Criminal Justice and Behavior 2010 Jewell 1086 113
-
Upload
corina-ica -
Category
Documents
-
view
213 -
download
0
description
Transcript of Criminal Justice and Behavior 2010 Jewell 1086 113
http://cjb.sagepub.com/Behavior
Criminal Justice and
http://cjb.sagepub.com/content/37/10/1086The online version of this article can be found at:
DOI: 10.1177/0093854810376815
2010 37: 1086Criminal Justice and BehaviorLisa M. Jewell and J. Stephen Wormith
Targeting Male Batterers: A Meta-AnalysisVariables Associated With Attrition From Domestic Violence Treatment Programs
Published by:
http://www.sagepublications.com
On behalf of:
International Association for Correctional and Forensic Psychology
can be found at:Criminal Justice and BehaviorAdditional services and information for
http://cjb.sagepub.com/cgi/alertsEmail Alerts:
http://cjb.sagepub.com/subscriptionsSubscriptions:
http://www.sagepub.com/journalsReprints.navReprints:
http://www.sagepub.com/journalsPermissions.navPermissions:
http://cjb.sagepub.com/content/37/10/1086.refs.htmlCitations:
What is This?
- Sep 10, 2010Version of Record >>
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
1086
CRIMINAL JUSTICE AND BEHAVIOR, Vol. XX, No. X, Month 2007 1086-XXXDOI: © 2007 American Association for Correctional and Forensic Psychology
1086
CRIMINAL JUSTICE AND BEHAVIOR, Vol. 37 No. 10, October 2010 1086-1113DOI: 10.1177/0093854810376815© 2010 International Association for Correctional and Forensic Psychology
AUTHORS’ NOTE: This research was supported by a Canada Graduate Scholarship awarded to the first author from the Social Sciences and Humanities Research Council of Canada. We would like to thank Shannon Durand for her assistance with this study. Correspondence should be addressed to Lisa M. Jewell, Department of Psychology, University of Saskatchewan, 9 Campus Drive, Arts Building, Saskatoon, Saskatchewan, Canada, S7N 5A5; phone: 306-966-1773; e-mail: [email protected].
VARIABLES ASSOCIATED WITH ATTRITION FROM DOMESTIC VIOLENCE TREATMENT PROGRAMS TARGETING MALE BATTERERS
A Meta-Analysis
LISA M. JEWELLJ. STEPHEN WORMITHUniversity of Saskatchewan
Attrition from domestic violence treatment programs is of concern to correctional treatment providers because batterers who do not complete treatment are at higher risk for recidivism. This meta-analysis was conducted to determine the extent to which various demographic, violence-related, and intrapersonal variables predict attrition from domestic violence treatment programs for male batterers. A total of 30 studies that focused on in-program attrition and were published in English between 1985 and 2010 were included in the meta-analysis. Several variables distinguished treatment completers from dropouts, including employment, age, income, education, marital status, race, referral source, previous domestic violence offenses, criminal history, and alcohol and drug use. Furthermore, the theoretical orientation of the treatment program (i.e., feminist psychoeducational vs. cognitive-behavioral therapy) was found to be an important moderating variable. Findings suggest that the variables that predict attrition tend to be the same variables that predict recidivism and are discussed in relation to the responsivity principle.
Keywords: domestic violence; male batterers; treatment; attrition; meta-analysis; responsivity
Domestic violence (defined as physical, sexual, and psychological abuse directed toward one partner in a romantic relationship by the other) affects thousands of
Canadians and Americans each year (Davidson et al., 2001; Plichta, 2004). In 2004, approximately 7% of Canadian women and 6% of Canadian men living in common-law or marital relationships had been the victims of domestic violence at some point in the previ-ous 5 years (Statistics Canada, 2006). Comparably, the Home Office reported that 6% of women and 4% of men living in England and Wales had been the victims of domestic vio-lence in 2008-2009, with 38% of these individuals experiencing victimization more than once during the 1-year period (Walker, Flatley, Kershaw, & Moon, 2009). In the United States, the National Violence Against Women Survey (Tjaden & Thoennes, 2000) investi-gated the lifetime prevalence of domestic violence and found that 22.1% and 7.4% of American women and men, respectively, had been abused by an intimate partner at some
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
Jewell, Wormith / DOMESTIC VIOLENCE TREATMENT ATTRITION 1087
point in their lives. In fact, domestic violence is a recognized worldwide problem; however, cultural norms surrounding abuse, discipline, and chastisement make it difficult to identify the prevalence and nature of abusive relationships in many cultures (Garcia-Moreno, Jansen, Ellsberg, Heise, & Watts, 2006).
Not surprisingly, numerous personal and societal costs are associated with domestic violence, including physical injuries, psychological distress, death, and the provision of health care, counseling, shelter, and criminal justice services (Greaves, Hankivsky, & Kingston-Riechers, 1995; Plichta, 2004). Offender treatment programs constitute one of the services offered by the criminal justice sector to address domestic violence and are designed to prevent individuals who have engaged in this form of violence from doing so in the future. The majority of treatment programs target male offenders because, even though intimate partner violence may be directed toward either female or male victims, women tend to suffer more severe injuries than men (Babcock, Green, & Robie, 2004; Tjaden & Thoennes, 2000). For instance, women are more likely to be the victims of mul-tiple assaults, suffer from the most severe forms of partner abuse, and experience negative emotional consequences (Statistics Canada, 2006). Consequently, many researchers have characterized domestic violence as a gendered form of violence (Anderson, 2005; Gill, 2006). In light of this perspective, the remainder of this article focuses explicitly on male batterer treatment programs.
Numerous studies have been carried out to explore the relationship between domestic violence treatment completion and recidivism. A recent meta-analysis conducted by Babcock et al. (2004) suggested that male batterer treatment programs were associated with a “small” reduction in domestic violence recidivism (d = .18). However, a limitation noted by Babcock et al. (2004) in interpreting their results is that many of the studies included in their meta-analysis (k = 22) were affected by program attrition or dropout, making it dif-ficult to estimate the effectiveness of treatment accurately. In general, domestic violence treatment programs are plagued by high attrition rates, with anywhere from 15% to 58% of individuals failing to complete treatment (Bennett, Stoops, Call, & Flett, 2007; Rondeau, Brodeur, Brochu, & Lemire, 2001). Wormith and Olver (2002) suggested that the client characteristics that cause offenders to be more at risk for failing to complete treatment may be the same characteristics that put offenders at risk for recidivism. In line with Wormith and Olver’s conjecture, numerous studies have reported that individuals who fail to com-plete domestic violence treatment are more likely than individuals who completed treat-ment to continue to abuse their partners (Babcock & Steiner, 1999; Bennett et al., 2007; Gordon & Moriarty, 2003). Furthermore, women are more likely to remain with men who are enrolled in domestic violence treatment programs (Gondolf, 1988). Thus, it is important to understand the offender characteristics associated with attrition to identify individuals who may be likely to drop out of treatment and to develop strategies to reduce treatment dropout.
Considerable research has been conducted to identify factors that predict domestic vio-lence treatment attrition. Typically, studies have focused on three broad categories of vari-ables: demographic variables, violence-related factors, and intrapersonal characteristics (Daly & Pelowski, 2000). The findings have been mixed with respect to all three of these categories. Thus, we conducted a meta-analysis to determine the extent to which various variables are associated with treatment completion and dropout.
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
1088 CRIMINAL JUSTICE AND BEHAVIOR
DEMOGRAPHIC VARIABLES
Demographic variables typically reflect “fixed” characteristics of an individual and are the most common variables that have been examined in relation to domestic violence treat-ment attrition (Daly & Pelowski, 2000). In fact, researchers exploring attrition are more likely to compare treatment completers and treatment dropouts on demographic variables than either violence-related or intrapersonal variables. Age, education, employment status, income, ethnicity, and marital status are among the variables most frequently investigated. Research has consistently shown that batterers who are older, employed, married, or Caucasian; who earn higher incomes; or who have more education tend to be more likely to complete treatment than batterers who are younger, unemployed, single, or a minority-group member; who earn lower incomes; or who have less education (Daly & Pelowski, 2000). However, the relative strength of each variable in differentiating between treatment completers and dropouts remains unknown. A meta-analysis will be helpful in determining which variables are most strongly related to attrition.
VIOLENCE-RELATED VARIABLES
The second category of variables that is associated with domestic violence treatment attrition relates to batterers’ experiences with violence, and here the evidence has been inconsistent. Frequently, researchers have assessed batterers’ criminal history in terms of any prior arrests or convictions or any domestic-violence-specific arrests or convictions (Cadsky, Hanson, Crawford, & Lalonde, 1996; Dalton, 2001). Treatment dropouts tend to have more prior arrests and/or convictions than treatment completers (Bowen & Gilchrist, 2006; Rondeau et al., 2001). However, some studies have found treatment completers to have more prior arrests and/or convictions (Dalton, 2001; Daly, Power, & Gondolf, 2001). Furthermore, some studies exploring batterers’ prior domestic violence arrests have found first-time offenders to be most likely to complete treatment (Babcock & Steiner, 1999; Bennett et al., 2007), whereas others have found offenders with greater histories of domes-tic violence to be more likely to complete treatment (Dalton, 2001). Research also has explored whether offenders who are court mandated to treatment are more likely to com-plete treatment than men who are not mandated, but findings are mixed with respect to whether treatment referral source is related to attrition (Dalton, 2001; Faulkner, Cogan, Nolder, & Shooter, 1991).
In addition, violence-related factors include batterers’ personal experiences with being victims of abuse and witnessing violence in their families of origin. Although batterers’ personal experiences of abuse have been examined infrequently, the evidence suggests that experiencing abuse is related to treatment dropout (Cadsky et al., 1996). The witnessing of violence between one’s parents has been investigated more often, and some research indi-cates that batterers who have witnessed violence as a child or adolescent also tend to be less likely to complete treatment (Cadsky et al., 1996; Chang & Saunders, 2002). However, other studies do not support this conclusion (Bowen & Gilchrist, 2006; Daly et al., 2001). In addition, the severity of abuse that men direct toward their partners has been associated with treatment attrition. The few studies exploring the severity of violence have generally found that men who inflict more severe forms of physical or psychological abuse are more likely to drop out of treatment than men who use less severe forms of violence (Carney,
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
Jewell, Wormith / DOMESTIC VIOLENCE TREATMENT ATTRITION 1089
Buttell, & Muldoon, 2006; Rooney & Hanson, 2001). Given the inconsistent results associ-ated with many of the violence-related variables, a meta-analysis will provide a more com-prehensive perspective on their ability to predict attrition by statistically establishing the overall direction and magnitude of these effects (Durlak & Lipsey, 1991).
