By
R. Karl Hanson & Suzanne Wallace-Capretta
Department of the Solicitor General Canada

The views expressed are those of the author and are not necessarily those of the Department of the Solicitor General of Canada. This document is available in French. 
Ce rapport est disponible en français sous le titre: Étude de divers programmes de traitement pour hommes violents

Also available on the Solicitor General Canada’s Internet Site 
http://www.sgc.gc.ca 

Public Works and Government Services Canada
Cat. No. JS42-90/2000-E
ISBN: 0-662-28855-6

This study examined the relative effectiveness of four treatment programs for abusive men (n = 230). Outcome was assessed by new arrests for violence after an average 58 month follow-up period. There was little difference in recidivism rates across programs despite substantial differences in treatment philosophies (cognitive-behavioural, humanistic, pro-feminist, eclectic). The highest recidivism rate was observed in the program that had the weakest program implementation. In general, the batterer recidivists tended to have the same lifestyle problems associated with recidivism among general offenders (substance abuse, frequent moves, prior convictions). Batterers who failed to complete treatment were at increased risk to recidivate, but most of the effect can be attributed to high risk offenders being the most likely to drop-out.

Male violence against their female partners is a serious social problem. Victimisation surveys routinely find between 3% and 14% of women reporting assaults by their male partners during the previous year (Johnson & Sacco, 1995; Smith, 1987; Straus & Gelles, 1986). A number of policies have been implemented in order to control male battering, including public education campaigns, mandatory arrest, and court-mandated treatment. The use of court-mandated treatment is now so routine that many jurisdictions have detailed standards for male batterer treatment programs. These standards can stipulate, for example, the required number of sessions (12 to 52 weeks), treatment philosophy (e.g., feminist, cognitive-behavioural), and treatment methods (e.g., groups) (Austin & Dankwort, 1999; Dankwort & Austin, 1999).

Standards have developed, however, in the absence of strong research evidence. There is only weak evidence supporting the efficacy of any form of batterer treatment, and even less evidence that one treatment approach is superior to another. Although treatment effects have been found in some studies (e.g., Palmer, Brown & Barrera, 1992), other studies have found no significant differences between the treatment and control conditions (e.g., Dunford, 1997). A recent meta-analytic review concluded that "the effect of treatment is small at best" (Levesque, 1998, p. 29). Multi-site comparison studies have typically found similar recidivism rates for participants in diverse treatment programs (Gondolf, 1999; Saunders, 1996).

Gondolf (1999) interpreted the similarity in results across four programs as indicating that treatment is likely to be equally effective given that the programs meet minimal standards. Although the programs examined by Gondolf (1999) varied in length and the amount of adjunctive services (e.g., alcohol counselling), all the programs met state standards, used a similar treatment philosophy (cognitive-behavioural) and appeared to have been implemented with high levels of treatment integrity. The programs selected were all well-established and well-respected, being those that conducted training and supervision of new and branch programs in their respective communities. Whether the same outcomes would be observed across a more diverse range of programs remains unanswered.

The research on the treatment of male batterers can be viewed as a subset of the research on the treatment of general offenders (e.g., thieves, drunk drivers, generally violent men). Meta-analytic reviews have concluded that the most effective forms of treatment for general offenders are those that use behavioural or cognitive-behavioural techniques and address relevant treatment targets (Andrews et al., 1990; Lipsey, 1995). The research on general offenders has also found that treatment given to offenders with a high likelihood of reoffending tends to result in greater treatment effects than when treatment is provided to low risk offenders (Andrews et al., 1990; Andrews & Bonta, 1998).

The present study examined the recidivism rates of participants in four diverse treatment programs for male batterers. The programs selected were those that had an adequate sample size (100 referrals per year) and were willing to participate in the evaluation. A smaller, fifth program withdrew when it was unable to maintain the necessary number of research participants. The programs were not intended to be exemplary, but to represent the range of programs typically available. Consequently, they varied in length (12 to 25 weeks), treatment approaches (existential, cognitive-behavioural, eclectic) and treatment integrity. In three of the settings, the programs appeared to have acceptable levels of treatment integrity and were implemented in a manner consistent with their official program philosophy; in one setting, however, the need for additional staff training and supervision was recognised by both external evaluators and the program itself. Consequently, this unintended variation provided a natural experiment on the importance of treatment integrity.

