Does Change in Hostility Predict Sexual Recidivism?

Abstract

The purpose of the study was to examine whether scores on a widely used measure of hostility—the Buss–Durkee Hostility Inventory (BDHI)—and change on this measure predicted sexual recidivism in a sample of 120 adult male incarcerated sexual offenders. Pre- and posttreatment scores, simple difference scores, and clinically significant change were examined. The majority of participants had functional scores on the BDHI prior to treatment. Of those who had dysfunctional pretreatment scores, the majority remained unchanged. Higher posttreatment scores on the Assault and Verbal Hostility subscales significantly predicted sexual recidivism. The remaining pre- and posttreatment scores as well as change scores and classifications did not significantly predict sexual recidivism. Our findings suggest that the Assault and Verbal Hostility subscales may be useful for predicting sexual recidivism but were not clearly consistent with the notion that the BDHI assesses a dynamic risk factor(s) for sexual recidivism. Due to a number of limitations of the current study, however, more rigorous research is needed before firm conclusions can be drawn.

Keywords

hostility sexual offenders recidivism treatment change clinical significance

Hostility has been conceptualized as a cognitive style characterized by an interpretation of the social world as adversarial and dangerous (Polaschek & Ward, 2002; Wakeling & Barnett, 2011; Ward & Keenan, 1999). Hostile individuals tend to view those around them as consistently and intentionally malicious (Berkowitz, Jaffee, Jo, & Troccoli, 2001; Mann & Beech, 2003) even in the relative absence of concrete evidence (i.e., hostile attribution bias; Gannon, Collie, Ward, & Thakker, 2008). Hostility is also strongly related to grievance thinking (Thornton, 2002), involving excessive rumination on perceived injustices (e.g., thoughts of having been wronged, having been deprived of what one is owed, and of having suffered disproportionately compared with others (Malamuth, 1998; Thornton, 2002; Wakeling & Barnett, 2011). In short, hostility can be defined as a cluster of negative evaluations of, and feelings toward people that is fuelled by a perception of others as antagonistic and threatening (Ramírez & Andreau, 2006). Hostile individuals are thought to be hypersensitive to provocation and have a tendency to respond to perceived slights with physical, verbal, or indirect aggression (e.g., deliberately spreading malicious gossip; Buss & Durkee, 1957). High levels of hostility have been linked with borderline and antisocial personalities (Fossati et al., 2004; Hatzitaskos, Soldatos, Sakkas, & Stefanis, 1997; Paris, Zweig-Frank, Bond, & Guzder, 1996; Sinha & Watson, 2006), domestic assault, and more general violence (Gunn & Gristwood, 1975; Maiuro, Cahn, Vitaliano, Wagner, & Zegree, 1988).

Most multi-factor theories of sexual offending include hostility as a contributing factor to sexual offending (Gannon et al., 2008; Hall & Hirschman, 1991). For example, grievance thinking can support sexual aggression as a means to personal gratification and vindication (Thornton, 2002; Wakeling & Barnett, 2011). Hostile cognitions may also interfere with processes that could otherwise inhibit interpersonal aggression, such as empathy (Marshall & Moulden, 2001), or fuel affective states such as anger, which in turn can facilitate violence (Buss & Perry, 1992; Langton & Marshall, 2001; Pawlak, 1994; Polaschek & Ward, 2002; Ward, Hudson, Johnston, & Marshall, 1997). For example, perceived hostile intent (e.g., the belief that a woman declined the offering of a date to humiliate the man) may generate anger, ultimately facilitating reactive aggression in the form of sexual assault (Howells, Day, & Wright, 2004). Hostile cognitions, including viewing others as dangerous and grievance thinking, may also contribute to sexual interest in persons perceived as less threatening, such as children (Marziano, Ward, Beech, & Pattison, 2006). For some offenders, children may also provide the recognition and respect of which these men feel they have been unjustly deprived (Ward, 2000).