INTRAPERSONAL CHARACTERISTICS
The last category of variables related to domestic violence treatment attrition consists of intrapersonal characteristics. The most common characteristics are alcohol and drug use, both of which have consistently been correlated with dropping out of treatment (Bowen & Gilchrist, 2006; Dalton, 2001; Faulkner et al., 1991; Hamberger, Lohr, & Gottlieb, 2000). However, as with the demographic variables, the relative strength of these variables for predicting treatment attrition is unknown.
A few studies have examined anger as a correlate of treatment completion. These studies tend to find that higher levels of anger are associated with treatment completion (Bowen & Gilchrist, 2006; Cadsky et al., 1996; Chang & Saunders, 2002). The relationship between depression and domestic violence treatment attrition also has been assessed, but these studies have not generally demonstrated a significant relationship (Faulkner et al., 1991; Saunders & Parker, 1989). Other intrapersonal variables known to be associated with treat-ment attrition in general, such as risk level, motivation, psychopathology, and therapeutic alliance, have not been studied frequently enough with male batterers to allow for com-mentary about their impact on treatment dropout (Daly & Pelowski, 2000; Wormith & Olver, 2002).
RELATIONSHIP BETWEEN PREDICTORS OF ATTRITION AND RECIDIVISM
It appears that many of the variables thought to be predictive of attrition from domestic violence treatment programs also are known predictors of domestic violence recidivism. For instance, factors such as younger age, lower socioeconomic status, unemployment, educa-tion, being unmarried, substance abuse, antisocial personality and conduct, violence in one’s family of origin, history of relationship conflict or abuse, prior severity of assaults, and previous arrests for partner abuse have all been identified as predictors of partner abuse recidivism (Babcock & Steiner, 1999; Cattaneo & Goodman, 2003; Hilton & Harris, 2005; Hilton, Harris, & Rice, 2001). Many of these variables are not unique to the prediction of domestic violence recidivism as they also predict general recidivism. For instance, a meta-analysis conducted by Gendreau, Little, and Goggin (1996) found that some of the most reliable predictors of male adult offender recidivism included age, race, criminal history, family factors, criminogenic need factors (i.e., antisocial personality, companions, interper-sonal conflict, substance abuse), and social achievement. Furthermore, it has been found that general offender risk assessments can be used to sufficiently predict partner abuse recidi-vism (Hanson & Wallace-Capretta, 2004; Hilton et al., 2001). Thus, it may be possible to predict domestic violence treatment attrition and recidivism using the same variables.
ADDITIONAL ISSUES RELATED TO ATTRITION TO BE CONSIDERED
A number of general issues must be considered when thinking about domestic violence treatment dropout beyond the types of variables that have been associated with attrition.
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
1090 CRIMINAL JUSTICE AND BEHAVIOR
First, it is necessary to consider the type of attrition that is reported in individual studies. Specifically, attrition may occur at three different points in time: (a) after referral and before assessment (i.e., postreferral), (b) after assessment but before beginning treatment (i.e., postassessment), and (c) during treatment (i.e., in program; Gondolf & Foster, 1991).
Researchers studying domestic violence treatment programs have only rarely looked at attrition that occurs between referral and assessment (see Gondolf & Foster, 1991, as a notable exception). The majority of research focuses exclusively on in-program attrition (Carney et al., 2006; DeMaris, 1989; Faulkner et al., 1991; Rondeau et al., 2001), whereas a handful of studies examine the differences between treatment completers and batterers who drop out between assessment and treatment (Cadsky et al., 1996; Chang & Saunders, 2002; Hamberger et al., 2000; Rooney & Hanson, 2001; Saunders & Parker, 1989). Unfortunately, some researchers have grouped individuals who have dropped out of treat-ment after assessment with those who dropped out during treatment, which obscures any unique differences between these two types of dropouts (Babcock & Steiner, 1999; Bowen & Gilchrist, 2007; Daly et al., 2001). Others have failed to identify the type of attrition with which they were concerned (Beldin, 2008; Bennett et al., 2007; Taft, Murphy, Elliott, & Keaser, 2001). Therefore, the current study focused specifically on in-program attrition because this type of attrition is most often reported and allowed for a sufficient number of studies to be included in the meta-analysis. Studies that collapsed postassessment and in-program attrition and those that did not specify the nature of attrition were included in the meta-analysis, as their samples likely reflect the more common in-program attrition. However, the type of attrition assessed in each study was coded as a moderator variable, and additional analyses were conducted to determine whether unclear definitions of attri-tion affected the results of the meta-analysis. It should be noted that treatment attrition also may be imposed on participants who are asked to terminate a program by staff, returned to jail, or transferred to another facility. However, this kind of attrition has rarely been discussed in the domestic violence treatment literature and is not addressed in this article.
A second issue that must be considered with respect to attrition is the definition of treat-ment completion employed across studies. Currently, there is no agreed-on definition of attrition. Some studies used definitions that were very demanding (e.g., batterers who missed one or more treatment sessions were considered dropouts; Beldin, 2008), whereas others were much more forgiving (e.g., batterers who attended fewer than 67% of treatment ses-sions were considered treatment dropouts; DeHart, Kennerly, Burke, & Follingstad, 1999). In some cases, attrition also has been recorded as a continuous variable (e.g., Rooney & Hanson, 2001). The lack of standardized definitions across studies likely contributes to the wide range of attrition rates observed in the literature. In addition, inconsistent conceptu-alizations of attrition may limit the extent to which it is possible to draw conclusions about the variables associated with treatment dropout and may result in lower effect sizes in the meta-analysis.
Finally, in considering the results of the meta-analysis, it is necessary to keep in mind the characteristics of the treatment programs under investigation. In addition to employing different definitions of treatment attrition, there also are differences across interventions in terms of program length and treatment modality. Batterer intervention programs may range from 8 to 36 weeks of group treatment (Faulkner et al., 1991; Gondolf & Foster, 1991). Consequently, attrition may be related to program length because there is less opportunity for offenders attending 8-week programs to drop out than there is for offenders who attend
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
Jewell, Wormith / DOMESTIC VIOLENCE TREATMENT ATTRITION 1091
36-week programs. Therefore, program length was an important variable to consider in the meta-analysis.
With respect to treatment modality, there are two commonly accepted treatment modali-ties employed in batterer intervention programs: feminist psychoeducational treatments and cognitive-behavioral therapy. Feminist psychoeducational treatments tend to focus on the belief that domestic violence is a direct consequence of living in a patriarchal society that sanctions men’s use of power and control over women (Babcock et al., 2004). Consequently, men are challenged to change their attitudes and behaviors toward women that stem from their assumed positions of male privilege. In contrast, cognitive-behavior therapy is based on the notion that violence is a learned response that allows the batterer to accomplish a specific goal (Babcock et al., 2004). Thus, to change the repertoire of behaviors that batter-ers can draw on to lead nonviolent lives, cognitive-behavior therapy often involves teaching them communication, assertiveness, and social skills, as well as anger management tech-niques. In reality, most treatment programs integrate elements of the feminist psychoeduca-tional model with the principles of cognitive-behavioral therapy, which can make it difficult to distinguish these treatment modalities from each other. A recent meta-analysis conducted by Babcock et al. (2004) suggested there was no difference in terms of either modality’s effectiveness in reducing domestic violence recidivism. However, it is possible that treatment attrition may be related to the type of programming offered to the batterers.
On the basis of the literature reviewed, it seems that several variables may potentially be related to domestic violence treatment attrition. However, the many nonsignificant and inconsistent findings make it difficult to establish which variables are the best predictors of treatment attrition (Cadsky et al., 1996). Thus, this meta-analysis explored the extent to which various demographic, violence-related, and intrapersonal variables were able to dis-tinguish between treatment completers and dropouts. It was expected that demographic variables (i.e., being older, employed, married, or White) would be positively associated with treatment completion, whereas violence-related variables (i.e., having a history of prior domestic violence offenses or other crimes, being the victim of abuse, witnessing family violence, or being court-referred) and intrapersonal variables (i.e., using alcohol and drugs or being depressed) would be negatively related to treatment completion.
METHOD
STRATEGY FOR SEARCHING THE LITERATURE
A literature search was conducted using PsycINFO, Sociological Abstracts, and Criminal Justice Abstracts to identify studies examining correlates of domestic violence treatment attrition. The keywords domestic violence, intimate partner violence, partner abuse, and batterer were used and cross-referenced with the words treatment, intervention, attrition, and dropout. In addition, the reference sections from two review articles (Babcock et al., 2004; Daly & Pelowski, 2000) were examined for potential articles, as was the reference section of each article included in the meta-analysis. Furthermore, Social Science Citations was used to identify recent studies that cite the articles included in the meta-analysis. Finally, to avoid issues associated with publication bias, several researchers who have conducted research on domestic violence treatment attrition were contacted as a means of
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
1092 CRIMINAL JUSTICE AND BEHAVIOR
identifying technical reports, conference presentations, and other studies published in the gray literature. On the basis of this search strategy, 39 published studies from 36 published journal articles and 2 unpublished dissertations were found (1 journal article contained 2 separate studies).
INCLUSION CRITERIA
Studies were required to meet numerous criteria to be included in the meta-analysis. First, the treatment program had to be directed primarily toward male batterers, and men had to compose the great majority (at least 90%) of the study’s participants. This resulted in one study being excluded (Tollefson, Gross, & Lundahl, 2008). In addition, three studies that treated male and female couples as the unit of analysis were also excluded (Brown & O’Leary, 2000; Brown, O’Leary, & Feldbau, 1997; Heyman, Brown, Feldbau-Kohn, & O’Leary, 1999). Second, in-program attrition had to serve as the study’s primary focus. Studies that collapsed postassessment and in-program attrition were included in the meta-analysis because individuals who dropped out postassessment generally composed a minority of these samples. Similarly, studies that did not explicitly specify the nature of attrition assessed tended to imply they were concerned with in-program attrition and also were included. However, studies investigating only postreferral or postassessment attrition were excluded (i.e., Gondolf & Forster, 1991). Third, studies had to employ a quasiexperi-mental or experimental design comparing treatment dropouts to treatment completers. However, correlational studies were included if they specifically related variables to treat-ment attrition and attrition was defined as a continuous variable. Fourth, studies had to be published between 1985 and April 2010 and reported in English. Fifth, studies had to assess treatment programs targeted toward men who had directed violence toward a female inti-mate partner. Sixth, the sample had to comprise adult participants. Finally, it had to be possible to calculate an effect size statistic (r) for relevant variables from data either included in the article or provided directly from authors. Four studies (DeHart et al., 1999; Gondolf, 2008; Huss & Ralston, 2008; Stalans & Seng, 2007) were excluded on the basis of this criterion. Ultimately, 30 separate studies (from 27 published articles and 2 unpub-lished dissertations) were included in the meta-analysis.