Based on the research on general offenders, it was expected that the strongest effects would be found in the programs that used well-implemented cognitive-behavioural techniques. Furthermore, it was expected that there would be an interaction between risk level and treatment approach. The offenders at high risk for recidivism should benefit most from the structured cognitive-behavioural treatment, and should benefit least from unstructured group psychotherapy. Such a pattern would parallel a similar interaction found by Saunders (1996) in which the antisocial batterers benefited most from the cognitive-behavioural program, whereas the dependent batterers benefited most from the unstructured process-oriented groups.

The four programs were located in different regions of Canada, which could reasonably be expected to have different policies concerning the type of men referred for treatment. Consequently, the study attempted to control for pre-existing differences in recidivism risk. A number of risk scales for male batterers have been proposed, but none had been validated at the time the study was designed (1992). Consequently, pre-existing risk was assessed using a well-established measure of risk for general recidivism, namely, the Level of Service Inventory – Revised (LSI-R; Andrews & Bonta, 1995). The LSI-R is one of the most accurate predictors of general recidivism among general offender populations (Gendreau, Little & Goggin, 1996). It contains a number of items related to unstable, antisocial lifestyle - features that research has linked to the most severe forms of partner abuse (Hanson, Cadsky, Harris, & Lalonde, 1997). An additional aim of the present study was to examine the accuracy of the LSI-R for predicting spousal assault recidivism.

The features of the four treatment programs are presented in Table 1. The information was based on reviews of written materials, observation of treatment sessions (3-4 per site), and taped interviews with most of the treatment providers and some participants. The site visits were all conducted by the first author, except for the observation of one site, Program A, which was conducted by a contract researcher. Additional information available for Program A included tape-recorded observations of 20 consecutive treatment sessions and interviews with seven female partners of the group participants. The following information is condensed from longer program descriptions that were reviewed by supervisors at each program (complete descriptions are available upon request). The program descriptions apply only to the period of the study (1993-1995).

All the programs provided men with group treatment intended to reduce further partner abuse. At each site, the groups met for approximately 2 hours weekly. The stated treatment targets were similar across the programs (e.g., decrease attitudes tolerant of spouse assault, promote respectful relationships).

This program operated as an independent agency with the mandate to provide intervention for abusive men. Their 25-week treatment program was eclectic, drawing on techniques and principles from feminist, cognitive-behavioural, psychoanalytic and systemic forms of treatment. The actual delivery of the program was unstructured, with the therapists using the methods they considered most suited to the problems presented by the men each week.

The program was delivered by six regular staff and periodic student trainees. Most of the therapists had Master’s or Bachelor degrees in social sciences, and received on-site training. Students and less experienced therapists were paired with senior staff members. In addition to the regularly scheduled supervision, staff frequently met to discuss cases. The program also organised periodic research and treatment conferences on family violence and related topics.

The observed sessions had a respectful, cohesive tone. Although the content varied across sessions, most of the interventions focused on developing non-aggressive communications skills and promoting satisfying intimate relationships. Role plays were common. Feminist issues, such as patriarchy and negative stereotypes of women, were rarely discussed. Instead, the emphasis was on self-awareness, personal responsibility and self-control.

This program operated as part of a generic counselling agency that provided services to individuals, families and couples. The program was modelled after the psycho-educational groups developed by the Duluth program (Pence & Paymar, 1993), although they used a somewhat more flexible format. The program was relatively new and changed during the course of study. Initially, men were expected to complete two cycles of 12 sessions, but in 1994 the program consolidated into a single series of 18 sessions. The content of the program was explicitly pro-feminist, with an emphasis on attitude change and, to a lesser extent, the improvement of relationship skills.

The program was delivered by two full-time counsellors and four contract staff under the administrative supervision of a part-time program manager. There was considerable staff turnover during the course of the study. The original program manager, who had no clinical training and did not provide clinical supervision, was replaced by a trained counsellor working for the host counselling agency. Most of the treatment providers had Bachelor degrees in social services, and variable experience working with abusive men (i.e., some were highly trained, some were just beginning). The contract service providers received no ongoing clinical supervision, and rarely communicated with each other. The need for improved supervision and staff training was recognised by the host agency and implemented near the end of the time period covered in this study. Consequently, the observations should not be generalised to program functioning in subsequent years.