Consistent with these theories, hostility has been found to successfully distinguish sexual offenders from nonsexual offenders (e.g., Lee, Pattison, Jackson, & Ward, 2001) and nonoffenders (e.g., d = 0.30, 95% confidence interval [CI] = [0.05, 0.55]; see Whitaker et al., 2008, for a review) in a number of empirical studies. To put this and the effect sizes reported below in context, d values of 0.20, 0.50, and 0.80 are generally considered small, medium, and large effects, respectively (Cohen, 1988). A number of researchers have found that sexual offenders endorse more statements consistent with hostile masculinity—an approach to women characterized by coercive dominance (Malamuth, Sockloskie, Koss, & Tanaka, 1991)—and are more hostile toward women than nonsexual offenders (Marshall & Hambley, 1996; Marshall & Moulden, 2001; Milner & Webster, 2005; Rice, Chaplin, Harris, & Coutts, 1994). Hostility and grievance thinking tend to be present in offence accounts of sexual offenders against children (Marziano et al., 2006) and sexual offenders against adults (Mann & Hollin, 2007, 2010).

A number of studies have also found support for the contention that hostility is relevant to the prediction of sexual recidivism (Firestone, Nunes, Moulden, Broom, & Bradford, 2005; Wakeling, Freemantle, Beech, & Elliott, 2011). For example, recent meta-analytic findings suggest that hostility may be a modest but significant predictor of sexual recidivism (d = 0.17, 95% CI = [0.04, 0.31]; k = 9), any violent recidivism (d = 0.21, 95% CI = [0.08, 0.34]; k = 6), and general recidivism (d = 0.31, 95% CI = [0.19, 0.43]; k = 5; Hanson & Morton-Bourgon, 2004) among sexual offenders. Using one of the best known and commonly used measures of general hostility—the Buss–Durkee Hostility Inventory (BDHI; Buss & Durkee, 1957)—Kingston, Firestone, Wexler, and Bradford (2008) found support for hostility as a predictor of both sexual recidivism (d = 0.51, p < .05) and violent recidivism (d = 0.67, p < .01). However, findings have been equivocal across studies. For example, Wakeling and Barnett (2011) found that high scores on a measure of grievance thinking (i.e., the persistent rumination on perceived wrongdoings and thoughts of vengeance; Thornton, 2002) significantly predicted general, but not sexual recidivism. It is worth noting, however, that the very low base rate of sexual recidivism in Wakeling and Barnett’s sample may have contributed to the null findings (i.e., approximately 4.0% of the sample recidivated sexually [n = 12]).

If hostility is a dynamic risk factor for sexual recidivism, it should not only be predictive of recidivism and be changeable, but changes in levels of hostility should also be related to changes in sexual recidivism (Andrews & Bonta, 2010; Douglas & Skeem, 2005; Harris & Rice, 2003; Kraemer et al., 1997; Quinsey, Harris, Rice, & Cormier, 2006; Seto, 2008). Thus, if changes in hostility do not predict sexual recidivism, the causal role of hostility in the maintenance of sexual offending may be questioned (also see Nunes, Pettersen, Hermann, Looman, & Spape, 2014). Studies examining change in self-reported hostility over the course of treatment as a predictor of sexual recidivism are relatively rare and results have not been consistent (Beggs & Grace, 2011; Wakeling & Barnett, 2011). For example, Beggs and Grace (2011) examined whether treatment change on a battery of self-report measures predicted sexual recidivism among sexual offenders against children. To partial out the effects of the general tendency to report improvement at posttreatment, as well as the greater opportunity to show positive change among offenders with more dysfunctional scores at pretreatment, Beggs and Grace went beyond using simple difference scores (i.e., simply subtracting the posttreatment scores from the pretreatment scores) by also controlling for pretreatment scores. Decreases in self-reported general anger/hostility were associated with significantly lower rates of sexual recidivism (r = −.19, p < .01; Area Under the Receiver Operating Characteristic Curve [AUC] = .66, 95% CI = [0.55, 0.76]). Briefly, the AUC represents the probability that a randomly selected recidivist will have a higher score on a measure of anger/hostility than a randomly selected nonrecidivist, with a value of .50 indicating prediction of recidivism at a chance level, and a value of 1 indicating perfect prediction (Mossman, 2013). The AUC of .66 reported by Beggs and Grace is a medium effect size, corresponding to a Cohen’s d of 0.59 (Rice & Harris, 2005). In contrast to Beggs and Grace’s study, Wakeling and Barnett (2011) found that change in general hostility from pre- to posttreatment was not significantly related to sexual or general recidivism. However, as mentioned above, there were very few sexual recidivists in Wakeling and Barnett’s study, likely due to the relatively short follow-up period.

