The Dynamic Factors Identification and Analysis (DFIA) Component of the Offender Intake Assessment (OIA) Process: A Meta-Analytic, Psychometric and Consultative Review

2005 N^o R-164

Shelley L. Brown
and
Laurence L. Motiuk

Research Branch
Correctional Service of Canada

March 2005

ACKNOWLEDGMENTS

We wish to express our sincere appreciation to the Offender Programs and Reintegration Branch of the Correctional Service Canada (CSC) for their active collaboration in this project. Specifically, we are grateful to Linda McLaren for coordinating the consultation process as well as for providing operational insight regarding how best to enhance the Dynamic Factors Assessment component of the Offender Intake Assessment (OIA) process.

From the Research Branch, we would also like to thank Kelley Blanchette, Christa Gillis and Nicole Epprecht for their assistance in facilitating the consultation process. In addition, we must acknowledge Gil Taylor, Brian Grant, Craig Dowden, Melanie Gates, Shelley Trevethan and Catherine Wilson for their varied contributions to this project. Also from the Research Branch, we would like to mention Mark Nafekh, Colette Cousineau, Ben Vuong, Nicole Crutcher and Krista Richard for their assistance regarding data extraction and analysis. A special thanks to Justin Gileno for his work in facilitating this project.

We would also like to acknowledge the contributions made by the correctional programs staff: Denis Barbe, Sylvie Carriere, Alain Pelchat, Michael Bettman and Lynn Stewart . Special thanks to the Women Offender Sector and the Aboriginal Initiatives Branch, particularly Randy Mason, Doreen Smith and Millard Beane for providing suggestions on how best to contextualize the Dynamic Factors Assessment process for women and Aboriginal offenders. CORCAN was also instrumental in guiding changes to the employment domain. Additionally, Bea Fisher provided helpful comments in regards to educational matters.

Of course, we are grateful to the external experts (Paul Gendreau, Claire Goggin, Glen Gray, Elizabeth Oddone Paolucci, Claudio Violato, Mary Ann Schofield, Fred Boland, Kathy Henderson, Jan Baker, David Robinson, Frank Porporino, Christopher Beal and Moira Law) for providing comprehensive and thoughtful reviews that greatly enhanced this initiative.

As well, we must acknowledge the work of Drs. Andrews, Bonta, Prochaska, Diclemente, Serin, Kennedy and Bettman and Ms. Claude Tellier. Their theoretical and empirical contributions to offender motivation and responsivity greatly influenced portions of this review.

Lastly and most importantly, we would like to extend a heartfelt thanks to the front-line staff across the Service without whom we could not advance corrections.

EXECUTIVE SUMMARY

Offender reintegration has been defined as all correctional and programming activity conducted to prepare an offender to return safely to the community as a law-abiding citizen. Programming must be directly linked to meeting offender needs, and particularly those needs which, if addressed, will result in reducing the likelihood of re-offending. Consequently, the Correctional Service of Canada (CSC) must be able to efficiently and effectively identify needs and match levels of programming to these needs.

Shortly after admission to federal custody, all offenders undergo an intake assessment to determine their level of risk and their programming needs. Known as the Offender Intake Assessment (OIA) process, a comprehensive evaluation of each individual is based on input from a variety of sources including community offices, courts, police reports and victims, as well as psychological, educational, vocational and substance abuse testing. For more than a decade, the results of the OIA process have been used by parole officers as the basis on which to develop correctional plans for each offender. Crafted in consultation with program delivery staff and the offender, individualized correctional plans are designed to address the factors which have been identified as contributing to an offender's criminal behaviour.

Programming needs must be prioritized so that interventions can be logical, sequenced and effective. Progress towards addressing identified needs and in reducing the risk of re-offending is the main correctional objective, and its ongoing assessment is central to the management of the offender throughout the sentence. Consequently, it has been recommended by various internal task forces and working groups as well as outside stakeholders that the Service's primary offender needs assessment process, the Dynamic Factor and Identification Analysis (DFIA) component of OIA, be thoroughly reviewed to ensure it continues to identify and prioritize those offender needs requiring correctional intervention. In response, the Research Branch in active collaboration with the Offender Programs and Reintegration Branch launched a major research initiative designed to enhance and better align the needs assessment component of OIA with the case planning process and program delivery.

The research strategy for the DFIA review involved an integrated series of initiatives which encompassed three major activities: 1) meta-analytic evaluations of the literature by external experts, 2) psychometric testing of the reliability and validity of the DFIA and 3) consultation with correctional practitioners. The consultation process engaged numerous field staff from across the country as well as various internal partners from national headquarters (NHQ). Women-specific consultations were also conducted with staff. Lastly, further consultations were also conducted with external agencies including the Canadian Human Rights Commission.

The purpose of the DFIA review is seen to be threefold. First, present the results of the meta-analyses, then the findings of psychometric testing and gather the perspectives from the various consultative exercises. Second, synthesize the results of all three aforementioned initiatives and then ultimately present a new DFIA-Revised (DFIA-R). Third, discuss directions for future research as well as the operational implications of the recommended changes.

Briefly, the existing DFIA component of OIA is comprised of seven dynamic factors: employment, marital/family, associates/social interaction, substance abuse, community functioning, personal/emotional orientation, and attitudes. Each factor is subsequently divided into principal components that are further broken down into sub-components. Moreover, a series of yes/no self-reference indicators and, in some cases, 'help messages' accompany the indicators to enhance rating clarity. In total, the DFIA consists of seven need domains, 35 principal components, 94 sub-components, and 197 indicators.

The methodological approach and result summary for each step of the DFIA review follows:

Step 1 involved a review of the DFIA conducted by various correctional experts, the majority of whom were external to the CSC. Each expert was instructed to complete three tasks: First, to review the literature pertaining to the predictive validity of one of the seven factors comprising the DFIA; second, to provide proposals for enhancing the domain; and third, to highlight research findings pertaining to women and Aboriginal offenders and other visible minorities.

Results: The meta-analytic reviews confirmed the content validity of the DFIA. Specifically, the reviews confirmed that DFIA contains factors that have been identified in the research literature as strong predictors of criminal re-offending. Similarly, with few exceptions (e.g., denial and minimization of crime), the reviews confirmed that the DFIA represents all factors considered moderate predictors of criminal re-offending. The review also revealed, however, that the DFIA contains several indicators deemed to be weak predictors of criminal behaviour. Furthermore, the DFIA review identified few empirical studies devoted exclusively to women, Aboriginal or minority group offenders.

The review experts provided various content-oriented proposals designed to enhance the efficiency of the DFIA. Some of the more significant proposals included: 1) retention of indicators with strong or moderate research support and deletion of indicators with weak or no research support, 2) simplification of the overall structure of the DFIA (e.g., grouping conceptually similar indicators), 3) reliability enhancement (e.g., increase level of detail associated with existing help messages), and 4) improve the distinction between contributing (‘criminogenic') and non-contributing (non-criminogenic) factors.

Step 2 involved a psychometric review of the DFIA. The reliability and predictive validity of the DFIA was examined separately for three release cohorts: men ( N  = 15,479), women ( N = 765) and Aboriginal ( N = 2,593) offenders. Study participants included all federal offenders who had been assessed using the DFIA protocol and who had subsequently been released as of September 2000. The predictive validity of the DFIA was assessed by examining how well it predicted readmission to federal custody during a three-year, fixed follow-up period. Internal consistency (i.e., the extent to which individual indicators comprising the DFIA are related to one another) was used to measure reliability.

Results: Overall, the statistical reliability (as measured via internal consistency) of each domain ranged from acceptable to superior within all three release cohorts: men, women and Aboriginal offenders. The reliability estimates for the community functioning domain, however, were relatively low compared to the other domains, particularly for women and Aboriginal offenders. Additionally, supplementary analyses revealed that the subgroup of indicators that assess marital quality possessed low reliability across all three release cohorts.

In general, predictive validity for all DFIA domains (e.g., employment, marital/family, associates, community functioning, substance abuse, personal/emotional, attitudes) ranged from moderate to strong across all three release cohorts, meaning that individuals who were rated as having more serious problems across the seven domains were significantly more likely to be readmitted to federal custody during the three-year, fixed follow-up period.

Women-specific highlights. All seven of the dynamic factor domain ratings as well as a significant number of the individual indicators within the domains predicted readmission for women offenders. While indicators pertaining to educational achievement, parenting skills, drug abuse, leisure activities, coping abilities, and manipulation predicted readmission for both men and women, the strength of the relationship was particularly stronger for women relative to men on these indicators. The trend was particularly predominant in regards to parenting-related indicators. Among the attitudes and associates indicators that significantly predicted readmission, the magnitude of the results were equivalent across gender.

Aboriginal-specific highlights. All seven of the dynamic factor domain ratings as well as a significant number of the individual indicators within the seven domains predicted readmission for Aboriginal offenders. The number of indicators that evidenced moderate to strong predictive validity was smaller within the Aboriginal release cohort relative to offenders in general. This finding was particularly evident in regards to the marital/family and personal/emotional domains. Notably indicators pertaining to chronic unemployment, criminal friends, criminal attitudes, impulsivity, time management, accommodation instability, and drug abuse predicted readmission for Aboriginal offenders to the same degree found within the general offender population.

Step 3 was qualitative in nature and involved extensive consultations with CSC staff and outside agencies . Internally, CSC staff members from the field as well as national headquarters were included. Two separate field consultations were conducted. The first set of consultation sessions included staff from across the country that provide direct services to all offender groups including men, women, Aboriginal and other ethno-cultural groups. While this consultation process addressed issues pertaining to women, Aboriginal offenders and ethno-cultural diversity, the results were largely male-specific. Consequently, a second set of consultations targeted staff members who work exclusively with women offenders.

During the CSC focus groups, participating staff members were specifically asked how to improve the efficiency of the DFIA. Specifically, staff provided general recommendations as well as more detailed suggestions in terms of whether or not each item comprising the DFIA (currently, there are 197 indicators) should be kept, dropped, or modified. Additionally, staff members were asked whether or not they agreed with the various suggestions proposed by the external experts. The individual meetings conducted with NHQ staff focused largely on the operational impacts of the revisions in a number of areas such as program referrals and cultural diversity.

Results: Overall, the staff reported being satisfied with the content of the DFIA. Some recommended adding new domains that addressed the following areas: survivors of trauma, responsivity factors (e.g., English as a second language; cultural diversity) and, more specifically, current motivation level and commitment to treatment. As well, it was recommended that a new component be added specifically devoted to past treatment performance. It was also proposed that positive and/or protective factors be underscored in addition to offender weaknesses.

A consistent and strongly voiced message was that, above all else, the clarity and meaning of the individual indicators and help messages must be enhanced to promote consistency across time, regions and people. Specifically, the field emphasized a need for objective and clear scoring guidelines that would accompany each and every indicator. Both women-specific and Aboriginal-specific recommendations were made and are detailed in the body of this report.

Dynamic Factors Identification and Analysis-Revised (DFIA-R)

In sum, the following changes are being proposed for the DFIA-R:

• It is proposed that the existing DFIA structure be changed. Recall that the existing DFIA is comprised of seven dynamic factor domains, 35 principal components, 94 sub-components, and 197 indicators. For DFIA-R, it is proposed that there be eight assessment domains (the existing seven dynamic factors plus a new responsivity domain), 27 principal components and 107 indicators. Noteworthy here is the deletion of the individual sub-components and the nearly 50% reduction in the number of indicators from 197 to 107.