Observation of the groups suggested that program implementation was inconsistent during the study period (1993-95). The groups were primarily educational (i.e., men took notes and presented homework assignments). Efforts to engage the men in group process, however, were less successful. Some of the therapists had trouble guiding the discussions, which, at times, involved extended periods of self-justification and victim-blaming. Problems with therapeutic cohesion were noted in two of the three groups observed.

Since the men waited several months between intake and treatment, Program B was the only program to have significant levels of pre-program attrition (for further details see Rooney & Hanson, in press).

This male batterer treatment program operated out of the same building as a women’s shelter and both programs reported to the same executive director. Men referred from probation received a 12 week program, after which time they could become voluntary clients. Voluntary clients attended open-ended groups (potentially for several years) with the goal of long-term personality change. The program philosophy was humanistic/existential as described by Yalom (1985). Behavioural and educational methods were used by some of the therapists in the early parts of treatment, but most of the treatment focused on ongoing process groups. The groups focused on themes such as low-self esteem, "father wounds", and childhood trauma. Therapeutic methods included those drawn from Gestalt therapy (e.g., 2 chairs), experiential regression, and the interpretation of within-session behaviour (see Greenberg, Rice & Elliott, 1993). By re-experiencing pivotal childhood events within therapy, the therapists believed that their clients could then make new decisions that would liberate them from their victim stance and promote a nonabusive lifestyle.

The program was delivered by three full-time and two contract therapists under the direction of a full-time clinical supervisor. Most of the therapists had Master’s degrees in social work or psychology. The staff were enthusiastic and committed, believing that their approach was the preferred method for changing the behaviour of abusive men.

Observation of groups indicated that the sessions were conducted in a manner consistent with the program philosophy. Group cohesion was moderate to high, such that the men and group leaders appeared to be working together to end the abusive behaviour. The men frequently intervened with other group members (e.g., "Are you trying to tell me, Bill, that it was really all her fault?").

This program has been offered since 1981 by a generic counselling centre. The 14 week program was based on an explicit cognitive-behavioural model of change as exemplified by Bandura (1973) and Meichenbaum (1977). Aggression was considered a dysfunctional learning pattern and therapists aimed to teach clients the attitudes and skills necessary for a non-abusive lifestyle. The program targeted both abusive men and a small number of abusive women, who participated in the same groups as the men. The abusive women were not included in the current evaluation.

The program was delivered by 6 to 8 contract counsellors under supervision of a program manager (a psychologist). The counsellors had Bachelor or Master’s degrees in social services along with training in general counselling. Most had received training on spouse abuse through the host agency. In some cases, successful program clients had obtained formal training and become paid group leaders. Program delivery was closely monitored, with each session described in a manual.

Observations of the sessions indicated that the program was delivered as intended. The groups showed moderate to high levels of cohesion and the participants appeared to be working together to end their abusive behaviour. Each session contained a mixture of formal teaching/exercises and open group discussions. The group leaders varied in their attention to group process, but in no case did the groups lose their constructive focus or degenerate into victim-blaming.

Characteristics of the men who attended at least one treatment session are displayed in Table 2. Overall, the men were approximately 35 years old with median annual income $30,000. Between 50% and 80% had been legally married, with the fewest married men in the Program A (46%). Most of the men were not under court order to attend counselling, although the proportion of court-ordered men ranged considerably from 5% in Program D to 41% in Program B. Overall 24% of the sample had a prior conviction for assault (excluding index offence), and 54% had been convicted of any offence (including the index offence).

Recidivism risk was assessed using a self-report version of the Level of Service Inventory – Revised (Andrews & Bonta, 1995; Motiuk, Motiuk, & Bonta, 1992). The LSI-R was designed to assess the risk and needs of offenders on community supervision. Based on a social learning model of crime, the LSI-R subscales address factors such as criminal history, problems with employment/education, relationships, substance abuse and attitudes tolerant of crime. Previous research on the LSI-R has demonstrated acceptable internal consistency (alpha = .72), inter-rater reliability (r = .94) and temporal stability over three months (r = .80) (Andrews, 1982; Andrews, Kiessling, Mickus, & Robinson, 1986). It is one of the best measures of general criminal recidivism (Gendreau et al., 1996), although its relationship to assault recidivism among male batterers has yet to be examined.

Note: Standard deviations in parentheses. Values with different superscripts are different from each other using post hoc tests. Dichotomous variables were analysed using _X ² and Bonferroni post hoc test; continuous variables were analysed using ANOVA (F) and Scheffé’s post hoc test.