Clinical Significance

There are several different methods for assessing change over the course of treatment including simple difference scores and clinical significance. As noted above, a (simple) difference score is the subtraction of an individual’s posttreatment score from his pretreatment score. Clinical significance refers to examining whether a client has reached some target level of functioning during the course of treatment, and whether the amount of change is larger than what would be expected by chance alone (Jacobson, Follette, & Revenstorf, 1984; Nunes, Babchishin, & Cortoni, 2011). Although some sexual offender treatment studies have examined whether clinically significant change predicts sexual recidivism (Barnett, Wakeling, Mandeville-Norden, & Rakestrow, 2013; Beech, Fisher, & Beckett, 1999; Beech & Ford, 2006; Keeling, Rose, & Beech, 2006, 2007; Mandeville-Norden, Beech, & Hayes, 2008), to the best of our knowledge no published studies have reported the relationship between clinically significant change in hostility and sexual recidivism.

The method most commonly used to examine clinical significance was developed by Jacobson and colleagues (1984) and involves (a) defining a cutoff point to separate dysfunctional from functional scores, and (b) evaluating the magnitude of the change to verify that any improvement exceeds the margin of measurement error. Clinical significance has advantages over simple difference scores (for a review, see Nunes et al., 2011; Nunes et al., 2014). Although simple difference scores indicate the magnitude of change from pre- to posttreatment, they do not speak to the level of relative function or dysfunction associated with any given score. In the context of treatment change among sexual offenders, the utilization of difference scores enables one to address whether, and in what direction, change has occurred. However, whether treatment change on a psychological variable could be expected to relate to behavioural outcomes such as recidivism arguably depends in part on the level of dysfunction (i.e., treatment need) at pretreatment. For example, a sexual offender against adults who demonstrates significant treatment needs in the area of cognitive distortions, and who subsequently improves over the course of treatment, may still engage in severely distorted thinking about sex and women at posttreatment. An offender who improves over the course of treatment, but who nevertheless remains dysfunctional may therefore still be at increased risk of recidivism relative to an offender who demonstrates little or no change but who had less severe cognitive distortions at pretreatment. In short, the level of dysfunction at pretreatment and the relative function or dysfunction at posttreatment may be important to consider when examining the relationship between change on a predictor variable (e.g., hostility) and recidivism.

Current Study

To the best of our knowledge, the relationship between clinically significant change in hostility and sexual recidivism has not yet been reported in the published literature. The goal of the current study was to examine whether clinically significant change in hostility over the course of treatment predicted sexual recidivism. We examined the relationship between hostility and sexual recidivism using pre- and posttreatment scores, change as indexed by simple difference scores, as well as clinically significant change. We expected that greater hostility would be associated with higher rates of sexual recidivism and that clinically significant treatment gains in hostility would be associated with significantly lower rates of sexual recidivism.