CODING PROCEDURES
Each of the remaining 30 studies was coded for variables related to study context, treat-ment and methodological characteristics, attrition, and sample characteristics. With respect to study context, the year of publication, country, type of publication (i.e., peer-reviewed journal article, unpublished dissertation, etc.), and first author’s professional affiliation (i.e., social work, psychology, etc.) were recorded. The methodological quality of the study was rated using the scientific method score developed by Sherman et al. (1997), where studies employing correlational designs are considered to be weak, quasi-experiments are considered to be moderate designs, and randomized experiments are considered to be strong designs. Various treatment characteristics were also coded, such as the theoretical orienta-tion of the treatment program (i.e., feminist psychoeducational or cognitive-behavioral), loca-tion of the treatment (i.e., institutional or community based), program format (i.e., group or individual therapy), referral source (i.e., court mandated or self-referral), and program
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
Jewell, Wormith / DOMESTIC VIOLENCE TREATMENT ATTRITION 1093
length (defined as the number of weeks in group therapy). In addition, each study’s defini-tion of treatment attrition was recorded. Finally, overall sample characteristics were coded for each study, including the overall sample size compared; the proportion of the sample that was male, employed, married, Caucasian, and a treatment completer or dropout; and the mean age and income of the sample.
To ensure the reliability of the coding, both the first author and a trained undergraduate research assistant coded the studies independently. To assess the reliability of the coding process, the percentage agreement was determined for each of the variables coded, as was the intraclass correlation coefficient (ICC). With the exception of three variables (i.e., over-all treatment completion rate, overall percentage employed, and overall percentage married), there was at least 80% agreement between raters on all variables coded. Other researchers in the field (Dowden, Antonowicz, & Andrews, 2003; Parhar, Wormith, Derkzen, & Beauregard, 2008) have used 80% agreement among raters as an indicator of satisfactory interrater reliability. Furthermore, all variables (with the exception of theoretical orienta-tion and program length) obtained absolute ICCs ranging from .90 to 1.0. The absolute ICCs for theoretical orientation and program length were .83 and .79, respectively. In gen-eral, the raters consistently rated each variable. Slight discrepancies were observed in the calculation of overall treatment completion rates, percentage employed, and percentage married because of the misleading way in which these statistics were presented in the original articles. For example, some studies used a larger sample to calculate the overall demographic characteristics than they did to carry out analyses related to treatment attri-tion, but this was not always clearly stated. For the few instances in which inconsistent ratings occurred, each discrepancy was reviewed jointly by the coders, and consensus about the correct rating was determined on the basis of the operational definition employed.
EFFECT SIZE CODING
In addition to coding various study procedures, effect sizes were coded for comparisons between treatment completers and dropouts for each of the following variables (if present in a given study): age, education, employment, income, marital status, race, referral source, criminal history, first-time domestic violence offender, personal experience of abuse, pres-ence of violence in the family of origin, severity of psychological abuse inflicted, severity of physical abuse inflicted, anger, alcohol use, drug use, and depression. Where possible, the effect size was calculated from raw data provided in the article (e.g., proportions, means and standard deviations, correlations) to allow for more accurate effect sizes (Durlak & Lipsey, 1991). When these statistics were not available, the effect size was estimated from the test statistics that were reported (e.g., c2, t test). In addition, the direction of the results (i.e., whether the data favored treatment completers or treatment dropouts) was coded. All raw data and test statistics used to estimate effect sizes were entered in Comprehensive Meta-Analysis Version 2.0 (Borenstein, Hedges, Higgins, & Rothstein, 2005), and this program was then used to calculate the effect sizes for each study as well as the summary effect sizes.
EFFECT SIZE INDICATOR AND META-ANALYTIC STRATEGY
A correlation coefficient (r) was used as the effect size indicator to allow for a common effect size to be calculated across variables because some variables were treated dichotomously by some researchers and continuously by others (Durlak & Lipsey, 1991). The correlation
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
1094 CRIMINAL JUSTICE AND BEHAVIOR
coefficient also has the advantage of being a widely used and understood statistic that is relatively easy to interpret (Gendreau & Smith, 2007; Wilson, 2001). Effect sizes were cal-culated using the formulas provided by Borenstein, Hedges, Higgins, and Rothstein (2009).
A random (rather than a fixed) effects model was employed to analyze the data, as this model provides a more conservative estimate of effect size and is thought to be appropriate for most criminal justice meta-analyses (Wilson, 2001). Because of presumed differences in effect size from one setting or jurisdiction to another, random effects models attempt to esti-mate a true population effect size and assume that the true effect size varies from study to study (Borenstein et al., 2009). In other words, random effects models assume that some por-tion of the study-level variability is unexplained by the study features included in the effect size statistic, and the goal of this model is to estimate the mean distribution of these effects (Borenstein et al., 2009; Wilson, 2001). In contrast, fixed effects models are based on the assumption that one true effect underlies all of the studies included in a meta analysis and estimates derived using this model are applicable only to the population of studies included in the meta-analysis (Borenstein et al., 2009; Hanson, Bourgon, Helmus, & Hodgson, 2009).
A number of additional steps were taken to ensure the meta-analytic results obtained would be accurate and useful. To establish the boundaries of an effect size, 95% confidence intervals (CIs) are reported for the grand mean effect size. An effect size can be considered to be statistically significant at the p < .05 level if the CI does not contain zero (Babcock et al., 2004). Furthermore, given that effect sizes based on small samples tend to be less precise than those based on large samples, all analyses were conducted using weighted effect sizes (Wilson, 2001). In addition, the Q statistic was examined for each analysis to determine whether the mean effect size reflects effects that are consistent across studies (i.e., are statistically homogenous; Borenstein et al., 2009; Wilson, 2001). Significant Q statistics indicate that there is more variability in the effect sizes than can be accounted for by sampling error and that they may not reflect a common population (Durlak & Lipsey, 1991; Wilson, 2001). As a result, further explorations were conducted when significant Q statis-tics were obtained to determine if any moderators could account for the observed hetero-geneity. Furthermore, outliers (defined as an effect size greater than three standard deviations from the mean) were removed from the meta-analytic calculations because the presence of outliers distorts the grand mean effect size. Finally, the effect sizes that are reported in this meta-analysis reflect independent comparisons. That is, several meta-analytic calculations were computed in the present study, and each calculation employs only one effect size from each study (Borenstein et al., 2009).
RESULTS
SUMMARY INFORMATION ABOUT THE STUDIES
Table 1 describes the overall characteristics of the 30 studies that were included in the meta-analysis. Most studies were published between 2000 and 2010 and were conducted in the United States. Treatment programs were equally as likely to employ feminist psych-oeducational models and cognitive-behavioral therapy, and nearly all programs provided group treatment in a community setting. Most studies focused solely on in-program attri-tion (70%), and 57% used samples consisting of a combination of participants who were court referred or self-referred to the program. Furthermore, 90% of the studies employed a
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
Jewell, Wormith / DOMESTIC VIOLENCE TREATMENT ATTRITION 1095
TABLE 1: Overall Characteristics of the Studies Included in the Meta-Analysis
Characteristic Number of Studies Proportion of Studies
Type of publicationJournal article 28 .93Unpublished dissertation 2 .07
Date of publication1989–1999 7 .232000–2010 17 .57
Country of originUnited States 25 .83Canada 4 .13United Kingdom 1 .03
Professional affiliation of first authorPsychology 12 .40Social work 10 .33Medicine 2 .07Sociology 1 .03Criminal justice 1 .03Nursing 1 .03Public health 1 .03Women’s and gender studies 1 .03Unknown 1 .03
Theoretical orientation of treatmentFeminist psychoeducational 12 .40Cognitive behavioral 12 .40Other or unknown 6 .20
Treatment formatGroup treatment 29 .97Unknown 1 .03
Location of treatmentCommunity setting 28 .93Unknown 2 .07
Type of attrition focused onIn-program attrition only 21 .70In-program and postassessment attrition collapsed 6 .20Unspecified 3 .10
Study designCorrelational 3 .10Nonequivalent comparison group 27 .90
Referral sourceCourt mandated 12 .40Combination of court mandated and self-referral 17 .57Unknown 1 .03
Program length (in weeks of group treatment)16 or fewer weeks 14 .4717 or more weeks 14 .47Unspecified 2 .07
Note. k = 27.
nonequivalent comparison group quasiexperimental design; however, three studies (Catlett, Toews, & Walilko, 2010; Daly et al., 2001; Taft, Murphy, Elliott, & Keaser, 2001) employed correlational designs. Programs ranged in length from 8 to 36 group sessions, and program completion rates ranged from 22% to 78%. Sample sizes for the studies varied from N = 34 to N = 3,460, and a total of 11,665 participants are reflected in the meta-analytic results. Table 2 provides more specific sample and program characteristics of each coded study.
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
1096
TAB
LE
2:
Sp
ecif
ic C
har
acte
rist
ics
of
Eac
h S
tud
y In
clu
ded
in t
he
Met
a-A
nal
ysis
Stu
dy
Sam
ple
D
efin
ition
of
Pro
gram
Com
plet
ion
Pro
gram
Com
plet
ion
Rat
e (%
)P
rogr
am L
engt
h an
d T
heor
etic
al
Orie
ntat
ion
Bab
cock
& S
tein
er,
1999
N =
355
91.8
% m
ale
M =
32.
