Skip to content | Skip to institutional links

Common menu bar links

Institutional links

Research Reports

Warning This Web page has been archived on the Web.

Revalidation of the Security Reclassification Scale for Women (SRSW)

2008 N^o R-191

Renée Gobeil

Correctional Service Canada

September, 2007

ACKNOWLEDGMENTS

The contributions of a number of individuals were essential to the successful completion of this study. The author is grateful to Jim Bonta, Kendra Delveaux, Brigitte Lavigne, and Kelley Blanchette who volunteered their time to participate in an advisory committee during the planning stages of this revalidation. All provided invaluable assistance in framing and defining the most relevant issues, as well as methodological suggestions. A special thanks to Brigitte Lavigne who provided additional guidance with her careful fact-checking and her suggestions regarding phrasing; Anne Kelly also provided assistance in this area. Virginia McGowan provided editorial suggestions at this stage.

Appreciation is also extended to Collette Cousineau and Ben Vuong who provided important assistance with regards to the preparation, coding, and analysis of quantitative data. Kelley Blanchette was also available to answer innumerable questions regarding the finer points of the scale.

The comments and suggestions of CSC field staff ensured a comprehensive review was possible. The feedback from primary workers, parole officers, and team leaders who have used the Security Reclassification Scale for Women was essential. A special thanks to those who took the time to share their remarks and observations.

EXECUTIVE SUMMARY

The Security Reclassification Scale for Women (SRSW; Blanchette & Taylor, 2005), a gender-informed instrument used in reviews of women offenders’ security classification, was implemented nationally in June 2005. Anchored in actuarial methods, the scale provides a recommended security level that caseworkers consider jointly with their clinical appraisals in arriving at security placement recommendations. Results of the scale’s field-test indicated that the SRSW had acceptable statistical reliability and validity, with the SRSW either equalling or outperforming the preceding clinically-based method. Notwithstanding, it has been two years since the implementation of the SRSW, and the current revalidation study allowed verification of whether the SRSW continues to function as desired despite changes in the inmate population.

Analyses involved 443 consecutive security reviews completed between the instrument’s June 2005 implementation and February 2007. Data included all SRSW recommendations, caseworker recommendations, and actual security placements; scores on measures of risk, need, and reintegration potential; and information regarding institutional behaviour, discretionary release, and post-release returns to custody. Feedback was also gathered from seven caseworkers with extensive experience using the SRSW.

Overall, findings demonstrated that the SRSW continues to be of considerable utility in assisting to make security review decisions. The scale was determined to reliably measure what it purports to measure, as well as to differentiate women according to institutional adjustment and risk. Women recommended to lower security classifications by the SRSW were better adjusted (as indicated by lower frequencies of incidents, lower need ratings, and higher reintegration potential) and lower risk (as demonstrated by lower risk ratings and higher frequencies of discretionary release) than their counterparts recommended to higher security classifications. Moreover, there were no differences in the distribution of scale recommendations associated with Aboriginal ethnicity, and the scale’s ability to predict institutional adjustment and the granting of discretionary release was equal or greater when considering only Aboriginal women.

Notably, however, there were high rates of inconsistency between SRSW and caseworker recommendations, even after accounting for the use of the scale’s built-in discretionary range. In most cases, inconsistencies represented caseworker recommendations which were higher than SRSW recommendations. The most common reasons for clinical dissention with the SRSW were linked to institutional behaviour or attitude, or insufficient progress on correctional programming. However, SRSW recommendations were found to be more predictive of both institutional misconduct and the granting of discretionary release than were caseworker recommendations. It is recommended that further research be conducted on inconsistencies between caseworker and scale recommendations. Doing so would ideally situate the Correctional Service of Canada to continue to respond to concerns alleging women’s over-classification (e.g., Canadian Association of Elizabeth Fry Agencies, 2004; Canadian Human Rights Commission, 2003), thereby allowing the Service to maintain its position as a fore-runner among correctional agencies in meeting the call for the empirical, gender-informed development, validation, and revalidation of classification instruments.

ACKNOWLEDGMENTS
EXECUTIVE SUMMARY
TABLE OF CONTENTS
LIST OF TABLES
LIST OF FIGURES
LIST OF APPENDICES
INTRODUCTION
METHOD
- Samples
- Data Sources
RESULTS
DISCUSSION
REFERENCES

INTRODUCTION

This report summarizes the results of a revalidation of the Security Reclassification Scale for Women (SRSW; Blanchette, 2005). These results are intended to supplement the information obtained during the instrument’s field test by examining whether security reclassification decisions made since the instrument’s implementation classify women more appropriately than did recommendations under the previous system. Moreover, the present examination examined the impacts of the implementation of the instrument, as well as its reliability and validity. In order to provide context for the revalidation, a brief review of offender security classification will first be provided.

Offender Classification

Offender classification is a process used to identify the degree of risk posed by individual offenders, both within an institution and upon release to the community, and is intended to inform the management of this risk. In practice, offender classification influences security placements, supervision requirements, discretionary release, and program placement (Motiuk, 1997).

Given that offender classification has such far-ranging implications, it is important that classification systems be fair and efficient. In recent decades, there has been a movement towards actuarial classification models, which have been found to be more equitable and defensible than their clinical counterparts (Austin & Hardyman, 2004; Brennan, 1987). Research indicates that subjective methods result in higher rates of over-classification than do actuarial ones (Bonta & Motiuk, 1990; Austin, 1983). Security classifications derived from actuarial methods tend to be lower and to include fewer false positives (predictions that an offender is high risk when this is not the case) than do clinical methods (Buchanan, Whitlow, & Austin, 1986). These findings have important consequences, as over-classification can result in offenders being detained in overly secure environments and/or denied discretionary release. Conversely, under-classification can result in very serious consequences, such as in cases where a high-risk offender escapes or is inappropriately released to the community. It is clear that an effective and equitable system is needed, and that an actuarial method is likely to meet these requirements.

Actuarial methods, due to their objective and reproducible character, also lend themselves well to evaluations and validation studies (Alexander & Austin, 1992). Such research allows correctional officials to ensure that classification methods are meeting functional, policy, and legislative goals. Moreover, such studies can underline areas where improvement is possible in terms of efficiency and predictive ability. For these reasons, evaluation through revalidation studies is an important part of the development of classification instruments.

Classification of Women Offenders

Most actuarial classification methods are developed using samples representative of the majority prison population; in other words, such methods are developed using samples of Caucasian men (Shaw & Hannah-Moffat, 2000). Consequently, women offenders are commonly classified using systems developed for a very different group. Though some attention has been focused on this issue, the tendency to use ‘gender-neutral’ methods has been slow to change. A 1991 study found that the majority of American states (40 of the 48 states surveyed) used the same classification system for men and women (Burke & Adams, 1991), but a more recent review found that only minimal changes had occurred, with 38 of 50 states using the same system for men and women offenders (Hardyman & Van Voorhis, 2004).

The use of the same instruments for both men and women is of concern because research indicates that classification systems developed for men offenders, even if purportedly gender-neutral, can result in misclassification - often over-classification - of women offenders. This misclassification is due, perhaps, to the instruments’ focus on variables which may not be relevant to women offenders (Hardyman & Van Voorhis, 2004; Van Voorhis & Presser, 2001). When a classification instrument yields a security level recommendation that is inconsistent with the caseworker’s clinical appraisal, the caseworker must assign his or her appraised level (i.e., override the scale). Though this practice is sometimes necessary with any classification instrument, in the Hardyman and Van Voorhis (2004) study, clinical overrides took place at inappropriately high rates in nine of the 38 states which used the same classification system for men and women offenders - in one case, 70% of the time. These rates suggest that ‘gender-neutral’ classification systems are perceived by users as ineffective for women offenders.

A further difficulty lies in the lack of validation studies concerning classification instruments used with women offenders. Validation studies of instruments used to classify both men and women are usually based on participants who reflect the majority of the prison population - men. Moreover, a number of instruments developed to classify women offenders specifically have not been validated, often due to insufficient sample sizes (Hannah-Moffat, 1999). For instance, in their review of American states’ classification procedures, Hardyman and Van Voorhis (2004) found that only about a quarter of the states (14 of 50) had validated their classification instruments with women offenders. Without validation of instruments, it is impossible to conclude that they are, in fact, classifying offenders appropriately.

The above findings underline that it is inappropriate to assume either that non-gendered or non-validated methods appropriately classify women. Consequently, there have been a number of strong suggestions that new security classification instruments for women must be developed (e.g., Canadian Association of Elizabeth Fry Societies, 2004; Canadian Human Rights Commission, 2003), that such instruments must include empirically derived gender-informed criteria, and that they should be subject to evaluation and validation (e.g., Shaw & Hannah-Moffat, 2000).

Security Classification in Canada

Initial Classification

In Canada, security classification is guided by the Corrections and Conditional Release Act (CCRA, 1992). According to this legislation, federal offenders are to be provided with a security classification of minimum, medium, or maximum according to their institutional adjustment, escape risk, and risk to the public in the event of an escape. The CCRA further requires that Correctional Service Canada "use the least restrictive measures consistent with the protection of the public, staff members and offenders" (s. 4[d]) in assigning this security rating.

Upon admission, each federal offender undergoes a routine admission process, which includes the administration of the Custody Rating Scale (Solicitor General of Canada, 1987), an actuarial measure that informs his or her initial security classification. As an offender’s institutional adjustment and / or security risk are assessed as more problematic, the security level recommendation increases. The scale is to be used in conjunction with professional judgment; in instances where the instrument’s recommendation is at odds with the caseworker’s clinical appraisal, however, the caseworker must provide a rationale for the assigned recommendation. Though the measure was originally developed with male offenders, several studies have assessed the reliability and validity of this scale among women and Aboriginal offenders, and have found it to be both reliable and valid (Blanchette & Motiuk, 2004; Blanchette, Verbrugge, & Wichmann, 2002; Grant & Luciani, 1998; Luciani, Motiuk, & Nafekh, 1996). Notwithstanding these findings, given the drive for gender-specific actuarial assessments, a contract has been tendered for the development of a new gender-informed tool for initial security classification.

Reclassification

Offenders’ security classifications can (and should) shift during incarceration. The CCRA provides specific guidelines for the periodic reclassification of federal offenders: for most offenders, reviews occur at a minimum of once a year.[1] Correctional Service Canada policies further shorten this minimum period. Commissioner’s Directive 710-6 indicates that additional security reviews are necessary whenever there is reason to believe that the appropriate security level for an offender has changed, or before making a recommendation for any decision, including transfers, temporary absences, work release, or parole (CSC, 2006b). Moreover, this Directive indicates that the security classifications of women rated as maximum security must be reviewed every six months because of the restrictive nature of this security level. These policies facilitate the timely reintegration of offenders to lower levels of security, and ultimately, to the community.

A reclassification instrument, the Security Reclassification Scale, was developed in 1998 for use with male offenders. This reclassification instrument includes mainly dynamic variables reflecting institutional adjustment, escape risk, and risk to the public in the event of an escape. Again, the instrument provides an actuarial score which anchors the security reclassification, but is to be used in conjunction with caseworkers’ professional appraisals. Where the actuarial score is inconsistent with clinical appraisal, the overall assessment must provide a rationale to justify the assigned security level.

Due to previous critiques about the use of instruments developed for men offenders with women offenders (e.g., Brennan, 1998; Farr, 2000; Hannah-Moffat & Shaw, 2001), the Security Reclassification Scale was not used with women offenders. Instead, a second instrument, the Security Reclassification for Women (SRSW), was developed specifically for this group (Blanchette & Taylor, 2005). The SRSW is similar to the instrument developed for male offenders in that it concentrates on dynamic variables and allows for professional discretion. This instrument was statistically derived using a sample of women offenders, however, and includes distinct variables and variable weights.