Arguably, the utility of clinical significance analyses depends in part on the accuracy of the cutoff score that is used to categorize clients as functional or dysfunctional on the construct of interest. Recognizing the importance of accurate norms that reflect the range found in functional populations, some researchers have meta-analyzed norms of various subpopulations rather than simply basing the cutoff score on a single, homogeneous functional sample (Babchishin, Pettersen, Nunes, & Cortoni, 2011; Nunes et al., 2014). To obtain the most representative norms and reliability data on which to base the clinical significance analyses, we conducted a meta-analysis of available norms and reliability coefficients for the BDHI (Buss & Durkee, 1957).

Method

Participants

An archival data set of 431 federally incarcerated sexual offenders who had been assessed and/or treated at the Regional Treatment Centre Sex Offender Treatment Program (RTCSOTP) in Ontario, Canada, from 1996 to 2010 was used in the current study. Of these 431 participants, 120 participants had pre- and posttreatment scores on the BDHI, as well as recidivism data. Participants’ release dates ranged from 1997 to 2011. Participants consented to the use of their data for both clinical and research purposes. On average, participants had a high risk of sexual recidivism as estimated by a validated actuarial scale (Static-2002; Hanson & Thornton, 2003; see Table 1 for sample characteristics).

Table 1.

Characteristics of the Total Sample, and by Sexual Recidivism.

	Total sample (N = 120)		Sexual recidivism
	Total sample (N = 120)		No (n = 94)		Yes (n = 26)
	N	M (SD) or %	N	M (SD) or %	n	M (SD) or %
Age	113	37.49 (10.0)	94	38.31 (10.29)	26	34.09 (8.04)
Offenders against children	120	51.7%	94	51.1%	26	53.8%
Offenders against adults	120	42.5%	94	44.7%	26	34.6%
Mixed offenders	120	5.8%	94	4.3%	26	11.5%
Prior convictions
Sexual	113	4.56 (3.95)	91	4.66 (4.26)	22	4.14 (2.27)
Violent (any)	113	6.41 (4.42)	91	6.64 (4.63)	22	5.45 (3.29)
Static-2002*^a	88	7.93 (1.84)	67	7.63 (1.82)	21	8.90 (1.55)
Average follow-up (years)	120	7.82 (3.63)	94	7.49 (3.79)	26	9.0 (2.74)

Note. Recidivists had significantly higher estimated risk of recidivism than nonrecidivists. There were no other significant differences between recidivists and nonrecidivists. Prior violent convictions include sexual offences. CI = confidence interval.

d = −0.72, 95% CI = [−1.22, −0.22].

p < .05.

Measures

The BDHI

The BDHI is a 66-item true/false self-report measure of hostility and aggression (Buss & Durkee, 1957). The BDHI consists of seven subscales: Assault (physical violence), Indirect Hostility (undirected—for example, slamming doors—and indirect relational aggression—for example, spreading gossip), Irritability (anger and reactive aggression), Negativism (oppositional behaviour and defiance of authority), Resentment (perceptions of injustice, envy, and jealousy), Suspicion (projection of hostility onto others), and Verbal Hostility (verbal expression of negative affect; Buss & Durkee, 1957). Total scores can range from 0 to 66, with higher scores reflecting more hostility. Previous studies have generally found acceptable internal consistency for the BDHI total score (Vassar & Hale, 2009; 75-item version including the Guilt scale), with Cronbach’s alpha ranging from .67 to .96. The meta-analytically derived internal consistency coefficients for the standard 66-item BDHI total and subscales scores also suggested adequate internal consistency (see Table 2). In the current sample, item-level data were only available for a minority of offenders (24.2%). Thus, internal consistency of the BDHI total score was examined using the scores of all offenders with complete item-level data on the BDHI, regardless of whether they also had available recidivism data (α = .87; n = 86), as well as using only the subset of offenders included in the remaining analyses (α = .84; n = 29).

Table 2.

Meta-Analytically Determined Functional Mean Scores and Standard Deviations, Internal Consistency Coefficients, and Computed Cutoff Scores and Standard Errors of the Difference Between Post- and Pretreatment Scores (S_diff).

