchewan Youth Edition [LSI-SK]²) has demonstrated good and equivalent predictive ability across several meta-analyses that incorporate studies with single-wave designs (e.g., Gendreau, Goggin, & Smith, 2002; Gendreau et al., 1996; Olver, Stockdale, & Wormith, 2014).

The ability of the LSI to continue to predict an offender’s recidivism risk after initial assessment is centered on its ability to capture factors in an offender’s life that can vary (i.e., dynamic factors). Several studies have demonstrated that the LSI tools are, in fact, sensitive to change (e.g., Andrews & Robinson, 1984; Arnold, 2007; Motiuk, 1991; Motiuk et al., 1990). For example, Schlager and Pacheco (2011) examined a sample of 179 parolees involved in a community-based corrections program and who were assessed using the LSI-R on two occasions. They found that LSI-R total scores at Time 2 were lower than scores at Time 1, and noted that there were significant decreases in seven of eight subscales but no significant differences in the Substance Use subscale. Similarly, Raynor (2007) examined 360 offenders from the British Isles who were assessed using the LSI-R on two occasions over a 24-month period. Using both assessment scores, he divided offenders into two groups: (a) those whose LSI-R scores increased at Time 2 and (b) those whose LSI-R scores decreased at Time 2. Results indicated that those with increasing scores had significantly higher reconviction rates than those with decreasing scores, as changes in the LSI-R total score reflected changes in absolute recidivism over time. Therefore, the detection of risk-related change increased the predictive validity of the LSI-R. Although more research into risk-related change is needed (see Brown, 2003), there is evidence to support the dynamic, predictive validity of the LSI tools. What remains relatively unknown is the differential impact of subscales on the total risk score and the dynamic relationship among them.

The use of a dynamic risk assessment instrument, such as the LSI-OR, in effective case management requires an identification of changeable areas of need that should take prominence in treatment and will yield the greatest overall effect on a given offender’s behavior, particularly with moderate- to high-risk offenders (Douglas & Skeem, 2005). With evidence that dynamic factors predict recidivism and can be addressed, how do justice practitioners decide which criminogenic needs take priority? Some may make decisions simply based on the dynamic factors’ empirical link to recidivism. For example, probation officers may choose to address antisocial peers prior to employment, because the former is a stronger predictor of recidivism. However, what if addressing an offender’s peer associations serves only to affect that risk factor, whereas assisting an offender gain employment reduces the scores on other dynamic factors and, in turn, more significantly reduces their overall risk? How is the influence of changing subscale scores reflected in the LSI-OR? Do all subscales typically increase or decrease with the total score? Which subscales influence the others, if any? Effective offender management requires the exploration of these questions, yet few studies have attempted to answer them.

Given the paucity of research on change in offender risk over more than 2 time periods and the relative importance of dynamic risk factors to the overall risk score, the present study aimed to explore these issues by using multi-wave LSI-OR data to investigate change in LSI-OR total and subscale scores for moderate- to high-risk offenders over a 15-year period. We examined the general growth curve of LSI-OR total risk and subscale scores among a sample of adult male offenders to explore the dynamic relationship between individual subscales and the overall risk score.

Method

Study Participants

Participants for this study came from a larger sample of 764 male offenders who had served a youth sentence between January 1, 1986 and December 30, 1997 at one of two open custody facilities in Toronto, Canada. As reported elsewhere (Day et al., 2010; Day et al., 2012; Ward et al., 2010), the criminal activity for these individuals was followed for an average of about 16 years, beginning in late childhood/early adolescence³ into adulthood. The follow-up period ended in September, 2007. For the purpose of the study, in 2011, all available LSI-OR data for the sample were requested from the (Ontario) Ministry of Community Safety and Correctional Services (MCSCS), which maintains an electronic database of offense-related information for all offenders with involvement in the provincial correctional system. Data were provided for N = 469 offenders (61.3%) for whom at least one LSI-OR was available.⁴ Risk assessment scores were available over a 15-year period between January 5, 1996 and February 15, 2011. The number of assessments for each offender ranged from 1 (N = 202) to 17 (N = 1), with a mean of 2.50 (SD = 2.0; median = 2.0; mode = 1.0) for a total of 1,174 recorded LSI-OR assessments for the sample. For offenders with multiple LSI-ORs, the time between each LSI-OR assessment varied from less than a month to 5.5 years (M = 32.09 months; SD = 30.64; median = 21.55). Table 1 presents the frequency of offenders with the corresponding number of LSI-ORs and the cumulative frequency of individuals in the sample with the corresponding number of LSI-OR assessments they contributed to the data set. Also presented are the mean total risk scores and standard deviations on the LSI-ORs over time and the percentage of individuals classified as high or very high risk levels.⁵

Table 1.