The Security Reclassification Scale for Women

The Security Reclassification Scale for Women (SRSW) is an objective, gender-informed classification instrument developed between 1998 and 2000 for federally sentenced women. It was field tested from 2000 to 2003, and implemented nationally in June 2005. Using actuarial methods, the instrument provides a recommended security placement for women. This recommendation anchors the caseworker’s clinical appraisal of the offender’s risk. In cases where the instrument’s recommendation and the final clinical appraisal are inconsistent, caseworkers must document their rationale for their reclassification recommendation. Similar to the measure developed for male offenders, the SRSW focuses predominantly on dynamic factors including those associated with escape risk, risk to the public if an escape were to occur, institutional adjustment and behaviour, focusing on changes in the offender’s behaviour since the preceding review.

Development

The development of the SRSW began in 1998, soon after that of the Security Reclassification Scale for male offenders. Based on reviews of previous empirical research and consultation with individuals involved in developing classification instruments, 176 variables thought to be predictive of risk for women offenders were identified. Among these variables were both historical risk factors and dynamic variables, including institutional behaviour, program performance, and substance use.

Analyses of the relationships between these variables and the outcomes of 285 security reclassification events for women offenders (using the previous clinically-based method) reduced the sample of potentially predictive variables considerably. The remaining items were then entered into a forward stepwise regression in order to further reduce the number of variables and retain only those most strongly related to clinical security reclassification decisions. Using this methodology, nine variables were selected for inclusion in the SRSW:

progress against correctional plan / offender’s motivation during review period;
pro-social / positive family contact during review period;
serious disciplinary offences during review period;
number of recorded incidents during review period;
current institutional pay level
involuntary segregation due to being a danger to others or to the institution during review period;
number of successful escorted temporary absences during review period;
history of having been unlawfully at large from a temporary absence, work release, or supervision; and,
Custody Rating Scale incident history.

Accepted statistical procedures were used to assign weights to each variable; cutoff values were then established in order to recommend a security level of minimum, medium, or maximum. Notably, a 10% discretionary range was provided at each cutoff, allowing flexibility in assigning security classifications to those women whose scores fall near the threshold values. Professional discretion was also recognized in developing this instrument. Though the actuarial instrument’s score must anchor security reclassification decisions, when mitigating circumstances or additional information lead to the clinical appraisal being at odds with the instrument’s recommendation, the caseworker may recommend an alternative security classification.[2] In such cases, a comprehensive explanation is required in the Assessment for Decision.The final decision-making authority at each institution (i.e., the warden or the Kikawinaw at the Healing Lodge) also has the opportunity to assign a security classification which differs from either the SRSW recommendation or the caseworker recommendation.

In practice, the SRSW is scored by a caseworker using a computer application that automatically pulls a portion of the relevant data from the automated Offender Management System. The application then guides the caseworker through the steps where input is needed and provides a recommendation and the opportunity to accept or change the recommendation based on the clinical appraisal.

Field test

Between 2000 and 2003, the SRSW was field tested with a sample of 580 security reclassification reviews (involving a total of 323 women; Blanchette & Taylor, 2005).[3] During this period, security reclassification reviews were conducted using the previous method (clinical appraisal only), but the SRSW was completed shortly afterward to allow comparison between the reclassification decisions produced by each method. Data were also collected for each woman on variables used in validating the new reclassification instrument, including demographic data, offence history, institutional misconducts, discretionary release, and post-release outcome.

Analyses indicated that the SRSW had acceptable statistical reliability and validity; for all analyses, the SRSW either equalled or outperformed the preceding clinically-based method. The SRSW also classified a larger proportion of offenders at a minimum security level than did the preceding method. Recommended SRSW levels were found to be positively related to other indicators of risk, demonstrating that as risk and need increased, so did SRSW classification. SRSW classification levels were also found to be related to involvement in minor and major misconducts and to the granting of discretionary release. The researchers conducting the field test concluded that the instrument provides Correctional Service Canada with "a gender-informed security classification tool with practical utility that meets legal requirements for the least restrictive measures of confinement. The SRSW assists in allaying concerns regarding women’s over-classification… and provides a nationally standardized, objective approach and an accountability framework for both inmates and staff" (Blanchette & Taylor, 2005, p. 52-53).

Implementation

Based on the results of this field test (and approval from an external panel of experts), the SRSW was implemented nationally by Correctional Service Canada in June 2005. Commissioner’s Directive 710-6 indicates that the SRSW must now anchor all security reclassification decisions for women offenders (CSC, 2006b).[4] Since its implementation, over 500 security reclassification events have occurred using the SRSW. This number of events provides a sufficient sample for the instrument’s revalidation.

Purpose of Study

An important component of the use of classification instruments is the periodic review of their applicability to those for whom they are intended. This review process allows verification of whether the instrument is continuing to function as desired despite any changes to the target group. Given the number of security reclassification decisions reached for women offenders since the SRSW’s implementation, a revalidation study was deemed appropriate.

In identifying the most salient revalidation issues, published guidelines for the evaluation of classification instruments (Alexander & Austin, 1992) were used as a starting point. Feedback was also solicited from an advisory committee composed of persons knowledgeable in the fields of classification, evaluation, and women offender issues, as well as the instrument’s developer. Three key research questions arose during this process:

Does the instrument result in more appropriate classifications than those produced by the previous method?
What were the impacts (if any) of implementation of the measure?
Does the instrument continue to measure what it purports to measure (i.e., is it valid)?

METHOD

Samples

The original sample of reclassification events comprised 525 consecutive security reviews for adult women offenders completed from the instrument’s June 2005 implementation to February 2007. In 32 cases wherein more than one scale was completed for the same reclassification decision (for example, due to an error in the original completion), the scales not corresponding to the final decision were eliminated. A further seven cases were eliminated as they were not associated with any reclassification decision. Additionally, the 52 cases wherein the review period was shorter than 90 days[5] were manually verified to ensure it was appropriate for them to have been completed; 43 of these were eliminated.[6] The final sample of reclassification events therefore comprised 443 reviews. Certain offenders had two or more records as their security classification had been reviewed more than once between June 2005 and February 2007. As such, the 443 security reclassification events involved a total of 296 individual women.

A sample of caseworkers also provided qualitative feedback. During fall 2006, comments were solicited from primary workers, parole officers, and team leaders who routinely use the SRSW. This feedback sample comprised a total of seven participants, with one participant from each of the six regional facilities and the Regional Psychiatric Centre.

Data Sources

Most data used in this study were extracted retrospectively from existing offender data records. The principal measure was the Security Reclassification Scale for Women and its components. As described previously, this scale includes nine items, seven of which are dynamic in nature, which were previously found to be predictive of security reclassification outcome. Scores on this scale span approximately a 30-point range, with higher scores representing higher assessed risk and resulting in higher security classification recommendations. The range of possible SRSW scores, including cutoffs for each security recommendation and their respective discretionary ranges, are represented in Figure 1.

Figure 1. SRSW Scores and Security Recommendations

Figure 1. SRSW Scores and Security Recommendations

The study also incorporated Offender Management System data, which include extensive information for each offender in the federal correctional system. Scores on measures of risk, need, and reintegration potential, as well as information regarding institutional behaviour, discretionary release, and post-release returns to custody were extracted for all women for whom a security review had been completed during the study period.

Interviews were used to gather feedback from staff members who have extensive experience with the SRSW. This data collection method permitted very rich information to be gathered, and allowed unexpected issues to surface. A detailed protocol for the interviews is presented in Appendix A. A semi-structured format was used, allowing participants to guide the interaction and comment only on areas relevant to their experience.

RESULTS

Sample

Security Reclassification Event Sample

The sample was composed of 443 consecutive security reviews conducted from June 1, 2005 to February 1, 2007. This sample of security reclassification events involved 296 federally incarcerated adult women for whom the average age was 33.78 years (SD = 9.29), ranging from 19 to 69. Table 1 presents a summary of ethnicity and marital status for this sample. Notably, the proportion of Aboriginal women within this sample (49%) was much higher than the proportion of inmates of Aboriginal origin within the female federally sentenced inmate population (27%).[7]

Table 1. Demographic Information

Demographic Variable	Number (%) of Offenders
Ethnicity
Caucasian	127 (43)
Aboriginal	145 (49)
Black	11 (4)
Other visible minority	8 (3)
Unknown	5 (2)
Marital Status
Single / Widowed / Divorced	202 (68)
Married / Common-law	91 (31)
Unknown	3 (1)

Note. N = 296.

Forty-seven (16%) of the women in this sample were serving life sentences; of those with determinate sentences (n = 249; 84%), the mean term aggregate sentence length was 3.96 years (SD = 2.69). Offender files were examined to ascertain the offences for which the women were convicted. Over three quarters (77%) of the women had perpetrated at least one violent offence for their current conviction, while slightly more than half (56%) had perpetrated at least one non-violent offence. A quarter of the women were homicide offenders. Current offence information for the sample is presented in Table 2.

Table 2. Offence Information

Present Conviction(s)	Number (%) of Offenders
Violent Offences
Homicide (murder, manslaughter)	74 (25)
Attempt murder / Conspire to commit murder	4 (1)
Assault (major / minor)	80 (27)
Robbery (with / without weapon)	72 (24)
Kidnapping / Forcible confinement	16 (5)
Sexual assault	6 (2)
Arson	21 (7)
Utter threats	5 (2)
Weapon offences	16 (5)
Non-violent Offences
Drug offences (importing, trafficking)	39 (13)
Break and enter	17 (6)
Fraud	3 (1)
Obstruct justice	22 (7)
Other non-violent	147 (50)

Note. N = 296. Numbers sum to more than 296 because many women were convicted of more than one offence.

Caseworker Sample

A total of seven institutional staff experienced in completing the SRSW, including primary workers, parole officers, and team leaders, provided feedback on the measure. All women’s regional facilities were represented in this sample.

SRSW Descriptive Information

Review Periods

As a scale with a dynamic focus, for most items, SRSW instructions indicate that the behaviours being scored must be considered only for the period under review. This period is calculated backwards from the current date and is defined as the length of time from the date of the current review to the date of the last SRSW security review occurring at least six months prior. If there is no SRSW security review which occurred at least six months before the current date, the review period is defined as the length of time from the date of the current review to the date of the last administration of the Custody Rating Scale. Lastly, if there is no Custody Rating Scale on file, the review period is defined as the length of time from the date of the current review to the offender’s date of admission on the current term.

Within the present sample, the average review period was 290 days (SD = 170.34 days). Given that CSC policies indicate that security reviews occur at different intervals based on security classification, it is not surprising that this average review period differs by pre-review classification. Consistent with expectations, this period was longer for those women with a pre-review classification of minimum security (M = 378.77; SD = 385.94) than it was for those classified pre-review as medium security (M = 287.99; SD = 136.85) or maximum security (M = 265.87; SD = 99.75), F (2, 435) = 6.67, p < .001. There was no statistical difference, however, between the mean review periods for the cases corresponding to pre-review classifications of medium and maximum security.

SRSW Scores

The mean SRSW score for this sample of women was 0.31 (SD = 7.31), with a range of -10.10 to 21.65. This score corresponds to the low end of the range for medium security classification, and is slightly lower than the mean score of 3.12 found during the SRSW field test.

SRSW Security Recommendations

As can be seen in Table 3, which presents the pre-review security classifications and SRSW security recommendations obtained for this sample of security reviews, more than half (59%) resulted in a recommended change in security classification. Of those where a change was recommended, a greater proportion were recommendations of reduced security (84%) than were recommendations of increased security (16%). Using the sign test, this directional difference was found to be statistically significant, z = 10.77, p < .0001 (two tailed).

Table 3. Pre-Review Security Classifications by SRSW Security Recommendations

Pre-Review Classification	SRSW Recommendation
	Min.		Med.		Max.		Total
	n	%	n	%	n	%	n	%
Minimum ^a	21	(5)	18	(4)	-	-	39	(9)
Medium	151	(34)	113	(26)	25	(6)	289	(66)
Maximum	9	(2)	58	(13)	43	(10)	110	(25)
Total	181	(41)	189	(43)	68	(16)	438

^a The lower total number of reclassification events conducted for cases with pre-review classifications of minimum security is attributable to the longer time period and the relative infrequency of reviews at this security level.

