FORUM on Corrections Research

Affecting detention referrals through proper selection

Patricia Nugent and Edward Zamble¹
Collins Bay Institution, Correctional Service of Canada

This article briefly summarizes a study on the use of Detention Legislation in Ontario between April 1996 and August 1998. Detailed pre-release information was collected on 78 detained and 64 non-detained offenders. Recidivism rates were collected over a 2.5-year period on 122 released offenders. Detained offenders had lower rates of general recidivism and more arrests for a new violent offence. Several structured risk assessments designed to predict recidivism were used to assess each offender’s risk of recidivism.

Introduction

Detention legislation in Canada was designed to protect the public by incapacitating offenders who are dangerous and most likely to inflict serious physical harm on a future victim. Federal law specifies three criteria, which must be met in order to refer an offender for detention under section 129 (2)(a) of the Corrections and Conditional Release Act (CCRA). The legislation clearly implies than an offender should be assessed as high risk to commit a future violent offence.

To select dangerous individuals for detention effectively, one must accurately predict future violent behaviour. Advancements in criminological research support the fact that actuarial and structured risk assessments are superior to clinical and lay judgements in predicting future behaviour.² A number of validated risk assessment measures that predict general³ and violent recidivism⁴ are now available. Logic dictates that detained offenders should have higher risk scores on these instruments. Furthermore, if offenders are accurately detained, they may also be expected to have higher rates of recidivism upon release, particularly with respect to violent or serious offences.

Contrary to expectations, previous studies have found that detained offenders had lower rates of recidivism than offenders released on parole.⁵ Furthermore, these lower rates of violent recidivism were not attributed to the increased time served on detention.⁶ However, these studies lacked sufficient detail to determine what factors were effecting detention decisions and recidivism rates. The current study provides detailed comparisons of risk levels and recidivism rates between detained and non-detained offenders. It also examines factors that effect detention decisions, and contribute to the accurate prediction of recidivism among high-risk offenders.

Method and procedure

Data were collected on 142 male offenders, from medium and maximum-security facilities. Only detained offenders or those being released on Statutory Release were approached and assessed approximately 3 months prior to release. Detailed information was obtained from files, a semi-structured interview, several psychometric tests, and the following actuarial and structured risk assessments: General Statistical Information on Recidivism (GSIR), Level of Supervision Inventory-Revised (LSI-R), Psychopathy Checklist-Revised (PCL-R), and the Violence Risk Appraisal Guide (VRAG). One hundred and twenty two offenders were eventually released and followed up for 2.5 years. Recidivism was defined as an official conviction for any new offence or a new violent offence. Violent offences included assault, murder, armed robbery, use of a weapon, threatening, sexual assault and any sexual contact with children, but excluded non-violent sexual acts such as exhibitionism and voyeurism. Results were analyzed using a variety of statistical tests.

Results

The results replicated early findings that detained offenders had significantly lower rates of overall recidivism and remained offence free for longer periods in the community than non-detained offenders (see Figure 1). Detained offenders also had a lower rate of violent offending, although this difference was not statistically significant. However, detained offenders showed significantly lower rates of arrest for new violent offences (Figure 2). These results raise the following issues: Are the highest risk offenders selected for detention? If the risk of recidivism does not determine detention status, then what other factors are affecting detention decisions? How can the detention process be improved?

Figure 1
Differences in Survival Time between Detained and
Non- Detained Offenders for General Recidivism

Figure 1
Differences in Survival Time for Detained and Non- Detained
Offenders Prior to an Arrest for a New Violent Offence

Comparisons of risk levels

Detained and non-detained offenders were compared on four commonly used risk measures. Table 1 shows that detained offenders had significantly lower risk scores on the GSIR while the LSI-R, PCL-R, and the VRAG did not discriminate between the two groups. Detained offenders consistently had lower risk scores than non-detained offenders with the exception of Factor 1, the personality component, of the PCL-R. Interestingly, several studies have suggested that Factor 1 may not be a good predictor of recidivism.7 The current data support the conclusion that detained offenders, as a group, were not higher risk than the group of non-detained offenders.

Table 1

Differences between Detained and Non- Detained Offenders on Risk Measures
Risk Assessment		Detained ( n = 78)	Non-Detained (n = 64)	t	df	p
GSIR	M (SD)	0.94 (10.55)	- 6.13 (10.18)	4.03	140	.000
LSI	M (SD)	29.69 (9.77)	31.73 (7.56)	-1.40	140	.163
PCL- R Total Score	M (SD)	23.13 (7.43)	22.16 (5.96)	0.86	140	.391
Factor 1	M (SD)	9.85(3.70)	7.94 (3.33)	3.91	140	.002
Factor 2	M (SD)	10.04 (4.87)	11.25 (3.48)	-1.73	138	.085
VRAG	M	7.64	8.88	-0.74	138	.462

The results suggest that the difference in general recidivism will likely hold for violent recidivism and may become more apparent with a longer follow-up time.