Number of LSI-OR Assessments Included in the Analyses and Sample Characteristics on the LSI-OR.

Number of LSI-ORs	Frequency (%)^a	Cumulative frequency (%)	M LSI-OR score (SD)^b	% classified as high or very high risk on the LSI-OR
1	202 (43.1)	469 (100)	20.98 (8.1)	56
2	94 (20.0)	267 (56.9)	22.98 (7.8)	67
3	65 (13.9)	173 (36.9)	24.98 (7.0)	77
4	41 (8.7)	108 (23.0)	26.47 (7.2)	79
5	31 (6.6)	67 (14.3)	27.48 (7.4)	80
6	16 (3.4)	36 (7.7)	27.89 (6.7)	86
7	8 (1.7)	20 (4.3)	30.75 (4.2)	95
8	6 (1.3)	12 (2.6)	28.33 (4.9)	92
9	1 (0.2)	6 (1.3)	31.67 (1.6)	100
10	1 (0.2)	5 (1.1)	33.60 (1.8)	100
11	2 (0.4)	4 (0.9)	33.00 (2.2)	100
12	1 (0.2)	2 (0.4)	32.50 (2.5)	100
13	0 (0.0)	1 (0.2)	34.00 (0.0)	100
14	0 (0.0)	1 (0.2)	32.00 (0.0)	100
15	0 (0.0)	1 (0.2)	32.00 (0.0)	100
16	0 (0.0)	1 (0.2)	35.00 (0.0)	100
17	1 (0.2)	1 (0.2)	33.00 (0.0)	100

Note. LSI-OR = Level of Service Inventory–Ontario Revised.

This column represents the end point in the number of LSI-OR assessments the offenders in the sample had.

Mean scores, standard deviations, and percentages are based on the cumulative frequency.

Measures

LSI-OR

The LSI-OR is a widely used risk assessment instrument that yields a total risk score (i.e., Risk/Need Summary score) and eight subscale scores.⁶ Scores are based on a file review and offender interview. The total score is the sum total of 43 items, scored from either yes/no or 0 = need for improvement to 3 = no need for improvement (which are then dichotomized for the final score), across the eight subscales. Scores on the eight subscales reflect (a) prior offenses, including juvenile record (“Criminal History”); (b) education level and current employment status (“Education/Employment”); (c) family relationships and marital circumstances (“Family/Marital”); (d) leisure activities and recreational involvement (“Leisure/Recreation”); (e) criminal behavior of acquaintances (“Companions”); (f) attitude toward criminal behavior (“Procriminal Attitude/Orientation”); (g) substance abuse problems (“Substance Abuse”); and (h) general pattern of antisocial personality and behavior (“Antisocial Pattern”). Aside from Criminal History, all subscales include both static and dynamic risk items. For example, the Education/Employment subscale consists of level of education, which is unlikely to change for adults or may only increase, and the current employment status, which may change between LSI-OR assessments. The LSI-OR has good psychometric properties across various offender populations. Internal consistency for the total score has been reported to be excellent, with Cronbach’s alphas reported between .88 and .94, and moderate to excellent for subscale scores, with Cronbach’s alphas reported between .39 and .85 (Andrews, Bonta, & Wormith, 2004). Variability in the internal reliabilities has been attributed to the length of the subscales, with subscales with fewer items (e.g., Leisure/Recreation) yielding smaller alphas than subscales with more items (Andrews et al., 2004; Girard & Wormith, 2004).⁷ Last, the scale has good predictive validity in terms of recidivism, as well as concurrent validity with other risk assessment instruments (Andrews et al., 2004).

Analytic plan

Latent growth curve analyses (LGCAs) were conducted as this approach allows for all available longitudinal data in the sample to be used to build the statistical models. Moreover, participants are not required to have the same number of measurement points over time, and data points do not need to be equally spaced across and within individuals (Singer & Willett, 2003). For example, individuals in the sample with a single LSI-OR assessment contributed to the intercept but not to the slope of the growth curve. This multi-level modeling technique accounts for individual trajectories while estimating an overall growth curve for the sample. Analyses were performed using the R statistical software program (Long, 2012). Parameters in the model were estimated using a maximum-likelihood approach.