Note. N = 438; pre-review security classification data could not be retrieved for five cases. Recommendations concordant with pre-review classifications appear on the diagonal.

Bonferroni-corrected follow-up analyses were conducted to compare SRSW security recommendations for groups defined by ethnicity, age, region, sentence length, and offence type. SRSW security level recommendations for cases corresponding to Aboriginal and non-Aboriginal women are presented in Table 4; these were found not to differ significantly, χ² (2, N = 437) = 5.64, ns.

Table 4. Percentage of Reviews at Each SRSW Level, by Ethnicity

SRSW Recommendation	Aboriginal (n = 225)	Non-Aboriginal ^a (n = 207)
	Ethnicity
Minimum	48	37
Medium	39	46
Maximum	13	17

^a Does not include reviews corresponding to those whose ethnicity was unknown.

SRSW recommendations were also examined for differences based on age, and are presented in Table 5. The age groupings established in the field test were also used here. Analyses indicated that relative to their younger counterparts, cases corresponding to older women tended to have fewer recommendations to maximum security and more recommendations to minimum security, χ² (4, N = 443) = 31.81, p < .0001.

Table 5. Percentage of Reviews at Each SRSW Level, by Age

SRSW Recommendation	18 - 27 years (n = 160)	28 - 35 years (n = 127)	36 or more years (n = 156)
	Age Group
Minimum	29	44	53
Medium	45	45	40
Maximum	26	11	8

Significant between-group differences were found by region, χ² (8, N = 443) = 30.37, p < .001. The proportions of women offenders receiving each security reclassification recommendation in each region are presented in Table 6. A greater proportion of women were classified as maximum security in the Atlantic and Quebec regions, while the greatest proportion of women were classified as minimum security in the Prairies.

Table 6. Percentage of Reviews at Each SRSW Level, by Region

SRSW Recommendation	Atlantic (n = 81)	Quebec (n = 57)	Ontario (n = 52)	Prairies (n = 215)	Pacific (n = 38)
	Region
Minimum	23	35	37	52	37
Medium	52	40	50	37	55
Maximum	25	25	13	11	8

Note. The higher number of security reviews completed in the Prairies region is attributable to the fact that two of the regional facilities, in addition to the Regional Psychiatric Centre, fall within this region.

To assess SRSW recommendations by sentence length, groups of offenders sentenced to life, more than three years, and up to three years were compared (see Table 7). No significant differences were found, χ² (4, N = 443) = 4.39, ns.

Table 7. Percentage of Reviews at Each SRSW Level, by Sentence Length

SRSW Recommendation	Life (n = 67)	Up to 3 years (n = 177)	More than 3 years (n = 199)
	Sentence Length
Minimum	36	47	39
Medium	49	38	46
Maximum	14	15	15

Finally, between-group differences on SRSW recommendations by offence type were examined. Differences were found in SRSW recommendations when comparing women convicted of violent offences to their counterparts convicted solely of non-violent offences, χ² (2, N = 443) = 9.79, p < .001, with violent offenders receiving higher security recommendations than non-violent offenders. In further disaggregating the offence types for violent offenders, significant differences were found in SRSW security recommendations for the cases corresponding to women convicted of assault and non-assault offences, χ² (2, N = 347) = 21.05, p < .0001, though not for those convicted of homicide and non-homicide offences, χ² (2, N = 347) = 9.01, ns. As shown in Table 8, a greater proportion of cases corresponding to women convicted of violent offences specifically, and of those corresponding to women convicted of assault offences specifically, resulted in maximum security recommendations.

Table 8. Percentage of Reviews at Each SRSW Level, by Offence Type

SRSW Recommendation	Violent (n = 347)	Non-Violent (n =96)
	Offence Type
Minimum	38	54
Medium	44	39
Maximum	17	7
	Homicide (n = 126)	Non-Homicide (n = 221)
Minimum	46	40
Medium	44	43
Maximum	10	17
	Assault (n = 145)	Non-Assault (n = 202)
Minimum	27	49
Medium	47	41
Maximum	27	10

Discretionary Ranges

For those SRSW scores which fall within 10% of a threshold between security levels, case workers can choose to assign either of the security levels bounding the threshold (see Figure 1). In the current sample of 443 security reclassification events, 36 (8%) of the scale scores fell within one of these discretionary ranges. Slightly fewer of these scores fell near the medium-maximum security level boundary (n = 11) than near the minimum-medium security level boundary (n = 25). Discretion was applied in almost two-thirds of these cases (n = 22).

Inconsistencies in Security Reclassification Decisions

Rates of Inconsistency

The security reclassification process allows for two opportunities for inconsistency (and therefore three contrasts): between the SRSW and caseworker recommendations and between the caseworker recommendation and the actual security placement approved by the warden or Kikawinaw (Institution Head at the Okimaw Ohci Healing Lodge). Briefly, the rate of agreement between SRSW and caseworker recommendations was 67%, between caseworker recommendations and actual security placement was 91%, and between SRSW recommendation and actual security placement was 66%.[8]. Each of these will be addressed separately.

As shown in Table 9, in 33% of cases (144 of 443), the caseworker’s review of case-specific factors suggested that the scale’s recommendation was inappropriate, and the caseworker’s recommendation was inconsistent with that of the scale.[9]. In 83% of these inconsistent cases (n = 119), caseworker recommendations were higher than those produced by the SRSW,[10] while in 17% of cases (n = 25) they were lower. The proportion of inconsistencies corresponding to higher caseworker recommendations was significantly higher than that corresponding to lower caseworker recommendations, z = 7.75, p < .0001 (two tailed). Notably, in all cases (n = 5) where the inconsistency represented a difference of more than one security level, caseworker recommendations were higher than those produced by the SRSW.

Table 9. SRSW and Caseworker Security Recommendations

SRSW Recommendation	Min.		Med.		Max.		Total
	Caseworker Recommendation
SRSW Recommendation	n	%	n	%	n	%	n	%
Minimum	115	(26)	66	(15)	5	(1)	186	(42)
Medium	12	(3)	134	(30)	48	(11)	194	(43)
Maximum	-	-	13	(3)	55	(12)	68	(15)
Total	127	(29)	213	(48)	108	(24)	448

Note. Concordant security levels appear on the diagonal.

Second, contrasts of caseworker recommendations and actual post-review classifications revealed an inconsistency rate of 9% (34 of 384 cases; see Table 10). Actual classifications were higher than caseworker recommendations in 53% of these cases (n = 18) and lower than caseworker recommendations in 47% of cases (n = 16). These proportions were not statistically significant, z = 0.17, ns (two tailed)

Table 10. Caseworker Recommendations and Post-Review Security Classifications

Caseworker Recommendation	Post-Review Classification
	Min.		Med.		Max.		Total
	n	%	n	%	n	%	n	%
Minimum	94	(24)	12	(3)	-	-	106	(28)
Medium	7	(2)	174	(45)	6	(2)	187	(49)
Maximum	-	-	9	(2)	82	(21)	91	(24)
Total	101	(26)	195	(51)	88	(23)	384

Note. N = 384; post-review security classification data were not available for 59 offenders as security reviews were still being finalized. Concordant security levels appear on the diagonal.

Finally, Table 11 shows that there was inconsistency between SRSW recommendations and actual post-review classifications in 34% of cases (132 of 384). Notably, this rate of inconsistency was higher than that found in the original field test study (28%; Blanchette & Taylor, 2005). The directional difference in inconsistencies was significant, z = 7.22, p < .0001 (two tailed), with 82% of cases (n = 108) representing actual classifications which were higher than SRSW recommendations, and 18% (n = 24) representing actual classifications which were lower.

Table 11. SRSW Recommendations and Post-Review Security Classifications

SRSW Recommendation	Post-Review Classification
	Min.		Med.		Max.		Total
	n	%	n	%	n	%	n	%
Minimum	90	(23)	64	(17)	1	(0)	155	(40)
Medium	11	(3)	118	(31)	43	(11)	172	(45)
Maximum	-	-	13	(3)	44	(10)	57	(15)
Total	101	(26)	195	(51)	88	(23)	384

Note. N = 384; post-review security classification data were not available for 59 offenders as their security reviews were still being finalized. Concordant security levels appear on the diagonal.

Taken together, these data seem to suggest that most inconsistencies involve the SRSW recommendations; in other words, the rate of concordance between caseworker recommendations and final placement was quite high, and only those contrasts involving SRSW recommendations had relatively high rates of inconsistency. Both caseworker recommendations and actual placements tended to be higher security than SRSW recommendations.

Reasons for Inconsistency

When a caseworker’s clinical appraisal results in an actual security placement inconsistent with the SRSW recommendation (i.e., in the first scenario described above), the caseworker is required to indicate the reason for this discrepancy. Within the scale, a number of specific reasons (e.g., deportation order on file, escape history or escape on current sentence), as well as an ‘other’ option, are available. Caseworkers must endorse one or more of these options in order to make a security recommendation inconsistent with that of the SRSW. In cases where the other option is the only one selected, caseworkers must document the reasons for the offender’s actual security placement.

Of the 144 cases which represented an inconsistency between SRSW recommendations and caseworker recommendations, inconsistency reasons were available for 121 cases. Of these, the majority (90%; n = 110) included an endorsement of the other option. In these cases, the documented reasons for inconsistency were reviewed and coded qualitatively using content analysis. As can be seen in Table 12, the four non-exclusive themes identified, in order of prominence, were (1) current behaviour / attitude; (2) progress on programming / dynamic factors; (3) behavioural / attitudinal history; and, (4) requirement for structure or support on the offender’s part. A number of other reasons which did not fit in to these themes were also found. No patterns of responses relating to ethnicity were detected.

Table 12. Reasons for Inconsistency between SRSW and Caseworker Security Recommendations

Theme	Explanations Citing
Theme	n	%^a
Current Behaviour / Attitude	60	55
Poor attitude (e.g., confrontational, lack of insight)	15	14
Poor institutional behaviour	31	28
Positive behaviour must be maintained longer	14	13
Positive and stable institutional behaviour	10	9
Behavioural / Attitudinal History	15	14
Poor behaviour prior to incarceration (e.g., index offence)	7	6
History of poor institutional behaviour	7	6
History of positive institutional behaviour	2	2
Progress on Programming / Dynamic Factors	51	46
Insufficient / no programming completed	32	29
Interested in and responsive to programming	11	10
Failure to address dynamic needs	14	13
Woman Requires Structure / Support	16	15
Requires the greater structure of the Secure Unit	6	6
Requires the structure or opportunities of the SLE	1	1
Requires gradual integration into general population	2	2
Requires more support / assistance	6	6
Other	13	12
Risk to public safety (no additional information)	4	4
Escape risk (no additional information)	5	5
Mental health concerns	4	4
Management protocol designation prevents security reduction	1	1
Deferral to previous Custody Rating Scale recommendation	1	1

Note. Themes are non-exclusive.

^a Percentages represent the proportion of cases including any given reason for inconsistency among cases for which the other option was endorsed as the reason for inconsistency (i.e., n / 110).

The most frequently cited reason for assigning a security classification above the one recommended by the SRSW was poor institutional behaviour. For example, one caseworker wrote, "[Offender] continues to use her assaultive behaviour to control and intimidate." Another indicated, "[Offender] has been involved in two serious physical altercations in the last six weeks." Poor attitude was also often cited, as in the following excerpt: "[Offender] has indicated that if she is not successful in her parole hearing, she will no longer cooperate with programming and treatment plans outlined in her correctional plan."

Another common reason for assigning a higher security classification than the scale recommendation was that insufficient correctional programming had been completed by the inmate, as shown in this excerpt: "[Offender]’s risk to public safety cannot be lowered at this time as she has not completed any Core programming to date to lower her risk in this area." Similarly, failure to address dynamic needs was also a commonly found reason given to explain this pattern of inconsistency. For example, one caseworker wrote, "The absence of progress with respect to contributing factors results in [Offender] continuing to represent a serious risk to the public."