BDHI scale	Functional M and SD				Internal consistency			Cutoff B^a	S _diff ^b
BDHI scale	k	N	M ^c	SD ^d	α	k	n	Cutoff B^a	S _diff ^b
Total	13	964	28.7 [26.55]	8.41	.80	3	225	37.11	5.32
Assault	11	1,062	4.20 [4.06]	2.12	.73	2	333	6.32	1.85
Indirect hostility	11	1,062	4.08 [3.66]	1.56	.55	2	333	6.46	1.48
Irritability	11	1,062	4.81 [4.72]	1.11	.62	2	333	6.93	1.19
Negativism	11	1,062	2.47 [2.28]	2.12	.43	2	333	3.58	1.56
Resentment	11	1,062	2.65 [2.84]	1.34	.63	2	333	3.99	1.09
Suspicion	11	1,062	3.48 [3.66]	1.92	.67	2	333	5.40	1.65
Verbal hostility	11	1,062	6.25 [5.99]	2.18	.54	2	333	8.43	2.09

Note. BDHI = Buss–Durkee Hostility Inventory; RCI = reliable change index.

Functional mean score plus 1 standard deviation (a score below Cutoff B is functional).

Minimum change to be reliable (RCI > 1.64).

Random effects [fixed effects].

Fixed effects.

Static-2002

The Static-2002 is an actuarial risk appraisal measure of sexual and violent recidivism consisting of 10 static items (Hanson & Thornton, 2003). Total scores range from 0 to 12, with higher scores indicating greater risk of sexual recidivism. Recent meta-analytic results indicate that the Static-2002 has good predictive validity for sexual recidivism (AUC = .70; Helmus, Thornton, Hanson, & Babchishin, 2012). In the current study, available Static-2002 scores (n = 88) were included to examine the association between estimated risk of sexual recidivism, total, and subscale scores on the BDHI.

Sexual recidivism

Sexual recidivism data were coded from the Canadian Police Information Centre (CPIC) database. Sexual recidivism was coded as any postrelease conviction for a sexual offence (e.g., sexual interference, sexual assault, possession of child pornography). In addition, offences generally considered nonsexual would be coded as sexual recidivism if file information suggested that the offender’s motivation was sexual in nature. For example, a new conviction for trespassing would be considered a sexual offence if collateral information suggested that it was committed for the purpose of obtaining sexual gratification by secretly observing the woman living at the residence in question (i.e., “peeping”). Technical offences (e.g., breach of probation) were not considered instances of sexual recidivism. From this information, participants were categorized as sexual recidivists or nonrecidivists. Of note, postrelease convictions for historical sexual offences (i.e., pseudo-recidivism) were not coded as recidivism when adequate file information was available to distinguish them from actual recidivism, but we did not have access to such detailed information for all postrelease convictions; as a result, it is possible that some offenders may have been incorrectly classified as recidivists.

Clinically Significant Change Analyses

There are three types of cutoff scores that can be used in clinical significance analyses. Cutoff A is equal to the dysfunctional mean minus 2 standard deviations, Cutoff B is equal to the functional mean plus 2 standard deviations, and Cutoff C is the midpoint between functional and dysfunction means (for more detailed descriptions and formulas, see Jacobson et al., 1984; Jacobson & Truax, 1991; Nunes et al., 2011). Cutoff A and C are problematic for the current analyses because they require dysfunctional norms, but such norms are not readily available for the BDHI. Sexual offenders are heterogeneous (Beech, 1998; Knight & Prentky, 1990; Mandeville-Norden & Beech, 2009) and vary in areas and extent of dysfunction. As a result, not all sexual offenders would be expected to be dysfunctional on hostility (Howells et al., 2004). Thus, arguably, dysfunctional norms for the BDHI scales cannot be derived from average scores of sexual offender samples in the absence of collateral information. Using samples of violent offenders and individuals diagnosed with personality disorders would have drastically limited the number of studies meeting the criteria for inclusion outlined below (k = 4), with only one study providing both means and standard deviations for the BDHI total score (i.e., 66-item version with the Guilt subscale excluded). Moreover, even violent offenders may not consistently have dysfunctional levels of hostility. Varied results have been reported in the literature regarding the link between violent behaviour and hostility (Milner & Webster, 2005). For example, some studies using the BDHI found significant differences in total scores in the expected direction between violent and nonviolent groups (e.g., Lothstein & Jones, 1978), some found significant differences only on some of the subscales (e.g., Syvertson & Romney, 1985), while others found no group differences on the BDHI (e.g., Holland, Levi, & Beckett, 1983). With the limitations above in mind, Cutoff B appeared to be the most appropriate choice in the current context. Thus, we opted to use Cutoff B in the current analyses. We determined the functional mean scores through meta-analysis (described below).