We first examined the overall growth pattern in LSI-OR total scores as well as each of the eight subscales of the LSI-OR as dependent variables. We then used a “leave-one-out” approach to examine if a given LSI-OR subscale differentially affected a modified LSI-OR total risk score. With this approach, the LSI-OR subscale score in question was considered an independent variable along with time; the LSI-OR total score with the given subscale score subtracted from the total score (referred to as the “corrected LSI-OR total risk score”) was considered the dependent variable. For example, to examine how change in the Antisocial Pattern score was associated with the total risk score comprised of all the other subscale scores (i.e., the corrected LSI-OR total risk score), the Antisocial Pattern score was entered as an independent variable with time predicting the LSI-OR total score minus the Antisocial Pattern subscale score as the dependent variable. Although the leave-one-out analysis may be novel, it is appropriate when the research question is concerned with the way in which a part of a whole (i.e., the explanatory variable in a regression paradigm) affects the rest (i.e., the response variable). To avoid redundancy, we must residualize the response variable by removing the contribution of the explanatory variable before performing the regression.

The full models included the following variables: (a) time, (b) the subscale score in question, and (c) their interaction term predicting the corrected LSI-OR total risk score. Consistent with the recommendations of Bates (2010), we refined the models by using a backwards selection approach in which of the three variables (i.e., time effect, subscale effect, and interaction effect) in each of the eight models (corresponding to the eight subscale scores), the variable with the highest p value was removed, and the full and reduced models were compared with ANOVA F tests. At each step, reduced models that were not significantly different from the full models were retained.

In interpreting the models, a significant effect for time indicated that the corrected LSI-OR total risk score had a significant growth curve⁸ without the effect of the scale of interest. A significant effect for the subscale of interest indicated an overall relationship between the given subscale and the other subscales comprising the corrected LSI-OR total risk score. In other words, the subscale was integral to predicting the overall growth curve. A significant interaction term between the subscale of interest and time indicated that growth in the given subscale was associated with disproportionate growth in the corrected LSI-OR total risk score; for example, as this subscale increased, the sum of the remaining subscales increased more (i.e., had a steeper slope).

An important consideration for the LGCAs was the treatment of time (for further discussion of the issue of time in growth curve modeling, see George, 2009; Piquero, Monahan, Glasheen, Schubert, & Mulvey, 2013; Singer & Willett, 2003). There were two options for the time variable: (a) age of the offender at time of LSI-OR or (b) lapse of time since first LSI-OR. The former assumes that the criminal trajectory follows an age-based path, whereas the latter assumes that criminal behavior evolves from the point of first offense, regardless of age. To maintain consistency with the existing literature on criminal careers and the evidence suggesting that criminal activity tends to change with age (Piquero, 2008; Sampson & Laub, 2003), the age-based assumption was chosen for the current study.⁹ This approach is also in keeping with the nature of the data, because LSI-ORs were not necessarily assessed at the start of a criminal career since an offender may have had a criminal record before his first LSI-OR. With this assumption, the models calculate an intercept for time 0, which corresponds to age 0, which practically speaking is not interpretable. Given that 16.9 years was the youngest age at which an offender in the sample had his first LSI-OR, we interpreted the intercept (unless otherwise specified) to be age 17 years.

Results

Growth Curves of LSI-OR Total and Subscale Scores

Results of the LGCAs revealed that the average age of first LSI-OR total score occurred at 18.6 years and increased at a rate of 0.11 points every year over the course of the follow-up period (p = .014; see Figure 1). Likewise, as indicated in Table 2, the LSI-OR subscale scores for Family/Marital, Procriminal Attitude/Orientation, and Substance Abuse significantly increased over time. In contrast, the Antisocial Pattern score significantly decreased over time. There were no significant growth patterns for the remaining subscales (i.e., Criminal History, Education/Employment, Leisure/Recreation, and Companions) when analyzed separately.

Figure 1.

Multi-level model of LSI-OR trajectories (black line) compared with 20 random individual trajectories (gray lines).

Table 2.

Results of the Linear Multi-Level Analyses for LSI-OR Subscales.

LSI-OR subscale	Intercept (γ00)	Slope (γ10)
Criminal History	5.50	.009
Education/Employment	5.39	−.24
Family/Marital	0.55	.042**
Leisure/Recreation	1.57	−.003
Companions	1.93	.004
Procriminal Attitude/Orientation	0.822	.028**
Substance Abuse	0.60	.093**
Antisocial Pattern	2.40	−.024**

Note. LSI-OR = Level of Service Inventory–Ontario Revised.

p < .001.