• A responsivity domain is to be added to enhance reintegration efforts. The new domain will be comprised of 14 indicators that address person-specific factors such as language barriers, spirituality, and self-esteem.

• Some new indicators will be added that reflect cultural diversity (e.g., strong cultural identity, language barrier), women-specific issues (e.g., low self-esteem) or recent advancements in the research literature (e.g., interprets neutral situations as hostile).

• ‘Domain-specific' motivation ratings and past treatment performance ratings will also be added. In total, 24 rating scales will be incorporated (12 pertaining to past treatment performance and 12 pertaining to current motivation level). For example, within the personal/emotional domain there are three possible past treatment performance ratings: cognitive skills-related programming, general violence programming and sex offender programming and three corresponding motivational ratings: cognitive skills, violence programming, sex offender programming.

• Currently, each dynamic domain is accompanied by an overall three- or four-point rating (asset, no need, some need, or considerable need). One level has been added to the rating scale. Specifically, a ‘moderate' level has been inserted between ‘some need' and ‘considerable need'. This change was put in place to accommodate the Service's trend towards the development of programs that vary along three intensity levels.

• Help messages have either been enhanced or added to each indicator along with suggested interview guidelines. When relevant, specialized help messages and interview prompts will be incorporated for women, Aboriginal, and young offenders.

• Lastly, the language used to describe each individual indicator is worded to be more positive in nature.

In conclusion, this report recommends that the DFIA component of OIA be revised and field tested. The proposed changes should increase the efficiency of DFIA, address issues pertaining to gender and cultural diversity, and lastly, align the protocol with recent advancements in correctional treatment. Finally, this initiative clearly underscores the need for more research in a variety of areas including women, cultural diversity, and the role of protective factors in criminal desistence.

TABLE OF CONTENTS

ACKNOWLEDGMENTS

EXECUTIVE SUMMARY

TABLE OF CONTENTS

LIST OF TABLES

LIST OF FIGURES

INTRODUCTION

Offender Intake Assessment (OIA)

Development

How OIA Works

Dynamic Factors Assessment

Previous Research on DFIA

Purpose of DFIA Review

STEP 1: A META-ANALYTIC REVIEW

Methodology

Meta-Analytic Results

Overview of Findings

Employment Domain

Marital/Family Domain

Associates Domain

Substance Abuse Domain

Community Functioning Domain

Personal/Emotional Domain

Attitude Domain

External Expert Proposals

1. Keep the Strong, Keep the Moderate, but Drop the Weak

2. Less is More

3. Increase Objectivity, Reduce Subjectivity

4. Revitalize with New Additions

5. View the Supplementary Assessment as Friend Rather than Foe

6. Distinguish Non-criminogenic from Criminogenic Need

7. Maximize the Dynamic Nature of the Assessment Strategy

8. Assess the Need as well as Severity

9. Beware of Specialized Risk Factors for Specialized Offender Groups

10. More Research, Better Research

STEP 2: A PSYCHOMETRIC REVIEW

Methodology

Follow-up Period and Outcome Definition

Statistical Analysis

Psychometric Results

Reliability

Readmission Rates by Level of Dynamic Intervention

Readmission Rates for Each Dynamic Domain Factor Rating

Employment Domain: Indicator Endorsement Rates

Employment Domain: Relationship between Indicators and Rating

Employment Domain: Correlational Analyses

Employment Domain: Regression Analyses

Marital/Family Domain: Indicator Endorsement Rates

Marital/Family Domain: Relationship between Indicators and Rating

Marital/Family Domain: Correlational Analyses

Marital/Family Domain: Regression Analyses

Associates Domain: Indicator Endorsement Rates

Associates Domain: Relationship between Indicators and Domain Rating

Associates Domain: Correlational Analyses

Associates Domain: Regression Analyses

Substance Abuse Domain: Indicator Endorsement Rates

Substance Abuse Domain: Relationship between Indicators and Domain Rating

Substance Abuse Domain: Correlational Analyses

Substance Abuse Domain: Regression Analyses

Community Functioning Domain: Indicator Endorsement Rates

Community Functioning Domain: Relationship between Indicators and Rating

Community Functioning Domain: Correlational Analyses

Community Functioning Domain: Regression Analyses

Personal/Emotional Domain: Indicator Endorsement Rates

Personal/Emotional Domain: Relationship between Indicators and Rating

Personal/Emotional Domain: Correlational Analyses

Personal/Emotional Domain: Regression Analyses

Attitude Domain: Indicator Endorsement Rates

Attitude Domain: Relationship between Indicators and Rating

Attitude Domain: Correlational Analyses

Attitude Domain: Regression Analyses

STEP 3: A CONSULTATIVE REVIEW

Methodology

Revision Logic Model

Consultation Results

Response to Research Findings

Content Issues

Reliability

Scoring and Rating

Case Management Strategy (CMS) Groupings

Program Referrals

Generalizability

Training/Staffing Issues

Employment Domain

General Comments

General Comments: Women-specific

Response to Expert Proposals: General and Women-specific

Changes to the Employment Indicators

Education (Indicators 1-3)

Impediments to Learning and Employment (Indicators 4-12, 15)

Skill Area/Trade/Profession (Indicators 13-14)

Employment History (Indicators 16-20)

Job Performance (Indicators 21-24)

Dismissal/Departure (Indicators 25-27)

Rewards (Indicators 28-30)

Relations (Indicators 31-32)

Interventions (Indicators 33-35)

Summary of Changes to the Employment Domain

Marital/Family Domain

General Comments

General Comments: Women-specific

Response to Expert Proposals: General and Women-specific

Proposed Changes to the Marital/family Indicators

Family Background (Indicators 1-10)

Marital Status (Indicators 11-12)

Marital Quality (Indicators 13-16)

Spousal Abuse (Indicators 17-18)

Dependents (Indicator 19)

Parenting Skills (Indicators 20-26)

Child Abuse (Indicators 27-28)

Interventions (Indicators 29-31)

Summary of Proposed Changes to the Marital/Family Domain

Associates/Social Interaction Domain

General Comments

Response to Expert Proposals: General and Women-specific

Proposed Changes to the Associates Indicators

Socially Isolated (Indicator 1)

Associates with Substance Abusers (Indicator 2)

Criminal Acquaintances and Mostly Criminal Friends (Indicators 3-4)

Gang Affiliation (Indicator 5)

Criminogenic Area (Indicator 6)

Attachment to Community Groups (Indicator 7)

Predatory Relations (Indicator 8)

Victimized in Social Relations (Indicator 9)

Easily Influenced by Others (Indicator 10)

Difficulty Communicating with Others (Indicator 11)

Summary of Proposed Changes to the Associates Domain

Substance Abuse Domain

General Comments

General Comments: Women-specific

Response to Expert Proposals: General and Women-specific

Proposed Changes to the Substance Abuse Indicators

Abuses Alcohol (Indicator 1)

Began Drinking at an Early Age (Indicator 2)

Drinks on a Regular Basis (Indicator 3)

Has History of Drinking Binges (Indicator 4)

Combined Use of Drugs/Alcohol (Indicator 5)

Alcohol: Situations Sub-component (Indicators 6-8)

Alcohol: Interference Sub-component (Indicators 9-13)

Abuses Drugs (Indicator 14)

Began Using Drugs at an Early Age (Indicator 15)

Uses Drugs on a Regular Basis (Indicator 16)

History of Drug-taking Sprees (Indicator 17)

Has Combined the Use of Different Types of Drugs (Indicator 18)

Drugs: Situations Sub-component (Indicators 19-21)

Drugs: Interference Sub-component (Indicators 22-26)

Intervention (Indicators 27-29)

Summary of Proposed Changes to the Substance Abuse Domain

Community Functioning Domain

General Comments

General Comments: Women-specific

Response to Expert Proposals: General and Women-specific

Proposed Changes to the Community Functioning Indicators

Accommodation Stability (Indicator 1)

Accommodation Maintenance (Indicator 2)

Deportment and Health (Indicators 3-7)

Finance (Indicators 8-12)

Communication (Indicators 13-14)

Leisure (Indicators 15-16)

Unaware of Social Services (Indicator 17)

Social Assistance (Indicator 18)

Intervention (Indicators 19-21)

Summary of Proposed Changes to the Community Functioning Domain

Personal/Emotional Domain

General Comments

General Comment: Women-specific

Response to Expert Proposals: General and Women-specific

Proposed Changes to the Personal/Emotional Indicators

Personal (Indicators 1-2)

Family Ties (Indicator 3)

Ethnicity/religion (Indicators 4-5)

Gang Member (Indicator 6)

Cognition (Indicators 7-16)

Aggressive (Indicator 17)

Assertion problem (Indicator 18)

Copes with stress poorly (Indicator 19)

Poor conflict resolution (Indicator 20)

Manages time poorly (Indicator 21)

Gambling is problematic (Indicator 22)

Has low frustration tolerance (Indicator 23)

Hostile (Indicator 24)

Worries unreasonably (Indicator 25)

Takes risks inappropriately and Thrill-seeking (Indicator 26-27)

Non-reflective (Indicator 28)

Is not conscientious (Indicator 29)

Manipulative (Indicator 30)

Sexual Behaviour (Indicators 31-34)

Mental Ability and Mental Health (Indicators 35-37)

Interventions (Indicators 38-46)

Summary of Proposed Changes to the Personal/Emotional Domain

Attitude Domain

General Comments

General Comments: Women-specific

Response to Expert Proposals: General and Women-specific

Proposed Changes to the Attitude Indicators

Justice (Indicators 1-6)

Convention (Indicators 7-12)

Elderly, Women, Men and Minorities (Indicators 13-17)

Property (Indicators 18-20)

Violence (Indicators 21-22)

Lifestyle (Indicators 23-24)

Summary of Proposed Changes to the Attitudes Domain

Responsivity Domain

Program Referral Matrices

DISCUSSION

REFERENCES

BIBLIOGRAPHIC INVENTORY

APPENDIX A: CURRENT DYNAMIC FACTOR IDENTIFICATION AND ANALYSIS COMPONENT

APPENDIX B: DYNAMIC FACTOR IDENTIFICATION AND ANALYSIS-REVISED

APPENDIX C: REVISED HELP MESSAGES, INTERVIEW PROMPTS, AND PROGRAM REFERRAL FLAGS

LIST OF TABLES

Table 1: Summary of Meta-Analytic Results

Table 2: Dynamic Factor Principal Component Summary

Table 3: Employment Domain: Meta-Analytic Review

Table 4: Marital/Family Domain: Meta-Analytic Review

Table 5: Associates Domain: Meta-Analytic Review

Table 6: Substance Abuse Domain: Meta-Analytic Review

Table 7: Community Functioning Domain: Meta-Analytic Review

Table 8: Personal/Emotional Domain: Narrative Literature Review

Table 9: Attitude Domain: Meta-Analytic Review

Table 10: Release Cohort Characteristics: Gender and Ethnic Origin

Table 11: Offence History

Table 12: Release Type and Releasing Region

Table 13: Internal Consistency for Each Dynamic Factor Domain

Table 14: Readmission Rates by Level of Dynamic Intervention Rating

Table 15: Readmission Rates for Each Dynamic Domain Rating

Table 16: Endorsement Rates for Employment Indicators

Table 17: Correlations between Employment Indicators and Rating

Table 18: Correlations between Employment Indicators and Readmission

Table 19: Employment Domain: Regression Analyses

Table 20: Marital/Family Domain: Indicator Endorsement Rates

Table 21: Correlations between Marital/Family Indicators and Rating

Table 22: Correlations between Marital/Family Indicators and Readmission

Table 23: Marital/Family Domain: Regression Analyses

Table 24: Endorsement Rates for Associates Indicators

Table 25: Correlations between Associates Indicators and Rating

Table 26: Correlations between Associates Indicators and Readmission

Table 27: Associates Domain: Regression Analyses

Table 28: Endorsement Rates for Substance Abuse Indicators

Table 29: Correlations between Substance Abuse Indicators and Rating

Table 30: Correlations between Substance Abuse Indicators and Readmission

Table 31: Substance Abuse: Regression Analyses

Table 32: Endorsement Rates for Community Functioning Indicators

Table 33: Community Functioning Domain: Correlations between Indicators and Rating .

