The Statistical Information on Recidivism (SIR) Scale is a static actuarial risk assessment tool developed and employed within the Correctional Service of Canada for the assessment of risk of general recidivism with adult male federal offenders (serving sentences of a minimum 2-year duration). This article discusses the development, applications, and validation research of the SIR Scale, including its use with special populations.
Development and Item Content
Developed initially in 1982 by Joan Nuffield, the SIR Scale was constructed and validated on a combined sample of 2,475 adult male federal offenders who were released from 1970 through 1972 and followed up for 3 years in the community. The SIR Scale was developed through what has become known as the Burgess method, that is, by differentially weighting predictor variables shown to successfully discriminate recidivists from nonrecidivists. For each category or value of a predictor demonstrating a 5% increase or decrease in recidivism from the overall rate, a weighting of +1 or −1, respectively, is assigned. The weights are increased in either direction based on the degree of difference from the overall recidivism rate associated with a given category or value on a variable. In this manner, stronger predictors (i.e., variables with the best discriminating power) are given the heaviest numerical weights. As the SIR Scale was developed in a purely data-driven manner in the absence of theory, it is termed an Empirical Actuarial Scale.
In the original Nuffield study, the construction sample had a mean 43.9% recidivism rate (i.e., rearrested for a new offense), or conversely, a 56.1% success rate (i.e., were not rearrested for a new offense). The Nuffield scoring of the SIR was conducted in terms of success, or nonrecidivism. For instance, having one or two prior imprisonments was associated with success rates of 55.1% and 53.6%, respectively; as this is less than a 5% departure from the overall success rate in the sample, values of 1 or 2 were assigned a score of 0. Having 0 prior imprisonments, however, was associated with a considerably higher rate of success, 79.1%; given that this is a 23% increase from the overall rate of success, this was given a weighting of 23/5 = 4.6 or “−4” (values were rounded down). By contrast, having three or four prior imprisonments was associated with success rates of 48.5% and 51.1%, respectively, which are each associated with 7.1% and 5% decreases in success, so they were assigned +1 points each. For this item, this was done up to a maximum of five to eight previous imprisonments, which was associated with success rates ranging from 43.3% to 40.5% (about a 12–15% decrease), and for consistency, a weighting of +2 was assigned.
In total, 15 variables shown to discriminate recidivists from nonrecidivists were identified and given differential weights in this manner. These variables include
- age at admission,
- number of previous imprisonments,
- previous breach of parole supervision or mandatory supervision,
- number of previous escapes,
- security classification,
- age at first adult conviction,
- number of previous convictions for assault,
- marital status,
- interval at risk,
- number of dependents,
- aggregate sentence,
- number of previous convictions for violent sexual offenses,
- number of previous convictions for break and enter,
- employment status at time of arrest for commitment offense, and
- type of commitment offense.
Once scored, the items were simply summed to generate an overall score that was associated with probability of success on release. The SIR was scored in a manner originally, such that higher scores were associated with increased recidivism (or decreased success) and lower scores were associated with increased success on release (or nonrecidivism), with possible scores ranging from −27 to +30. This has since been reversed so that possible scores range from −30 to +27, with lower scores indicating poorer prospects for success upon release. The SIR was divided into five score bands associated with different rates of success or nonrecidivism. In a validation sample of 1,237 offenders, these score bands with associated rates of success were as follows:
- Group 1: Very Good (−6 to −27) 84.5% success
- Group 2: Good (−1 to −5) 67.9% success
- Group 3: Fair (0 to +4), 50.8% success
- Group 4: Fair to Poor (+5 to +8) 41.9% success
- Group 5: Poor (+9 to +30) 31.6% success
Applications
The SIR Scale has undergone minor revisions in the years subsequent to Nuffield’s development of the tool, with the SIR-Revised-1 developed in 2002 and still in use. Note, the term SIR Scale is used throughout this article for consistency in language. Experimental variations on the SIR and SIR-Revised-1, termed SIR proxies, have been developed with parallel item content and similar item scoring and weighting for examination with special populations for which the SIR Scale had not been originally validated; most notably, female and Aboriginal offenders (discussed below). It is standard operating procedure with the Correctional Service of Canada for parole officers to administer the SIR at the point of intake in the offender’s admission to custody in order to evaluate potential for success on release to the community. It is not, however, administered to women or Aboriginal offenders.
The SIR Scale is one important piece of information the Parole Board of Canada uses in evaluating an offender’s suitability for conditional release. As the SIR is made up of static items (i.e., historical, and generally unchanging), it cannot assess changes in risk from positive progress made on the individual’s correctional plan or resulting from other credible change agents (e.g., specialized treatment completion, improvement in supports, major deterioration in health). For this reason, the SIR is typically supplemented with other measures that assess dynamic (i.e., changeable) risk markers to assess suitability for conditional release. Moreover, although subsequent validations of the SIR have provided some support for the ability of scores on the tool to predict other forms of recidivism (e.g., return to custody for a sexual or violent offense), the SIR Scale is used primarily to forecast probable success in the community or new returns to custody for any new offense in general (i.e., general recidivism).