Leave-One-Out Analyses

First, results of the leave-one-out analyses revealed significant time and subscale effects for Companions, Employment/Education, and Antisocial Pattern subscales (see Table 3). The significance of the time effect indicated that these subscales did not account, by themselves, for the change in the corrected LSI-OR total risk, as the significant value indicated that, even when controlling for the Focus subscale, the corrected LSI-OR total risk score still increased or decreased. The significant subscale values indicated that there was some association between the Focus subscale scores and the corrected LSI-OR, in that the Focus subscale did not behave independently of the whole. For example, when the corrected LSI-OR score was high, so too was the Focus subscale score.

Table 3.

Results of the Full Multi-Level Models for LSI-OR Subscales.

LSI-OR subscale	Intercept (γ₀₀)	Time effect (γ₁₀)	Subscale effect (γ₂₀)	Interaction (γ₃₀)
Criminal History	5.59	ns	1.40**	.02*
Employment/Education	7.65	.17**	1.06**	ns
Family/Marital	16.49	ns	2.17**	ns
Leisure/Recreation	17.41	ns	1.44*	.09**
Companions	10.78	.14**	3.07**	ns
Procriminal Attitude/Orientation	14.26	ns	3.44**	−.04*
Substance Abuse	16.01	ns	0.88**	ns
Antisocial Pattern	7.14	.23**	3.85**	ns

Note. LSI-OR = Level of Service Inventory–Ontario Revised.

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

Second, there were significant subscale-only effects for the Substance Abuse and Family/Marital subscales. These results indicated that information provided by these two subscales, irrespective of time, predicted the corrected LSI-OR total risk score. In other words, when each of these subscales was removed from the total LSI-OR risk score, there was no significant time growth in the corrected LSI-OR total risk score.

Last, there were significant interaction terms for Procriminal Attitude/Orientation, Criminal History, and Leisure/Recreation subscales. The interaction effect indicated that a change in the Focus subscale score was associated with the corresponding change in the rate of growth of the corrected LSI-OR total risk score (i.e., the slope of the change in the LSI-OR). The negative value found for the Procriminal Attitude/Orientation subscale indicated that as these attitudes increased over time, there was increasingly less risk according to the corrected LSI-OR total risk score. The positive interaction terms for the other subscales indicated that increases in these subscales were associated with greater risk over time according to the corrected LSI-OR total risk score.

Discussion

Examining the issue of change over time in risk to re-offend is an important issue for the criminal justice system. Decisions need to be made about how best to allocate limited resources to bring about significant and meaningful reductions in the likelihood of committing another offense.¹⁰ Across the range of risk factors that may influence an offender’s overall risk level, such as those represented on the LSI-OR, it is an empirical question as to whether all risk factors exert the same influence on the rate of change on the overall risk level or whether some factors have greater impact than others. Unfortunately, there is currently very little research examining this question. Therefore, the present study used a sample of moderate- to high-risk offenders to examine the dynamic relationship among LSI-OR subscales by assessing the degree to which different subscales affect the slope of the growth curve of the total risk scores. Therefore, an important goal of this study was to begin to identify the LSI-OR subscales that were related to the most change on the overall risk score.

Change on LSI-OR Total Risk and Subscale Scores

With regard to overall change on the total risk scores, the analyses revealed a statistically significant and positive slope for the overall LSI-OR total score. This suggests that, for the offenders in the sample, their LSI-OR total scores increased over time. This finding is consistent with the sample chosen, as individuals were of interest if they continued to be involved in the criminal justice system and greater contact with the justice system often results in higher scores on non-dynamic items (e.g., Criminal History subscale). With regard to individual subscale effects, we found a diverse pattern of effects across the eight LSI-OR subscales. For three of the subscales, Family/Marital, Procriminal Attitude/Orientation, and Substance Abuse, the growth curves showed an upward trend. For the Antisocial Pattern scale, the growth curve showed a downward trend. No change was observed for the Criminal History, Education/Employment, Leisure/Recreation, and Companions scores. As these analyses were only completed as part of the growth curve analyses that allow us to investigate dynamic change among subscales over time and are limited by the very sample selected, these results should not be interpreted as representing change in risk over time.