Table 34: Correlations between Community Indicators and Readmission

Table 35: Community Functioning Domain: Regression Analyses

Table 36: Endorsement Rates for Personal/Emotional Indicators

Table 37: Correlations between Personal/Emotional Indicators and Rating

Table 38: Correlations between Personal/Emotional Indicators and Readmission

Table 39: Personal/Emotional Domain: Regression Analyses

Table 40: Endorsement Rates for Attitude Indicators

Table 41: Correlations between Attitude Indicators and Attitude Rating

Table 42: Correlations between Attitude Indicators and Readmission

Table 43: Attitude Domain: Regression Analyses

Table 44: Degree of Research Support Rating Criteria

Table 45: Streamlining Decision Rules

Table 46: Proposed personal/emotional domain reorganization

Table 47: Substance Abuse: Program Referral Matrix

Table 48: Parenting Skills: Program Referral Matrix

Table 49: Family Violence: Program Referral Matrix

LIST OF FIGURES

Figure 1: Associates and Social Interaction

Figure 2: Streamlining Decision Model

INTRODUCTION

During the early 1990s, the number of federal offenders being granted discretionary conditional release had steadily declined (Correctional Service of Canada, 1997). As a result, the Correctional Service of Canada (CSC) assembled a Task Force on Offender Reintegration with the explicit mandate to isolate the causes of the decline as well as to provide suggestions for enhancing the reintegration process. After a thorough review, the Task Force provided a series of recommendations designed to meet these objectives. Task Force recommendations were numerous and varied, ranging from assessment related issues to policy matters (CSC, 1997). Further, the Task Force recommended that the design and application of the Correctional Service's primary mechanism for assessing offender needs be reviewed. Specifically, the Task Force recommended that the Dynamic Factor Identification and Analysis (DFIA) component of the Offender Intake Assessment (OIA) process, formerly known as Case Needs Identification and Analysis (CNIA) be “ reviewed to ensure it identifies and prioritizes only offender needs related to criminal behaviour” (CSC, 1997, p.30). In addition, a recent 2003 CSC working group also generated recommendations for increasing the efficiency of the OIA process. Once again, one of the recommendations focused on the offender needs or DFIA component of the OIA process. More specifically, it was recommended that the existing DFIA be reviewed to ensure it reflects up-to-date research and technological advancements.

Consequently, the Research Branch, in collaboration with the Offender Programs and Reintegration Branch launched a three-step research initiative in 1998 designed to address this recommendation. The strategy involved a series of external reviews of the DFIA, an internal statistical review of the reliability and validity of the DFIA, and an exhaustive consultation process. The consultation process included numerous field staff from across the country as well as various partners from national headquarters.

Offender Intake Assessment (OIA)

Development

A series of public inquiries and internal task forces conducted throughout the 1980s underscored the need for improved offender assessment and information-sharing within and between components of the criminal justice system. Not surprisingly, much attention concentrated on the decision-making policies and risk assessment procedures utilized by the Service and the National Parole Board. In August 1991, CSC launched an ambitious Correctional Strategy Initiative. This initiative set about putting into place a framework for establishing program priorities, implementing programs, and allocating resources to meet the needs of offenders (CSC, 1991). As renewed emphasis had been placed on the safe reintegration of offenders, the Service recognized the need for a comprehensive and integrated process to assess offenders upon admission to federal custody. Consequently, the Service assembled a National Offender Intake Assessment Working Group equipped with the mandate “to design and develop a systematic approach to offender assessment upon admission to federal corrections” (Motiuk, 1993). After numerous consultation sessions, the Working Group's efforts culminated in a national assessment model, the Offender Intake Assessment (OIA) process, intended to standardize and integrate offender risk and needs assessment throughout the CSC.

At the time of the Correctional Strategy Initiative, the tool that was being used to conduct assessments upon admission, the Force-field Analysis of Needs, was deemed to be inadequate for profiling the risk and needs of federal offenders. As a result, the National OIA Working Group constructed a new scheme to improve the assessment of criminal risk and identify offender needs at time of admission. The development of a new intake assessment protocol purposefully followed, and expanded upon, existing assessment tools, namely the Case Management Assessment Interview, the Force-field Analysis of Needs and the Community Risk/Needs Management Scale. Combined with a new Criminal Risk Assessment protocol (e.g., criminal history, offence severity, sex offence history, etc.), a new Case Needs Identification and Analysis (CNIA) protocol was developed that collapsed the 12 need areas of the Community Risk/Needs Management Scale into seven need dimensions or target domains. These included: employment, marital/family, associates/social interaction, substance abuse, community functioning, personal/emotional orientation, and attitude. Rating guidelines were developed for each of the seven domains and self-reference statements or indicators were crafted within each of the needs areas. The CNIA was first piloted as one component of the OIA process in 1992 and then implemented across the Service in November 1994 (Motiuk, 1997). Subsequent operational reviews of the OIA process during the mid-90s led to a name change from CNIA to Dynamic Factors Identification and Analysis (DFIA) for two reasons: first, to put focus on those offender needs assessed to be contributing factors to crime; and, second, to emphasize that these factors are capable of reflecting change, thereby dynamic in nature and promising targets for correctional intervention.

How OIA Works

The OIA and correctional planning processes begin immediately after the offender is sentenced. Parole officers are responsible for co-ordinating the collection of all relevant information pertaining to the offender's case such as criminal records, police reports, court transcripts, crown briefs, judges' comments, pre-sentence reports, community assessments and victim impact statements. During the information gathering phase, the parole officer screens the offender for immediate physical health concerns, security issues (personal and others' safety), mental health and suicide concerns. Next, the offender proceeds to the two core components of the assessment process: Assessment of Static Factors and Assessment of Dynamic Factors. The static assessment component emphasizes historical factors such as criminal history, offence severity, sex offence history and the probability of future re-offending as measured by the Statistical Information on Recidivism Scale – Revised 1 (SIR-R1) [Nuffield, 1982; Standard Operating Practice 700-04, ANNEX 700-04B; Correctional Service of Canada, 2003]. In contrast, the dynamic factors assessment component considers factors that can change in response to programming such as employment skills, substance abuse and attitude. Although the parole officer is responsible for completing the static and dynamic assessments, additional information including psychological reports, behavioural observations by unit staff and supplementary assessments pertaining to education, vocation and substance abuse is also incorporated. Thus, the process is multi-method in that the parole officer relies upon a variety of information sources (e.g., official police reports, offender self-report, collateral reports from community) as well as varied assessment strategies (e.g., interviews, self-report questionnaires, standardized assessments protocols, behavioural observation). Additionally, the process utilizes a multi-disciplinary team comprised of institutional and community parole officers, psychologists, vocational experts and front-line unit staff. In the end, an automated report is generated that includes information pertaining to criminal and social history, static intervention level, dynamic intervention level, treatment priorities, motivation level, reintegration potential, potential for detention referral, a security classification, an institutional placement decision, and, lastly, a sentence-wide correctional plan (Motiuk, 1997; CSC, 2003).

Dynamic Factors Assessment

The protocol used to conduct the dynamic factors assessment is called ‘Dynamic Factors Identification and Analysis (DFIA)'. It is comprised of seven dynamic factors: employment, marital/family, associates and social interaction, substance abuse, community functioning, personal and emotional orientation, and attitude. Each factor is subsequently divided into principal components that are further broken down into sub-components. Moreover, a series of yes/no indicators and, in some cases, 'help messages' accompany the indicators to enhance rating clarity. In total, the instrument consists of seven dynamic factor domains, 35 principal components, 94 sub-components, and 197 indicators. The primary objective of the DFIA is to provide a straightforward yet systematic means for identifying dynamic factors that inform the correctional plan. Specifically, the DFIA identifies and prioritizes factors that are directly linked to an offender's criminal behaviour. The DFIA also serves two additional functions, however. First, certain indicators assist in the program referral process. For example, indicators within the personal and emotional domain such as 'unable to recognize problem areas' and 'unable to generate choices' are used as flags for identifying individuals in need of one of the Service's correctional programs, namely Reasoning and Rehabilitation , formerly, Cognitive Skills. Second, certain indicator combinations are used to classify offenders according to Case Management Strategies (CMS) (Lerner, Arling & Baird, 1986). The CMS categorizes offenders into one of five groups. Each group is defined by a unique needs profile, interaction style, and criminal offence pattern. For example, individuals classified as 'Selective Intervention' are generally first-time offenders who have pro-social values. They value structure and lifestyle stability. In contrast, individuals classified as 'Limit Setters' have extensive and varied criminal histories, are characterized by highly internalized criminal attitudes, and are often motivated to commit crime for money, power, and excitement. The supervision strategy then varies as a function of the assigned classification.

Parole officers use pre-established guidelines outlined in the Service's Standard Operating Practices (SOP 700-04, CSC, 2003) to complete the DFIA. The guidelines specify that two of the domains (substance abuse and personal/emotional) be rated on a three-point scale ranging from ‘no need for improvement' to ‘considerable need for improvement'. The remaining five domains are rated on a four-point scale ranging from ‘factor seen as an asset to community adjustment' to ‘considerable need for improvement'. An overall dynamic intervention rating of either low, medium, or high is determined by integrating information regarding the severity and number of identified dynamic factors, the post-sentence community assessment and the initial assessment of immediate medical, health, and suicide concerns. The rating guidelines state that offenders with either no or relatively few identified dynamic factors are to be assigned an overall 'low' dynamic intervention rating. Offenders who have few identified dynamic factors but are rated 'considerable need for improvement' in those areas receive a high dynamic intervention rating. Similarly, offenders identified as multi-need, regardless of severity are also rated high. Lastly, parole officers are instructed to assign an overall high dynamic intervention rating to any offender exhibiting serious medical concerns, mental health problems or suicide risk potential (CSC, 2003). The organization of the “associates and social interaction” domain is provided in Figure 1 as an illustrative example. The entire DFIA protocol is presented in Appendix A.