Consistent with Wakeling et al. (2011), we modified Cutoff B to equal 1 standard deviation above the functional mean (vs. 2 standard deviations above the functional mean), thereby producing a more conservative cutoff by narrowing the operational definition of “functional.” This was necessary given that most offenders in our sample fell within the functional range even before treatment when the standard cutoff was used (see Table 2).

Reliable change index (RCI)

The RCI was computed using the formula presented by Jacobson and Truax (1991; also see Nunes et al., 2011, for a review). The RCI is an individual’s pretreatment score subtracted from his posttreatment score, divided by the standard error of difference. Following convention (e.g., Jacobson et al., 1984), the RCI was used to classify offenders as recovered, improved, unchanged, or deteriorated. To be classified as recovered, the individual’s posttreatment score had to fall below the functional cutoff and the amount of change from pre- to posttreatment had to be reliable. Offenders who did not fall within the functional range at posttreatment (scores above the functional cutoff) but who nevertheless showed reliable change in the desired direction (i.e., lower BDHI scores at post- than at pretreatment) were classified as improved. Offenders who did not show reliable change were classified as unchanged. Finally, offenders who showed reliable change but in the “wrong” direction (i.e., higher BDHI scores at post- than at pretreatment) were classified as deteriorated.

Participants with functional pretreatment scores were excluded from the clinically significant change analyses. Clinically significant change was originally designed for research on treatment change among psychotherapy clients. Consequently, these analyses generally rest on the assumption that all clients are dysfunctional on the construct of interest at pretreatment. In contrast, sexual offenders are heterogeneous and vary in area and extent of dysfunction at pretreatment but nevertheless participate in the same treatment program. This means that some offenders will likely have problematically high levels of hostility whereas others will not. The categories typically used in clinically significant change analyses cannot be applied to participants who are already functional on the construct at pretreatment, as some of these categories are impossible when participants start treatment in the functional range (i.e., improved category). Consequently, we have excluded these offenders from the clinically significant change analyses.

For modified Cutoff B, the standard error of difference was computed using the aggregated reliability coefficient and the aggregated standard deviation for the functional group from our meta-analysis (described below). RCIs greater than ±1.64 indicate significant change (p < .10). Given that the internal consistency of the BDHI total score was appreciably better in the current study relative to the meta-analyzed internal consistency coefficient, we used an RCI of larer than +1.64 or less than −1.64 (p < .10), rather than larger than +1.96 or less than −1.96 (p < .05), to determine thresholds for reliable change that would more accurately reflect the data at hand.¹

Meta-Analysis of Norms and Reliability Coefficients

Selection of studies

The following databases were systematically searched for empirical articles; PsycINFO, PubMed, and Web of Science using terms such as Buss–Durkee Hostility Inventory, BDHI, hostility, hostile attitudes, hostile cognition, and hostile personality. Additional articles were found using the reference lists of collected articles. The search was terminated in January 2013. To be included in the meta-analysis, the studies had to be published and report means and standard deviations or internal consistency coefficients for at least one of the BDHI subscales or for the total score. Only samples of adult men were included.