7 ag
e50
% e
mpl
oyed
41%
Cau
casi
an31
% m
arrie
dC
ourt
man
date
dB
oth
post
asse
ssm
ent
and
in-p
rogr
am a
ttriti
on
Com
plet
ed 2
4 or
mor
e se
ssio
ns29
.926
wee
kly
sess
ions
; 6 m
onth
ly
sess
ions
; FP
E
Bel
din,
200
8N
= 1
84C
ompl
eted
all
trea
tmen
t se
ssio
ns55
.424
or
36 w
eekl
y se
ssio
ns; u
nkno
wn
100%
mal
eM
= 3
1.7
age
81%
em
ploy
ed67
.9%
Cau
casi
an15
.2%
mar
ried
Cou
rt m
anda
ted
Uns
peci
fied
type
of
attr
ition
Ben
nett
et a
l., 2
007
N =
549
Uns
peci
fied
75.2
Unk
now
n; u
nkno
wn
100%
mal
e74
.9%
em
ploy
ed34
.8%
Cau
casi
an41
.9%
mar
ried
Cou
rt m
anda
ted
Uns
peci
fied
type
of
attr
ition
Bow
en &
Gilc
hris
t, 20
06N
= 1
2010
0% m
ale
M =
34.
95 a
ge84
.8%
Cau
casi
an24
% m
arrie
dC
ourt
man
date
dB
oth
post
asse
ssm
ent
and
in-p
rogr
am a
ttriti
on
Com
plet
ed a
t le
ast
16 o
f 24
gro
up
sess
ions
67.5
24 g
roup
ses
sion
s, 1
indu
ctio
n se
ssio
n, 5
mon
thly
follo
w-u
p; F
PE
(con
tiune
d)
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
1097
TAB
LE
2:
(co
nti
nu
ed)
Stu
dy
Sam
ple
D
efin
ition
of
Pro
gram
Com
plet
ion
Pro
gram
Com
plet
ion
Rat
e (%
)P
rogr
am L
engt
h an
d T
heor
etic
al
Orie
ntat
ion
But
tell
& C
arne
y, 2
002
N =
137
100%
mal
eM
= 3
4.0
age
89%
em
ploy
ed49
.6%
Cau
casi
an39
.4%
mar
ried
Cou
rt m
anda
ted
In-p
rogr
am a
ttriti
on
Com
plet
ed a
ll se
ssio
ns56
.216
wee
kly
sess
ions
, in
clud
ing
2 or
ient
atio
n, 1
2 ps
ycho
educ
atio
nal,
and
2 te
rmin
atio
n cl
asse
s; C
BT
But
tell
& C
arne
y, 2
008
N =
1,7
0210
0% m
ale
M =
34.
8 ag
e91
.8%
em
ploy
ed49
.1%
Cau
casi
an59
.6%
mar
ried
Cou
rt m
anda
ted
In-p
rogr
am a
ttriti
on
Com
plet
ed a
ll se
ssio
ns49
.926
wee
kly
sess
ions
, in
clud
ing
2 or
ient
atio
n, 2
0 ps
ycho
educ
atio
nal,
and
4 te
rmin
atio
n cl
asse
s; C
BT
But
tell
& P
ike,
200
2N
= 8
3C
ompl
eted
all
sess
ions
67.0
12 w
eekl
y se
ssio
ns; C
BT
100%
mal
eM
= 3
1.0
age
42.2
% C
auca
sian
40.9
% m
arrie
dC
ourt
man
date
dIn
-pro
gram
attr
ition
Cad
sky
et a
l., 1
996
N =
218
100%
mal
eC
ourt
man
date
d an
d se
lf-re
ferr
alIn
-pro
gram
attr
ition
Com
plet
e at
leas
t 8
of 1
0 se
ssio
ns61
.010
wee
kly
sess
ions
; CB
T
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
1098
Car
ney
et a
l., 2
006
N =
114
Com
plet
ed a
ll 16
ses
sion
s49
.116
wee
kly
sess
ions
; CB
T10
0% m
ale
M =
32.
0 ag
e46
.0%
Cau
casi
an43
.2%
mar
ried
Cou
rt m
anda
ted
In-p
rogr
am a
ttriti
onC
atle
tt et
al.,
201
0N
= 1
5410
0% m
ale
M =
34.
4 ag
e38
.3%
Cau
casi
anC
ourt
man
date
dIn
-pro
gram
attr
ition
Com
plet
ed a
t le
ast
16 o
f 20
se
ssio
ns54
.5In
take
inte
rvie
w; b
imon
thly
as
sess
men
t se
ssio
ns u
ntil
asse
ssm
ent
com
plet
e; 1
6 gr
oup
sess
ions
; FP
E
Cha
ng &
Sau
nder
s, 2
002
N =
178
100%
mal
eM
= 3
2.0
age
77.0
% C
auca
sian
8.0%
mar
ried
Cou
rt m
anda
ted
and
self-
refe
rral
Com
plet
ed a
t le
ast
16 o
f 20
se
ssio
ns62
.020
wee
kly
sess
ions
; FP
E a
nd C
BT
In-p
rogr
am a
ttriti
onD
alto
n, 2
001
N =
85
Com
plet
ed 1
4 or
mor
e se
ssio
ns70
.616
wee
kly
sess
ions
; FP
E92
% m
ale
M =
32.
0 ag
e84
% e
mpl
oyed
54%
Cau
casi
an44
% m
arrie
dC
ourt
man
date
d an
d se
lf-re
ferr
alIn
-pro
gram
attr
ition
(con
tiune
d)
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
Car
ney
et a
l., 2
006
N =
114
Com
plet
ed a
ll 16
ses
sion
s49
.116
wee
kly
sess
ions
; CB
T10
0% m
ale
M =
32.
0 ag
e46
.0%
Cau
casi
an43
.2%
mar
ried
Cou
rt m
anda
ted
In-p
rogr
am a
ttriti
onC
atle
tt et
al.,
201
0N
= 1
5410
0% m
ale
M =
34.
4 ag
e38
.3%
Cau
casi
anC
ourt
man
date
dIn
-pro
gram
attr
ition
Com
plet
ed a
t le
ast
16 o
f 20
se
ssio
ns54
.5In
take
inte
rvie
w; b
imon
thly
as
sess
men
t se
ssio
ns u
ntil
asse
ssm
ent
com
plet
e; 1
6 gr
oup
sess
ions
; FP
E
Cha
ng &
Sau
nder
s, 2
002
N =
178
100%
mal
eM
= 3
2.0
age
77.0
% C
auca
sian
8.0%
mar
ried
Cou
rt m
anda
ted
and
self-
refe
rral
Com
plet
ed a
t le
ast
16 o
f 20
se
ssio
ns62
.020
wee
kly
sess
ions
; FP
E a
nd C
BT
In-p
rogr
am a
ttriti
onD
alto
n, 2
001
N =
85
Com
plet
ed 1
4 or
mor
e se
ssio
ns70
.616
wee
kly
sess
ions
; FP
E92
% m
ale
M =
32.
0 ag
e84
% e
mpl
oyed
54%
Cau
casi
an44
% m
arrie
dC
ourt
man
date
d an
d se
lf-re
ferr
alIn
-pro
gram
attr
ition
(con
tiune
d)
1099
TAB
LE
2:
(co
nti
nu
ed)
Stu
dy
Sam
ple
D
efin
ition
of
Pro
gram
Com
plet
ion
Pro
gram
Com
plet
ion
Rat
e (%
)P
rogr
am L
engt
h an
d T
heor
etic
al
Orie
ntat
ion
Dal
y et
al.,
200
1N
= 2
2010
0% m
ale
M =
31.
8 ag
e82
.2%
em
ploy
ed32
.1%
Cau
casi
anC
ourt
man
date
d an
d se
lf-re
ferr
alB
oth
post
asse
ssm
ent
and
in-p
rogr
am a
ttriti
on
Com
plet
ed 1
8 or
mor
e se
ssio
ns
and
two
eval
uatio
n se
ssio
ns48
.018
wee
kly
sess
ions
; 1 in
divi
dual
ev
alua
tion,
1 g
roup
orie
ntat
ion;
F
PE
DeM
aris
, 19
89N
= 2
9510
0% m
ale
M =
32.
0 ag
e74
% e
mpl
oyed
24%
Cau
casi
an71
% m
arrie
dC
ourt
man
date
dIn
-pro
gram
attr
ition
Com
plet
ed a
t le
ast
11 o
f 12
se
ssio
ns72
.012
gro
up s
essi
ons;
FP
E
Dup
lant
is,
Rom
ans,
& B
ear,
2006
N =
313
100%
mal
eM
= 3
3.0
age
66.1
% C
auca
sian
Cou
rt m
anda
ted
and
self-
refe
rral
Bot
h po
stas
sess
men
t an
d in
-pro
gram
attr
ition
N =
34
Uns
peci
fied
Unk
now
n24
gro
up,
2 dr
ug a
nd a
lcoh
ol
educ
atio
n, 1
con
sulta
tion,
and
2
inta
ke s
essi
ons;
FP
E a
nd C
BT
Faul
kner
et
al.,
1991
Uns
peci
fied
52.9
8 se
ssio
ns; C
BT
100%
mal
eM
= 3
1.5
age
52.9
% e
mpl
oyed
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
1100
55.9
% C
auca
sian
35.3
% m
arrie
dC
ourt
man
date
d an
d se
lf-re
ferr
alIn
-pro
gram
attr
ition
Ger
lock
, 20
01N
= 6
110
0% m
ale
M =
38.
8 ag
e54
.8%
Cau
casi
anC
ourt
man
date
dIn
-pro
gram
attr
ition
Bat
tere
rs a
re s
atis
fact
orily
re
habi
litat
ed37
.726
wee
kly
sess
ions
; 4 in
trod
uctio
n m
eetin
gs; F
PE
Gor
don
& M
oria
rty,
200
3N
= 1
32C
ompl
eted
all
sess
ions
61.4
20 o
r 24
wee
kly
sess
ions
; FP
E10
0% m
ale
M =
34.
8 ag
e80
% e
mpl
oyed
58%
Cau
casi
an56
% m
arrie
dC
ourt
man
date
dIn
-pro
gram
attr
ition
Ham
berg
er &
Has
tings
, 19
89N
= 1
5610
0% m
ale
M =
30.