Following recommendations of the scale’s authors, the version of the LSI-R used in the current study was slightly adapted to focus on the problems associated with spousal assault. For example, rather than assessing attitudes tolerant of all kinds of crime, the men were specifically asked about attitudes tolerant of wife assault (e.g., "Hitting your wife or partner a few times is not something that should be against the law"). The men completed the LSI-R either at intake or in specially scheduled group sessions just prior to commencing treatment.

In the current sample, the internal consistency of the LSI-R was .90 (alpha). The overall level of risk for criminal recidivism was in the low to moderate range (0–13 is considered "Low Risk/Needs" for male inmates). Nevertheless, there was some variability across sites, with the highest risk cases coming from Program C (mean = 16) and the lowest risk cases coming from Program B (mean = 10) and Program D (mean = 12).

Recidivism information was obtained from the Royal Canadian Mounted Police (R.C.M.P.) criminal history records on February 11, 1999. These records include arrest and conviction information reported from all police departments across Canada. Although the R.C.M.P. records are the most comprehensive records available in Canada, and are the records used for official purposes, they omit some criminal history information (particularly older, summary offences). As well, official criminal history records would not be expected to include all cases of physical abuse, particularly since most of the men were not convicted for the offences that brought them into treatment. It is likely, however, that the cases involving the most serious victim injury would result is police attention. Consequently, a new arrest was considered a good indicator of serious violent recidivism (high specificity), but the lack of a new arrest does not necessarily mean that the men have been abuse free (low sensitivity).

Of the 250 men who started treatment, valid criminal history records were obtained for 114. Of the 136 without records, 116 were considered to never have had criminal records, and 20 cases (8.0%) were considered missing. Records were judged to be missing if the men admitted to prior convictions for violence or current convictions (for anything). The distribution of missing cases was similar across the four sites (p > .22). Overall, valid follow-up information was available for a combined sample of 230.

Recidivism information was based on both charges and convictions for three categories: assault, any violence, and any reoffence. "Assault" included violent offences in which an intimate partner was the most probable victim. These included all assault offences, as well as threats, criminal harassment and harassing telephone calls. Weapons offences (e.g., pointing a firearm) would also be included in this category if they were paired with one of the above offences.

"Any violent recidivism" included assault offences along with other violent offences such as assaults on police officers, assault with attempt to resist arrest, cruelty to animals, arson, robberies, and associated weapons offences. The one reoffender convicted of manslaughter was grouped in this category since the identity of the victim was unknown. The category of "any recidivism" included the above violent offences along with non-violent offences (e.g., theft) and violations of conditional release.

In order to control for variable periods at risk, recidivism probabilities were estimated using survival analysis (Allison, 1984). Survival analysis calculates the probability of recidivating for each time period given that the offender has not yet reoffended. Once offenders recidivate, they are removed from the analysis of subsequent time periods. The survival start date was the date of initial assessment, which, in most cases was a week or two before the start of treatment. The end date of the survival analysis was the date that the records were received (February 11, 1999). In two cases, the survival end date was earlier since the offenders have received long-sentences limiting their opportunities for additional reoffences. Most recidivism events, however, resulted in little or no incarceration.

Of the 230 men for whom recidivism information was available, 36 (15.7%) recidivated with a new assault offence, 39 (17.0%) with any violent offence, and 56 (24.3%) with any offence. The follow-up period ranged from 39 to 73 months (average of 57.6, SD = 7.7). Very few men recidivated with a violent offence that were clearly unrelated to partner abuse (2 cases); consequently, subsequent analyses report only the findings for any violent recidivism. Readers can assume, however, that analyses using the more restrictive criteria of assault recidivism would yield equivalent results.

Overall, the recidivism rates were similar across the four programs (see Table 3). The program with the weakness implementation, Program B, showed slightly higher recidivism rates than the other programs, but the overall differences were not significant for violent (Wilcoxon _X ² [3] = 1.22, p > .70) or general recidivism (Wilcoxon _X ² [3] = 1.37, p > .70).

As can be seen in Table 4, there were a number of individual characteristics associated with recidivism risk. In general, the recidivists tended to be young, low income, single, court-referred, have prior criminal records and relatively high LSI-R scores. The same factors predicted both violent and general recidivism, with the curious exception that the relationship between violent recidivism and prior assault convictions was not significant (r = .10, p > .11).