Figure 1: Associates and Social Interaction Domain

Previous Research on DFIA

During the early '90s the DFIA, then known as Case Needs Identification and Analysis (CNIA), was adapted from the Community Risk/Needs Management Scale and Force-field Analysis of Needs and Case Management Strategies Interview and piloted in the Ontario region for community corrections. Motiuk and Brown (1993) examined the ability of the CNIA to predict post-release suspension outcome in a sample of 604 federally sentenced adult offenders (573 males; 31 females). The results demonstrated that each of the seven dynamic factors was significantly related to the likelihood of having a suspension warrant issued during the first six months of release. Furthermore, the following indicators demonstrated strong predictive validity: unstable job history, criminal friends and acquaintances, associates with drinkers/drug abusers, relations with others are exploitative, poor financial management, unable to set goals, low empathy, impulsive, difficulty controlling temper, copes poorly with stress/frustration, and unable to work towards life goals.

In 1998, Motiuk examined the predictive validity of each DFIA domain in a sample of 3,380 male offenders who had been at risk in the community, for on average 250 days. Chi-square and correlational analyses revealed that each domain rating was significantly related to return to federal custody. Pearson r 's correlation coefficients ranged from .09 for the attitudinal domain to .17 for the associates and employment domains. The correlation coefficients for the remaining domains (personal/emotional domain, community functioning, substance abuse, marital/ family) ranged from .11 to .15.

In 1997, Motiuk presented additional evidence for the validity of the DFIA and demonstrated that the number of positively endorsed indicators within a given domain (i.e., the number of indicators scored ‘yes') was strongly associated with the domain rating. For example, the number of positively endorsed indicators in the substance abuse domain was correlated .78 with the substance abuse domain rating. In sum, all correlations ranged from a low of .54 for community functioning to .78 for substance abuse. Thus, this analysis demonstrated that the DFIA was being used in the manner in which it was originally intended; that the individual indicator endorsements would guide the parole officer's domain rating.

Motiuk and Blanchette (2000) examined the gender differences between women (n = 469) and men (n = 12,265) offenders who underwent OIA and found that, at time of admission, women offenders were more likely to have had difficulties in the area of associates/significant others. Men offenders were more likely to have been experiencing problems in substance abuse and attitude. Interestingly, there appeared to be no statistically meaningful differences between women and men with respect to employment, community functioning, or personal/emotional orientation. In other words, women and men were equally as likely to be experiencing difficulties in these areas.

At the indicator level of the DFIA, Motiuk and Blanchette (2000) reported gender-specific issues that emerged for women offenders relative to their male counterparts. As a group, women offenders were more likely than men to be underemployed or unemployed and to have experienced relational (marital, maternal, sibling, and other relative) difficulties. Also, women offenders tended to have more family members involved in crime, have been victims of spousal abuse, be single, have parenting responsibilities, and be easily influenced by others. Motiuk and Blanchette (2000) also found that women offenders' drug use was more likely to interfere with their health, and there was a greater likelihood of dietary problems among them. Financially, women offenders were more likely than men to have relied on social assistance and lack credit. Also, relative to their male counterparts, women offenders were more likely to be unaware of consequences, take more risks, worry unreasonably, and have problems asserting themselves. Further, women offenders were more likely than men to have been diagnosed as mentally disordered, to have been prescribed medication for mental health issues, and to have been hospitalized for mental health problems in the past.

Post-implementation of the OIA, perhaps the most compelling research on the DFIA has been to produce meaningful and accurate profiles of specific types of offenders and offender characteristics: offender employment patterns (Motiuk, 1996), sex offenders (Motiuk & Belcourt, 1996), violent offenders (Motiuk & Belcourt, 1997), offenders on conditional release (Motiuk, 1998; Trevethan & Rastin, 2003), older offenders (Uzoaba, 1998), women offenders (Motiuk & Nafekh, 1999; Bell, 2004; Law, 2004; Taylor & Flight, 2004), young adult offenders (Motiuk & Latimer, 1999), Aboriginal offenders (Motiuk & Nafekh, 2000; Trevethan, Moore & Rastin, 2002; Moore & Trevethan, 2002; Moore, 2003; Trevethan, Moore, Naqitarvik, Watson & Saunders, 2004 ), long-term offenders (Motiuk & Nafekh, 2000), substance-abusing offenders (Motiuk & Nafekh, 2001), dangerous and long-term supervision order offenders (Trevethan, Crutcher & Moore, 2002), women offenders who engage in self-harm (Wichman, Serin & Abracen, 2002), gang affiliation (Nafekh, 2002; MacKenzie & Johnson, 2003; Nafekh & Stys, 2004), offender physical health (Motiuk, 2002) and mental health (Boe & Vuong, 2002). Presently, comprehensive DFIA profiles are available Service-wide via a Climate Indicators and Profiling System (CIPS) in two scenarios – “on any given day” (a snapshot of who is in custody) and “in any given year” ( a flow of new admissions). Presently, these DFIA profiles are available by region, institution or area office; for men, women and Aboriginal men (and more recently Aboriginal women); and for both the institution and community supervision populations (Boe, 2003).

Purpose of DFIA Review

The purpose of this review is threefold: first, to present the results of each step of the DFIA project, namely the results of the meta-analytic and narrative reviews, the psyc home tric review, and, lastly, the results of the consultative exercises; secondly, to integrate the results of all three initiatives and propose how the dynamic factors assessment component should be revised in accordance with the findings; and third, to discuss directions for future research and to review the operational implications of the proposed changes. The methodology and results associated with each study are presented separately followed by a brief discussion of how best to enhance the program referral process.

STEP 1: A META-ANALYTIC REVIEW

Methodology

The first step in the DFIA review involved a meta-analytic review of the DFIA domains and indicators. The process was external in that CSC-derived studies were excluded from the review. Moreover, correctional experts, the majority of whom were external to the Service, conducted the reviews. Each expert was instructed to complete three tasks: first, to review the literature pertaining to the predictive validity of a given domain and its corresponding components and indicators in adult offender populations; second, to provide recommendations for enhancing the domain; and third, to highlight research findings pertaining to women and Aboriginal offenders as well as other visible minorities.

Each review involved a quantitative meta-analysis, a qualitative narrative review, or a combination of the two. Briefly, a meta-analysis is a statistical technique that allows researchers to objectively aggregate the size of a relationship between two variables (e.g., criminal associates and recidivism) across numerous studies in the form of an effect size or correlation coefficient. In contrast, a narrative review involves a qualitative examination of a given area whereby a researcher reviews all relevant literature and renders a summary statement based on his/her subjective interpretation. Arguably, both strategies are required for theory building, identification of new research directions, and ascertaining the current state of the literature (McGuire & Priestley, 1995).

Step 1 presents the meta-analytic review results. Although general proposed changes provided by the external experts are discussed in Step 1, domain-specific proposals are reserved for Step 3. This strategy was adopted to reduce duplication. More specifically, Step 3 presents the field consultation results including the field's response to specific external proposals. Consequently, given that it was necessary to discuss how the field responded to the expert domain-specific proposals, it was decided to reserve discussion of the specific expert proposals for Step 3. Lastly, summaries of the various meta-analytic reports can be found in Volume 10, Number 3 Forum on Corrections Research as well as in various research reports published by the Research Branch (see www.csc-scc.gc.ca) .

Meta-Analytic Results

Overview of Findings

Meta-analytic findings were available for each domain with the exception of personal/emotional orientation; the external reviewers conducted a narrative rather than meta-analytic review for this domain. Table 1 summarizes the general findings from the meta-analyses and can be interpreted as follows. Reading from the left of row 1, the employment meta-analysis was based on 69 studies involving 128,018 offenders that generated 214 separate effect sizes. Overall, the average unweighted effect size or correlation between the employment domain and general recidivism was .13. The strength of this relationship did not change when the effect size was weighted for sample size variations. Analyses revealed that each weighted and unweighted effect size was statistically significant. Further, not only was the magnitude of the various relationships consistent with past research (Gendreau, Little and Goggin, 1996) but it was also consistent with the Step 2 results. In sum, the results confirm the overall criminogenic relationship between each domain and recidivism.

Table 1: Summary of Meta-Analytic Results

^aThe marital/family results are based on the combined findings of two separate meta-analyses (Brown & Dowden, 1999, and Gendreau et al., 1996) that used overlapping studies. Consequently a weighted effect size could not be calculated.

^bThe review of the personal/emotional domain was narrative in nature, and consequently effect sizes were not available.

Each reviewer undertook a more in-depth analysis to determine which factors within a given domain were most strongly related to recidivism. Originally, attempts were made to organize each review around the principal components, sub-components and indicators of the DFIA. In most cases, however, the predictor categories examined in the existing literature did not permit this type of classification. Consequently, most reviewers used naturally occurring predictor categories deemed close approximations to DFIA principal components and sub-components. Moreover, predictive studies for certain components of the DFIA were unavailable. Analysis at the indicator level was often unavailable or deemed unreliable.

Considering all reviews simultaneously, most predictor categories produced moderate correlations with recidivism (weighted effect sizes between .10 and .19). As Table 2 demonstrates, few predictor categories demonstrated strong (weighted effect sizes equal to or greater than .20) or weak (weighted effect sizes less than .10) relationships with recidivism. Lastly, the aforementioned results are consistent with previous research and prominent theories of criminal conduct (Gendreau et al., 1996; Andrews & Bonta, 2003 ).

Table 2: Dynamic Factor Principal Component Summary

^aThe personal/emotional review was not a meta-analysis, so the size of the empirical relationship was estimated conservatively based on the narrative review.

^bThis result is based on samples of known sex offenders.

In sum, the meta-analytic review process revealed that the DFIA contains all predictor categories identified by external experts as strong predictors of criminal recidivism. Further, with the exception of denial and minimization of crime, the external reviews confirmed that the DFIA adequately represents all predictor categories identified as having moderate empirical support. Thus, the meta-analytic reviews provide support for the content validity of the DFIA; specifically, the DFIA includes the best predictors of criminal behaviour as identified in the literature. It is important to emphasize that the meta-analytic review process also revealed that the DFIA contains several factors deemed weak predictors of criminal recidivism. Additionally, the meta-analytic review process identified few studies devoted exclusively to women, Aboriginal offenders or minority groups.

Employment Domain

Gendreau, Goggin and Gray (1998; 2000) conducted a meta-analytic review of employment factors and recidivism among adult offender populations. The DFIA statistical results were not included in the review. Overall, the authors identified 67 studies that generated 200 individual effect sizes with recidivism. The major findings are presented in Table 3. The results confirm that employment history and employment needs at discharge were predictive of criminal recidivism. Although education was predictive of recidivism, the strength of the relationship was considerably less than that for employment.

Gendreau et al. (1998; 2000) identified few studies devoted exclusively to women or Aboriginal offenders. Further, inconsistent results were often observed for women. Moreover, the strength of the relationship between employment-related factors and recidivism was typically stronger among non-Aboriginal than Aboriginal offenders. The authors conclude that additional research is required involving unique samples of women and Aboriginal offenders before firm conclusions can be reached.

Table 3: Employment Domain: Meta-Analytic Review

Note: k = effect sizes per predictor domain. N = subjects per predictor domain.

Mr = mean Pearson r (average unweighted effect size); SD = Standard Deviation ; Mz + = [( z r )x( N – 3)] ÷ ( N – 3)½] where N = number of subjects per effect size (average weighted effect size); CI = confidence interval about the mean Pearson r and mean z + ; * p < .05.