Effects of Each Subscale on the Corrected LSI-OR Total Risk Scores

The main objective set out by this article concerned the dynamic relationship between subscale scores and the overall risk score over time. Therefore, the last set of analyses examined how the change pattern of each of the eight subscales was related to the overall (corrected) risk assessment score (i.e., the LSI-OR total risk minus the subscale score of interest). The question we wished to investigate was, “If the total LSI-OR risk score is meant to be changeable and dynamic, which specific subscales are driving that change and what is the nature of those effects?”

The findings revealed a diverse pattern of effects across the eight subscales. First, for the Companions, Employment/Education, and Antisocial Pattern subscales, both subscale and time effects were found. These effects suggest that the subscale changes with the sum of the other subscales but does not account for the entire growth curve. Therefore, these subscales may heuristically be referred to as indicative variables, as the size of their score indicates the size of the score for the other categories, but they do not affect the growth curve of the LSI-OR score. Consider the Employment/Education subscale as an example. The significant time effect indicates that the growth of Employment/Education subscale cannot effectively explain the growth of the corrected LSI-OR total risk score. When this subscale was removed from the LSI-OR total risk score, a positive linear growth in scores was still observed. However, the level of Employment/Education does to some extent indicate the risk level of the individual offender; when Employment/Education scores are high, the other subscales are also likely to be high. The interpretations for Companions and Antisocial Pattern are identical. The effect for the Antisocial Pattern subscale would be expected, given that this subscale is composed of items taken from other subscales.

Second, the results indicated significant subscale effects for the Substance Abuse and Family/Marital subscales. In other words, when these subscales were each removed from the total LSI-OR risk score, there was no significant time growth in the corrected LSI-OR total risk scores. This suggests that these subscales were not only an indicator of the overall risk, but were also characteristic, in that its growth represents a main source of the total observed growth in corrected LSI-OR total risk scores. In other words, even though the overall LSI-OR total risk scores increased over time, when the Substance Abuse and Family/Marital subscale scores were independently removed from the overall total risk score, no such significant increase was found. It is important to note that this does not imply that the scores on the Substance Abuse and Family/Marital subscales change the growth curve, merely that their growth accounted for a significant portion, but not all, of the overall growth.

Third, results indicated that the Criminal History, Leisure/Recreation, and Procriminal Attitude/Orientation subscales had statistically significant interaction terms. This suggests that these subscales were clearly associated with changes in the overall trend. When these subscale scores were high, the corrected LSI-OR total risk scores were exacerbated, in that the rate of growth of the total score increased (or decreased). It is important to distinguish this effect from the previous effect. The interaction effect indicates that a change in the Focus subscale score was associated with the corresponding change in the rate of growth of the corrected LSI-OR total risk score (i.e., the slope of the change in the LSI-OR), rather than the general value of the LSI-OR total score as seen with simply a subscale effect previously discussed. In other words, observing high scores on Criminal History, Leisure/Recreation, and Procriminal Attitude/Orientation suggests not only that the LSI-OR total risk score was high but also that it was increasing (or decreasing, in the case of the Procriminal Attitude/Orientation subscale) more rapidly than expected.

The significant effect for the Leisure/Recreation score suggests that targeting this dynamic variable may have a positive effect on the overall level of risk of offenders. It may be that high-risk leisure activities are at least partly responsible for the decline in stability of other risk areas leading to an increased growth in overall risk. Providing access and troubleshooting obstacles to positive recreational activities that are structured, expose offenders to positive role models, and provide opportunities for skills building may stabilize the family/marital circumstances, reduce the opportunities for substance use and association with negative peer interactions, and provide opportunities for developing job skills, leading to gainful employment, and so forth (M. B. Jones & Offord, 1989; Mahoney & Stattin, 2000; Morris, Sallybanks, Willis, & Makkai, 2003). For example, Gottfredson, Gerstenblith, Soulé, Womer, and Lu (2004) found that the effectiveness of an after-school program to reduce delinquent behavior in young people was mediated by positive peer associations and a reduced likelihood of using drugs.