The reason cited most often for assigning a security classification below the one recommended by the SRSW was responsiveness to programming. For example, one file included the notation, "[Offender] has fully complied with her correctional plan and has maintained a high level of motivation to address her identified need areas." Another caseworker wrote, "[Offender] has shown the ability needed to take full advantage of the required programming." A second commonly cited reason for this pattern of inconsistency was positive or stable institutional behaviour. For example, one caseworker indicated, "Her behaviour in the Secure Unit has been stable," while another wrote, "As a result of [Offender]’s prolonged period of stable behaviour, the CMT [case management team] feels that she is still manageable as a minimum security offender."

A notable cited reason was mental health issues (n = 4). For example, one file included the following indication: "Due to her mental health issues, [Offender] requires an environment which offers considerable supervision and support, such as that offered by the secure unit." In all cases, this reason was cited jointly with others, such as a requirement for greater supervision or structure or poor institutional behaviour. It is worth mentioning that in all cases where this reason was documented, the final assigned security level was higher than that recommended by the SRSW.

Caseworker Feedback on Inconsistency

During the qualitative interviews, feedback from caseworkers was also solicited with regards to the frequency and reasons for inconsistencies between recommendations produced by the SRSW and those of the caseworkers themselves. Six of the interviewed staff members shared their perceptions of the proportion of cases wherein such inconsistencies were present: of these, five suggested that the proportion was ten percent or less, while the last respondent indicated that such inconsistencies were present in about half of cases. This staff member further suggested that virtually all inconsistencies stemmed from a scale recommendation of minimum security and a subsequent placement to medium security. Based on the quantitative data reviewed above, it seems that many caseworkers underestimate the frequency of inconsistencies. Notably, one participant mentioned that inconsistencies between SRSW recommendations and clinical appraisals encourage a verification of clinical appraisals for errors, thereby acting as a quality control mechanism.

Principle reasons for inconsistencies were threshold events (e.g., assaults, positive urinalysis results; reported by half of respondents), limited correctional plan progress (33%), public safety risk (33%), and generally having more confidence in one’s clinical judgment (33%) than in the mathematically-derived recommendation. Deportation issues were also mentioned (17%), but are accounted for in the SRSW.[11] Finally, the respondent who indicated that the rate of inconsistencies was about half indicated that most of these inconsistencies corresponded to the cases of Aboriginal women, and suggested that the scale may be less sensitive with this group than with other women. As mentioned earlier, however, there were no differences found in the rates of inconsistency by ethnicity.

Finally, respondents indicated whether they felt that the rates of inconsistency were appropriate. The respondents indicating that only a low proportion of cases result in inconsistencies felt that this rate was acceptable. On the other hand, the respondent who reported a discrepancy rate of about half felt, understandably, that this rate was too high. When asked what changes to the scale could reduce the proportion of cases resulting in inconsistencies, most respondents (75%) felt that no changes were required. One said, "for threshold events and one-time circumstances, there’s no way to develop a scale to capture every significant possibility." Together, these data demonstrate that the majority of respondents feel both that the SRSW’s recommendations are generally consistent with clinical appraisals, and that the frequency of inconsistency is acceptable.

Face Validity

One of the key criteria in ensuring that any assessment scale be accepted by the staff responsible for its administration is that the staff believe that the scale include appropriate items. As such, face validity, which assesses whether scale items are perceived to be measuring targeted concepts, was examined by soliciting feedback from caseworkers. Encouragingly, most respondents indicated both that all relevant items were included in the scale (60%) and that there were no irrelevant items included (67%). One respondent (17%) indicated that the item assessing the quality of family contact should be removed, as many women react emotionally to this item and it provides little useful information. Another (17%) suggested that the unweighted section of the scale might include questions about Aboriginal women’s progress on their Elder-assisted healing plans.

The remaining respondents suggested that certain SRSW items required clarification (40%) and that the weighting of certain items could be adjusted (33%). Specifically, respondents indicated that they had difficulty understanding the specific meaning of three items: history of having been unlawfully at large (weighted question); Custody Rating Scale incident history (weighted question); and, presence of positive community contact (unweighted question). For each of these, it was suggested that the scoring instructions might be clarified, and perhaps expanded to include examples, to ensure full understanding.

With regards to weighting, respondents suggested both that some unweighted items should be weighted and that the weighting of some weighted items should be modified. With respect to the former, one respondent suggested specifically that the current drug or alcohol rating item should be weighted, as substance abuse can often be a contributing factor in offenders’ criminal patterns. Similarly, another respondent suggested that the extent of community contact should be a weighted factor, as, in this caseworker’s opinion, community contact and family contact should both be considered. In terms of modifying weights, two suggestions were made: first, that minor charges should be ascribed greater weight, as an offender who receives many minor charges should be recognized as representing high risk. Second, that in considering the escorted temporary absence item, the type of temporary absence should be considered. One respondent explained, "We have a fairly limited number of family contact and personal development ETAs [escorted temporary absences], but medical are huge. Is that given more emphasis than perhaps it deserves? Because we tend to think of medical ETAs as nothing, they just go out with a staff, whip out to the hospital and they’re back - family contact, that’s big."

Collectively, this feedback demonstrates that many staff feel that the items included in the SRSW are appropriate and comprehensively evaluate key issues in security reclassification. One respondent provided this feedback: "I can’t think of anything on the scale that I would change because these are the kinds of questions that we have always been encouraging the primary workers to be asking prior to doing any kind of security reclass[ification]." Those staff who did provide suggestions for improvement to the scale focused predominantly on clarification of scoring and instructions and item weighting, as opposed to item inclusion.

Reliability

The SRSW’s internal consistency was also examined. Item-total correlations are presented for each of the scale’s nine items in Table 13, which also includes the mean and standard deviation for each item. The mean item-total correlation was r = .51, which demonstrates a moderate to strong association between total scores and individual items. Only one of the correlations fell below the commonly accepted threshold value of r = .30. Specifically, the item-total correlation for item 5 (Current institutional pay level) reached only r = .15. It is notable that in the field test, the item-total correlation for this item was r = .30, while those of two other items did not reach this threshold - item 2 (Positive family contact; r = .36 in the current revalidation) and item 8 (Custody Rating Scale incident score; r = .44 in the current revalidation). This pattern indicates differences in the distribution of both SRSW scale items and total scores in contrasting the field test and the revalidation samples.

Table 13. Mean, Standard Deviation, and Item-Total Correlation for SRSW Scale Items

SRSW Scale Item	M	(SD)	R
1. Correctional plan progress / motivation	-0.28	(2.13)	.63
2. Positive family contact during review	0.16	(0.62)	.36
3. Serious disciplinary offences during review	-0.16	(1.73)	.63
4. Number of recorded incidents during review	0.12	(1.88)	.67
5. Current institutional pay level	-0.60	(0.60)	.15
6. Involuntary segregation during review	1.02	(2.49)	.45
7. Successful ETAs during review	-0.03	(1.00)	.79
8. History of being unlawfully at large	0.00	(0.54)	.44
9. Custody Rating Scale incident history score	0.07	(1.25)	.51

Note. r = Item-total correlation. All item-total correlations significant at p < .001.

Inter-item correlations were also calculated. As can be seen in Appendix B, the mean inter-item correlation was r = .20. This is very similar to the mean inter-item correlation of r = .19 found in the first validation study of the SRSW (Blanchette & Taylor, 2005). Interestingly, however, the patterns of inter-item correlations were somewhat different; while item 8 was found to be virtually uncorrelated with the other items during the field test, it was uncorrelated with only one item - item 5 - in the present revalidation. Conversely, in this study, item 5 was primarily uncorrelated with the other items, which was not the case in the field test.

An alternative analysis was used to assess to what extent the scale items are homogeneous - an exploratory principal components analysis with varimax rotation. Two sets of scale items emerged. Of the nine scale items, eight loaded on the first factor with loadings of .45 or greater; the last, item 5, had a loading of .17. This item had a loading of .84 on the second factor, and was the only item to load importantly on this factor. However, both the eigenvalue of 1.10 for this second factor and scree plot examination suggested that this factor not be retained. This suggests that the scale’s items, with the possible exception of item 5, represent a homogeneous set.

Finally, Cronbach’s alpha was calculated for the scale, and was found to be r = .69, the same value as was obtained during the field test. This is very slightly below the commonly acceptable range for social science research, but is satisfactory for a scale with few items (Cortina, 1993). Excluding item 5, the internal consistency was r = .71. Taken with the results of the item-total and inter-item correlations and the principal components analysis, these results indicate that, broadly, the SRSW is a reliable and homogeneous scale. Among scale items, item 5 (Current institutional pay level) seems the least consistent with the scale as a whole. Notwithstanding this, the scale remains reliable, and its indices of reliability have not changed substantially since its field test.

Convergent Validity

The relations between SRSW scores and conceptually related measures - specifically, measures of risk, need, and reintegration potential - were calculated in order to assess the scale’s convergent validity. As shown in Table 14, mean SRSW scores differed significantly with the levels of each of these measures; higher SRSW scores were associated with higher ratings of risk and need and lower ratings of reintegration potential.

Table 14. Mean SRSW Scores by Ratings on Related Measures

Measure	Low	Medium	High	F
	Rating
Risk	-1.77^a	-1.47^a	2.38^b	16.84****
Need	-2.12^{a, b}	-2.40^a	1.64^b	15.78****
Reintegration Potential	3.92^a	-1.16^b	-4.20^c	47.97****

Note. Means with different superscripts are significantly different at p < .05 according to Tukey’s HSD criterion for pair-wise comparisons.

**** p < .0001.

Predictive Validity

Institutional Misconducts

In order to assess whether SRSW classification recommendations were able to predict involvement in institutional misconducts, each offender’s file was verified for involvement in such misconducts after completion of each reclassification review. A three-month follow-up period was used for these analyses; the 84 cases for which data were not available for this period were excluded, leaving a sample of 359 reclassification events for these analyses. Misconducts were coded as minor or major;[12] 12% of reviews were associated with a subsequent minor misconduct, 12% with a major misconduct, and 23% with any (i.e., minor or major) misconduct.

As illustrated in Table 15, reviews involving a higher SRSW security classification recommendation were associated with greater involvement in institutional misconducts than were reviews involving a lower security classification recommendation. This difference was significant for each of any misconducts, minor misconducts, and major misconducts. Offenders’ actual post-review security classifications were also significantly related to involvement in institutional misconducts.

Table 15. Misconduct Rates by Security Classification

Security Classification	Any	Minor	Major
	Percentage of Reviews Associated with Misconduct
SRSW Recommendation/th>
Minimum	9	5	3
Medium	28	15	17
Maximum	48	24	24
χ²	36.93****	14.66***	23.68****
Actual Classification
Minimum	15	8	5
Medium	18	9	11
Maximum	42	21	22
χ²	21.76****	8.69*	12.14**

* p < .05. ** p < .01. *** p < .001. **** p < .0001.

In examining these results, it was seen that the proportion of reviews corresponding to offenders actually classified as minimum and medium security who engaged in institutional misconducts were similar. Post-hoc chi-square analyses were therefore computed to determine whether involvement in misconducts differed when considering only those offenders classified as minimum and medium security. Considering SRSW recommendations, each of any misconducts, χ²(1, N = 305) = 17.00, p < .0001, minor misconducts, χ²(1, N = 305) = 7.54, p < .01, and major misconducts, χ²(1, N = 305) = 18.17, p < .0001, continued to differ significantly by security level. In considering actual classifications, however, none of any misconducts, χ²(1, N = 261) = 0.34, ns, minor misconducts, χ²(1, N = 261) = 0.10, ns, or major misconducts, χ²(1, N = 261) = 3.30, ns, differed by security level. In other words, in considering only women classified as minimum and medium security, the proportion of cases involved in incidents differed by security level when the classification was determined by SRSW recommendations, but did not when classification was determined by actual placement. This finding suggests differences in the discriminative ability of SRSW recommendations and actual placements, with SRSW recommendations more able to differentiate between women classified as minimum and medium security.