9 ag
e68
.6%
em
ploy
ed81
.4%
Cau
casi
an41
.7%
mar
ried
Cou
rt m
anda
ted
and
self-
refe
rral
Bot
h po
stas
sess
men
t an
d in
-pro
gram
attr
ition
Com
plet
ed a
ll ev
alua
tion
and
inte
rven
tion
sess
ions
56.4
12 g
roup
, 3
indi
vidu
al a
sses
smen
t, an
d 1
post
inte
rven
tion
eval
uatio
n se
ssio
n; C
BT
Ham
berg
er e
t al
., 20
00N
= 4
8210
0% m
ale
Com
plet
ed a
ll se
ssio
ns57
.112
wee
kly
and
3 in
divi
dual
se
ssio
ns; C
BT
(con
tiune
d)
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
TAB
LE
2:
(co
nti
nu
ed)
Stu
dy
Sam
ple
D
efin
ition
of
Pro
gram
Com
plet
ion
Pro
gram
Com
plet
ion
Rat
e (%
)P
rogr
am L
engt
h an
d T
heor
etic
al
Orie
ntat
ion
M =
31.
5 ag
e77
.6%
em
ploy
ed69
.1%
Cau
casi
an32
.6%
mar
ried
Cou
rt m
anda
ted
and
self-
refe
rral
In-p
rogr
am a
ttriti
onM
cClo
skey
, S
itake
r, G
rigsb
y, &
M
allo
y, 2
003
N =
216
100%
mal
eM
= 3
5.0
age
76%
em
ploy
ed69
% C
auca
sian
35%
mar
ried
Cou
rt m
anda
ted
and
self-
refe
rral
Bot
h po
stas
sess
men
t an
d in
-pro
gram
attr
ition
Com
plet
ed a
ll se
ssio
ns22
.216
or
26 w
eekl
y se
ssio
ns; 3
ps
ycho
educ
atio
nal c
lass
es; 1
in
take
inte
rvie
w; F
PE
Ron
deau
et
al.,
2001
N =
286
Com
plet
ed a
ll se
ssio
ns42
.314
to
25 g
roup
ses
sion
s; u
nkno
wn
100%
mal
e62
.6%
em
ploy
ed34
.3%
mar
ried
Cou
rt m
anda
ted
and
self-
refe
rral
In-p
rogr
am a
ttriti
onR
oone
y &
Han
son,
200
1N
= 3
0610
0% m
ale
M =
35.
5 ag
e77
% e
mpl
oyed
Cou
rt m
anda
ted
and
self-
refe
rral
In-p
rogr
am a
ttriti
on
Uns
peci
fied
50.7
12 t
o 25
wee
kly
sess
ions
; mul
tiple
pr
ogra
ms
1101 at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
1102
Rot
hman
, G
upta
, P
avlo
s, D
ang,
&
Cou
tinho
, 20
07N
= 3
,460
100%
mal
eM
= 3
3.3
age
55.4
% C
auca
sian
Uns
peci
fied
refe
rral
sou
rce
Uns
peci
fied
type
of
attr
ition
N =
196
Com
plet
ed a
ll re
quire
d se
ssio
ns,
paid
all
fees
, an
d fu
lfille
d pr
ogra
m r
equi
rem
ents
51.8
Unk
now
n; u
nkno
wn
Sar
tin,
2004
Com
plet
ed a
ll se
ssio
ns59
.224
wee
kly
sess
ions
; FP
E10
0% m
ale
M =
35.
0 ag
e89
.2%
em
ploy
ed80
.1%
Cau
casi
an34
.5%
mar
ried
Cou
rt m
anda
ted
and
self-
refe
rral
In-p
rogr
am a
ttriti
onS
aund
ers
& P
arke
r, 19
89—
Stu
dy 1
N =
138
Uns
peci
fied
Unk
now
n12
gro
up s
essi
ons;
CB
T10
0% m
ale
M =
29.
9 ag
e70
% e
mpl
oyed
74.8
% C
auca
sian
27.9
% m
arrie
dC
ourt
man
date
d an
d se
lf-re
ferr
alIn
-pro
gram
attr
ition
Sau
nder
s &
Par
ker,
1989
—S
tudy
2N
= 1
33U
nspe
cifie
d78
.212
gro
up s
essi
ons;
CB
T10
0% m
ale
M =
30.
8 ag
e78
% e
mpl
oyed
88.2
% C
auca
sian
33.1
% m
arrie
dC
ourt
man
date
d an
d se
lf-re
ferr
alIn
-pro
gram
attr
ition
(con
tiune
d)
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
1103
TAB
LE
2:
(co
nti
nu
ed)
Stu
dy
Sam
ple
D
efin
ition
of
Pro
gram
Com
plet
ion
Pro
gram
Com
plet
ion
Rat
e (%
)P
rogr
am L
engt
h an
d T
heor
etic
al
Orie
ntat
ion
Sco
tt, 2
004
N =
308
100%
mal
eM
= 3
5.4
age
65%
em
ploy
ed83
% C
auca
sian
Cou
rt m
anda
ted
and
self-
refe
rral
In-p
rogr
am a
ttriti
on
Com
plet
ed a
ll se
ssio
ns38
.617
wee
kly
sess
ions
; 1 in
take
se
ssio
n; F
PE
Taft,
Mur
phy,
Elli
ott,
& K
ease
r, 20
01N
= 1
0110
0% m
ale
M =
33.
6 ag
e80
% e
mpl
oyed
60.4
% C
auca
sian
33%
mar
ried
Cou
rt m
anda
ted
and
self-
refe
rral
In-p
rogr
am a
ttriti
on
Com
plet
ed a
t le
ast
75%
of
sess
ions
69.3
16 w
eekl
y se
ssio
ns; C
BT
Taft,
Mur
phy,
Elli
ott,
& M
orre
l, 20
01N
= 1
8910
0% m
ale
M =
34.
2 ag
e60
.0%
Cau
casi
an30
.0%
mar
ried
Cou
rt m
anda
ted
and
se
lf-re
ferr
alIn
-pro
gram
attr
ition
Com
plet
ed a
t le
ast
75%
of
sess
ions
(12
of
16 s
essi
ons)
76.7
16 w
eekl
y se
ssio
ns; C
BT
Not
e. k
= 3
0. F
PE
= fe
min
ist
psyc
hoed
ucat
iona
l tre
atm
ent;
CB
T =
cog
nitiv
e-be
havi
oral
the
rapy
.
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
1104 CRIMINAL JUSTICE AND BEHAVIOR
DEMOGRAPHIC VARIABLES
Table 3 summarizes the grand mean effect size obtained for each variable assessed using meta-analytic techniques. Overall, the demographic variables were better able to distin-guish between treatment completers and treatment dropouts than the violence-related and intrapersonal variables. In general, individuals who were employed were 20% more likely to complete treatment than individuals who were unemployed, batterers who were older were 16% more likely to complete treatment than those who were younger, and men who had higher incomes were 13% more likely to complete treatment than those with lower incomes. Modest effect sizes also were observed for the relationship between attrition and education (r = .08), marital status (r = .08), and race (r = .09). Examinations of the Q sta-tistic for each of the demographic variables revealed that several of the variables (i.e., employment, age, and education) demonstrated significant heterogeneity. As such, further analyses were conducted to determine whether any of the moderator variables could account for the observed variability.
Theoretical orientation of the treatment program moderated the relationship observed between age and treatment completion (after one outlier was removed). Specifically, com-pared to individuals who were younger, older individuals were significantly more likely to complete cognitive-behavioral programs (r = .18, 95% CI = .12 to .23, k = 10) or unspeci-fied programs (r = .13, 95% CI = .01 to .24, k = 4) than feminist psychoeducational pro-grams (r = .02, 95% CI = –.10 to .13, k = 7). In addition, there was a significant difference with respect to program length wherein men who were older were more likely to complete programs that were of a shorter duration (i.e., 16 or fewer weeks; r = .18, 95% CI = .13 to .24, k = 11) than men attending longer programs (r = .08, 95% CI = .02 to .16, k = 9). Theoretical orientation also moderated the relationship between education and attrition. Batterers who had less education were more likely to drop out of feminist psychoeduca-tional treatment programs (r = .14, 95% CI = .08 to .20, k = 6) than cognitive behavioral programs (r = .10, 95% CI = .05 to .15, k = 11) or unspecified programs (r = .01, 95% CI = –.05 to .06, k = 6) compared to men who had more education. In addition, significantly higher rates of treatment completion among men who had more education were observed within studies that employed samples consisting of both postassessment and in-program dropouts (r = .17, 95% CI = .09 to .24, k = 3) than studies that examined only in-program attrition (r = .08, 95% CI = .03 to .12, k = 17) or did not specify the nature of attrition (r = .03, 95% CI = –.03 to .11, k = 3). None of the potential moderator variables (i.e., type of attrition, referral source, theoretical orientation, program length, sample size, country, pro-fessional affiliation, and publication type) explained the heterogeneity observed with respect to employment.
VIOLENCE-RELATED VARIABLES
Of all the violence-related variables, referral source was most strongly associated with attrition. Batterers who were court mandated were 16% more likely than batterers who were not court mandated to complete treatment. Furthermore, being a first-time domestic violence offender and having a previous criminal history were both modestly associated with completion. Specifically, men attending treatment after their first domestic violence offense were approximately 14% more likely to complete treatment than batterers who had
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
1105
TAB
LE
3:
Eff
ect
Siz
es o
f Var
iab
les
Rel
ated
to
Att
riti
on
Ave
rage
Cor
rela
tions
and
95%
Con
fiden
ce I
nter
vals
(C
Is)
Ran
dom
Effe
cts
Mod
el
Var
iabl
eC
orre
latio
n (r
)95
% C
IQ
Sta
tistic
kR
ange
of
ES
(r)
NIn
terp
reta
tion
Dem
ogra
phic
var
iabl
esE
mpl
oym
ent
.20
.15
to .
2441
.89*
21–.
02 t
o .5
14,
419
TC
mor
e lik
ely
to b
e em
ploy
edA
ge .
16.0
8 to
.24
370.
70*
22–.
42 t
o .8
38,
621
TC
mor
e lik
ely
to b
e ol
der
Inco
me
.13
.10
to .
169.
7212
–.08
to
.33
3,64
8T
C m
ore
likel
y to
hav
e hi
gher
in
com
esM
arita
l sta
tus
(mar
ried)
.09
.04
to .
1322
.14
15–.
14 t
o .2
54,
697
TC
mor
e lik
ely
to b
e m
arrie
dR
ace
(Cau
casi
an)
.09
.04
to .