Factors affecting detention

If detained offenders were not higher risk, then what other factors might be influencing the decision to detain certain offenders? Table 2 lists some of the variables where significant differences occurred between detained and non-detained offenders. In summary, detained offenders were older, and typically had been convicted of a sexual crime. Consequently, detained offenders had more child victims, were more likely to have known their victims, more likely to have a diagnosis of paraphilia, and had a higher incidence of reported childhood sexual abuse. Detained offenders had a greater number of current violent offences, and their current offences were rated as more severe. Detained offenders also were more likely to deny their offence, and use more socially desirable responses such as rationalization, denial, and impression management, in order to minimize their actions. Finally, detained offenders were more likely to have refused or quit treatment, and showed less progress in institutional treatment programs.

Table 2

Differences between Detained and Non- detained Offenders
Variable	Detained	Non- Detained	p
Sexual Offence1	80.8%	23.4%	***
Child Victim1	48.9%	15.0%	***
Paraphilia1	21.8%	3.1%	**
Childhood Sexual Abuse1	46.1%	21.3%	*
Stranger as a Victim2	1.87%	2.27%	*
Rationalization2	1.63%	1.29%	*
Denial of Offence2	2.22%	1.50%	*
Refusal of treatment2	0.77%	0.41%	***
Progress in treatment2	4.79%	7.20%	*
Age2	38.90%	32.73%	**
No. Current Violent Offence2	3.15%	1.95%	**
Offence Severity2	3.14%	2.18%	**
No. Past offences2	10.56%	18.81%	***
No. Past Youth Offences2	1.05%	4.05%	***
Criminal Associates2	1.45%	2.17%	***
No months employed2	61.66%	29.31%	*
Alienation2	5.20%	7.55%	***
Interpersonal problems2	6.86%	8.78%	*
Impulse expression2	6.86%	8.22%	***
p < .01, ** p < .001, *** p < .0001. 1. Numbers reflect percentage of offender in category. 2. Number reflects mean score.

Detained offenders scored lower on many factors previously shown to predict recidivism. For instance, they had significantly fewer past offences, fewer youth offences, fewer criminal associates, and more stable employment records. On personality measures, detained offender showed less alienation from society, had fewer interpersonal problems, and better impulse control. These latter characteristics are generally indicative of a lower risk of recidivism.

In summary, detained offenders could be described as socially reprehensible, dislikable, and uncooperative. Actual recidivism rates suggest these characteristics do not necessarily correspond to an increased risk of offending. Moreover, in some cases they may predict lower rates of recidivism. It is easy to understand how detained offenders have been unsuccessful in endearing themselves to staff and associates within the correctional system. It is important to recognize that decisions based on personal reactions may not provide better protection to the public.

Table 3

Partial Correlation between Structured Risk Assessment Measures and a New Conviction and a New Violent Conviction
Risk Assessment	New Conviction (n=120)	New Violent Conviction (n=120)
GSIR	-.38***	-.23*
LSI	.27**	.21*
VRAG	.26**	.23*
PCL-R	.15ns	.17ns
Factor 1	-.05ns	.0 4ns
Factor 2	.24**	.20*
*p < .05, ** p < .01, *** p < .001, ns = non significant. ( Note: The above r values reflect partial correlation which controlled for the time at risk.)

Improving prediction of recidivism

The final part of this study looked at variables that successfully predicted recidivism. Initially, several actuarial and structured risk assessment tools, commonly used within the Service, were compared on their ability to predict recidivism. Table 3 shows the partial correlation, controlling for time at risk, between risk assessment measures and recidivism. The GSIR had the strongest correlation with both overall recidivism and violent recidivism, although the LSI-R and VRAG were equally good predictors. This would be expected since these three instruments were specifically designed for the actuarial prediction of recidivism. While the PCL-R was not developed to predict recidivism, it is commonly used in conducting risk assessments. Factor 2 of the PCL-R predicted recidivism, but Factor 1 did not and consequently reduced the overall effectiveness of the PCL-R. The data indicate that the actuarial risk assessment measures were strongly associated with recidivism and imply that accurate prediction of both general and violent recidivism within this population is possible.