In order to further investigate the ability of SRSW recommendations to predict rates of misconduct, receiver operating characteristic (ROC) curves were calculated. This analytic technique provides an estimate of the overall predictive ability of a scale or measure - that is, it combines estimates of likelihood of a measure classifying correctly and avoiding classifying incorrectly. Though ROC curves provide extensive information, the easiest to interpret is the area under the curve (AUC), which can be described as the likelihood that the measure will assign a higher score to an offender who will display the outcome of interest (in this case, institutional misconduct) than to one who will not. AUCs range from .50 to 1.00, with higher values representing greater predictive accuracy. A generally accepted guideline is that an AUC of at least .70 indicates acceptable prediction (e.g., Swets, 1988).

As displayed in Table 16, SRSW recommendations were able to adequately predict involvement in incidents of major and any institutional misconduct. Minor incidents were less well predicted. Prediction based on actual placements did not reach the .70 threshold, most notably minor misconducts. Each pair of predictive results was compared (using the method outlined by Hanley & McNeil, 1983); AUCs for the SRSW recommendations and the actual security classification were not found to differ significantly for minor misconducts, z = 1.55, p < .05, or major misconducts, z = 1.41, ns, though they did for any misconducts, z = 2.16, ns.[13]

Table 16. Prediction of Institutional Misconduct from SRSW Recommendation and Actual Security Classification

Model	AUC	95% C.I.
Any Institutional Misconduct
SRSW Recommendation	.69***	.64 - .75
Actual Security Placement	.63***	.57 - .71
Minor Institutional Misconduct
SRSW Recommendation	.66***	.59 - .74
Actual Security Placement	.61*	.51 - .70
Major Institutional Misconduct
SRSW Recommendation	.70***	.63 - .77
Actual Security Placement	.65***	.57 - .74

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

The predictive ability of the SRSW recommendations was then assessed using survival analysis. This type of analysis examines whether there are differences, associated with security recommendation or classification, in the average number of days that elapse before an offender is involved in an institutional misconduct. This technique also accounts statistically for differing follow-up periods (or time at risk) and for offenders who do not become involved in misconducts. Given these features, the entire sample of 443 reclassification events (as opposed to the sample for which three-month follow-up periods were available) was used in these analyses. The follow-up period varied by case, and was defined as the period from the date of the security review to the next security review, release, or February 1, 2007 (the end of the study period), whichever came first.

Analyses were conducted for any, minor, and major misconducts. In all cases, the relationships between time to failure and security level were significant for both SRSW recommendations and actual security classifications. Mean time to failure (i.e., to misconduct) was shortest for maximum security cases, though similar, in most cases, for minimum and medium cases. Table 17 summarizes these results for any misconduct; analyses for minor and major misconducts, being substantively similar, are reported in Appendix C.

Table 17. Survival Time for Any Misconduct by Security Classification

Security Level	N	n Failed (%)	Total	Uncensored	χ²
			Mean Survival (Days)
SRSW Recommendation
Minimum	183	23 (5)	118	78	41,81****
Medium	192	74 (17)	115	79
Maximum	68	34(8)	83	60
Actual Security Placement
Minimum	101	19 (19)	129	66	34.42****
Medium	195	59 (30)	135	97
Maximum	88	44 (50)	80	58

**** p < .0001.

The relationships between time before involvement in any institutional misconduct and security classification for the SRSW recommendations are also presented graphically in Figure 2. This figure presents the proportion of cases of each security classification for which no participation in institutional misconduct was recorded at each point in time. (For ease of presentation, the data in this figure are truncated at 330 days.) Again, the difference between security classifications is clear, with maximum security cases engaging in more misconduct sooner, and minimum security cases engaging in less misconduct, and taking longer to do so. These results are consistent with those found using the ROC analyses.

Figure 2. Time to Failure for Any Misconduct by SRSW Security Recommendation

Escape Incidents

It was not possible to analyse the relationship between security classifications and escape incident as there were no escapes during the study period. Of the women included in the sample, however, seven had committed escapes prior to the commencement of the study period (two of these women committed two escapes each; the remainder committed one escape each).

Discretionary Release

145 of the 296 women in this sample had been released at least once by the date of data collection (February 1, 2007).[14] In order to be consistent with the previous study, only each woman’s first release following her date of inclusion in the study sample was considered. The releases were coded as being discretionary (day and full parole; n = 61), or non-discretionary (statutory release, warrant expiry; n = 80). Since the granting of discretionary release suggests that risk to the public is considered manageable, the proportions of women at each security classification receiving this type of release (in contrast to non-discretionary release) were compared. As can be seen in Table 18, rates of discretionary release differed by security classification for both SRSW recommendations and actual security placements. In both cases, discretionary release was granted to a higher proportion of women with a minimum classification, to a lower proportion of those with a medium classification, and to none of those with a minimum classification. This suggests that the proportion of women classified as minimum security considered to be of manageable risk to the public is higher than the proportion classified as medium security, which is in turn higher than the proportion classified as maximum security.

Table 18. Discretionary Release by Security Classification

Security Classification	n / N	%	χ²
	Rate of Discretionary Release
SRSW Recommendation
Minimum	47 / 72	65	31.49****
Medium	14 / 57	25
Maximum	0 / 12	0
Actual Security Placement
Minimum	31 / 46	67	28.80****
Medium	21 / 63	33
Maximum	0 / 20	0

**** p < .0001.

In order to further investigate the ability of security classification to predict the granting of discretionary release, ROC curves were calculated. The areas under the curve for both SRSW recommendation (AUC = .75, p < .001; CI = .67 - .83)[15] and actual security placement (AUC = .75, p < .001; CI = .67 - .84) exceeded the conventional threshold of .70, indicating that both methods satisfactorily predict the granting of discretionary release. The AUC values derived using SRSW recommendations and actual security were virtually identical, demonstrating that neither is superior to the other in predicting the granting of such releases.

Post-Release Outcome

For those offenders who had been released (either on discretionary or non-discretionary release), the relationship between SRSW security classification recommendations and return to custody was also examined. Due to the relatively short time period encompassed in the current study, a three-month follow-up period was used; there were 106 offenders for whom this follow-up was possible. Of these, 38 (36%) returned to custody during the study period.[16] Table 19 presents the proportion of these offenders at each security classification, both for SRSW recommendations and by actual security placements. In both cases, a greater proportion of women classified as maximum security than as minimum or medium security returned to custody. These differences, however, were not significant, and can therefore not be considered reliable.

Table 19. Return to Custody by Security Classification

Security Classification	n / N	%	χ²
	Rate of Return to Custody
SRSW Recommendation
Minimum	19 / 60	32	2.04^ns
Medium	14 / 37	38
Maximum	5 / 9	56
Actual Classification
Minimum	11 / 39	28	1.21^ns
Medium	17 / 44	39
Maximum	6 / 15	40

^nsnon-significant.

Due to the low rates of return to custody, indices of predictive accuracy were not calculated. As such, though these data demonstrate that the rates of return to custody do not significantly differ according to security classification, a full analysis of this area was not possible.

Impacts of the SRSW

Implementation

During the qualitative interviews conducted with primary workers, parole officers, and team leaders, a series of questions were asked regarding the implementation of the SRSW. Staff were uniformly positive about the approach used in training staff on the SRSW, the support provided during this process, the way in which the scale was implemented, and the process of completing the SRSW. Respondents commented on the scale’s simple structure (67%) and user-friendliness (50%), as well as the support provided at both the institutional (33%) and national (50%) level during implementation. One staff member said, "From day one, my experience has been that the SRSW has been really well brought forward and implemented. We’ve had very few problems using it." It should be noted, however, that one participant mentioned that some staff were uncomfortable with the SRSW, asking, "Well, is it [the implementation of the SRSW] because they don’t trust us and is it because they don’t have faith in us to be able to know our clients?"

When asked for suggestions to improve the scale, half (50%) of staff respondents indicated that no improvements were necessary. The remaining caseworkers provided a number of suggestions, among which was the inclusion of an illustrative schematic of score ranges (available in the training manual and reproduced in Figure 1) in the computerized application in order to facilitate interpretation of the final SRSW score provided for each offender. Suggestions were also made with regards to formatting issues associated with the scale (which were resolved since), and the provision of refresher training for staff members. One staff member explained, "They [caseworkers] don’t do a lot of scales, so they’ll do one, and then they won’t do another for months and months, so it kind of wears off, the training."

Impacts on Reclassification Process

Caseworkers were also asked to comment on the impacts of the addition of the SRSW to the security reclassification process both generally and in terms of difficulty, time requirements, decisional comfort and decisional confidence. Most feedback in this area was positive, with respondents indicating that the scale provides a guide or framework for decision making (40%), is objective (40%), and serves to increase consistency (20%). One caseworker’s explanation expands on these points: "I like working on the [SRSW] tool in conjunction with the clinical analysis, because it does provide a framework, it does give you a very step-by-step approach to what kind of things you should be assessing in terms of security level, and what is considered risk, and what isn’t risk and is only annoying behaviour." Participants also mentioned that the SRSW recommendations can assist when caseworkers are called upon to defend their recommendations, either at offender management review boards (60%) or to the offender herself (20%). One participant shared an example of the latter: "It was a lot easier to explain to the inmate why medium is being recommended when she [a co-worker] had the scale, because she was able to say, ‘See, you’re even coming up medium on the scale, and I have no information which I could use to override the scale.’"

Moreover, most respondents felt that having a recommendation produced by the SRSW can increase the caseworker’s levels of comfort (60%) and confidence (50%), this being especially the case when the clinical appraisal and the SRSW provide the same recommendation. One respondent indicated that the requirement that the scale be completed adds to primary workers and parole officers’ workloads, saying, "There is a fair amount of additional work involved." This was the exception, however, as the remainder indicated that the scale was neither particularly time consuming ("If you know your case, you can get it [SRSW] done in less than five minutes") nor difficult to complete. Moreover, two staff members mentioned their appreciation of working with a gender-specific tool. One respondent summarized as follows: "I think the SRSW is a great tool, I’d like to continue using it, and I want to see both of them [the SRSW and the clinical appraisal] kept together.

Other Impacts

Caseworkers were asked whether there were any differences in the distribution of security reclassifications since the SRSW had been implemented. Most (67%) indicated that no such differences were apparent. The remaining two staff members, however, reported opposite perceptions - one felt that the addition of the SRSW coincided with an overall decrease in security classifications, while the latter perceived the opposite. The first said, "I think it gave a much clearer picture, security-wise, using the scale, because we did have a lot more women coming out minimum, where they rightfully should be, whereas in the past, a big chunk of women were sitting in medium when they probably didn’t need to be." The other, however, felt that Aboriginal women were often classified by the scale as medium when they were more appropriately considered minimum security.

The qualitative interview also included a question examining the extent to which the implementation of the SRSW, with its very specific scoring criteria, had influenced the women offenders’ understanding of how their behaviours impact security classification. Though most respondents (67%) indicated that there has been no influence in this regard, a proportion (33%) did think there had been a minimal effect. One summarized this impact as follows, "There’s a very small percentage of our population that will actually look at that [the SRSW’s items] and try to look at the information; most of them just look at the bottom line." Another respondent also mentioned, however, that the use of the SRSW helps offenders understand that the decision is based on objective criteria (even if the specific criteria aren’t considered), and is therefore less subject to personal bias. An example illustrates this point: "The inmate was already actually saying to her team leader that the PW [primary worker] was against her, she wasn’t supporting her for anything, she had done all that she had been asked to do... But once she saw the scale, then she realized, ‘Okay, well maybe there’s validity to what my PW is telling me. Maybe it’s not just that she’s against me."