Marital/Family Domain

Results from three separate reviews (Brown & Dowden, 1999; Gendreau et al., 1996; Oddone-Paolucci, Violato & Schofield, 1998; 2000) were consulted to obtain an external position regarding the relationship between the marital/family domain of the DFIA and recidivism. While Oddone-Paoluci et al. (1998; 2000) conducted a narrative review, Brown and Dowden (1999) and Gendreau et al. (1996) conducted meta-analytic reviews, the results of which are presented in Table 4. While there was a paucity of external studies at the indicator level, the meta-analytic results demonstrated that family background characteristics, marital quality, marital status, and parenting skills were moderately related to recidivism. It should be noted that only three studies examined parenting ability in relationship to recidivism. Additionally, no Aboriginal-specific studies regarding the role of marital/family factors and criminal re-offending were located.

In regards to women offenders, Dowden and Andrews' (1999) meta-analytic review of treatment effects on criminal recidivism demonstrated that programs addressing family process issues ( n =9 studies) yielded the strongest reductions in reoffending among women offender samples. Loucks and Zamble (2000), however, demonstrated that family cohesiveness was not predictive of recidivism in a sample of federally sentenced women offenders. In regards to parenthood, Bonta, Pang, and Wallace-Capretta (1995) demonstrated that single-parents recidivate at higher rates than those with partners. Lastly, Rettinger (1998) showed that parenthood in and of itself was not predictive of recidivism in a sample of women offenders.

Table 4: Marital/Family Domain: Meta-Analytic Review

Note: Results pertaining to the first three predictor variables are taken from Gendreau et al., 1996. Results from the last three predictor variables are derived from Brown & Dowden, 1999. k = effect sizes per predictor domain. N = subjects per predictor domain. Mr = mean Pearson r (average weighted effect size); Mz + = [( z r )x( N – 3)] ÷ ( N – 3)½] where N = number of subjects per effect size (average weighted effect size).

Associates Domain

Goggin, Gendreau and Gray (1998a; 1998b) conducted a meta-analytic review of the relationship between criminal associates and recidivism among adult offender samples. Overall, 35 predictive studies generated 75 effect sizes with recidivism. As Table 5 demonstrates, criminal companions, criminal family, and crime neighbourhood were all moderate to strong predictors of criminal recidivism. Notable was the weighted mean effect size of .21 for criminal companions.

The external reviews located two independent studies involving Aboriginal offenders (Bonta, 1989; Bonta, LaPrairie & Wallace-Capretta, 1997). Both studies demonstrated that criminal associates was equally predictive of criminal recidivism among Aboriginal and non-Aboriginal offender groups. The authors did not identify any studies that exclusively examined the role of criminal associates and criminal recidivism among women offender samples, although more recent meta-analytic research by Dowden and Andrews (1999) reported that ‘associates-based' correctional programming was strongly related to reduced re-offending in studies involving predominantly or exclusively women offender samples.

Table 5: Associates Domain: Meta-Analytic Review

k = number of effect sizes per predictor domain; N = number subjects per predictor domain.

Mr (SD) = mean Pearson r; SD = standard deviation (average unweighted effect size); CI = confidence interval about the mean Pearson r and mean z + ; Mz + = [( z r ) x ( N – 3)] ÷ ( N – 3)½] where N = number of subjects per effect size (average weighted effect size); * p < .05.

Substance Abuse Domain

Two independent substance abuse reviews were conducted. First, Boland, Henderson and Baker (1998) conducted a narrative review that examined existing substance abuse assessment approaches as well as the appropriateness of the content and assessment guidelines associated with the substance abuse domain. Additionally, Boland et al. (1998) provided a series of process-, as well as content-orientated suggestions for enhancing the substance abuse domain. Dowden and Brown (1998; 2002) conducted the second review. This review focused specifically on establishing the magnitude of the relationship between the indicators comprising the substance abuse domain and criminal recidivism. A meta-analytic approach was used to meet this objective. Originally, the authors attempted to organize the meta-analytic predictor categories around the principal components and sub-components comprising the substance abuse domain. However, this proved difficult due the nature of the existing literature. Consequently, the existing research literature was grouped into conceptually similar predictor categories that approximated the principal components and sub-components of the substance abuse domain as much as possible.

To determine whether or not substance abuse was differentially predictive across gender, the mean effect size for each predictor category was re-calculated separately for both men and women. Overall, substance abuse was slightly more predictive for women ( Mz + = .14) than men ( Mz + = .10) (see Table 6). Overlapping confidence intervals rendered the effect negligible, however. Interestingly, the only predictor category to yield a significant between-group difference was parental substance abuser. More specifically, the weighted mean effect size for males ( Mz + = .09) was significantly smaller than the weighted mean effect size for women ( Mz + = .20). This difference was further confirmed by the presence of non-overlapping confidence intervals (see Table 40).

Table 6: Substance Abuse Domain: Meta-Analytic Review

Note: k = number of effect sizes per predictor domain; N = number subjects per predictor domain.

Mr = mean Pearson r (average unweighted effect size); Mz + = [( z r ) x ( N – 3)] ÷ ( N – 3)½] where N  = number of subjects per effect size (average weighted effect size); CI = confidence interval about the mean z + . * p < .05, **p < .01

In sum, Boland et al. concluded that “although the DFIA requires some modification in terms of scoring and item restructuring, it addresses the major content areas deemed necessary to screen accurately for substance abuse and to assess the severity of the problem” (p. 34, Boland et al., 1998). Also noteworthy are two additional observations made by Boland and colleagues in regards to the validity of self-report data and the multi-purpose nature of assessment. First, Boland et al. conclude that, in most cases, self-reported information tends to be relatively accurate and consequently should be considered a valid source of information, though it is advisable to incorporate collateral sources of information when over- or under-reporting is suspected. Second, Boland et al. remind us that assessment measures in general, and more specifically, substance abuse measures in particular, exist along a five-dimensional continuum: 1) those that screen for the presence or absence of a problem; 2) those that elaborate on the nature and severity of the problem; 3) those that help establish treatment and relapse prevention targets but also assess changes associated with treatment; 4) comprehensive batteries that serve multi-functions including screening, problem severity assessment, multiple need identification, treatment planning and research database construction; and 5) laboratory assessments. It should be noted that Boland's assessment framework guided the DFIA revisions. Specifically, all proposed changes were performed under the assumption that the DFIA is a level 1 and level 2 assessment process. Thus, not only should it identify whether or not a problem is present or absent, but it also identifies the nature and severity of that problem.

Community Functioning Domain

Gates, Dowden and Brown (1998) conducted a meta-analytic review that examined the predictive relationship between community functioning variables and adult recidivism (e.g., readmission, new charges/convictions, revocation). Seventeen studies were identified yielding 80 effect sizes. It should be noted that the meta-analysis identified studies that focused on community functioning variables extraneous to the DFIA. As a result, two additional categories were included in the analyses: living companions (living alone versus living with others) and childhood community functioning (accommodation stability during childhood).

In sum, the community functioning variables generated an overall weighted mean effect size of .10 with recidivism. The leisure component produced the strongest effect size followed by accommodation and finance (see Table 7). Further analyses revealed the results were consistent across gender and ethnic origin. This finding was also confirmed when the data were disaggregated by gender (12 women-specific studies) in that the overall weighted mean effect size for women remained unchanged ( Mz+ = .09; C. Dowden, August 15, 2000, personal communication cited in Blanchette, 2001).

Table 7: Community Functioning Domain: Meta-Analytic Review

^agiven that only one effect size comprised this predictor category confidence intervals were not calculated.

Note: k = number of effect sizes per predictor domain; N = number subjects per predictor domain.

Mr = mean Pearson r (average unweighted effect size) ; Mz + = [( z r ) x ( N – 3)] ÷ ( N – 3)½] where N  = number of subjects per effect size (average weighted effect size); CI = confidence interval about the mean z + . * p < .05, ** p < .01.

Personal/Emotional Domain

Robinson, Porporino, and Beal (1998; 1999) conducted a detailed narrative review of the personal/emotional domain. In sum, the review supports the continued use of the cognitive and behavioural principal components and the elimination of the self-concept, mental ability, mental health and intervention principal components. A detailed summary of the results are presented in Table 8. The authors' recommendations for either retaining or eliminating a given indicator or component are empirically as well as theoretically based. Aboriginal- and women-specific results were not discussed in detail.

Blanchette (2001) concludes that the majority of conclusions reached by Robinson et al. (1998) apply equally to women offenders. Additionally, she notes that three additional variables (i.e., self-esteem, personal victimization, and self-harm/parasuicide) have received considerable attention in the literature in regards to women offenders. For example, Larivière (1999) conducted a meta-analytic review of the relationship between self-esteem and criminality. He identified six studies comprised exclusively of women offender samples. Overall, the results indicated a strong negative correlation between self-esteem and antisocial behaviour among women offenders ( r = -.38). Interestingly, this correlation was substantially lower for men ( r = -.17). It is important to note, however, that the meta-analysis included both predictive and post-dictive studies. For example, only one of the six women offender studies was predictive in nature. Additionally, the women offender studies were largely over-represented by individuals who had engaged in child abuse.

Table 8: Personal/Emotional Domain: Narrative Literature Review

Table taken from Robinson, Porporino & Beal (1998). A Review of the Literature on Personal/Emotional Need Factors . Research Report (R-76), Correctional Service of Canada. R = Retain as indicator; D = Delete as indicator; E = Empirical support available; T = Theoretical support.

In regards to victimization, Blanchette (2001, p. 74) concludes:

Collectively, the research to date suggests that victimization, although very common amongst female offenders, is not a criminogenic need. Collectively, study results suggest that, while victimization experiences appear to play a role in the onset of criminal offending, they are not associated with recidivism. Despite this, the astonishingly high prevalence of survivors in the correctional system signals a requirement for service providers to address this issue. Although some authors argue that past victimization represents an important treatment target for female offenders, this does not necessarily infer that 'victimization' represents a need that is criminogenic (Gray et al., 1995; Koons et al., 1997). Rather, it has been suggested that it is the psychological sequelae to victimization experiences that contributes to behaviours such as substance abuse, criminal offending, adjustment while incarcerated, and recidivism (McLean, 1998). Many women suffer from post-traumatic symptoms that can potentially impede progress in addressing other (criminogenic) need areas. It is therefore suggested that female offenders' victimization histories are an important part of a holistic approach to case-based classification for effective correctional intervention.

Lastly, Blanchette's review identified three studies (i.e., Bonta et al., 1995; Blanchette & Motiuk, 1995; Rettinger, 1998) that examined the relationship between self-injury and/or attempted suicide among women offenders and criminal recidivism. Each study demonstrated that self-injury and/or attempted suicide was predictive of future criminal offending including violent re-offending, thereby supporting the argument that self-injurious behaviour is indeed criminogenic.

Attitude Domain

Law (1998) conducted a meta-analytic review examining the ability of the criminal attitude domain and its various components to predict criminal recidivism. The meta-analysis included 32 studies yielding 112 correlations with recidivism. Overall, the meta-analysis indicated that the justice, violence and lifestyle components of the DFIA were moderately related to recidivism. In contrast, the society component was only weakly related to recidivism. The predictive power of the property component could not be assessed given that no studies examined its relationship to future criminal behaviour. The strongest indicator was ‘non-conforming attitudes'. In contrast, the weakest indicators included attitudes toward employment, marital/family, and personal/emotional stability. One construct currently not included in the DFIA but identified in the external literature as being criminogenic was denial and minimization (techniques used by offenders to minimize the severity and nature of their offending). See Table 9 for an overview of the results. Women- and Aboriginal-specific results were not reported.