Other variables that contributed to the prediction of recidivism among this population were examined with multiple regression techniques. Table 4 shows that the strongest predictors of overall recidivism included: the GSIR, age, the total number of months incarcerated prior to the current offence, and the number of institutional charges incurred during the two years before release. Four personality traits, as measured by the Basic Personality Inventory, also predicted recidivism, including measures of anxiety, alienation, impulsiveness, and interpersonal problems. Other variables such as criminal self-efficacy, criminal associates, antisocial family members, and poor employment records also indicated an increased risk of recidivism. Note that age, offence severity, and convictions for sexual offences negatively related to recidivism and that age was the only variable negatively correlated with violent recidivism.

Table 4

Partial Correlation between Predictive Variables and a New Conviction and a New Violent Conviction
Risk Assessment	New Conviction (n=120)	New Violent Conviction (n=120)
GSIR	-.34***	-.23*
Previous Months Incarcerated	.36***	.28**
Age	-.31***	-.20*
Institutional Charges	.46***	.37***
Personality Factors ( anxiety, alienation, impulsiveness, interpersonal problems)	.31**	.18ns
Criminal Self- Efficacy	.24**	.11ns
Antisocial Family	.23*	.24*
Criminal Associates	.30**	.21*
Employment Rating	.30**	.23*
Offence Severity	-.24**	-.05ns
Sexual Offence	-.34***	-.17ns
* p < .05. ** p < .01. *** p < .001, ns = non significant. ( Note: The above r values reflect partial correlations which controlled for the time at risk.)

Statistical techniques can be used not only to determine combinations of variables that predict recidivism, but can also be used to compare the accuracy of predictions. A statistic called the Common Language Effect Size Statistic (CLES) can compare the accuracy of actual detention decisions to decisions that would have been made using the GSIR. The CLES represents the probability that an offender sampled from a high-risk group would be a recidivist and an offender sampled from a low-risk group would not. The CLES for the GSIR in this study was 72.24 for general recidivism and 65.54 for violent recidivism. This means that over 72% of offenders were correctly classified with respect to general recidivism and 65% for violent recidivist. In contrast, detention decisions made by the parole board correctly classified 34% for general recidivism and 42% for violent recidivism. The accuracy of predicting future recidivism can be further improved using a combination of predictor variables, such as shown in Table 4. These numbers provide strong support for the use of actuarial risk assessment methods and the assessment of criminogenic risk factors in order to improve the accuracy of detention decisions.

Conclusions

Current detention decisions appear to be influenced by several factors, such as the nature of the offence, denial, refusal of treatment, and undesirable personality traits. Human reactions to repulsive crimes and individuals appear to play a large role in influencing detention decisions. Although these reactions may be entirely justified, they are not necessarily predictive of future criminal behaviour. Detention decisions based on actuarial or structured risk assessment measures were more accurate than the detention decisions currently being made. The former approach would both protect the public more effectively and provide financial savings by reducing the cost of over-incarceration.

---

1. Collins Bay Institution, Box 190, Kingston, Ontario K7L 4V9.

2. Mossman, D. (1994). Assessing predictions of violence: Being accurate about accuracy. Journal of Consulting and Clinical Psychology, 62, 783-792. See also, Zamble, E., & Quinsey, V. L. (1997). The criminal recidivism process. Cambridge, England: Cambridge University Press.

3. Nuffield, J. (1982). Parole decision-making in Canada: Research towards decision guidelines. Ottawa, ON: Supply and Services Canada. See also Andrews, D. A., & Bonta, J. (1995). The Level of Service Inventory —Revised (LSI-R). Toronto, ON: Multi-Health Systems Inc.

4. Harris, G. R., Rice, M. E., & Quinsey, V. L. (1993). Violent recidivism of mentally disordered offenders: The development of a statistical prediction instrument. Criminal Justice and Behavior, 20, 314-335. See also Hare, R. D. (1991). Manual for the Revised Psychopathy Checklist.
Toronto, ON: Multi-Health Systems Inc.

5. Grant, B. A. (1996). Inmates referred for detention (1989/90 to 1993/94): A comparative analysis. Research Report R-45. Ottawa, ON: Research Branch, Correctional Service of Canada.

6. Motiuk, L. L., Belcourt, R., & Bonta, J. (1995). Managing High-Risk Offenders: A Post-Detention Follow-up. Research Report R-39. Ottawa, ON: Correctional Service of Canada.

7. Serin, R. C. (1996). Violent recidivism in criminal psychopaths. Law and Human Behavior, 20, 207-217. See also Hart, S. D., Kropp, R. D., & Hare, R. D. (1988). Performance of male psychopaths following conditional release from prison. Journal of Consulting and Clinical Psychology, 56, 227-232.