Means and standard deviations

Comprehensive meta-analysis, Version 2.0 (CMA; Borenstein, Hedges, Higgins, & Rothstein, 2005), was used to conduct the meta-analysis of the means and standard deviations. CMA produces weighted means and standard errors for both fixed effects and random effects models. Fixed effects models do not account for variation across studies as a function of sample size or other study design characteristics because they assume that any between-study variability is due to sampling error (Hedges & Vevea, 1998). Random effects models assume that there is natural variability between studies based on sample and study design that is not due to measurement error and incorporate this variability in the error term (Hedges & Vevea, 1998). In the current study, there may be “true” variability between groups that were included in the calculation of the functional means (e.g., university students and community men from a variety of cultural contexts may respond differently, despite both groups being included in the weighted mean of the functional group). For this reason, the weighted means of the random effects model were selected. The weighted standard errors were selected from the fixed effects model. CMA provides average standard errors that are weighted by the inverse of the variance. However, CMA does not provide the equivalent standard deviations. To obtain an average standard deviation, the weighted standard error was multiplied by the square root of the total sample size (i.e., number of units included in the original calculation of that standard error) using Microsoft Excel.

Coefficient alpha

The meta-analysis of coefficient alpha was conducted with SPSS, using fixed effects model syntax provided by Rodriguez and Maeda (2006). There is no evidence to suggest that the internal consistency of the BDHI (Buss & Durkee, 1957) varies consistently between samples and, thus, the fixed effects model was most appropriate. The mean coefficient ( $\bar{T}$ .) was then converted back to alpha by subtracting the coefficient, raised to the power of three, from 1.

Participant populations

For the derivation of functional norms (i.e., average scores from samples that represent populations considered functional on hostility), we excluded samples expected to have elevated levels of hostility or subtraits thereof (e.g., assault), such as violent offenders, patients with personality disorders associated with aggression (e.g., antisocial or borderline personality disorder, paranoid disorder, psychopathy; Reinehr, Swartz, & Dudley, 1984; Sinha & Watson, 2006), chronic substance and alcohol abusers (e.g., Moss, 1989), and domestic abusers (Maiuro et al., 1988; Valliant, De Wit, & Bowes, 2004). A total of eight studies that met initial inclusion criteria were excluded from meta-analyses for the following reasons: (a) standard scoring procedures of the BDHI had been altered (k = 2), (b) the Guilt subscale was included in the total score and subscale scores were not reported separately (k = 1), (c) the sample overlapped with that of an already included study (k = 1), or (d) reported means without standard deviations (k = 3). One additional study was excluded because it was an extreme outlier (i.e., the mean total scores reported in the study far exceeded those reported for populations expected to have very high levels of hostility, such as domestic abusers and violent psychiatric patients; see Hanson & Bussiére, 1998, for criteria used to determine the exclusion/inclusion of an outlier).

For the BDHI total score, a total of 13 studies (16 samples) met the above criteria and contained means, standard deviations, or internal consistency coefficients for functional populations, whereas 11 studies (14 samples) contained scores or coefficients for one or more subscale(s). Of note, only 2 studies of functional populations (e.g., university students) reported internal consistency coefficients. Thus, samples of men from populations otherwise excluded from the meta-analysis were included to obtain additional internal consistency estimates (see Table 2). There was considerable overlap between studies reporting internal consistency norms on subscales and on the total score (i.e., 3 studies reported internal consistency for one or several subscales only, while the remaining studies reported on both the subscales and total scores). Studies retained for analyses originated in Canada (k = 1), England (k = 1), Ireland (k = 1), Israel (k = 1), Italy (k = 2), Singapore (k = 1), and the United States (k = 8). Overall, 7 samples consisted of university students, 9 samples of community members, 3 of nonviolent offenders, and 1 of missionaries.