Dowden and Andrews (1999) conducted a meta-analytic review of effective treatment targets in adult offender populations. A composite treatment target, labelled ‘antisocial cognition', comprised of variables such as antisocial attitudes, values, and rationalizations proved to be an important treatment target for both men and women offenders. The results pertaining to women, however, were derived from a small number of effect sizes ( k = 5).

Table 9: Attitude Domain: Meta-Analytic Review

^aother to the small number of effect sizes confidence intervals were not calculated.

k = number of effect sizes per predictor domain; N = number subjects per predictor domain.

Mr = mean Pearson r (average unweighted effect size); Mz + = [( z r ) x ( N – 3)] ÷ ( N – 3)½] where N  = number of subjects per effect size (average weighted effect size); CI = confidence interval about the mean z +, * p < .05, ** p < .01.

External Expert Proposals

In addition to the empirical review, each reviewer provided a variety of proposals for enhancing the DFIA. In sum, ten themes emerged. Each theme is discussed in turn.

1. Keep the Strong, Keep the Moderate, but Drop the Weak

Most reviewers suggested that indicators demonstrating strong empirical support, strong theoretical support or moderate empirical support be retained, while indicators demonstrating either weak or no support be deleted.

2. Less is More

Currently, the DFIA is comprised of 7 domains, 35 principal components, 94 sub-components and 197 indicators. The most common theme presented by the external reviewers was the need to simplify the DFIA's structure. The proposed strategies to do this included the deletion of indicators and/or sub-components that were redundant, showed little or no correlation with recidivism or were represented in more than one domain. Additional suggestions included using highly specific indicators as operational definitions for more global constructs as well as creating a new domain called 'non-criminogenic factors' (e.g., non-contributing).

3. Increase Objectivity, Reduce Subjectivity

Fifty years of research overwhelmingly confirms the superiority of objective, statistically based prediction strategies over purely subjective, clinically based methods (Grove & Meehl, 1996; Grove, Zald, Lebow, Switz & Nelson, 2000; Hall, 1988; Mossman, 1994; Wormith & Goldstone, 1984). This conclusion was firmly echoed in several of the external proposals that recommended improved scoring guidelines as well as the development of concrete behavioural indicators.

4. Revitalize with New Additions

It was evident throughout the reviews that the DFIA could be enhanced by incorporating a few additional constructs, including intrinsic job motivation and denial and/or minimization of crime.

5. View the Supplementary Assessment as Friend Rather than Foe

There is some concern that specialized or supplementary assessments may result in over-programming for low-risk and/or low-need offenders. Two independent reviewers recommended their continued use for the substance abuse and employment domains, however.

6. Distinguish Non-criminogenic from Criminogenic Need

The DFIA must clearly distinguish which indicators are criminogenic (contributing factors - dynamic or changeable treatment targets that are directly related to criminal behaviour) versus those that are non-criminogenic (non-contributing factors – treatment targets that are not related to criminal behaviour but require intervention nonetheless).

7. Maximize the Dynamic Nature of the Assessment Strategy

Although the DFIA serves multiple functions, its ultimate purpose is to identify factors that can, in theory, be altered. Thus, unless it can be shown that a given static factor serves a specific operational function, every effort should be made to ensure that each component of the DFIA reflects current and/or recent circumstances, rather than events or circumstances from the past. Thus, while poor family functioning during childhood is important, what may be more crucial for effective correctional programming is current family functioning or, similarly, family functioning when offences are originally committed.

8. Assess the Need as well as Severity

The substance abuse assessment review introduced the notion of a stepped approach to assessment. The first step involves screening for the presence or absence of a problem; the second step elaborates on the nature and severity of the problem; and the third step establishes specific treatment and relapse prevention targets and assesses changes associated with treatment. Perhaps a similar framework could be applied to the DFIA. Steps 1 and 2 would be considered a mandatory component of the DFIA, and Step 3 could be incorporated during program delivery. This strategy might reduce both unnecessary programming and redundancy in assessment.

9. Beware of Specialized Risk Factors for Specialized Offender Groups

Deviant sexual preference and deviant sexual attitudes predict sexual recidivism among known sex offenders, but they do not predict general recidivism in the general male offender population. In fact, there is a slight negative correlation between these variables and general recidivism among male offenders (Robinson et al., 1998; 1999). Thus, it might be prudent to remove indicators related to sex offences from the DFIA and use them exclusively during assessments of sex offenders.

10. More Research, Better Research

Another common theme was the need for new and improved research, specifically in regards to women, Aboriginal offenders and the identification of unique predictors of violent recidivism. Additionally, the application of advanced statistical procedures coupled with the development of assessment strategies that incorporate concrete behavioural indicators was also suggested.

STEP 2: A PSYCHOMETRIC REVIEW

Methodology

All federal offenders who were assessed via the OIA and correctional planning process between November 1 st , 1994, and September 1 st , 2000, were identified as potential candidates for the review. Potential candidates were included in the analysis if they had been subsequently released as of September 1 st , 2000. This method generated 16,645 offenders, the majority of whom (approximately 99%) had been initially assessed when they were admitted on a Warrant of Committal. A small number of revocation cases ( n = 25) were included given that a new OIA had been conducted as a result of the revocation. Two hundred and twenty cases were excluded given that the variable concerning gender identity or ethnicity was missing. A further 181 cases were excluded given that more than 20% of the data was missing at the indicator level. Therefore all analyses are based on a total release cohort of 16,244 offenders that included 765 women offenders and 2,593 Aboriginal offenders.

The release cohort ranged in age from 17.7 to 81.6 years with a mean of 34.6 ( SD = 10.6). Women offenders ( M = 33.4, SD = 8.9) were found to be significantly younger than their male counterparts ( M = 34.6, SD = 10.7), t (877) = 3.4, p < .001. Aboriginal offenders ( M = 32.1, SD = 9.3) were also found to be significantly younger than their non-Aboriginal counterparts ( M = 35.0, SD = 10.8), t (4,032) = 14.3, p < .001.

As Table 10 illustrates, the majority of the release cohort is male (95.3%). In terms of ethnicity, the predominant group is Caucasian (69.4%) followed by Aboriginal (16.0%). The mean aggregate determinate sentence length was 3.7 years ( SD = 2.3), with a range from 6 months to 25 years. The mean aggregate determinate sentence length was significantly shorter for women ( M = 3.4 years, SD = 1.9) in comparison to men ( M = 3.7 years, SD = 2.3), t (882) = 4.7, p < .001. In addition, the mean aggregate sentence length was significantly shorter for Aboriginal offenders ( M = 3.5 years, SD  = 1.9) than non-Aboriginal offenders ( M = 3.7 years, SD = 2.4) t (4362) = 6.3, p  < .001.

Although OIA was not implemented until 1994, the domain level ratings had been retrospectively completed for all of those who were already incarcerated as of 1994. Consequently, the release cohort included a small number of lifers or dangerous offenders ( n = 30) who had been released as of September 1 st , 2000.

Table 10: Release Cohort Characteristics: Gender and Ethnic Origin

More than two-thirds of the release cohort had either been convicted in the past or were currently serving time for a violence-related offence. Aboriginal offenders were significantly more likely to have been convicted (past or present) for a violent offence (87.2%) in comparison to their non-Aboriginal counterparts (64.3%), X² (1, N = 15,629) = 501.2, p < .001. In contrast, women offenders (43.1%) were significantly less likely to have been convicted of a violent offence (past or present) compared to male offenders (69.2%), X² (1, N = 15,629) = 221.4, p < .001. More information regarding the nature of the offence history is provided in Table 11.

Table 11: Offence History

Note: ^aoffence categories are not mutually exclusive.

As Table 12 illustrates, the majority of the release cohort was released on some form of discretionary conditional release. The percentage of offenders released in each region is also presented in Table 12.

Table 12: Release Type and Releasing Region

Follow-up Period and Outcome Definition

Outcome information was extracted from the Offender Management System (OMS) on September 1 st , 2003. Given that offenders were only included in the study if they had been released as of September 1 st , 2000, this allowed for a minimum three- year follow-up period. Additionally, although some offenders were at risk for more than three years, the follow-up period was fixed at three years to ensure that each offender included in the study was afforded the same amount of time to recidivate. Recidivism was defined as a readmission to federal custody either pre or post warrant expiry. Thus, offenders who were readmitted for breaching their conditional release due to the commission of a new offence(s) or for technical violations (e.g., association with criminal others, substance abuse) were counted as failures. Additionally, offenders convicted of new offence(s) (sentence > two years) post warrant expiry were also considered failures. It should be noted that supplementary analyses using a more narrow definition of recidivism [i.e., returned to federal custody with new offence(s)] were also conducted. However, given that both sets of analyses produced similar trends only those pertaining to the broader definition of recidivism [i.e., readmitted with or without an offence(s)] will be presented.

Statistical Analysis

The reliability of the DFIA was assessed in terms of internal consistency. More specifically, Cronbach's alpha was calculated for each domain of the DFIA. The predictive validity of the DFIA domains and the individual indicators was assessed at the univariate and the multivariate level. The univariate analyses involved a series of correlational and chi square analyses. Additionally, the Bonferoni correction procedure was applied to control for Type 1 errors. It should be noted, however, that the magnitude of the effect sizes as well as their statistical significance guided the interpretation of the results due to the large sample size. It is important to consider the magnitude of the effect as well as the significance level given that large sample sizes often generate Type I errors (e.g., the probability of erroneously finding significant differences) ( see Kraemer et al., 1997, and Gendreau, Goggin, & Smith, 2001, for a discussion of the importance of considering magnitude as well as statistical significance). Consequently, significant univariate correlation coefficients that exceeded .20 were considered strong predictors of readmission, while significant coefficients between .10 and .19 were considered moderate. Lastly, correlation coefficients that ranged between .05 and .09 were considered weak predictors while those that were .04 or less were considered to have no predictive relationship with readmission.

Next, a series of stepwise regressions were conducted for each domain. For example, all of the substance abuse indicators were entered into a stepwise regression to determine which of the substance abuse indicators were most strongly related to readmission. This allowed for a relative comparison among all of the indicators comprising a given domain in terms of their ability to predict readmission. Due to the reasons discussed above, however, both ‘proportion of variance accounted for in outcome' and statistical significance (Bonferoni adjusted) were used to establish the entry and exit criteria.

All analyses were conducted separately for all three groups: men, women, and Aboriginal offenders. Lastly, missing data at the indicator level were replaced with the mode. Although the percentage of missing data ranged from 0% to 17.4% across all 197 indicators, it should be noted that the vast majority of the indicators (94% within the male release cohort, 96% within the Aboriginal offender cohort, and 98% within the women offender cohort) had no more than 3% of their data missing.

Psychometric Results

Reliability

As Table 13 indicates each domain demonstrated acceptable levels of internal consistency with alphas ranging from .62 to .96. Only one domain, criminal associates, resulted in alphas less than .70 for all three groups. This result is most likely attributable to the relatively small number of indicators (11) that comprise the associates domain. Also noteworthy was that the results were consistent for men, women and Aboriginal offenders, although, the community functioning alphas were under .70 for women (.62) and Aboriginal offenders (.65) but not so for men (alpha = .70).

