Classification without validity or equity: an empirical examination of the custody rating scale for federally sentenced women offenders in Canada.

• Author: Webster, Cheryl Marie

The [Correctional] Service's classification tools used to allocate necessary controls and program resources for women offenders have been validated and re-validated for initial security placement ... These offender classification instruments are critical to the effective and efficient risk management of women offenders while under federal sentence.

--Research Branch, Correctional Service of Canada (2003)

Introduction

Consistent with the wider shift within the correctional field toward increased reliance on actuarial techniques of risk management (Ericson and Haggerty 1997; Pratt 1997; Feeley and Simon 1992), the Correctional Service of Canada (CSC)--like prison services in many nations--has adopted an objective, statistical instrument to determine the initial penitentiary placement of offenders admitted to federal custody. Developed in 1987, the Custody Rating Scale (CRS) assesses the level of supervision and control required to safely detain inmates, assigning a security classification of minimum, medium, or maximum to all federally sentenced offenders. Seen by CSC researchers as an effective means of enhancing consistency and equity in internal decision making while also ensuring optimal public protection (Grant and Luciani 1998; Luciani, Motiuk, and Nafekh 1996), this tool has been widely adopted throughout federal correctional institutions for both men and women. In fact, if currently constitutes the foundation of the security classification system in Canada (Canadian Human Rights Commission 2003).

Despite playing this central role in federal correctional facilities, the CRS has not been without criticism. In particular, significant concern has arisen around the use of this classification tool for small subgroups of the inmate population. More specifically, a multitude of diverse entities--official commissions and government agencies as well as advocacy groups, practitioners, and academics--have questioned the applicability of the CRS to female offenders generally and to Aboriginal women especially (Canada, Auditor General 2003; McIvor and Johnson 2003; Office of the Correctional Investigator 2003; Hannah-Moffat 1999; Arbour 1996). (2) In essence, it has been argued that a classification scale which was designed for a homogeneous (white) male population will not be sensitive to differences based on gender, race/ethnicity, and social disadvantage. As a result, it may incorrectly assign (Aboriginal and non-Aboriginal) women to the various security levels (Canadian Human Rights Commission 2003; Monture-Angus 2000; Shaw and Hannah-Moffat 2000; Correctional Service of Canada 1990; see also, generally, Van Voorhis and Presser 2001).

This danger of unwarranted custody assignment is particularly disconcerting when one considers the powerful impact that the level of security has on both the offender and the correctional facility. On the one hand, classification decisions determine many of the inmate's living conditions, including supervision levels, (3) type of accommodation, (4) geographical location of incarceration, (5) use of restraints, (6) and inmate privileges. Similarly, security ratings dictate not only the kind of rehabilitative programming available to the offender (e.g., Aboriginal Healing Lodge, (7) work release, or community programs (8)) but also inmate eligibility for discretionary release (e.g., unescorted temporary absences (9) or conditional release (10)). On the other hand, the misclassification of offenders can have a detrimental impact on the institution's ability to fulfil its legislative and policy mandates. In particular, unjustified classification raises potential issues of inequity and discrimination, (11) as well as constituting a violation of CSC's obligation to place inmates in the least restrictive environment possible. (12) Equally problematic is the potential risk that it poses to the safety of the inmates, correctional staff, and the public, as well as to the most (cost-) effective use of limited resources.

Given these serious ramifications, this article extends the debate surrounding the use of the CRS for federally sentenced female offenders. More specifically, it reviews the empirical support presented by CSC for the validity of the CRS for initial security placement as applied to female offenders, in general, and to Aboriginal women offenders, in particular. While recent CSC characterizations (Blanchette, Verbrugge, and Wichmann 2002; Verbrugge and Blanchette 2002) of the overall Custody Rating Scale as a valid predictive instrument for the classification of women appear to be accurate at first glance, a more careful examination of the various components of the scale suggests a considerably more problematic conclusion. In particular, this article will show that the overall scale, one of its two sub-scales, and many of the individual items making up the CRS have weak or no predictive validity for Aboriginal and/or non-Aboriginal female offenders. Further, it will provide evidence that a substantial proportion of Aboriginal (relative to non-Aboriginal) women are, in fact, unjustly over-classified in higher security levels.