1327
.56
19–.
16 t
o .2
94,
879
TC
mor
e lik
ely
to b
e C
auca
sian
Edu
catio
n .
08.0
4 to
.12
35.0
7*23
–.10
to
.26
9,42
4T
C m
ore
likel
y to
hav
e m
ore
educ
atio
nV
iole
nce-
rela
ted
fact
ors
Ref
erra
l sou
rce
(cou
rt
man
date
d) .
16.0
7 to
.25
28.9
4*11
–.05
to
.39
1,82
0T
C m
ore
likel
y to
be
cour
t m
anda
ted
Firs
t-tim
e D
V o
ffend
er .
14.0
8 to
.21
3.35
5.0
3 to
.24
1,14
1M
ore
likel
y to
be
TC
’s f
irst
DV
of
fens
eC
rimin
al h
isto
ry–.
10–.
16 t
o –.
0537
.51*
17–.
34 t
o .1
07,
312
TD
mor
e lik
ely
to h
ave
a cr
imin
al
hist
ory
Fam
ily o
f or
igin
vio
lenc
e–.
06–.
14 t
o .0
1—
9–.
22 t
o .1
62,
370
Not
sig
nific
ant
Per
sona
l exp
erie
nce
of
abus
e–.
04–.
12 t
o .0
5—
7–.
22 t
o .0
61,
530
Not
sig
nific
ant
Sev
erity
of
psyc
holo
gica
l ab
use
infli
cted
–.03
–.11
to
.05
— 7
–.15
to
.21
2,58
9N
ot s
igni
fican
t
Sev
erity
of
phys
ical
abu
se
infli
cted
–.02
–.12
to
.07
—10
–.19
to
.31
3,17
9N
ot s
igni
fican
t
Intr
aper
sona
l var
iabl
esD
rug
use
–.12
–.18
to
–.07
20.0
0*12
–.40
to
.01
6,98
3T
D m
ore
likel
y to
use
dru
gsA
lcoh
ol u
se–.
09–.
13 t
o –.
0519
.96
17–.
45 t
o –.
107,
947
TD
mor
e lik
ely
to u
se a
lcoh
olA
nger
–.03
–.19
to
.14
— 6
–.42
to
.11
1,16
9N
ot s
igni
fican
tD
epre
ssio
n .
03–.
18 t
o .2
3—
7–.
24 t
o .5
784
5N
ot s
igni
fican
t
Not
e. k
= n
umbe
r of
stu
dies
incl
uded
in t
he a
naly
sis;
TC
= t
reat
men
t co
mpl
eter
; TD
= t
reat
men
t dr
opou
t; D
V =
dom
estic
vio
lenc
e.
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
1106 CRIMINAL JUSTICE AND BEHAVIOR
previously been arrested or convicted for domestic violence. Conversely, batterers who had a criminal history (i.e., they had been arrested or convicted for crimes separate from the domestic violence incident that placed them in treatment) were 10% more likely to drop out of treatment.
The effect sizes associated with referral source and having a criminal history were both significantly heterogeneous. Again, the theoretical orientation of the treatment program moderated the relationship between referral source and attrition, with men who were court mandated to treatment being more likely to complete feminist psychoeducational treatment programming (r = .31, 95% CI = .23 to .39, k = 3) than cognitive-behavioral programs (r = .10, 95% CI = .01 to .19, k = 6) or unspecified programs (r = .03, 95% CI = –.12 to .18, k = 2) compared to men who were not court referred. None of the potential moderator variables (i.e., type of attrition, referral source, theoretical orientation, program length, sample size, country, professional affiliation, and publication type) explained the heteroge-neity observed with respect to criminal history.
The effect sizes observed for all other violence-related variables (i.e., personal experi-ence of abuse, witnessing violence in one’s family of origin, and the severity of the physi-cal and psychological abuse inflicted) were nonsignificant, indicating that these variables are not significant predictors of attrition. The nonsignificant findings may be because of the relatively few studies included in the analyses (k = 10 for severity of physical abuse, k = 9 for family of origin violence, and k = 7 for personal experience of abuse and severity of psychological abuse).
INTRAPERSONAL VARIABLES
Both alcohol use and drug use were associated with treatment attrition, with treatment dropouts suffering from alcohol and drug problems being 12% and 10%, respectively, less likely to complete treatment than batterers not contending with these issues. A Q test indi-cated that the effect size calculated for alcohol use reflects a homogenous population of studies but that drug use does not. Additional analyses were conducted for drug use, and sample size was found to moderate its relationship with attrition. Specifically, studies that had samples of 200 or fewer participants reported greater differences between the treatment completion rates of drug users versus nondrug users (r = –.18, 95% CI = –.26 to –.11, k = 8) than samples consisting of more than 200 participants (r = –.06, 95% CI = –.10 to –.02, k = 4). The other two intrapersonal variables (i.e., anger and depression) did not signifi-cantly differentiate between treatment completers and dropouts.
DISCUSSION
The results from this meta-analysis suggest that numerous variables are related to treat-ment completion. The strongest predictors associated with treatment completion were employ-ment, age, and referral source (rs ≥ .16), wherein individuals who were employed, older, and court mandated were more likely to remain in treatment than were persons who were unem-ployed, younger, or not court mandated. Other correlates of attrition associated with modest differences between treatment completers and dropouts were previous domestic violence offenses, income, drug use, and criminal history (rs ≥ .10). Specifically, individuals who
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
Jewell, Wormith / DOMESTIC VIOLENCE TREATMENT ATTRITION 1107
were attending domestic violence treatment after their first partner assault offense or who had higher incomes were significantly more likely to complete treatment. In contrast, indi-viduals who had a criminal history prior to attending domestic violence treatment or had drug problems were more likely to drop out of treatment. Men who had more education, were married, were Caucasian, or did not have problems with alcohol also completed treat-ment at a somewhat higher rate than persons who were less educated, were unmarried, were members of a minority group, or had problems with alcohol (rs ≥ .08). Variables such as being a victim of abuse, witnessing violence within one’s family of origin, severity of physi-cal and psychological abuse, depression, and anger did not distinguish between treatment completers and dropouts.
The results obtained from the meta-analysis suggest the same variables that have been found to predict domestic violence recidivism also predict attrition from domestic violence treatment programs. These findings support Wormith and Olver’s (2002) suggestion that attrition and recidivism are predicted by the same variables. However, some variables are clearly better predictors than others, indicating that certain variables may be uniquely suited to predicting domestic violence attrition. A particularly interesting finding arising from the meta-analysis is that the theoretical orientation of the treatment program served as an important moderator variable that explained some of the heterogeneity in effect sizes for age, education, and referral source. Specifically, men who were more educated and court mandated were more likely to complete feminist psychoeducational programs than were men who were not as educated or court mandated, whereas batterers who were older completed cognitive-behavioral programs more readily than younger men. Although the theoretical orientation of a program may not have a significant impact on domestic violence recidivism (Babcock et al., 2004), our results suggest it may affect attrition.
The pattern of results obtained from the meta-analysis also fits nicely into the risk, need, and responsivity framework developed by Andrews, Bonta, and Hoge (1990). Briefly, this theory is based on the notions that treatment intensity should be matched to offenders’ risk level, treatment should focus primarily on criminogenic needs, and treatment should be deliv-ered in a way that best matches an offender’s ability and learning style (Andrews et al., 1990). In fact, many of the variables found to be related to domestic violence treatment attrition reflect the criminogenic needs (i.e., employment, drug and alcohol use, lack of familial ties) that put offenders at risk for recidivism. Of particular relevance to domestic violence treatment dropout, however, is the responsivity principle, which assumes that treatment strategies most likely to produce change should be selected and tailored to match each offender’s learning styles, motivation, aptitude, and ability (Andrews, 2000).
Indeed, many of the variables associated with treatment completion and attrition reflect the individual responsivity characteristics that should be taken into consideration when matching offenders to treatment, such as age, race, education, employment, income, and various personality characteristics (including depression and mental illness; Kennedy, 2000). The findings that individuals who are older, are employed, are married, earn higher incomes, and are first-time domestic violence offenders are more likely to remain in treat-ment may reflect that these individuals have more stable lifestyles, are more socially inte-grated, and consequently have more to lose (both economically and socially) by not completing treatment compared to those who have more unstable lifestyles (Babcock & Steiner, 1999). Thus, lifestyle instability may be an important responsivity issue to consider when treating male batterers, and it has implications for how treatment providers can tailor interventions
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
1108 CRIMINAL JUSTICE AND BEHAVIOR
to increase the likelihood that offenders who are unemployed, are unmarried, and have greater criminal histories remain in treatment (Cadsky et al., 1996). For instance, program providers may make treatment more flexible for clients at risk for dropping out because of their unstable lifestyles, provide extra services to these individuals, or make more of an effort to contact clients who fail to attend a group session to encourage them to remain with the program (Cadsky et al., 1996; Chang & Saunders, 2002). In light of the evidence sug-gesting that theoretical orientation is a moderator variable that can differentiate treatment completers from dropouts, we also suggest that, when case managers have the luxury of choosing between programs, they should refer offenders according to the match between personal characteristics (i.e., age, education, and referral source) and the theoretical orien-tation of the treatment programs.
The relationship observed between education and program completion also may be reflective of another central responsivity issue: cognitive functioning (Kennedy, 2000). Although little research has explicitly studied the relationship between male batterers’ intellectual and cognitive functioning and domestic violence program completion, the level of education obtained may be a proxy for lower levels of cognitive functioning. That is, individuals with less education may find it more difficult to “keep up” with domestic vio-lence treatment programs and, as a result, decide to drop out (Chang & Saunders, 2002). More research is needed to explore this potential implication. However, if less education is found to be a proxy for lower cognitive abilities, it may be possible to decrease attrition by ensuring that treatment is delivered to individuals with less education in a way that they are able to follow and understand or by providing additional assistance to these persons outside of the group meetings (Chang & Saunders, 2002; Rooney & Hanson, 2001).
Given that others (Wormith & Olver, 2002) have found that factors related to recidivism also predict attrition and that substance abuse is a known predictor of domestic violence recidivism (Hilton & Harris, 2005), it was not surprising to find that offenders who suffer from alcohol and drug abuse were less likely to persist in treatment. Therefore, it may be possible to increase batterers’ likelihood of attending and remaining in treatment by address-ing substance abuse (Daly & Pelowski, 2000). Similarly, the finding that individuals belonging to a minority group have a greater tendency to drop out of treatment suggests that the provision of culturally sensitive programs may be a means of increasing treatment completion among racial minority offenders (Chang & Saunders, 2002).