Finally, those respondents whose institutions regularly process Aboriginal offenders were asked to comment on the relative applicability of the SRSW to Aboriginal and non-Aboriginal offenders. One of the three respondents who addressed this question indicated that (as previously mentioned) rates of inconsistency between SRSW recommendations and clinical appraisal were consistently higher for Aboriginal women than for their non-Aboriginal counterparts. This respondent mentioned, however, that a larger proportion of non-Aboriginal offenders were accelerated parole review cases, and therefore suggested that ethnicity and accelerated parole status may be confounded. The other two respondents indicated that there were no apparent differences between Aboriginal and non-Aboriginal offenders in relation to SRSW recommendations. One of these participants, however, suggested that the scale could include, in its unweighted section, some items regarding preparation of and adherence to Elder-assisted healing plans.

Appropriateness of Security Reclassification Decisions

It was important to examine whether the security reclassification decisions resulting from the current system (i.e., SRSW scales completed jointly with a clinical appraisal) were more appropriate than those resulting from the previous system, which consisted solely a clinical appraisal. The measurement of appropriateness was quite specific, and involved consideration of two indicators: first, changes in reclassification relative to risk, examined through comparison of the reclassification decisions made with each system, using measures of risk and need for comparison; and second (if applicable), frequency of institutional misconducts. More appropriate classifications would be demonstrated by a pattern of lowered security classifications relative to risk after the implementation, in combination with unchanged (or lowered) frequencies of institutional misconduct. Since the mean security classification at admission of federally incarcerated women has been increasing in recent years (Correctional Service Canada, 2005), any other pattern of findings would not be conclusively interpretable. Specifically, other patterns could be attributable either to this trend in security classifications or to the implementation of the SRSW.

Interrupted time series analysis was used to explore this issue. This type of analysis is conducted on a number of consecutive observations of the same variable (in this case, security reclassification decisions), and involves comparison of the observations taken before and after an "interruption" (here, the implementation of the SRSW). The inclusion of a large number of observations over time allows the pre-existing trend in the data to be modeled, and helps to differentiate between changes over time due to an implementation effect and those due to these pre-existing trends.

For the analyses conducted in this study, the "interruption" was defined as the date of implementation of the SRSW, June 1, 2005; because it took time for all women to undergo a security reclassification using the SRSW,[17] the data were examined for a gradual effect. Observations included all women offenders incarcerated between January 2002 and January 2007 who had undergone at least one security review, regardless of outcome. The proportion of this sample of women classified at each security level was recorded on the first of each month for the period defined above. Figure 3 summarizes these data.[18] It is evident that there are considerable month-to-month fluctuations.

Figure 3. Proportion of Offenders at Each Security Classification

Figure 3. Proportion of Offenders at Each Security Classification

For each security level, separate models were fit to the data representing the proportion of offenders at each month during the study period. In each case, the June 2005 SRSW implementation date was input as an intervention parameter and Akaike Information Criterion (AIC) values were used to select the most parsimonious model. For minimum security classification, the interruption (i.e., the SRSW implementation) did not significantly influence the existing pattern, t(60) = 1.47, ns. Conversely, for medium and maximum security classifications, the SRSW implementation did influence the underlying trend. For medium security classifications, the interruption parameter was negative and of relatively small magnitude, β = -0.27, SE_β = 0.08, t(60) = 3.42, p < .01, while for maximum security classifications, it was positive and also of relatively small magnitude, β = 0.27, SE_β = 0.12, t(60) = 2.18, p < .05, respectively. This indicates a small increase in the proportion of maximum security classifications, a small decrease in the proportion of medium security classifications, and a statistically unchanged proportion of minimum security classifications, since June 2005. The proportions of variance accounted for by these models, however, were relatively low, at R² = .39, R² = .49, and R² = .76, for minimum, medium, and maximum security respectively. This suggests that additional unmeasured variables may be influencing these data.

The same procedure was followed for the proportion of women classified at each level of risk and at each level of need on the first of each month for the period defined above. Notably, these risk and need ratings correspond to the rating closest in time to (but before) the security review. As can be seen in Table 20, which presents the regression parameters and t-test values for the interruption component of the resulting models, along with R² values, the findings above are mirrored in those for the proportions of women classified as high and moderate risk. No statistical differences in the time series were detected with regards to the proportion of women classified as low risk, nor with respect to levels of need. These findings demonstrate that the differences in security classification identified above are consistent with those in assessed risk.

Table 20. Parameter Estimates for Levels of Risk and Need

Predictor	β	SE_β	t	R²
Risk
Low	-0.07	0.09	.85^ns	.66
Moderate	0.51	0.13	3.98***	.94
High	-0.01	0.00	4.97****	.92
Need
Low	-0.09	0.05	1.98^ns	.55
Moderate	0.02	0.01	1.57^ns	.61
High	-0.47	0.40	1.19^ns	.68

^ns non-significant. ***p < .001. ****p < .0001.

These contrasts, together with the fact that the above data are consistent with the general trend of increased security classifications among women offenders mentioned above, suggest that the increases in security classifications associated with security reviews occurring since June 2005 are not attributable to the SRSW’s implementation. This cannot be considered conclusive, however, as only the opposite pattern of findings would have been unambiguous. Due to this result, however, analyses were not conducted on the frequency of institutional misconducts. As explained earlier, the present pattern does not allow for assessment of the appropriateness of security reclassification decisions (as explained above). In considering these findings, moreover, it is important to recall the high rate of inconsistency between SRSW recommendations and actual security placements. Since the data analysed here correspond to actual placements, they attempt to measure the appropriateness of security reclassification decisions made considering SRSW recommendations and clinical appraisals jointly; they do not represent an assessment of the appropriateness of SRSW recommendations. Analyses were not conducted to determine the appropriateness of SRSW recommendations in and of themselves because it would have been impossible, methodologically, to differentiate the impact of SRSW recommendations from that of the environments associated with actual security placements.

DISCUSSION

This study comprised a revalidation of the Security Reclassification Scale for Women (SRSW). The objectives of this study were to examine the appropriateness of security reclassification decisions made using clinical appraisal in conjunction with the instrument’s recommendations, to investigate the impacts of implementation of the scale, and to verify whether the scale reliably measures what it purports to measure (i.e., to examine the scale’s psychometric characteristics). Each of these areas will be discussed separately.

SRSW Psychometric Characteristics

Indices of reliability for the scale, including item-total correlations, inter-item correlations, factor analysis, and calculation of Cronbach’s alpha, demonstrated that the SRSW is a reliable and homogeneous scale. Notably, one of the scale’s component items, Current institutional pay level, was less consistent with the scale as a whole than were the other items. This finding was not in harmony with those of the field test, however, and is therefore insufficient to suggest that removal of the item is warranted at the current time. Indices of reliability have not substantially changed since the scale’s field test, and these results therefore continue to support the scale’s reliability.

A number of indicators of scale validity were also examined. In assessing the extent to which the SRSW items were perceived by users to measure the constructs of interest (i.e., face validity), qualitative findings demonstrated that most surveyed caseworkers indicated that no changes to the scale were needed, that the SRSW included all the relevant items, and that it did not include any irrelevant items. Those respondents who did provide suggestions for scale improvement suggested clarification of scoring and instructions rather than changing the items themselves.

Arguably the most important indicators of validity are convergent and predictive validity, which measure the extent to which scale scores are related to similar constructs and are associated with relevant outcomes, respectively. If unsatisfactory results were found in these areas, the SRSW would be of limited value, despite acceptable findings in the areas of reliability and face validity. The present findings, however, were unambiguously and persuasively supportive of the scale’s validity in both of these areas. First, an examination of convergent validity demonstrated that higher scores on the SRSW were associated with higher ratings of assessed risk and assessed need, as well as lower ratings of assessed reintegration potential. These results are consistent with those of the field test, and indicate that the SRSW continues to converge appropriately with static and dynamic indices of risk.

Second, participation in institutional misconducts was found to differ according to SRSW security recommendations. During a three month follow-up period, those cases for which the scale produced a recommendation of maximum security engaged in the most institutional misconducts, variously defined as minor, major, and any type of misconduct. Similarly, those cases for which the scale produced a recommendation of minimum security engaged in the least institutional misconducts, and those for which the scale produced a recommendation of medium security fell between the other two groups. Moreover, results of ROC analyses and of Cox regressions were largely consistent with these findings, demonstrating a consistent relationship between SRSW recommendations and both participation in misconducts and time until participation in misconducts. Notably, the scale performed less well in terms of predicting minor misconducts (e.g., possession of unauthorized items) than it did in predicting major ones (e.g., homicide, assault, possession of drugs or weapons). This is appropriate for an instrument constructed to assess, in part, institutional adjustment, as major misconducts indicate more serious maladjustment. Moreover, it is notable that though differences in involvement in institutional misconducts were also found to be associated with actual security placements, the ability of actual placements to predict involvement in such misconducts did not reach satisfactory levels.

This difference in predictive ability suggests that it would be of value to reduce the rates of inconsistency between SRSW recommendations and actual placements. A greater proportion of placements consistent with SRSW recommendations would lead to both reduced incarceration costs (Correctional Service of Canada, 2002) and less restrictive environments for women who could manage well at lower security levels. Of course, professional discretion will continue to be necessary for those reclassification decisions with case-specific factors not measured by the SRSW. Increased staff training may be helpful in addressing this issue.

The association between the granting of discretionary release and SRSW security recommendations was also examined. Since guidelines for the granting of discretionary release indicate that such decisions must be "based on the risk presented by the offender and consistent with the goal of promoting the protection of the public" (NPB, 2007, para. 6), this information was used as a proxy for the assessment of estimated risk to the public. Both SRSW recommendations and actual security placements were found to be associated with discretionary release, with cases corresponding to maximum security recommendations or placements being uniformly denied discretionary release, and a greater proportion of those corresponding to minimum security, relative to those corresponding to medium security, being granted discretionary release. Furthermore, ROC analyses demonstrated that SRSW recommendations and actual placements were equally predictive of the granting of discretionary release. This latter finding is somewhat surprising, in that it would be expected that actual placement would be more predictive of discretionary release decision than would be SRSW recommendation. Previous research has determined that even after accounting for level of risk, offenders are less likely to be granted discretionary release when they have higher security classifications (Luciani, Motiuk, & Nafekh, 1996). Given this, the fact that SRSW recommendations were able to match the predictive ability of actual security classifications is remarkable.

Finally, post-release return to custody was examined for differences by SRSW recommendation and actual security placement. Within the relatively small sample of women for whom these analyses were possible, neither was found to be significantly related to return to custody. This is not surprising, given that the SRSW is not intended to predict post-release outcome, but rather to inform the management of risk within the institution.

Altogether, examination of the psychometric properties of the SRSW revealed that the scale does indeed measure what it purports to measure. Security classification and reclassification instruments are legislatively required to consider institutional adjustment, escape risk, and risk to the public in the event of an escape. The present results demonstrate that the SRSW was reliably able to predict outcome in two of these areas, while the third, escape risk, was of such low frequency that analyses were not possible.

Impacts of the SRSW

The current evaluation also examined the impacts of implementation of the scale. A number of staff mentioned their appreciation of the availability of a security reclassification scale specifically for women. Moreover, staff were pleased with the training, support, and procedures involved in the implementation of the scale. It should be noted, however, that one staff member suggested that the implementation of the scale led to a perception on staff’s part that their judgment was not trusted.

Most staff were satisfied with the process of scale completion, mentioning the scale’s user-friendliness and simplicity. They also appreciated the structure, objectivity, and increased consistency associated with use of the scale. A number of staff spoke of increased comfort and confidence in making security reclassification recommendations using the SRSW as an anchor.

When asked whether they perceived any shifts in the distribution of security classification corresponding to reclassification events since the SRSW has been implemented, most staff said that they did not. Nonetheless, this does not fully reflect the results of security recommendations stemming from the SRSW. Indeed, though almost two-thirds of reviews resulted in recommended changes to security classification, these recommendations were not consistently accepted. Of those recommendations, however, most corresponded to reductions of security classification, and compliance with these recommendations would have resulted in 53% more cases being classified as minimum security and 34% fewer being classified as maximum security. This is consistent with the results of the field test, which demonstrated that adherence to SRSW recommendations would have resulted in 20% fewer cases being classified as maximum security and more than a quarter more being classified as minimum security (Blanchette & Taylor, 2005).