Three domains (employment, marital/family, substance abuse) contain subsets of indicators that are automatically rated 'no' or 'yes' if they are not applicable to the offender in question. For example, in the marital/family domain offenders who do not have children are automatically rated 'no' for each indicator that specifically rates parenting abilities. Similarly, offenders who are rated as having 'no employment history' automatically receive default ratings of 'no' for those indicators that pertain directly to work performance (e.g., often shows up late, no initiative). Consequently, supplementary reliability analyses were conducted on each subset of indicators affected by the default scoring system.

In regards to the subset of employment indicators affected by the default system (16 indicators), the re-computed alphas were as follows: 1) male release cohort: alpha = .81, 2) women offender release cohort: alpha = .71, and 3) Aboriginal offender release cohort: alpha = .77. In regards to the marital domain two separate subsets of indicators, one pertaining to parenting abilities (9 indicators) and the other pertaining to marital relationship quality (6 indicators) are affected by the default system. The re-computed alphas for the relationship quality indicators were as follows: male release cohort: alpha = .61, women offender release cohort: alpha = .59, and Aboriginal offender release cohort: alpha = .55. Similarly, for the parenting abilities subset, the alphas were as follows: male release cohort: alpha = .81, women offender release cohort: alpha = .80, and Aboriginal offender release cohort: alpha = .79. The subset of alcohol abuse indicators affected by the default scoring system generated the following alphas: male release cohort: alpha = .83, women offender release cohort: alpha = .84, and Aboriginal offender release cohort: alpha = .85. Lastly, the drug abuse subset of indicators affected by the default scoring system also produced comparable alphas: male release cohort: alpha = .85, women offender release cohort: alpha = .80, and Aboriginal release cohort: alpha = .69. Thus, with the exception marital quality indicators, all of the indicator sub domains affected by the default system produced comparable results to those presented for the overall domains.

Table 13: Internal Consistency for Each Dynamic Factor Domain

Readmission Rates by Level of Dynamic Intervention

Overall, the readmission rate for the entire release cohort was 43.2%. While the base rate was lower for women (32.4%), more than half of the Aboriginal cohort (56.1%) was readmitted during the follow-up period. Readmission rates by the overall level of dynamic intervention are presented in Table 14. As expected, offenders identified with a higher level of dynamic intervention were significantly more likely to be readmitted than those with a lower dynamic intervention rating. This trend was observed for men [ X² (2, N = 15, 479) = 1061.3, p < .001], women [ X² (2, N = 765) = 81.1, p < .001], and Aboriginal offenders [ X² (2, N = 2,593) = 41.1, p < .001].

Table 14: Readmission Rates by Level of Dynamic Intervention Rating

Note: The relationship between the dynamic intervention rating and readmission was statistically significant for all three groups. The corresponding test statistic results are presented in the text.

Readmission Rates for Each Dynamic Domain Factor Rating

Readmission rates for each dynamic factor domain rating are presented in Table 15. As expected, offenders who received higher dynamic factor ratings were significantly more likely to be readmitted than those who received lower dynamic factor ratings. This general finding was consistent for men, women and Aboriginal offenders. It should be noted, however, that a perfect linear relationship in the expected direction (i.e., higher dynamic factor rating associated with a greater probability of readmission) was not always observed among the women and Aboriginal offender release cohorts. For example, among the women offender release cohort a linear relationship was not observed between readmission rates and the employment, marital/family, and community functioning domains. For example, while 37.2% of women offenders rated ‘no difficulty' on the marital/family domain were readmitted, a lower percentage (26.4%) of the women rated ‘some difficulty' were readmitted. This result is counterintuitive in that individuals rated ‘some difficulty' should evidence a greater likelihood of failure than individuals rated ‘no difficulty'. Similarly, the marital/family and personal/emotional domain ratings for Aboriginal offenders also did not evidence a clear linear relationship with readmission.

Table 15: Readmission Rates for Each Dynamic Domain Rating

Note: *** p < .001.

Employment Domain: Indicator Endorsement Rates

Table 16 displays the percentage of positively endorsed indicators (indicators rated 'yes') for each release cohort (men, women, Aboriginal). It is important to note that the results have been disaggregated by gender and Aboriginal status for descriptive purposes only. More specifically, the purpose of this phase of the analysis was not to conduct a comparative examination of the differences between groups. Rather, the purpose was to flag indicators with exceedingly low or high endorsement rates. Exceedingly uneven distributions within a dichotomous variable (e.g., 90%/10%) can artificially deflate the strength of the relationship between such an indicator and the criterion of interest. Consequently, further comparative analyses were not conducted.

Interestingly, the endorsement rates for the indicators 'has physical problems which interfere with learning' and 'has difficulty with co-workers' were exceedingly low (less than 6%) for all three groups. Also noteworthy was that three additional indicators (i.e., 'often shows up late for work', 'has poor attendance record', and 'has difficulty meeting workload requirements') generated low endorsement rates within the women offender release cohort.

Table 16: Endorsement Rates for Employment Indicators

Note: The endorsement rates for indicators 17-32 were rendered non-applicable for those cases identified as having no employment history (indicator 16). Consequently, the reported endorsement rates for these indicators were derived exclusively from cases identified as having an employment history (men: N  = 14,124; women: N = 628; Aboriginal offenders: N = 2,159).

Employment Domain: Relationship between Indicators and Rating

The relationship between each employment indicator and the overall employment dynamic factor rating was examined using a series of Pearson r correlations (see Table 17). The purpose of this analysis was to determine which indicators appeared to be driving the employment domain factor rating. As Table 17 illustrates, the majority of the indicators were significantly correlated with the employment rating for all three release cohorts.

The unemployment related indicators (e.g., ‘unemployed 50% or more', ‘unstable job history') along with ‘lacks a skill, area, trade or profession' were most strongly associated with the employment rating (i.e., r 's > .30). Interestingly, this trend was observed for men, women and Aboriginal offenders, although, only one indicator for the women offender release cohort exceeded .30 in comparison to nine and six indicators for the men and Aboriginal release cohorts, respectively.

Table 17: Correlations between Employment Indicators and Rating

* p < .001; family wise error rate ( p = .05/35 = .001).

Correlation coefficients for indicators 17-32 were rendered non-applicable for those cases identified as having no employment history (indicator 16). Consequently, the correlations for these indicators were derived exclusively from cases identified as having an employment history (men: N = 14,124; women: N  = 628; Aboriginal: N = 2,159).

Employment Domain: Correlational Analyses

The Pearson r correlations between each indicator and readmission are presented in Table 18 for men, women and Aboriginal offenders separately. Although most indicators were significantly correlated with readmission, the magnitude of r was quite small (i.e. less than .05) in some cases. In sum, unstable job history was a strong predictor of readmission for both women and men ( r 's > .20) and a moderate predictor for Aboriginal offenders ( r = .17). Moreover, all of the remaining unemployment related indicators (16-19) were moderately predictive of readmission ( r 's between .10 and .19) for men, women and Aboriginal offenders. Interestingly, the indicators ‘lacks a skill area/ trade/profession' and ‘dissatisfied with skill area /trade/profession' were also moderately predictive of readmission for all three release cohorts. Lastly, ‘concentration problems' and ‘learning difficulties' were moderately predictive of readmission among the women offender release cohort.

Table 18: Correlations between Employment Indicators and Readmission

* p < .001; familywise error rate ( p = .05/35 = .001). Correlation coefficients for indicators 17-32 were rendered non-applicable for those cases identified as having no employment history (indicator 16). Consequently, the correlations for these indicators were derived exclusively from cases identified as having an employment history (men: N = 14,124; women: N = 628; Aboriginal: N = 2,159).

Employment Domain: Regression Analyses

A series of stepwise regression analyses were conducted to allow the employment indicators to compete simultaneously for unique variance in outcome. First, three separate stepwise regression analyses were conducted for each release cohort. The model entry and exit criteria were set at p < .001 (.05/35 = .001) and, 1% of the variance accounted for'. Thus, only those indicators that were statistically significant and accounted for at least 1% of the explained variance in outcome were included in the final model.

As Table 19 indicates, a pattern of unemployment or unstable job history accounted for the majority of the variance for all three groups. Interestingly, ‘lacks grade 10' only entered the final equation for the women offender release cohort.

Supplementary regression analyses were also conducted on the sub-set of 16 indicators affected by the default system. These analyses once again revealed that unemployment-related indicators (e.g., ‘unemployed 50% or more', ‘unstable job history') accounted for the majority of explained variance in outcome across all three groups.

Table 19: Employment Domain: Regression Analyses

B = Beta weight; SE B = Standard Error of Beta; Model R² at Step 2 = .08 for the male offender release cohort; Model R² at Step 2 = .09 for the women offender release cohort; Model R² at Step 3 = .04 for the Aboriginal release cohort. No other indicators met the p < .001 and ‘1% of variance accounted for' criteria for entry into the model.

Marital/Family Domain: Indicator Endorsement Rates

Only six indicators produced endorsement rates that were exceedingly low across all three release cohorts. Further, all the indicators with low endorsement rates were concentrated in the parenting skills and intervention sub-components. For a complete overview see Table 20.

Table 20: Marital/Family Domain: Indicator Endorsement Rates

Note: The endorsement rates for indicators 13-18 were rendered non-applicable for those cases identified as having no relationship history (indicator 12). Consequently, the reported endorsement rates for these indicators were derived exclusively from cases identified as having a relationship history (men: N  = 11,843; women: N = 620; Aboriginal offenders: N = 2,085). Similarly, indicators 20-28 were rendered non-applicable for those cases with no dependents. Consequently, the reported endorsement rates for these indicators were derived exclusively from cases with dependents (men: N = 8,342; women: N = 480; Aboriginal offenders: N = 1,507).

Marital/Family Domain: Relationship between Indicators and Rating

The relationship between each marital/family indicator and the overall marital/family dynamic rating was examined using a series of Pearson r correlations (see Table 21). The purpose of this analysis was to determine which indicators appeared to be driving the marital/family rating. Interestingly, for the men offenders, only three indicators (i.e., ‘paternal relations negative as a child', ‘communication problem affect relationship', ‘perpetrator of spousal abuse‘) were strongly associated with the marital/family rating ( r 's > .30). In contrast, no indicators exceeded .30 among the Aboriginal release cohort while five indicators did so among the women offender release cohort.

Table 21: Correlations between Marital/Family Indicators and Rating

*p < .002. (family wise error rate = p = .05/31 = .002).

Correlation coefficients for indicators 13-18 were rendered non-applicable for those cases identified as having no relationship history (indicator 12). Consequently, correlations for these indicators were derived exclusively from cases identified as having a relationship history (men: N = 11,843; women: N = 620; Aboriginal offenders: N = 2,085). Similarly, indicators 20-28 were rendered non-applicable for those cases with no dependents. Consequently, correlations for these indicators were derived exclusively from cases with dependents (men: N = 8,342; women: N = 480; Aboriginal offenders: N = 1,507).

Marital/Family Domain: Correlational Analyses

The Pearson r correlations between each indicator and readmission are presented in Table 22. In sum, several indicators (e.g., childhood lacked family ties) were moderately related to readmission ( r 's between .10 and .20) across all three groups. Interestingly, only one indicator, ‘unable to handle parenting responsibilities' generated a strong predictive relationship with outcome ( r = .27), and, this strong association was only observed for the women offender release cohort.