Precisely by demonstrating the indefensible nature of the CRS based on quantitative measures provided by CSC itself, this article adds an empirical dimension to the mounting theoretical criticisms surrounding the validity and equity of this tool in classifying federally sentenced women offenders. In fact, our examination of the CRS is intentionally limited in scope, demonstrating that one can accept--a priori--the underlying (theoretical, methodological, and statistical) decisions made by CSC when it carried out its evaluation of the validity of the security classification tool for federally sentenced offenders and still find the CRS to be remarkably flawed, indeed, our critique is based solely on CSC's own data and findings. In this way, our conclusion of the unjustified use of the CRS for women inmates constitutes a powerful complement to prior (theoretically driven) concerns. On the one hand, it provides plausible empirical support for these criticisms, increasing their persuasiveness. On the other hand, the continued acceptance of the CRS by those who negate the legitimacy of these theoretical concerns (and of the recommendations derived from them) is rendered problematic by CSC's own findings, which demand the same rejection of this classification tool.

Assessing the predictive validity of the CRS

In assessing the value of a particular scale, one of the principal criteria of evaluation is its validity. This term refers to the accuracy of the measure or, more simply, its ability to measure what it is intended to measure. In this case, the CRS is designed to assess the propensity of an offender to cause harm (Blanchette et al. 2002). In particular, it is expected to quantify an inmate's probability of escape and risk to public safety, as well as her need for supervision within the penitentiary. More simply, this scale is intended to assess two types of risk: (1) risk to the public in the event of escape; and (2) risk to staff, other inmates, and self caused by problems relating to institutional adjustment (Canadian Human Rights Commission 2003).

Presumably in order to correspond to these two dimensions of the principal construct, the CRS consists of two independently scored subscales: a five-item Institutional Adjustment (IA) scale and a seven-item Security Risk (SR) scale. The individual items within each sub-scale are scored using a statistical weighting scheme and subsequently summed to provide a total score for each of the sub-scales. Cut-off points divide each sub-scale into three groups, producing ratings of minimum, medium, and maximum security (Blanchette et al. 2002). In the event of disagreement between the two sub-scales whereby an offender is classified atone level of custody according to one of the sub-scales but at another under the second sub-scale, the final CRS security level designation is determined by the scale assigning the higher classification rating (Luciani et al. 1996).

In assessing the extent to which this statistical process accurately measures the intended concept of risk, the most appropriate "yardstick" against which to judge its success or failure is the scale's predictive validity (Blanchette et al. 2002). Indeed, the logical framework of the classification system is essentially predictive in nature (Kane 1986), as the CRS is expected to foresee or anticipate future (risky or harmful) behaviour. Therefore, the extent to which the CRS is accurate in classifying women offenders in terms of predicting relevant future behaviour provides a valuable measure of the (predictive) validity of the scale (Luciani et al. 1996). To put it differently, to the degree that the numerical scores on the CRS are indicative of differing probabilities of offenders' risk and consequently discriminate among the three security classification groups, it can be considered a valid (predictive) instrument.

In evaluating the predictive validity of the CRS, one is concerned primarily with the relationship or correlation between the overall classification scale (and its various components) and the outcome measure--that is, the problem behaviour that the instrument is designed to predict (Kane 1986). In our case, CSC--in its most recent evaluation of the validity of the CRS for female offenders (Blanchette et al. 2002)--has operationalized the construct of "risk" as institutional misconduct (as measured by institutional incident data within a six-month follow-up period (13). Hence, one would expect that the overall scale--as well as its sub-scales and individual scale items--would be a statistically significant predictor of institutional incidents. Indeed, evidence that the scale as a whole is associated with institutional incidents does not demonstrate the predictive power of the sub-scales or of individual factors (Buchanan, Whitlow, and...