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

Because the risk factors would be expected to be highly intercorrelated, step-wise logistic regressions were conducted to identify a reduced set of variables providing the most unique variance. Due to missing data, 146 men were available for these analyses. The two best predictors of violent recidivism were age and LSI-R scores (_X ² [2] = 22.5, p < .001); for any recidivism, the strongest predictors were any criminal convictions and LSI-R scores (_X ² [2] = 26.1, p < .001). No other variables contributed significant variance once the above variables were considered.

Given the pre-existing differences in recidivism risk across the four treatment sites, the following analyses considered whether site differences would emerge after controlling for risk. These analyses were conducted using Cox regression (see Cox & Oakes, 1984), a version of survival analysis that can control for pre-existing subject characteristics, as well as for time at risk.

A plot of the violent recidivism survival function that controls for pre-existing risk factors is presented in Figure 1. The lines in the figure are smoothed to conform to the proportional hazard assumption required of Cox regression; in the unadjusted, raw data, the survival curves overlapped during the first 2-3 years. The specific contrast between Program B and the other programs approached significance (Wald = 3.50, df = 1, p = .06). The overall site differences in violent recidivism remained non-significant (Wald = 3.78, df = 3, p = .29). Similar results were found using any recidivism as the dependent variable and prior criminal convictions and LSI-R scores as covariates. The contrast between Program B and the other programs approached significance (Wald = 2.84, df = 1, p = .09) and the overall between site comparison was non-significant (Wald = 4.00, df = 3, p = .26).

Contrary to expectation, the well-implemented cognitive-behavioural program, Program D, showed no superiority over the other programs (Walds < 1 for both violent and any recidivism). As well, the predicted interaction between program structure and risk was not found. Offenders with LSI-R scores of 13 or less were considered low risk, whereas those with scores of 14 or more were considered high risk. Scores in the 14 to 23 range are still considered relatively low risk in comparison to general criminal samples (Andrews & Bonta, 1995), but this lenient cut-off was selected to retain adequate cell sizes. Analyses using alternate cut-offs did not change the findings.

The high risk offenders showed similar recidivism rates in the cognitive-behavioural program (45% any, 30% violent, n = 20, Program D) as in the two programs that offered unstructured group psychotherapy (40% any, 29% violent, n = 58, Programs A & C). The low risk offenders had very similar recidivism rates in both the structured (11% any, 7% violent, n = 45, Program D) and unstructured programs (9% any, 8% violent, n = 66, Programs A & C). Any potential trends in the data were, if anything, opposite to the pattern predicted. Program B was excluded from this analysis since problems with program implementation made it difficult to determine what type of treatment was actually delivered.

Of the 230 who started treatment, 138 completed and 92 dropped-out. The rates of within-program attrition were similar across the sites (_X ² [3] = .58, p > .90). The men who failed to complete treatment showed higher recidivism rates for violence (26.1% versus 10.9%; Wilcoxon _X ² [1] = 8.65, p = .003) and any reoffending (32.6% versus 18.5%; Wilcoxon _X ² [1] = 5.68, p = .02) than did treatment completers. The effects of attrition were substantially reduced, however, after controlling for pre-existing differences in risk (high risk men were most likely to drop-out, see Rooney & Hanson, in press). When controlling for age and LSI-R scores, the effect of attrition on violent recidivism reduced to Wald = 2.92 (df = 1), p = .09; controlling for prior convictions and LSI-R scores reduced the effect of attrition on any recidivism to Wald = 2.48 (df = 1), p = .12.

This study examined the relative effectiveness of four different treatment programs for abusive men. Despite substantial differences in program philosophy and implementation, there were relatively few differences in the recidivism rates of program participants. Men who received unstructured, humanistic group psychotherapy did as well as the men who received structured, cognitive-behavioural interventions. The only trend suggesting group differences seemed more related to program integrity than program content: the men most likely to recidivate were those attending a program with noticeable weaknesses in staff training and supervision (although the effect was only marginally significant).

When no differences between treatment approaches are found, it is difficult to tell whether the programs are equally effective or equally poor. The findings could be interpreted as further support for the Dodo Bird hypothesis, i.e., "everybody wins and all must have prizes" – see Wampold, (1997). Alternately, the lack of group differences can be considered evidence that we have yet to discover what really works with abusive men. Any interpretation of the present results is difficult given the lack of a untreated comparison group. A further complication is that the programs differed in any number of ways that could potentially influence their effectiveness. Not only did they differ in program philosophy and treatment integrity, but they also differed in length, funding, and support from related community agencies. As well, regional variation in criminal justice processing of wife assault cases could obscure any real differences in treatment effectiveness. The relatively small sample sizes (low statistical power) also makes it difficult to make statements about the lack of group differences.