Subsequent Validations
Subsequent validations of the SIR Scale have reaffirmed its properties for predicting post-release returns to custody. One of the largest validations was performed in the early 1990s by federal government researchers from the Ministry of the Solicitor General of Canada. In this study, 3,267 male prison inmates released from 1983 to 1984 and rated on the SIR Scale were followed up for a minimum 3 years post-release. The SIR Scale in this new sample demonstrated similar rates of success and return to custody across the five score bands as in the original Nuffield evaluation. It also displayed impressive predictive accuracy for general recidivism, with an area under the receiver operating curve (AUC) value of .76. With possible values ranging from 0 to 1.0, an AUC value of .76 would be interpreted to mean that there is a 76% chance that a randomly selected recidivist (for any new offense) would have a worse score on the SIR Scale than a randomly selected nonrecidivist.
As the volume of research on the SIR has accumulated, meta-analyses have been possible to examine how well the SIR and its variants predict different forms of recidivism, although most of these reviews have examined how the SIR has fared in specific offender populations, such as sexual or violent offenders. Two meta-analyses that examined the prediction of violent recidivism in violent offenders found the SIR Scale fared reasonably well, with AUC values ranging from approximately .63 (17 studies) to .68 (three studies). A further major meta-analysis published in 2009 of recidivism prediction among men who committed sexual offenses yielded similar findings. Specifically, AUC equivalents were obtained for the prediction of sexual recidivism (AUC = .64, five studies), violent recidivism (AUC = .71, three studies), and general recidivism (AUC = .76, four studies). In all, the existing research to date has supported the predictive accuracy of SIR scores for return to custody, particularly when it comes to assessing risk of any (i.e., general) recidivism.
Research With Special Populations
In 1988, it became policy within the Correctional Service of Canada to not administer the SIR Scale to persons of Aboriginal ancestry owing in part to the limited research available at the time. At least four evaluations to date conducted by Canadian federal government researchers have examined the properties of the SIR Scale or the SIR-Proxy with Aboriginal men and Aboriginal women federal offenders. Two early evaluations of the SIR with Aboriginal men in 1989 (n = 49) and 1993 (n = 269) found the SIR Scale had acceptable accuracy for predicting general recidivism with AUC values in the low .70s. Much larger subsequent examinations of the SIR-Proxy with Aboriginal men in 2002 (n = 1,211) and 2012 (n = 2,560) found that it predicted sexual (AUC = .60 and .62, respectively), violent (AUC = .63 and .57, respectively), and general (AUC = .68 and .63, respectively) recidivism. Of note, the effect sizes for the latter two evaluations tended to be higher for non-Aboriginal offenders in the prediction of general and violent recidivism but were higher for Aboriginal offenders in the prediction of sexual recidivism.
The latter two studies benefited from a large sample size, which enabled them to examine the SIR-Proxy among female offenders, including Aboriginal and non-Aboriginal subgroups. In the subset of 342 women offenders from the 2002 evaluation, the SIR-Proxy showed good predictive accuracy for general (AUC = .77) and violent (AUC = .73) recidivism. The larger 2012 evaluation similarly found the SIR-Proxy to predict general and violent recidivism among 246 Aboriginal (AUC = .65 and .70, respectively) and 666 non-Aboriginal (AUC = .74 and .82, respectively) women. Although prediction effect sizes among male and female offenders did not appear to differ substantively, again, prediction magnitudes were somewhat higher for non-Aboriginal offenders with respect to violent and general recidivism.
Final Thoughts
The SIR Scale continues to be a primary risk assessment tool to assist in appraising suitability for conditional release and prospects of post-release success among adult male, non-Aboriginal federal offenders in Canada. As a static actuarial tool, the SIR is limited in its ability to assess change but is a useful complement to other general and specialized dynamic measures that can track progress and change. Although psychometric research to date demonstrates that the SIR Scale and its variants can predict recidivism among special populations, such as women and Aboriginal offenders, at this current juncture, the tool is not used with these groups, and alternative measures are used instead to aid in the assessment and management of risk.
References:
- Barnum, G., & Gobeil, R. (2012). Prediction of re-offence using the SIR-R1 and a proxy (Corrections Research Report No. R-281). Ottawa, Canada: Correctional Service of Canada.
- Bonta, J., Harman, W. G., Hann, R. G., & Cormier, R. B. (1996). The prediction of recidivism among federally sentenced offenders: A re-validation of the SIR scale. Canadian Journal of Criminology, 38, 61–79.
- Nafekh, M., & Motiuk, L. L. (2002). The Statistical Information on Recidivism-Revised (SIR-R1) Scale: A psychometric examination (Corrections Research Report No. R-126). Ottawa, Canada: Correctional Service of Canada.
- Nuffield, J. (1982). Parole decision-making in Canada. Ottawa, Canada: Solicitor General Canada.