The significant interaction effect for the Criminal History subscale, a static variable that can only increase, suggests that an upward growth in risk on this factor, perhaps as a result of a new set of charges, new period of incarceration, or parole or probation violation since the previous LSI-OR assessment, results in an increase in the rate of growth of the overall risk score. This effect should not be surprising, given that criminal history is regarded as one of the strongest predictors of recidivism (Andrews et al., 2004). Finally, the negative interaction term for the Procriminal Attitude/Orientation subscale suggests that as these attitudes increased over time, there was increasingly less risk according to the corrected LSI-OR total risk score. This appears to be a paradoxical effect—as criminal attitudes become more procriminal, the corrected LSI-OR total risk score decreases. This finding was unexpected and difficult to explain. Although speculative, it may be that as these offenders become more procriminal in their attitudes, their ability to create an overall positive impression of other areas of their life for their probation officer (or whomever is completing the LSI-OR) is enhanced (Davis, Thake, & Weekes, 2012; Mills & Kroner, 2006). For example, in a study of 11,370 newly incarcerated male offenders, Davis and colleagues (2012) examined the relationship between impression management and antisocial attitudes, criminal offenses, and sentence length. They found that individuals who had committed more serious offenses and had longer sentences scored higher on a measure of impression management than offenders who committed less serious offenses. The authors concluded that more procriminal individuals (i.e., those with longer prison sentences) have a greater motivation to manage their impression, as they have more to gain to convince correctional staff that they are “decent, upstanding individuals” (p. 29). In the present study, although offenders may have had difficulty concealing their procriminal attitudes, they may have had more success in convincing their probation officer that they are functioning well in other areas of their lives captured by the LSI-OR. Alternatively, unlike the other subscales, where information for scoring may be derived from collateral sources, the Procriminal Attitude/Orientation subscale is often based on information gathered directly from the interview with the offender. Therefore, as offenders experience repeated admissions into prison (resulting in higher-risk scores), they may become more adept at presenting with a prosocial attitude to criminal justice personnel.

Limitations

A number of limitations of the study need to be acknowledged. First, although we had multi-wave LSI-OR data that extended beyond the two time periods often seen in other studies, our sample size decreased as the number of available LSI-ORs increased. As a result, offenders with more LSI-ORs, suggesting greater involvement in the criminal justice system and, therefore, higher risk, contributed more to the overall growth trends than offenders with fewer LSI-ORs. Second, as previously noted, our data set did not include information about treatments received. Therefore, we are unable to comment on the extent to which the observed growth curves were the result of involvement (or lack thereof) in rehabilitation or treatment services. This was unfortunate, as LSI-OR scores are meant to inform decisions about which programs to attend, and the aim of this study was to identify which subscale scores appear to have largest impact on the growth trajectory in overall risk. The included sample consisted of individuals who continued to have contact with the criminal justice system over time; therefore, the degree to which the pattern of findings can be generalized to lower-risk offenders and offenders with less penetration into the system is unclear. Last, we did not have access to other scores on the LSI-OR, including the strength scores or the specific risk/need factors, which also could have been used to predict the risk growth curve (Andrews et al., 2004).

Practical Implications

The above limitations notwithstanding, the results of this study have important practical implications for informing key areas to target for rehabilitative interventions by both community supervision officers and within the prison system. As Probation, Parole, and Correctional Officers are often tasked with identifying the most pressing treatment needs, particularly for high-risk and high-rate offenders, their ability to distinguish which criminogenic needs will effect the greatest overall improvement in overall risk level is of utmost importance. The results of this study may provide a first step in assisting with this task. It is recognized that justice personnel draw on personal experience and other factors to inform decisions about prioritizing treatment targets to yield maximum benefits. These factors might include offender motivation, availability of treatment programs, or a hypothesized sequencing of factors to target for maximum benefit. It is suggested here that Probation/Parole Officers using the LSI-OR as a guide for making decisions about rehabilitation could take a more holistic approach to the array of criminogenic risk factors and consider how change in one area may, directly or indirectly, effect change in a positive direction in other risk factors, rather than view each subscale as a discrete target for treatment. As Andrews and colleagues (2004) noted, for offenders with multiple needs, probation officers “must prioritize resources to address the more pressing criminogenic needs” (p. 31). However, they also note that there is insufficient research to guide such critical decisions.

We contend that the analytical strategy presented here provides a unique approach to begin to examine the longitudinal, dynamic interrelationships among these criminogenic need factors. Further research of this kind is needed and may inform the important practical decisions of prioritizing resources for maximum impact. As the current study capitalized on offending patterns occurring naturally over time, another way to examine this issue through experimental research designs would be to examine change in overall risk score on the LSI-OR, or other dynamic, actuarial risk assessments, in response to specific treatment targeting one particular risk factor (e.g., procriminal attitudes, leisure time). This would significantly contribute to this area by providing data on the ways in which targeting one criminogenic need through programming can significantly affect other need areas, potentially resulting in positive improvement in overall risk.