The responsivity principle also brings to our attention additional individual characteris-tics that may be related to attrition and should be considered in future research. Specifically, motivation is thought to be an important factor for predicting whether participants will remain in treatment (Daly & Pelowski, 2000; Kennedy, 2000). Some researchers (e.g., Cadsky et al., 1996) have used court referral as a proxy for measuring motivation and have argued that individuals who volunteer for treatment will be more intrinsically motivated to attend and complete treatment. However, others have suggested that individuals who are extrinsically motivated by the courts will be more likely to complete treatment (Faulkner et al., 1991; Hamberger & Hastings, 1989). A meta-analysis by Parhar et al. (2008) exam-ined the relative impact of being court mandated and self-referred on treatment success and found that court-mandated treatment tends to be ineffective, whereas voluntary treatment tends to bring about more successful results. The results from our meta-analysis suggest that individuals who are court mandated to treatment are more likely to complete domestic violence treatment, suggesting that extrinsic motivation may be sufficient for this particular
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
Jewell, Wormith / DOMESTIC VIOLENCE TREATMENT ATTRITION 1109
group of offenders. However, given that referral source is not an ideal measure of motivation and that these results are in contrast to the meta-analysis conducted by Parhar et al., the cur-rent finding should be considered carefully. In the future, researchers should explicitly meas-ure motivation as a means of determining the extent to which it is a predictor of attrition.
Other factors related to responsivity that have not been studied in relation to domestic violence treatment attrition are therapist characteristics (Kennedy, 2000). Specifically, a therapist’s ability to establish a therapeutic alliance and a therapist’s ability to enhance offenders’ motivation have been identified as important factors in preventing program attri-tion (Dalton, 2001; Kennedy, 2000; Wormith & Olver, 2002). However, little research has focused on the relationship between therapist characteristics and domestic violence treat-ment attrition. This question should be addressed in future studies (Chang & Saunders, 2002). Aside from the current meta-analytic findings, the congruence between the client and treatment also has not been widely explored in relation to domestic violence treatment attrition (Cadsky et al., 1996; Rooney & Hanson, 2001). For instance, minimal research has investigated whether congruence between batterers’ self-identified problems and the treat-ment that is provided leads to higher rates of program completion (Cadsky et al., 1996). In addition, the relationship between batterers’ learning style and likelihood of completing treatment is largely unknown (Rooney & Hanson, 2001). Given our finding that the theo-retical orientation of a program may influence attrition, it seems that program characteris-tics are important issues to consider when assigning batterers to treatment programs. Clinicians and researchers are encouraged to explore the congruence between client and treatment characteristics with attrition as we have begun to show that responsivity issues may be used to minimize treatment attrition.
CLINICAL AND POLICY IMPLICATIONS
Several implications pertaining to policy can be derived from these meta-analytic findings. With respect to clinical policy, the results underscore the importance of using demographic, violence-related, and intrapersonal client characteristics to determine which individuals may be at risk for attrition from domestic violence treatment programs. On identifying individuals who are at risk for dropping out, it may be possible to minimize their attrition by providing them with additional supports to enhance their ability to attend group sessions and to under-stand the content of the programming. Therapists might also consider “pretreatment sessions” with identified offenders using motivational interviewing techniques to prepare the clients for the formal programming to follow (Miller & Rollnick, 2002).
The meta-analysis also has implications for sentencing. When developing conditions for sentencing, it may be valuable to take particular client characteristics (e.g., age and educa-tion) into account to ensure that clients are directed to the programs from which they are most likely to benefit. Although there is considerable debate about both the ethics and effec-tiveness of mandated versus voluntary treatment of offenders (Parhar et al., 2008), using court sentences to mandate batterers to treatment programs may be one way to increase program completion. Overall, the meta-analysis indicates that one type of batterer treat-ment programming does not fit all offenders. Batterer intervention agencies are encouraged to improve successful completion of programming by offering treatment in different for-mats rather than insisting that, for instance, all batterers must attend feminist psychoeducational treatment. Moreover, it should not be concluded from this analysis that offenders who are prone to attrition should be restricted from programming because they are less likely to
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
1110 CRIMINAL JUSTICE AND BEHAVIOR
complete treatment. In fact, it is these same individuals who are the most likely to reoffend, and therefore, it is important to direct additional resources toward them to increase the likelihood they will complete treatment as per the “risk” principles (Andrews et al., 1990; Babcock & Steiner, 1999; Bennett et al., 2007; Gordon & Moriarty, 2003).
LIMITATIONS
The present meta-analysis has resulted in the identification of numerous variables that can be used to distinguish domestic violence program completers and dropouts. However, the findings should be interpreted with the following caveats in mind. First, variations in definitions of attrition, program content (particularly with respect to theoretical orienta-tion), and program length compromise our ability to compare results from different studies and introduce increased variance into the meta-analytic findings. Future researchers may consider attrition as a continuous variable (i.e., number of sessions attended) to avoid issues associated with defining attrition. Second, the current study focused only on in-program attrition; postreferral and postassessment attrition may each be characterized by a different constellation of factors. Furthermore, only studies that were offered in community- based settings using a group format were included in this meta-analysis. Different variables may predict attrition from programs offered in institutional settings or that employ other formats. Third, the fact that it was not possible to account for the variation associated with employment and criminal history suggests there may be additional variables that have not yet been studied in relation to domestic violence treatment attrition (e.g., motivation, therapeutic alliance) that may account for treatment dropout. Additional research is needed to uncover these variables. Similarly, more research is needed on variables such as the severity of abuse inflicted, personal experience of abuse, witnessing of family violence, anger, and depression, as the small number of studies exploring these variables limited our ability to detect effects in this meta-analysis. Finally, the meta-analysis was limited to English-language documents and could be enhanced by including non-English research.
A considerable body of research has focused on the relationship between domestic vio-lence treatment attrition and particular demographic, violence-related, and intrapersonal variables. At this point, it would be valuable for researchers to turn to lesser studied vari-ables (e.g., therapist characteristics, batterers’ learning styles and cognitive abilities) asso-ciated with treatment responsivity to establish whether any of these currently unstudied factors also are related to treatment attrition. Many of the limitations associated with this meta-analysis will be resolved as more evidence is gathered. Until that time, however, this study reflects an important first look at the existing evidence to establish what is currently known about the client and program factors associated with domestic violence treatment dropout. It also provides direction to providers of domestic violence treatment programs that target male batterers about how to minimize dropout by matching client characteristics with the mode of service delivery.
REFERENCES
*References marked with an asterisk indicate studies included in the meta-analysis.Anderson, K. L. (2005). Theorizing gender in intimate partner violence. Sex Roles, 52, 853-865.Andrews, D. A. (2000). Principles of effective correctional programs. In L. L. Motiuk & R. C. Serin (Eds.), Compendium
2000 on effective correctional programming (pp. 9-17). Ottawa Ontario, Canada: Correctional Service Canada.
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
Jewell, Wormith / DOMESTIC VIOLENCE TREATMENT ATTRITION 1111
Andrews, D. A., Bonta, J., & Hoge, R. D. (1990). Classification for effective rehabilitation: Rediscovering psychology. Criminal Justice and Behavior, 17, 19-52.
Babcock, J. C., Green, C. E., & Robie, C. (2004). Does batterers’ treatment work? A meta-analytic review of domestic vio-lence treatment. Clinical Psychology Review, 23, 1023-1053.
*Babcock, J. C., & Steiner, R. (1999). The relationship between treatment, incarceration, and recidivism of battering: A program evaluation of Seattle’s Coordinated Community Response to Domestic Violence. Journal of Family Psychology, 13, 46-59.
*Beldin, K. L. (2008). Social information processing, program completion, and recidivism: One court’s referrals to a batterer intervention program (Unpublished doctoral dissertation). Case Western Reserve University, Cleveland, OH.
*Bennett, L. W., Stoops, C., Call, C., & Flett, H. (2007). Program completion and re-arrest in a batterer intervention system. Research on Social Work Practice, 17, 42-54.
Borenstein, M., Hedges, L. V., Higgins, J. P. T., & Rothstein, H. R. (2005). Comprehensive meta-analysis version 2.0 [Computer software]. Englewood, NJ: BioStat.
Borenstein, M., Hedges, L. V., Higgins, J. P. T., & Rothstein, H. R. (2009). Introduction to meta-analysis. West Sussex, England: Wiley.
*Bowen, E., & Gilchrist, E. (2006). Predicting dropout of court-mandated treatment in a British sample of domestic violence offenders. Psychology, Crime & Law, 12, 573-587.
Brown, P. D., & O’Leary, K. D. (2000). Therapeutic alliance: Predicting continuance and success in group treatment for spouse abuse. Journal of Consulting and Clinical Psychology, 68, 340-345.
Brown, P. D., O’Leary, K. D., & Feldbau, S. R. (1997). Dropout in treatment program for self-referring wife abusing men. Journal of Family Violence, 12, 365-387.
*Buttell, F. P., & Carney, M. M. (2002). Psychological and demographic predictors of attrition among batterers court ordered into treatment. Social Work Research, 26, 31-41.
*Buttell, F. P., & Carney, M. M. (2008). A large sample investigation of batterer intervention program attrition: Evaluating the impact of state program standards. Research on Social Work Practice, 18, 177-188.
*Buttell, F. P., & Pike, C. K. (2002). Investigating predictors of treatment attrition among court-ordered batterers. Journal of Social Service Research, 28, 53-68.
*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.
*Carney, M. M., Buttell, F. P., & Muldoon, J. (2006). Predictors of batterer intervention program attrition: Developing and implementing logistic regression models. Journal of Offender Rehabilitation, 43(2), 35-54.
*Catlett, B. S., Toews, M. L., & Walilko, V. (2010). Men’s gendered constructions of intimate partner violence as predictors of court-mandated batterer treatment drop-out. American Journal of Community Psychology, 45, 107-123.
Cattaneo, L. B., & Goodman, L. A. (2003). Victim-reported risk factors for continued abusive behavior: Assessing the dan-gerousness of arrested batterers. Journal of Community Psychology, 31, 349-369.