Nevertheless, the lack of superiority for cognitive-behavioural treatment for the high risk men runs counter to the conclusions of meta-analyses on the effectiveness of treatment for general criminal offenders (Andrews et al., 1990; Andrews & Bonta, 1998). Several interpretations are possible. Firstly, there may have been insufficient number of high risk offenders to adequately test the hypothesis; almost all the men in the present study were relatively low risk compared to general criminal samples. Another possibility is that the "unstructured" therapies may have, nevertheless, contained the necessary elements of effective cognitive-behavioural intervention (e.g., attitude change and skill building). For example, direct observation of Program A’s groups showed that role plays were frequently used, even though such skill building was a minor part of their overall treatment philosophy.

It is also possible that abusive men should be considered a special type of offender that responds to unique types of treatment. The present findings, however, argue against such an hypothesis because there were strong similarities between the abusive men and other offender groups. In particular, the factors that predict recidivism among general criminal populations (see Gendreau et al., 1996) also predicted violent and general recidivism among the abusive men in this study. The recidivists tended to be young, single, have low incomes, and prior criminal convictions. The life problems associated with recidivism among general criminals (e.g., poor housing, substance abuse, family conflicts) also predicted violent recidivism among abusive men. In fact, the LSI-R (a combined measure of criminogenic risk/needs) was the single strongest predictor of recidivism.

Critics could argue that any apparent similarity between abusive men and general criminals is an artefact of using a common outcome criteria – official records of violent recidivism. Given the high rates of undetected abuse, the risk factors could be primarily related to the probability of criminal justice processing rather than to any relationship with ongoing abuse. The resolution of such debate requires data that was not available in the current study, namely, comparisons between the predictors of non-criminal justice indicators of abusive behaviour (e.g., partner reports) and the predictors of recidivism as indicated by official criminal records. Nevertheless, it seems likely that most of the detected offences were related to partner abuse because there were almost no violent offences that were clearly unrelated (two out of 39).

The present study found clear evidence that those offenders who failed to complete treatment were higher risk to recidivate than treatment completers. Among several potential explanations, the simplest explanation is that the highest risk men are those most likely to drop-out of treatment (see Cadsky, Hanson, Crawford & Lalonde, 1996; Rooney & Hanson, in press). In the current study, the effects of attrition on recidivism were no longer significant after controlling for pre-existing risk factors. Consequently, researchers and policy-makers should be extremely cautious when using completer/drop-out comparisons to judge the effectiveness of treatment programs. Whatever the explanation for the effect of drop-out, this finding has clear practical implications: service providers (and potential victims) should be particularly concerned about men who fail to complete treatment.

After controlling for risk levels, the adjusted violent recidivism rate of the men from Program B was approximately 21% compared to an adjusted rate of 12% for the remaining programs. Although this difference did not meet the conventional .05 test of statistical significance, it is consistent with the small treatment effects found in other studies. In Levesque’s (1998) meta-analysis, the average recidivism rate using official records was 14.3% for the treatment groups, which was significantly lower than the 21.8% recidivism rate found for the comparison groups (based on 11 studies). The effect size in the current study (Cohen’s h = .24) was, in fact, slightly higher than the average effect size of h = .19 found by Levesque (1998). Critics of treatment could argue that none of the programs were effective and that the poorly implemented program actually made men worse. Given the small positive treatment effects found in other studies, however, another plausible interpretation of the current results is that it provides further evidence that adequately implemented treatment can reduce the recidivism rates of abusive men. The essential elements of effective intervention, however, remain unknown.

Allison, P. D. (1984). Event history analysis: Regression for longitudinal event data. Beverly Hills, CA: Sage.

Andrews, D. A. (1982). The Level of Supervision Inventory (LSI): The first follow-up. Toronto, ON: Ministry of Correctional Services of Ontario.

Andrews, D. A., & Bonta, J. (1995). LSI-R: The Level of Service Inventory -Revised. Toronto, ON: Multi-Health.