Footnotes

Acknowledgements

We thank Kathy Underhill at the Ontario Ministry of Community Safety and Correctional Services (MCSCS) for providing the risk assessment data. We are also grateful to the reviewers of this manuscript for their helpful comments.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

References

Andrews

D. A.

Bonta

Wormith

J. S.

(1995). Level of Service Inventory–Ontario Revision (LSI-OR): Interview and scoring guide. Toronto, Canada: Ontario Ministry of the Solicitor General and Correctional Services.

Andrews

D. A.

Bonta

Wormith

J. S.

(2004). Level of Service/Case Management Inventory (LS/CMITM): An offender assessment system. Toronto, Ontario, Canada: Multi-Health Systems.

Andrews

D. A.

Bonta

Wormith

J. S.

(2006). The recent past and near future of risk and/or need assessment. Crime & Delinquency, 52, 7-27. doi:10.1177/0011128705281756

Andrews

D. A.

Robinson

(1984). The Level of Supervision Inventory: Second report. Toronto, Ontario, Canada: (Ontario) Ministry of Correctional Services.

Arnold

T. K.

(2007). Dynamic changes in Level of Service Inventory–Revised (LSI-R) scores and the effects on prediction accuracy (Unpublished master’s thesis). St. Cloud State University, MN.

Bates

D. M.

(2010). lme4: Mixed-effects modeling with R. Retrieved from http://lme4.r-forge.r-project.org/book

Beech

Friendship

Erikson

Hanson

R. K.

(2002). The relationship between static and dynamic risk factors and reconviction in a sample of UK child abusers. Sexual Abuse: A Journal of Research and Treatment, 14, 155-167.

Brown

S. L.

(2003). The dynamic prediction of criminal recidivism: A three-wave prospective study 1995-2002 (Unpublished doctoral dissertation). Queen’s University, Kingston, Ontario, Canada.

Brown

Amand

M. D.

St. Zamble

(2009). The dynamic prediction of criminal recidivism: A three-wave prospective study. Law and Human Behavior, 33, 25-45. doi:10.1007/s10979-008-9139-7

10.

Cauldy

M. S.

Durso

J. M.

Taxman

F. S.

(2013). How well do dynamic needs predict recidivism? Implications for risk assessment and risk reduction. Journal of Criminal Justice, 41, 458-466.

11.

Davis

C. G.

Thake

Weekes

J. R.

(2012). Impression managers: Nice guys or serious criminals? Journal of Research in Personality, 46, 26-31.

12.

Day

D. M.

Nielsen

J. D.

Ward

A. K.

Rosenthal

J. S.

Sun

Bevc

. . . Samuels

(2010). Criminal trajectories of two subsamples of adjudicated Ontario youths. Final research report submitted to the National Crime Prevention Centre (NCPC), Ottawa, Ontario. Retrieved from http://www.publicsafety.gc.ca/cnt/rsrcs/pblctns/djdctd-ntr-yth/djdctd-ntr-yth-eng.pdf

13.

Day

D. M.

Nielsen

J. D.

Ward

A. K.

Sun

Rosenthal

J. S.

Duchesne

. . . Rossman

(2012). Long-term follow-up of criminal activity with adjudicated youth in Ontario: Identifying offence trajectories and predictors/correlates of trajectory group membership. Canadian Journal of Criminology and Criminal Justice, 54, 377-413.

14.

Douglas

K. S.

Skeem

J. L.

(2005). Violence risk assessment: Getting specific about being dynamic. Psychology, Public Policy, and Law, 11, 347-383. doi:10.1037/1076-8971.11.3.347

15.

Gendreau

Goggin

Smith

(2002). Is the PCL-R really the “unparalleled” measure of offender risk? A lesson in knowledge cumulation. Criminal Justice and Behavior, 29, 397-426. doi:10.1177/0093854802029004004

16.

Gendreau

Little

Goggin

(1996). A meta-analysis of the predictors of adult offender recidivism: What works! Criminology, 34, 575-607.

17.

George

L. K.

(2009). Conceptualizing and measuring trajectories. In Elder

G. H.

Giele

J. A.

(Eds.), The craft of life course research (pp. 163-186). New York, NY: Guilford Press.

18.

Girard

Wormith

J. S.

(2004). The predictive validity of the Level of Service Inventory–Ontario Revision on general and violent recidivism among various offender groups. Criminal Justice and Behavior, 31, 150-181.