Indeed, there were high rates of inconsistency between SRSW and caseworker recommendations, as well as between SRSW recommendations and actual security placements, even after accounting for the 5% of cases where the built-in discretionary range was used. These inconsistency rates, of 33% and 34% respectively, are considerably higher than the 20% guideline suggested in the literature (Buchanan et al., 1986). In most cases, inconsistencies represented caseworker recommendations or actual placements higher than SRSW recommendations, and based on reviews of reclassification decisions, the most common reasons for these inconsistencies were institutional behaviour or attitude, or insufficient progress on correctional programming. Conversely, caseworker recommendations or actual placements which were higher than SRSW recommendations were often linked to responsiveness to treatment programs or stable positive institutional behaviour. Staff members’ perceptions of the reasons for inconsistency were consistent with these results, though it appears that staff do not recognize the frequency with which inconsistencies occur. Most indicated that such inconsistencies occurred in 10% or less of cases, which represents more than a threefold underestimate.

Overall then, the results of this study demonstrated that the impacts of the implementation were positive and that staff were generally responsive to and appreciative of the SRSW. Clearly, however, its recommendations were not consistently accepted; notably, SRSW recommendations were often to a lower security classification than the actual placement. This finding is particularly notable when considered jointly with the results regarding predictive validity presented above. These results are consistent with findings that, relative to clinical judgment, actuarial methods generally result in more liberal decisions (Buchanan et al., 1986) and are more predictive of outcomes of interest (Ægisdóttir et al., 2006).

However, it should be recalled that there are almost certainly some cases where actual placements inconsistent with SRSW recommendations resulted in the prevention of possibly serious institutional misconduct. For instance, high risk women may have benefited from the greater support associated with housing at higher security levels. Alternatively, some women may have found it more difficult to engage in certain behaviours in the more controlled environments associated with higher security classifications. Conversely, some researchers posit that the environment at higher security levels encourages more misconduct behaviours (Harer & Langan, 2001), while others’ findings suggest that it is classification level, rather than environment, that is most associated with institutional behaviour (Hanson, Moss, Hosford, & Johnson, 1983). As such, though it is very likely that some inconsistencies may have resulted in the prevention of institutional misconduct, the extent to which this may have been the case remains unclear.

Appropriateness of Security Reclassification Decisions

Since this revalidation study was the first to be completed after the SRSW’s implementation, it was the first to examine whether security reclassification decisions arising from consideration of SRSW recommendations jointly with caseworkers’ clinical appraisal resulted in decisions which were more appropriate than those stemming solely from clinical appraisal. Investigation of this issue required consideration of changes in the pattern of reclassification decisions since the scale’s implementation, relative to changes in the risk and need profiles of the same offenders during the same period.

Interrupted time series analyses examined whether there were changes in the pre-existing patterns in these areas which corresponded to a gradual effect beginning the day of the scale’s implementation. Results revealed small but significant increases in the proportions of women classified as maximum security and as high risk, as well as small but significant decreases in the proportions of women classified as medium security and moderate risk. There were no changes in the pattern of distribution of need levels during this time. Given that the patterns for the reclassification decisions were mirrored by those for risk levels, the findings suggest that changes in the pattern of security reclassification are not attributable to implementation of the scale; rather, they are consistent with evidence that women offenders’ security classification at admission is on the rise (Correctional Service of Canada, 2005). To add to the challenge of interpreting these findings, it should be recalled that actual reclassification decisions were in agreement with SRSW recommendations in only two-thirds of cases. As such, the present results reflect the appropriateness of actual classification decisions reached using both SRSW recommendations and caseworkers’ clinical appraisals, not the appropriateness of SRSW recommendations alone.

Limitations and Future Directions

In interpreting the present results, a number of limitations must be acknowledged. First, the sample used to collect qualitative data included only seven respondents, and it is unknown whether inclusion of extra caseworkers may have provided more insight. Second, the inter-rater reliability associated with the two SRSW items which are manually scored by caseworkers was not examined. Subsequent revalidations should explore this issue. Finally, the present study did not solicit feedback from the women inmates regarding the impacts of the SRSW’s implementation. Though one-third of the staff said that the SRSW allows women to better believe that the security reclassification process is objective, it is unknown whether this is consistent with the women’s own perceptions. Moreover, it is unknown whether the women feel that the objective and transparent nature of the scale allows them to better influence the results of future reclassification decisions.

Implications

The current study offers promising support for the use of the SRSW with federally incarcerated women. Moreover, it also led to the recognition of certain issues requiring further attention. First, this investigation revealed that there are a number of areas where some staff may require clarification regarding the SRSW. One of these areas is with regards to the contexts wherein use of the SRSW is appropriate. Verification of 52 cases corresponding to review periods of less than 90 days revealed that in 39 of these cases, the Custody Rating Scale should have been used. Since this time, however, the Deputy Commissioner for Women has issued a memo to the regional facilities in order to remind staff that the SRSW is not intended for use upon new admissions.

Another area requiring attention is the items included in the scale. Staff indicated that the interpretation of certain items - notably History of being unlawfully at large and Custody Rating Scale incident history among the weighted items - were unclear. They suggested that scoring instructions should be clarified, and should perhaps include examples. In order to ensure the SRSW be completed as appropriately as possible, implementation of this suggestion should seriously be considered. Moreover, some staff requested refresher training; such training may also be a possibility.

A second issue that has become clear as a result of this revalidation is the relatively high frequency with which SRSW recommendations are not consistent with actual placements. The findings summarized above demonstrate that, relative to actual security classifications, scale recommendations are equally or more predictive of both institutional misconduct and the granting of discretionary release. This is consistent with other recent findings that there is "no evidence that clinical overrides [i.e., inconsistencies] can improve on actuarial risk estimates and some evidence they make them worse" (Rice, 2007, p. 52). This position, however, seems excessive given the potential for caseworkers to identify cases where higher security classifications could contribute to the prevention of serious institutional incidents. Additionally, the inclusion of clinical appraisals is (and should continue to be) required under the policies guiding security reclassification (CSC, 2006b). These results, however, are supportive of Rice’s (2007) entreaty for further research to evaluate the impacts of "controlled versus unbridled deviation" (p. 60) from actuarial recommendations.

Within this context, understanding how environmental factors contribute to the interplay of the inconsistencies between SRSW recommendations and actual placements would be of utility. Given that there is currently little consensus regarding whether it is classification or placement which most influences institutional outcome, research in this area would be necessary to fully understand the potential impacts of restrictions on deviations from SRSW recommendations. Together, research on the impact of institutional environments and on the impacts of controlled or unrestrained inconsistencies would allow the Correctional Service of Canada to continue to respond to concerns alleging women’s over-classification (e.g., Canadian Association of Elizabeth Fry Societies, 2004; Canadian Human Rights Commission, 2003).

Finally, it is important that ongoing revalidations of the SRSW be completed. Not only is this good regular practice (e.g., Shaw & Hannah-Moffat, 2000), but it is also necessary to ensure that any actuarial measure remains appropriate given procedural and population changes. For instance, we know that there are a number of gradual changes occurring in the federal women offender population, in terms of risk and need profiles, offence categories, and sentence lengths (Correctional Service Canada, 2005). Ongoing revalidations would assure that the scale is equally appropriate for a population which incorporates these changing characteristics. Moreover, the implementation of a number of recommendations arising from a recent external review of two regional women’s facilities could lead to procedural changes. Specifically, the authors of these reports recommended that women classified as maximum security should have their security classifications reviewed at least every 90 days (HM Chief Inspector of Prisons for England and Wales, 2005a, 2005b). Given that, at present, the scale has not been validated for reviews of less than three months, implementation of such a recommendation would also have implications for ongoing scale revalidations.

Conclusion

Overall, the present study demonstrated that the Security Reclassification Scale for Women (SRSW) continues to be a strong instrument in assisting to make security review decisions. Women recommended to lower security classifications by the SRSW were better adjusted (as indicated by lower frequencies of incidents, lower need ratings, and higher reintegration potential ratings) and lower risk (as demonstrated by lower risk ratings and higher frequencies of discretionary release) than their counterparts recommended to higher security classifications. Moreover, it is notable that the distribution of SRSW recommendations for Aboriginal and non-Aboriginal women did not differ, and the scale’s ability to predict institutional adjustment and the granting of discretionary release was equal or greater when considering only Aboriginal women.

Clearly, the SRSW remains a consistent, reliable, and valid scale for use in anchoring security reclassification decisions with both Aboriginal and non-Aboriginal Canadian women offenders. The SRSW continues to play an important role in assisting the Correctional Service of Canada in providing legislatively required periodic security reclassifications. The scale’s actuarial nature also adds objectivity, transparency, and national consistency to this decision-making process. As such, the SRSW positions the Correctional Service of Canada as a fore-runner among correctional agencies in meeting the call for empirical, gender-informed development, validation, and revalidation of classification instruments.

REFERENCES

Alexander, J. & Austin, J. (1992). Handbook for evaluating objective prison classification systems. Washington, DC: National Institute of Corrections.

Austin, J. (1983). Assessing the new generation of prison classification models. Crime and Delinquency, 28, 561-576.

Austin, J., & Hardyman, P.L. (2004). Objective prison classification: A guide for correctional agencies. Washington, DC: National Institute of Corrections.

Blanchette, K. (2005). Field-test of a Gender-informed Reclassification Scale for Women Offenders. Unpublished doctoral dissertation, Carleton University, Ottawa, ON.

Blanchette, K. & Motiuk, L.L. (2004). Taking down the straw man: A reply to Webster and Doob. Canadian Journal of Criminology and Criminal Justice, 46, 621-630.

Blanchette, K. & Taylor, K. (2005). Development and field test of a gender-informed security reclassification scale for women offenders. Report R-167. Ottawa, ON: Correctional Service Canada.

Blanchette, K., Verbrugge, P., & Wichmann, C.G. (2002). The Custody Rating Scale, initial security level placement, and women offenders. Report R-127. Ottawa, ON: Correctional Service Canada.

Bonta, J. & Motiuk, L.L. (1990). Classification to halfway houses: A quasi-experimental evaluation. Criminology, 28, 497-506.

Brennan, T. (1998). Institutions classification of females: Problems and some proposals for reform. In R.T. Zaplin (Ed.), Female offenders: Critical perspectives and effective interventions (pp. 179-204). Aspen Publishers, Inc.

Brennan, T. (1987). Classification for control. In D.M. Gottfredson & M. Tonry (Eds.), Prediction and classification: Criminal justice decision making. Chicago: University of Chicago Press.

Buchanan, R.A., Whitlow, K.L., & Austin, J. (1986). National evaluation of objective prison classification systems: The current state of the art. Crime and Delinquency, 32, 272-290.

Burke, P. & Adams, L. (1991). Classification of women offenders in state correctional facilities: A handbook for practitioners. Washington, DC: National Institute of Corrections.

Canadian Association of Elizabeth Fry Societies (2004, June). Executive Director’s Report. Retrieved April 6, 2007 from http://www.elizabthfry.ca/areport/2003-01/english/edreport.htm

Canadian Human Rights Commission (2003). Protecting their rights: A systematic review of human rights in correctional services for federally sentenced women. Ottawa, ON: Author.

Correctional Service Canada (2007). [Climate Indicators and Profiling System snapshot of women offenders’ sentence length and initial security classifications, by Aboriginal ethnicity.] Unpublished raw data.

Correctional Service Canada (2006a). Commissioner’s directive 705-7: Security classification and penitentiary placement. Ottawa, ON: Author.

Correctional Service Canada (2006b). Commissioner’s directive 710-6: Review of offender security classification. Ottawa, ON: Author.

Correctional Service Canada (2005). The changing federal offender population: Profiles and forecasts, 2005. Ottawa, ON: Author.