*Chang, H., & Saunders, D. G. (2002). Predictors of attrition in two types of group programs for men who batter. Journal of Family Violence, 17, 273-292.
*Dalton, B. (2001). Batterer characteristics and treatment completion. Journal of Interpersonal Violence, 16, 1223-1238.Daly, J. E., & Pelowski, S. (2000). Predictors of dropout among men who batter: A review of studies with implications for
research and practice. Violence and Victims, 15, 137-160.*Daly, J. E., Power, T. G., & Gondolf, E. W. (2001). Predictors of batterer program attendance. Journal of Interpersonal
Violence, 16, 971-991.Davidson, L. L., Grisso, J. A., Garcia-Moreno, C., Garcia, J., King, V. J., & Marchant, S. (2001). Training programs for
healthcare professionals in domestic violence. Journal of Women’s Health and Gender-Based Medicine, 10, 953-969.DeHart, D. D., Kennerly, R. J., Burke, L. K., & Follingstad, D. R. (1999). Predictors of attrition in a treatment program for
battering men. Journal of Family Violence, 14, 19-34.*DeMaris, A. (1989). Attrition in batterers’ counselling: The role of social and demographic factors. Social Service Review,
63, 142-154.Dowden, C., Antonowicz, D., & Andrews, D. A. (2003). The effectiveness of relapse prevention with offenders: A meta-
analysis. International Journal of Offender Therapy and Comparative Criminology, 47, 516-528.*Duplantis, A. D., Romans, J. S. C., & Bear, T. M. (2006). Persistence in domestic violence treatment and self-esteem, locus
of control, risk of alcoholism, level of abuse, and beliefs about abuse. Journal of Aggression, Maltreatment, & Trauma, 13(1), 1-18.
Durlak, J. A., & Lipsey, M. A. (1991). A practitioner’s guide to meta-analysis. American Journal of Community Psychology, 19, 291-332.
*Faulkner, K. K., Cogan, R., Nolder, M., & Shooter, G. (1991). Characteristics of men and women completing cognitive/behavioural spouse abuse treatment. Journal of Family Violence, 6, 243-254.
Garcia-Moreno, C., Jansen, H. A. F. M., Ellsberg, M., Heise, L., & Watts, C. H. (2006). Prevalence of intimate partner vio-lence: Findings from the WHO multi-country study on women’s health and domestic violence. Lancet, 368, 1260-1269.
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
1112 CRIMINAL JUSTICE AND BEHAVIOR
Gendreau, P., Little, T., & Goggin, C. (1996). A meta-analysis of the predictors of adult offender recidivism: What works! Criminology, 34, 575-607.
Gendreau, P., & Smith, P. (2007). Influencing the “people who count”: Some perspectives on the reporting of meta-analytic results for prediction and treatment outcomes with offenders. Criminal Justice and Behavior, 34, 1536-1559.
*Gerlock, A. A. (2001). A profile of who completes and who drops out of domestic violence rehabilitation. Issues in Mental Health Nursing, 22, 379-400.
Gill, C. (2006). Understanding theories and their links to intervention strategies. In M. R. Hampton & N. Gerrard (Eds.), Intimate partner violence: Reflections on experience, theory, and policy (pp. 47-66). Toronto, Ontario, Canada: Cormorant Books.
Gondolf, E. W. (1988). The effect of batterer counselling on shelter outcome. Journal of Interpersonal Violence, 3, 275-289.Gondolf, E. W. (2008). Program completion in specialized batterer counselling for African-American men. Journal of
Interpersonal Violence, 23, 94-116.Gondolf, E. W., & Foster, R. A. (1991). Pre-program attrition in batterer programs. Journal of Family Violence, 6, 337-349.*Gordon, J. A., & Moriarty, L. J. (2003). The effects of domestic violence batterer treatment on domestic violence recidivism:
The Chesterfield County experience. Criminal Justice and Behavior, 30, 118-134.Greaves, L., Hankivsky, O., & Kingston-Riechers, J. (1995). Selected estimates of the costs of violence against women.
London, Ontario, Canada: Centre for Research on Violence Against Women and Children.*Hamberger, L. K., & Hastings, J. E. (1989). Counseling male spouse abusers: Characteristics of treatment completers and
dropouts. Violence and Victims, 4, 275-286.*Hamberger, L. K., Lohr, J. M., & Gottlieb, M. (2000). Predictors of treatment dropout from a spouse abuse abatement pro-
gram. Behavior Modification, 24, 528-552.Hanson, R. K., Bourgon, G., Helmus, L., & Hodgson, S. (2009). The principles of effective correctional treatment also apply
to sexual offenders: A meta-analysis. Criminal Justice and Behavior, 36, 865-891.Hanson, R. K., & Wallace-Capretta, S. (2004). Predictors of criminal recidivism among male batterers. Psychology, Crime &
Law, 10, 413-427.Heyman, R. E., Brown, P. D., Feldbau-Kohn, S. R., & O’Leary, K. D. (1999). Couples’ communication behaviours as predic-
tors of dropout and treatment response in wife abuse treatment programs. Behavior Therapy, 30, 165-189.Hilton, N. Z., & Harris, G. T. (2005). Predicting wife assault: A critical review and implications for policy and practice.
Trauma, Violence, & Abuse, 6, 3-23.Hilton, N. Z., Harris, G. T., & Rice, M. E. (2001). Predicting violence by serious wife assaulters. Journal of Interpersonal
Violence, 16, 408-423.Huss, M. T., & Ralston, A. (2008). Do batterer subtypes actually matter? Treatment completion, treatment response, and
recidivism across a batterer typology. Criminal Justice & Behavior, 35, 710-724.Kennedy, S. (2000). Treatment responsivity: Reducing recidivism by enhancing treatment effectiveness. In L. L. Motiuk &
R. C. Serin (Eds.), Compendium 2000 on effective correctional programming (pp. 30-36). Ottawa Ontario, Canada: Correctional Service Canada.
*McCloskey, K. A., Sitaker, M., Grigsby, N., & Malloy, K. A. (2004). Characteristics of male batterers in treatment: An example of a localized program evaluation concerning attrition. Journal of Aggression, Maltreatment, & Trauma, 8(4), 67-95.
Miller, W. R., & Rollnick, S. (2002). Motivational interviewing: Preparing people for change (2nd ed.). New York, NY: Guilford.
Parhar, K. K., Wormith, J. S., Derkzen, D. M., & Beauregard, A. M. (2008). Offender coercion in treatment: A meta-analysis of effectiveness. Criminal Justice and Behavior, 35, 1109-1135.
Plichta, S. B. (2004). Intimate partner violence and physical health consequences: Policy and practice implications. Journal of Interpersonal Violence, 19, 1296-1323.
*Rondeau, G., Brodeur, N., Brochu, S., & Lemire, G. (2001). Dropout and completion of treatment among spouse abusers. Violence and Victims, 16, 127-143.
*Rooney, J., & Hanson, R. K. (2001). Predicting attrition from treatment programs for abusive men. Journal of Family Violence, 16, 131-149.
*Rothman, E. F., Gupta, J., Pavlos, C., Dang, Q., & Coutinho, P. (2007). Batterer intervention program enrolment and com-pletion among immigrant men in Massachusetts. Violence Against Women, 13, 527-543.
*Sartin, R. M. (2004). Characteristics associated with domestic violence perpetration: An examination of factors related to treat-ment response and the utility of a batterer typology (Unpublished doctoral dissertation). University of Nebraska, Lincoln.
*Saunders, D. G., & Parker, J. C. (1989). Legal sanctions and treatment follow-through among men who batter: A multi-variate analysis. Social Work Research & Abstracts, 25, 21-29.
*Scott, K. L. (2004). State of change as a predictor of attrition among men in a batterer treatment program. Journal of Family Violence, 19, 37-47.
Sherman, L., Gottfredson, D., Mackenzie, D. L., Eck, J., Reuter, P., & Bushway, S. (1997). Preventing crime: What works, what doesn’t, what’s promising. Washington, DC: Office of Justice Programs.
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from
Jewell, Wormith / DOMESTIC VIOLENCE TREATMENT ATTRITION 1113
Stalans, L. J., & Seng, M. (2007). Identifying subgroups at high risk of dropping out of domestic batterer treatment: The buffering effects of a high school education. International Journal of Offender Therapy and Comparative Criminology, 51, 151-169.
Statistics Canada. (2006). Measure violence against women: Statistical trends 2006. Ottawa Ontario, Canada: Author.*Taft, C. T., Murphy, C. M., Elliott, J. D., & Keaser, M. C. (2001). Race and demographic factors in treatment attendance for
domestically abusive men. Journal of Family Violence, 16, 385-400.*Taft, C. T., Murphy, C. M., Elliott, J. D., & Morrel, T. M. (2001). Attendance-enhancing procedures in group counselling
for domestic abusers. Journal of Counseling Psychology, 48, 51-60.Tjaden, P., & Thoennes, N. (2000). Full report of the prevalence, incidence, and consequences of violence against women:
Findings from the National Violence Against Women Survey (NCJ183781). Washington, DC: U.S. Department of Justice, National Institute of Justice.
Tollefson, D. R., Gross, E., & Lundahl, B. (2008). Factors that predict attrition from a state-sponsored rural batter treatment program. Journal of Aggression, Maltreatment, & Trauma, 17, 453-477.
Walker, A., Flatley, J., Kershaw, C., & Moon, D. (2009). Crime in England and Wales 2008/09: Findings from the British Crime Survey and police recorded crime (Home Office Statistical Bulletin vol. 1). London, England: Home Office.
Wilson, D. B. (2001). Meta-analytic methods for criminology. Annals of the American Academy Political and Social Science, 578, 71-89.
Wormith, J. S., & Olver, M. E. (2002). Offender treatment attrition and its relationship with risk, responsivity, and recidivism. Criminal Justice and Behavior, 29, 447-471.
Lisa M. Jewell is a PhD candidate in applied social psychology at the University of Saskatchewan. She is currently carrying out research in the areas of intimate partner violence, homelessness, and prejudice and discrimination directed toward sexual minorities.
J. Stephen Wormith is a professor in the Department of Psychology at the University of Saskatchewan. His research inter-ests include risk assessment, offender treatment, sexual offenders, gangs, and crime prevention.
at University of Bucharest on August 5, 2014cjb.sagepub.comDownloaded from