Andrews, D. A., & Bonta, J. (1998). The psychology of criminal conduct (2^nd edition). Cincinnati, OH: Anderson.

Andrews, D. A., Kiessling, J. J., Mickus, S. G., & Robinson, D. (1986). The construct validity of interview-based risk assessment in corrections. Canadian Journal of Behavioural Sciences, 18, 460-470.

Andrews, D. A., Zinger, I., Hoge, R. D., Bonta, J., Gendreau, P., & Cullen, F. T. (1990). Does correctional treatment work? A psychologically informed meta-analysis. Criminology, 28, 369-404.

Austin, J. B., & Dankwort, J. (1999). Standards for batterer programs: A review and analysis. Journal of Interpersonal Violence, 14, 152-168.

Bandura, A. (1973). Aggression: A social learning analysis. Englewook Cliffs, NJ: Prentice-Hall.

Cadsky, O., Hanson, R. K., Crawford, M., & Lalonde, C. (1996). Attrition from a male batterer treatment program: Client-treatment congruence and lifestyle instability. Violence and Victims, 11, 51-64.

Cox, D. R., & Oakes, D. (1984). Analysis of survival data. London, UK: Chapman & Hall.

Dankwort, J., & Austin, J. B. (1999). Standards for batterer programs in Canada: A history and review. Canadian Journal of Community Mental Health, 18, 19-38.

Dunford, F. W. (1997, June). The research design and preliminary outcome findings of the San Diego Navy Experiment. Presentation at the 5^th International Family Violence Research Conference, Durham, New Hampshire.

Gendreau, P., Little, T., & Goggin, C. (1996). A meta-analysis of the predictors of adult offender recidivism: What works! Criminology, 34, 575-607.

Gondolf, E. W. (1999). A comparison of four batterer interventions systems: Do court referral, program length, and service matter? Journal of Interpersonal Violence, 14, 41-61.

Greenberg, L. S., Rice, L. N., & Elliott, R. (1993). Facilitating emotional change: The moment-by-moment process. New York: Guilford Press.

Hanson, R. K., Cadsky, O., Harris, A., & Lalonde, C. (1997). Correlates of battering among 997 men: Family history, adjustment and attitudinal differences. Violence and Victims, 12, 191-208.

Johnson, H., & Sacco, V. F. (1995). Researching violence against women: Statistics Canada's national survey. Canadian Journal of Criminology, 37, 281-304.

Levesque, D. A. (1998). Violence desistance among battering men: Existing interventions and the application of the transtheoretical model of change. Unpublished doctoral dissertation. Department of Psychology, University of Rhode Island.

Lipsey, M. (1995). What do we learn from 400 studies on the effectiveness of treatment with juvenile delinquents? In J. McGuire (Ed.), What works: Reducing reoffending (pp. 63-78). Chichester, England: Wiley.

Meichenbaum, D. H. (1977). Cognitive-behavior modification. New York: Plenum.

Motiuk, M. S., Motiuk, L. L., & Bonta, J. (1992). A comparison of self-report and interview-based inventories in offender classification. Criminal Justice and Behavior, 19, 143-59.

Palmer, S. E., Brown, R. A., & Barrera, M. E. (1992). Group treatment program for abusive husbands: Longterm evaluation. American Journal of Orthopsychiatry, 62, 276-283.

Pence, E., & Paymar, M. (1993). Education groups for men who batter: The Duluth model. New York: Springer.

Rooney, J., & Hanson, R. K. (in press). Predicting attrition from treatment programs for abusive men. Journal of Family Violence.

Saunders, D. G. (1996). Feminist-cognitive-behavioural and process-psychodynamic treatments for men who batter: Interaction of abuser traits and treatment models. Violence and Victims, 11, 393-413.

Smith, M. D. (1987). The incidence and prevalence of women abuse in Toronto. Violence and Victims, 2, 173-187.

Straus, M. A., & Gelles, R. J. (1986). Societal change and change in family violence rates from 1975 to 1985 as revealed by two national surveys. Journal of Marriage and the Family, 48, 465-479.

Wampold, B. E., Mondin, G. W., Moody, M., Stich, F., Benson, K., & Ahn, H. (1997). A meta-analysis of outcome studies comparing bona fide psychotherapies: Empiricially, "all must have prizes." Psychological Bulletin, 122, 203-215.

Yalom, I. D. (1985). The theory and practice of group psychotherapy (3^rd ed.). New York: Basic Books.