Table 22: Correlations between Marital/Family Indicators and Readmission

* p < .002 (family wise error rate = p = .05/31 = .002).

Correlation coefficients for indicators 13-18 were rendered non-applicable for those cases identified as having no relationship history (indicator 12). Consequently, correlations for these indicators were derived exclusively from cases identified as having a relationship history (men: N = 11,843; women: N = 620; Aboriginal offenders: N = 2,085). Similarly, indicators 20-28 were rendered non-applicable for those cases with no dependents. Consequently, correlations for these indicators were derived exclusively from cases with dependents (men: N = 8,342; women: N = 480; Aboriginal offenders: N = 1,507).

Marital/Family Domain: Regression Analyses

A series of stepwise regression analyses were conducted to allow the marital/family indicators to compete simultaneously for unique variance in outcome. Three separate stepwise regression analyses were conducted for each release cohort. To be included in the final model two outcomes were required: 1) the indicator had to be significant at p < .002 (Bonferoni correction applied .05/31 = .002) and the indicator had to account for at least 1% of the variance in outcome.

As Table 23 indicates, the top three predictor variables for men were: ‘parents relationship dysfunctional during childhood', ‘family members involved in crime' and ‘currently single'. For women, three indicators also entered the final equation: ‘unable to handle parenting responsibilities', ‘has no parenting responsibilities' and ‘family members involved in crime'. Noteworthy was the particularly strong contribution made by ‘unable to handle parenting responsibilities' (partial r 2 = .03), a finding echoed previously in the univariate correlational analysis. In contrast, this indicator was completely unrelated to readmission for Aboriginal offenders and was only mildly related to outcome for men.

Only two indicators, ‘arrested for incest' and ‘childhood lacked family ties', contributed unique variance in explained outcome among the Aboriginal release cohort. It is important to note, however, that ‘arrested for incest' was negatively related to outcome as indicated by the negative Beta value. Thus, Aboriginal offenders in general who have been arrested for incest are significantly less likely to be readmitted to a federal institution, although it should be noted that a history of sexual offending is predictive of sexual recidivism among known samples of sex offenders (Hanson & Morton-Bourgon, 2004).

Supplementary regression analyses were also conducted on the sub-set of indicators affected by the default system. Interestingly, none of the six marital quality indicators entered the final model for men. ‘Perpetrator of spousal abuse' was significantly related to outcome for women offenders (partial r 2 = .016, p < .002) while ‘sexual problems affect relationship' was significantly related to outcome for Aboriginal offenders (partial r 2 = .012, p < .002). In regards to the seven parenting skills default indicators, ‘arrested for incest' was significantly and negatively related to readmission for men (partial r 2 = .016, p < .002) while ‘does not participate in activities with the child' was significantly and positively related to readmission for men (partial r 2 = .013, p  < .002). Once again, ‘unable to handle parenting responsibilities' emerged as a significant predictor for women (partial r 2 = .07, p < .002). Lastly, ‘arrested for incest' emerged as a significant and negative predictor of readmission for Aboriginal offenders.

Table 23: Marital/Family Domain: Regression Analyses

B = Beta weight; SE B = Standard Error of Beta; Model R² at Step 3 = .05 for the male offender release cohort; Model R² at Step 3 = .07 for the women offender release cohort; Model R² at Step 2 for the Aboriginal release cohort = .03. No other indicators met the p < .002 & ‘1% of variance accounted for' criteria for entry into the model.

Associates Domain: Indicator Endorsement Rates

The endorsement rates for the associate indicators are presented in Table 24. Interestingly, only two indicators exhibited endorsement rates lower than 10%. For example, only 6.9% of the women offender cohort was identified as being affiliated with a gang while only 8.8% of women were identified as having predatory relationships. Interestingly, almost 90% of the Aboriginal release cohort associated with substance abusers in comparison to 65.9% and 58.4% of the men and women offender cohorts, respectively.

Table 24: Endorsement Rates for Associates Indicators

Associates Domain: Relationship between Indicators and Domain Rating

The following three indicators were strongly related to the overall associate domain rating: ‘associates with substance abusers', ‘has many criminal acquaintances', and ‘has mostly criminal friends'. ‘Gang involvement', ‘criminogenic neighbourhood', ‘unattached to community groups', and ‘easily influenced by others' also made substantial contributions to the domain rating. The results were consistent across all three release cohorts.

Table 25: Correlations between Associates Indicators and Rating

*p < .005 (family wise error rate ( p = .05/11 = .005).

Associates Domain: Correlational Analyses

Interestingly, the same pattern of results emerged in regards to which indicators were most predictive of readmission. Once again, ‘associates with substance abusers', ‘has many criminal acquaintances', and ‘has mostly criminal friends' were strongly related to readmission for all three groups while ‘unattached to community groups' and ‘resides in a criminogenic neighbourhood' demonstrated a moderate relationship with readmission. Once again, the trends were consistent for men, women and Aboriginal offenders. See Table 26 for a complete overview.

Table 26: Correlations between Associates Indicators and Readmission

*p < .005 (family wise error rate (p = .05/11 = .005).

Associates Domain: Regression Analyses

Three separate stepwise regression analyses were conducted for each release cohort. The model entry and exit criteria were set at p < .005 (Bonferoni correction = .05/11) and ‘1% of the variance accounted for'. Thus, only those indicators that accounted for at least 1% of the explained variance in outcome and were significant at .005 were included in the final model.

As Table 27 indicates, ‘associates with substance abusers' and ‘has mostly criminal friends' were uniquely and significantly related to readmission for both men and women. Interestingly, while ‘associates with substance abusers' did not emerge as a significant predictor among the Aboriginal release cohort, ‘has mostly criminal friends' and ‘has many criminal acquaintances' did. It is possible that ‘associates with substance abusers' did not emerge as a significant predictor due to restricted variance (i.e., almost 90% of the Aboriginal release cohort associate with substance abusers). Nonetheless, the stepwise regression results confirm the univariate results that associating with criminal associates, criminal friends and known substance abusers increases the likelihood of readmission regardless of gender or Aboriginal status.

Table 27: Associates Domain: Regression Analyses  

B = Beta weight; SE B = Standard Error of Beta; Model R² at Step 2 = .08 for the male offender release cohort; Model R² at Step 2 = .11 for the women offender release cohort; Model R² at Step 2 = .08 for the Aboriginal release cohort. No other indicators met the p < .005 and ‘1% of variance accounted for' criteria for entry into the model.

Substance Abuse Domain: Indicator Endorsement Rates

Overall, the endorsement rates were high, ranging between 27.0% and 90.5%. This finding was consistent across all three release cohorts, although there was a tendency for the Aboriginal offender release cohort to evidence higher endorsement rates in comparison to the women and the men (see Table 28).

Table 28: Endorsement Rates for Substance Abuse Indicators

Note: The endorsement rates for indicators 2-13 were rendered non-applicable for those cases scored ‘no' for ‘abuses alcohol' (indicator 1). Consequently, the reported endorsement rates for these indicators were derived exclusively from cases identified as alcohol abusers (men: N = 8,617; women: N = 289; Aboriginal offenders: N = 2,250). In addition, the reported endorsement rates for indicators 15-26 were rendered non-applicable for those cases scored ‘no' for ‘abuses drugs' (indicator 14). Consequently, the reported endorsement rates for these indicators were derived from cases identified as drug abusers (men: N = 9,525; women: N = 413; Aboriginal offenders: N = 1,949).

Substance Abuse Domain: Relationship between Indicators and Domain Rating

As Table 29 illustrates, the majority of the substance abuse indicators were strongly correlated with the overall substance abuse rating (most r 's > .30). Once again, this finding remained true for all release cohorts. Only 13 indicators for the Aboriginal release cohort evidenced correlations below .30, although all but six still exceeded .20. It is likely the slightly deflated magnitude is attributable to high endorsement rates among the Aboriginal release cohort.

Table 29: Correlations between Substance Abuse Indicators and Rating

* p < .002 (family wise error rate ( p = .05/29 = .002).

Substance Abuse Domain: Correlational Analyses

As Table 30 illustrates, several indicators were either strongly or moderately correlated with readmission for all three release cohorts. For example, three indicators were strongly related to readmission among the male release cohort and six indicators were strongly related to readmission among the women offender cohort. Interestingly, no indicators exceeded .20 within the Aboriginal release cohort. Lastly, 12 indicators were moderately related to readmission for both men and women, while only six indicators were moderately related to readmission for the Aboriginal release cohort. The fact that fewer indicators within the Aboriginal release cohort were significantly related to readmission is most likely attributable to restricted variance (i.e., Aboriginal offenders typically evidence significantly higher endorsement rates on the majority of substance abuse indicators thus deflating the magnitude of the correlational results).

Table 30: Correlations between Substance Abuse Indicators and Readmission

* p < .002 (family wise error rate ( p = .05/29 = .002).

The correlation coefficients for indicators 2-13 were rendered non-applicable for those cases rated ‘no' for ‘abuses alcohol' (indicator 1). Consequently, the reported correlations for these indicators were derived exclusively from cases identified as alcohol abusers (men: N = 8,617; women: N = 289; Aboriginal offenders: N = 2,250). In addition, the reported correlation coefficients for indicators 15-26 were rendered non-applicable for those cases rated ‘no' for ‘abuses drugs' (indicator 14). Consequently, the reported correlation for these indicators were derived from cases identified as drug abusers (men: N = 9,525; women: N = 413; Aboriginal offenders: N = 1,949).

Substance Abuse Domain: Regression Analyses

Once again three separate stepwise regression analyses were conducted for each release cohort. The model entry and exit criteria were set at p < .002 (Bonferoni correction .05/29) and ‘1% of the variance accounted for'. Thus, only those indicators that met both criteria were included in the final model.

As Table 31 indicates, drug use, specifically ‘early drug use' and ‘drug sprees', were the strongest unique predictors of readmissions for all three release cohorts. Interestingly, previous substance abuse assessment and treatment were also uniquely related to readmission for both men and women but not so for the Aboriginal release cohort.

Supplementary regression analyses were also conducted on the subset of alcohol and drug default indicators. These analyses once again revealed that early drug use and drug-sprees were the strongest unique predictors of readmission for men, women and Aboriginal offenders. When the analysis was restricted to the alcohol default indicators, ‘combined alcohol and drug use' was the only indicator that emerged as a significant unique predictor across all three groups. The only additional indicator that emerged as a significant predictor during the supplementary regression analyses was ‘alcohol use interferes with employment', but this finding was restricted to the women offender cohort. Thus, it would appear that drug abuse is a relatively stronger predictor of readmission than alcohol abuse.

Table 31: Substance Abuse: Regression Analyses

B = Beta weight; SE B = Standard Error of Beta; Model R² at Step 3 = .11 for the male offender release cohort; Model R² at Step 2 = .15 for the women offender release cohort; Model R² at Step 1 = .04 for the Aboriginal release cohort. No other indicators met the p < .002 & ‘1% of variance accounted for' criteria for entry into the model.

Community Functioning Domain: Indicator Endorsement Rates

Interestingly, nine of the 21 community functioning indicators demonstrated exceedingly low endorsement rates (i.e., below 10%). Moreover, this trend was consistent across each release cohort. See Table 32 for a detailed account.

Table 32: Endorsement Rates for Community Functioning Indicators