19.

Gottfredson

D. C.

Gerstenblith

S. A.

Soulé

D. A.

Womer

S. C.

(2004). Do after school programs reduce delinquency? Prevention Science, 5, 253-266.

20.

Hanson

R. K.

Harris

A. J. R.

(2000). Where should we intervene? Dynamic predictors of sexual offense recidivism. Criminal Justice and Behavior, 27, 6-35. doi:10.1177/0093854800027001002

21.

Jones

M. B.

Offord

D. R.

(1989). Reduction of antisocial behavior in poor children by nonschool skill-development. Journal of Child Psychology and Psychiatry, and Allied Disciplines, 30, 737-750.

22.

Jones

N. J.

Brown

S. L.

Zamble

(2010). Predicting criminal recidivism in adult male offenders: Researchers versus parole officer assessment of dynamic risk. Criminal Justice and Behavior, 37, 860-882. doi:10.1177/0093854810368924

23.

Long

J. D.

(2012). Longitudinal data analysis for the behavioral sciences using R. Los Angeles, CA: SAGE.

24.

Mahoney

J. L.

Stattin

(2000). Leisure activities and adolescent antisocial behavior: The role of structure and social context. Journal of Adolescence, 23, 113-127.

25.

Mills

J. F.

Kroner

D. G.

(2006). Impression management and self-report among violent offenders. Journal of Interpersonal Violence, 21, 178-192.

26.

Morris

Sallybanks

Willis

Makkai

(2003). Sport, physical activity and antisocial behaviour in youth: Trends and issues in crime and criminal justice (Research Report No. 249). Canberra: Australian Institute of Criminology. Retrieved from http://www.aic.gov.au/media_library/publications/tandi_pdf/tandi249.pdf

27.

Motiuk

L. L.

(1991). Antecedents and consequences of prison adjustment: A systematic assessment and reassessment approach (Unpublished doctoral dissertation). Carleton University, Ottawa, Ontario, Canada.

28.

Motiuk

L. L.

(1997). Classification for correctional programming: The Offender Intake Assessment (OIA) process. Forum on Corrections Research, 9, 18-22.

29.

Motiuk

L. L.

Bonta

Andrews

D. A.

(1990, June). Dynamic predictive criterion validity in offender assessment. A paper presented at the Canadian Psychological Association Annual Convention, Ottawa, Ontario.

30.

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, 156-176. doi:10.1037/a0035080

31.

Piquero

A. R.

(2008). Taking stock of developmental trajectories of criminal activity over the life course. In Liberman

A. M.

(Ed.), The long view of crime: A synthesis of longitudinal research (pp. 23-79). New York, NY: Springer.

32.

Piquero

A. R.

Monahan

K. C.

Glasheen

Schubert

C. A.

Mulvey

E. P.

(2013). Does time matter? Comparing trajectory concordance and covariate association using time-based and age-based assessments. Crime & Delinquency, 59, 738-763. doi:10.1177/0011128712459491

33.

Quinsey

V. L.

Jones

G. B.

Book

A. S.

Barr

K. N.

(2006). The dynamic prediction of antisocial behavior among forensic psychiatric patients. A prospective field study. Journal of Interpersonal Violence, 21, 1539-1565. doi:10.1177/0886260506294238

34.

Raynor

(2007). Risk and need assessment in British probation: The contribution of LSI-R. Psychology, Crime & Law, 13, 125-138.

35.

Sampson

R. J.

Laub

J. H.

(2003). Life-course desisters? Trajectories of crime among delinquent boys followed to age 70. Criminology, 41, 555-592.

36.

Schlager

M. D.

Pacheco

(2011). An examination of changes in LSI-R scores over time: Making the case for needs-based case management. Criminal Justice and Behavior, 38, 541-553.

37.

Singer

J. D.

Willett

J. B.

(2003). Applied longitudinal data analysis. New York, NY: Oxford University Press.

38.

Van Dieten

Robinson

(2007). The Service Planning Instrument (SPIn). Ottawa, Ontario, Canada: Orbis Partners.

39.

Ward

A. K.

Day

D. M.

Bevc

Sun

Rosenthal

J. S.

Duchesne

(2010). Criminal trajectories and risk factors in a Canadian sample of offenders. Criminal Justice and Behavior, 37, 1278-1300.

40.

Zamble

Quinsey

V. L.

(1997). The criminal recidivism process. Cambridge, UK: Cambridge University Press.