Correctional Service Canada (2002). Corporate results. Ottawa, ON: Author, Performance Measurement Branch.

Corrections and Conditional Release Act (1992).

Cortina, J.M. (1993). What is coefficient alpha? An examination of theory and applications. Journal of Applied Psychology, 78, 98-104.

Farr, K. (2000). Classification for female inmates: Moving forward. Crime and Delinquency, 46, 3-17.

Grant, B.A. & Luciani, F.P. (1998). Security classification using the Custody Rating Scale. Report R-67. Ottawa, ON: Correctional Service Canada.

Hanley, J.A. & McNeil, B.J. (1983). A method of comparing the areas under receiver operating characteristic curves derived from the same cases. Radiology, 148, 839-843.

Hannah-Moffat, K. (1999). Moral agent or actuarial subject: Risk and Canadian women’s imprisonment. Theoretical Criminology, 3, 71-94.

Hannah-Moffat, K., & Shaw, M. (2001). Taking risks: Incorporating gender and culture into the classification and assessment of federally sentenced women in Canada. Ottawa, ON: Policy Research, Status of Women Canada.

Hanson, R.W., Moss, C.S., Hosford, R.W., & Johnson, M.E. (1983). Predicting inmate penitentiary adjustment: An assessment of four classificatory methods. Crime and Delinquency, 10, 293-309.

Hardyman, P.L. & Van Voorhis, P.V. (2004). Developing gender-specific classification systems for women offenders. Washington, DC: National Institute of Corrections.

Harer, M.D. & Langan, N.P. (2001). Gender differences in predictors of prison violence: Assessing the predictive validity of a risk classification system. Crime and Delinquency, 47, 513-536.

HM Chief Inspector of Prisons for England and Wales (2005a). Report on announced inspection in Canada by HM Chief Inspector of Prisons for England and Wales: Grand Valley Institution for Women. London, UK: Author.

HM Chief Inspector of Prisons for England and Wales (2005b). Report on announced inspection in Canada by HM Chief Inspector of Prisons for England and Wales: Nova Institution for Women. London, UK: Author.

Luciani, F., Motiuk, L., & Nafekh, M. (1996). An operational review of the Custody Rating Scale: Validity, reliability, and practical utility. Report R-47. Ottawa, ON: Correctional Service Canada.

Motiuk, L.L. (1997). Classification for correctional programming: The Offender Intake Assessment (OIA) process. Forum on Corrections Research, 9(1), 18-22.

National Parole Board (2007). Core values. Retrieved June 15, 2007 from http://www.npb-cnlc.gc.ca/about/corev_e.htm

Rice, M. (June, 2007). Current status of violence risk assessment: Is there a role for clinical judgment? Invited presentation at the North American Correctional and Criminal Justice Psychology Conference, Ottawa, ON.

Shaw, M. & Hannah-Moffat, K. (2000). Gender, diversity, and risk assessment in Canadian corrections. Probation Journal, 47, 163-173.

Solicitor General of Canada (1987). Development of a security classification model for Canadian federal offenders. Ottawa, ON: Author.

Swets, J.A. (1988). Measuring the accuracy of diagnostic systems. Science, 240, 1285-1293.

Van Voorhis, P.V. & Presser, L. (2001). Classification of women offenders: A national assessment of current practices. Washington, DC: National Institute of Corrections.

Appendix A. Guide for Interview with Caseworkers

Implementation

1. Did your training occur with the National trainers in Ottawa, or on site?

2. What did you think of the train-the-trainers approach used?

3. What did you think of the implementation of the SRSW?

4. Did you find you had access to sufficient resources (e.g., people, time) in learning to use the SRSW?

5. What do you think of the process of completing the SRSW?

Prompt: Clarity / transparency, instructions, scoring?

6. How could the process be improved?

Prompt: Is there anything that would make the process easier?

Impacts

7. How would you compare the previously used OSL method and the new method including the SRSW?

8. Are there differences in the difficulty of completing the previous OSL method and the new method including the SRSW? If so, what differences are there?

9. Are there differences in time requirements in completing the previous and current OSL methods? If so, what differences are there?

10. Is your comfort level different when using the SRSW as compared to the previous method? How?

11. Is your confidence level different when using the SRSW as compared to the previous method? How?

12. Have you noticed any differences in the distribution of security reclassification since the SRSW has been implemented?

IF YES: Do you think the new distribution is more or less reflective of the women’s actual risk than the previous distribution?

13. To your knowledge, has the implementation of the SRSW impacted the women’s understanding of how to impact their security reclassification? If so, how? (e.g., Have they realized what behaviours will result in lowering or increasing of security classifications?)

Face Validity

14. Does the SRSW capture all the issues which you feel are important in security reclassification?

IF NO: What additional issues are important?

15. Does the SRSW include issues or items which you feel are irrelevant to security reclassification?

IF NO: What issues are irrelevant?

16. What reasons do you think make overrides necessary?

17. In what proportion of cases do you think overrides are currently necessary? Do you feel that this proportion is too high, too low, or appropriate?

18. What changes to the SRSW would reduce the need for overrides?

Strengths and Weaknesses

19. Have any advantages resulted from the implementation of the SRSW?

20. Have any disadvantages resulted from the implementation of the SRSW?

21. What are the SRSW’s strengths?

22. What are the SRSW’s weaknesses?

General Comments

23. Is there anything else you would like to add? Any suggestions, comments, feedback?

Appendix B. SRSW Scale Inter-Item Correlations

SRSW Scale Item	1.	2.	3.	4.	5.	6.	7.	8.
1. Correctional plan progress / motivation	--
2. Positive family contact during review	0.30****	--
3. Serious disciplinary offences during review	0.23****	0.08	--
4. Number of recorded incidents during review	0.17***	0.13**	0.39****	--
5. Current institutional pay level	0.05	0.01	0.10*	-0.07	--
6. Involuntary segregation during review	0.42****	0.22****	0.16***	0.08	0.18***	--
7. Successful ETAs during review	0.32****	0.16***	0.40****	0.54****	0.04	0.25***	--
8. History of being unlawfully at large	0.21*****	0.23****	0.18****	0.17****	0.05	0.30****	0.23****	--
9. Custody Rating Scale incident history score	0.24****	0.17***	0.19****	0.28****	0.12**	0.14**	0.28****	0.17***

Note. ETAs = Escorted temporary absences.

**p < .01. ***p < .001. **** p < .0001.

Appendix C. Survival Analyses for Misconducts

Table D1. Survival Time for Minor Misconducts by Security Level

Security Level	N	n Failed (%)	Total	Uncensored	χ²
			Mean Survival (Days)
SRSW Recommendation
Minimum	183	12 (7)	120	84	25.34****
Medium	192	46 (24)	131	88
Maximum	68	23 (34)	98	73
Actual Security Placement
Minimum	101	10 (10)	135	69	22.20****
Medium	195	33 (17)	146	104
Maximum	88	31 (35)	100	80

**p < .01. **** p < .0001.

Figure D1. Time to Failure for Minor Misconducts by SRSW Security Recommendation

Figure D1. Time to Failure for Minor Misconducts by SRSW Security Recommendation

Table D2. Survival Time for Major Misconducts by Security Level

Security Level	N	n Failed (%)	Total	Uncensored	χ²
			Mean Survival (Days)
SRSW Recommendation
Minimum	183	10 (5)	126	101	17.98****
Medium	192	39 (20)	133	82
Maximum	68	15 (22)	117	105
Actual Security Placement
Minimum	101	7 (7)	140	95	13.00**
Medium	195	33 (17)	149	94
Maximum	88	22 (25)	109	88

**** p < .0001.

Figure D2. Time to Failure for Major Misconducts by SRSW Security Recommendation

Figure D2. Time to Failure for Major Misconducts by SRSW Security Recommendation

[1] Offenders serving a minimum life sentence for first or second degree murder undergo security reviews at least every two years, while the reviews of those offenders classified as minimum security are event-driven.

[2] The SRSW also includes a number of unweighted items to assist caseworkers in considering additional factors. These include such items as current drug and alcohol rating, number of assaults during the review period, and number of positive urinalyses or refusals during the review period.

[3] These 580 security reclassification reviews represented 97% of all security reclassifications from July 2000 to June 2003.

[4] In order to facilitate this process, SRSW training was added to the primary worker continuous development training. At some sites, parole officers also complete security reclassifications and have received appropriate training.

[5] The 90 day threshold was chosen as the researchers who completed the field test cautioned that the instrument’s validity has not been demonstrated for periods of three months or less.

[6] In these cases, the SRSW was completed either in addition to or instead of the Custody Rating Scale upon admission (including readmission) of an offender. Comissionner’s Directive 705-7, however, states that the Custody Rating Scale (rather than the SRSW) is to be used "at the time of initial placement and on any readmission" (CSC, 2006a, sec. 28).

[7] This is likely attributable to the fact that, relative to non-Aboriginal women, a greater proportion of Aboriginal women are both sentenced to serve shorter terms (i.e., under three years) and classified as being maximum security (CSC, 2007). Both of these factors contribute to more rapid security reviews in an effort to promptly cascade women to lower security levels. The completion of such reviews sooner after admission, relative to women of other ethnicities, would result in Aboriginal women representing a larger proportion of reviews than would otherwise be the case. Notably, however, Aboriginal women do not undergo security reviews more frequently than do non-Aboriginal women, χ² (2, N = 291) = 1.71, ns.

[8] Each type of inconsistency was examined to see if there were differences in frequency for Aboriginal and non-Aboriginal women. No differences were found for each of SRSW and caseworker recommendations, χ² (1, N = 432) = 0.28, ns, caseworker recommendations and actual security placements, χ² (1, N = 432) = 2.05, ns, and SRSW recommendations and actual security placements, χ² (1, N = 432) = 0.49, ns.

[9] In order both to be consistent with the previous validation of the SRSW and to recognize the integrated flexibility of the instrument, the 10% discretionary ranges were applied to this and subsequent analyses. For example, if an offender’s score resulted in a recommendation to medium security but fell within the discretionary range for maximum security, and the actual security placement was maximum security, the SRSW recommendation was changed from ‘medium’ to ‘maximum’.

[10] Follow-up analyses revealed that this pattern was not explained by the so-called "two-year rule," whereby women convicted of first or second degree murder are to be classified to maximum security for the first two years of their sentence regardless of scale recommendation. In fact, only 12 cases in the current sample corresponded to a woman for whom this rule was relevant at the time of review.

[11] The scale includes a number of reasons for inconsistencies other than the other option, of which deportation order on file is one.

[12] In order to ensure comparability with the previous validation study, major misconducts were defined as the following: homicide, assault, sexual assault, fighting, threatening behaviour, hostage-taking, inciting to riot or strike, possession of drugs, possession of weapons, escape, or attempting escape. Minor misconducts were defined as any misconduct not defined as major.

[13] The analyses specific to the SRSW were also conducted separately for Aboriginal and non-Aboriginal offenders. Results demonstrated that the scale was equally or more predictive of each type of institutional misconduct for Aboriginal offenders than it was for non-Aboriginal offenders.

[14] Only releases occurring after a security review were considered. In other words, if an offender was released and subsequently returned to custody, this prior release, if occurring prior to the review under study, was not considered in these analyses.

[15] These analyses were also conducted separately for Aboriginal and non-Aboriginal women; no statistical differences by ethnicity emerged.

[16] Due to the small number of women in this study, analyses were not disaggregated by type of return (e.g., revocation with or without a new offence).

[17] As mentioned, reclassification events occur with varying frequency depending on pre-review security classification.

[18] It must be remembered that in over a third of cases, the actual security placement after a security review is not consistent with the SRSW recommendations, and therefore these data do not solely reflect the results of the SRSW implementation.

Common menu bar links

Institutional links

Subjects

Working at CSC

Resources

Proactive Disclosure

Revalidation of the Security Reclassification Scale for Women (SRSW)

Security Classification in Canada

Appendix B. SRSW Scale Inter-Item Correlations