Results

Bivariate correlations between pre- and posttreatment total and subscale scores on the BDHI, the Static-2002, as well as prior sexual and violent convictions were examined (Tables 3 and 4). With few exceptions, BDHI total and subscale scores were not significantly associated with estimated risk of sexual recidivism (i.e., Static-2002) or prior sexual or violent offence convictions. This was the case for both pre- and posttreatment scores. One exception was the Negativism subscale scores, which correlated significantly and positively with estimated risk of sexual recidivism (Static-2002 scores) at pretreatment (Table 3) and at posttreatment (Table 4). Contrary to expectations, higher pretreatment scores on the Assault and Verbal Hostility subscales were significantly associated with fewer prior sexual offence convictions (Table 3).

Table 3.

Bivariate Correlations Between Pretreatment BDHI Total and Subscale Scores, the Static-2002, and Number of Prior Convictions for Sexual and Violent Offences.

Measure	1	2	3	4	5	6	7	8	9	10	11
1. BDHI total	—	.65**	.65**	.83**	.62**	.71**	.69**	.67**	.10	−.06	−.01
2. Assault	.63**	—	.29**	.40**	.36**	.29**	.27**	.53**	−.01	−.24*	−.09
3. Indirect hostility	.64**	.26**	—	.51**	.29**	.40**	.35**	.38**	.02	.07	.12
4. Irritability	.82**	.38**	.52**	—	.51**	.62**	.52**	.41**	.12	.03	.03
5. Negativism	.61**	.33**	.29**	.52**	—	.35**	.39**	.40**	.23*	−.01	−.01
6. Resentment	.73**	.31**	.39**	.62**	.37**	—	.66**	.25**	.10	.01	−.06
7. Suspicion	.72**	.28**	.35**	.54**	.41**	.68**	—	.26**	.10	.11	.04
8. Verbal hostility	.65**	.51**	.37**	.39**	.37**	.24**	.25**	—	.02	−.26**	−.12
9. Static-2002	.10	.00	.08	.11	.25*	.08	.11	.04	—	.21	.18
10. PSO	−.00	−.24*	.15	.05	−.03	.07	.16	−.17	.23*	—	.81**
11. PVO	.05	−.05	.16	.09	−.01	−.01	.09	−.05	.17	.76**	—

Note. Pearson’s coefficients are above the diagonal line. Spearman’s rho correlations are below the diagonal line. BDHI = Buss–Durkee Hostility Inventory. PSO = prior sexual offences; PVO = prior violent offences. PSO is a subset of PVO. PSO and PVO are continuous variables (see Table 1 for the means and standard deviations).

p < .05.

p < .01.

Table 4.

Bivariate Correlations Between Posttreatment BDHI Total and Subscale Scores, the Static-2002, and Number of Prior Convictions for Sexual and Violent Offences.

Measure	1	2	3	4	5	6	7	8	9	10	11
1. BDHI total	—	.78**	.66**	.88**	.72**	.75**	.74**	.68**	.15	−.08	−.02
2. Assault	.75**	—	.44**	.62**	.54**	.45**	.41**	.58**	.12	−.16	−.04
3. Indirect hostility	.65**	.43**	—	.53**	.43**	.37**	.32**	.36**	.09	−.01	.06
4. Irritability	.90**	.60**	.55**	—	.59**	.63**	.61**	.58**	.10	−.06	−.02
5. Negativism	.71**	.54**	.42**	.61**	—	.53**	.52**	.36**	.25*	.02	.03
6. Resentment	.77**	.49**	.39**	.65**	.54**	—	.68**	.28**	.20	−.10	−.13
7. Suspicion	.74**	.38**	.32**	.62**	.53**	.68**	—	.30**	.02	.02	.00
8. Verbal hostility	.69**	.57**	.37**	.59**	.37**	.34**	.30**	—	.09	−.08	−.02
9. Static-2002	.15	.11	.09	.09	.22*	.19	.00	.13	—	.21	.18
10. PSO	−.02	−.15	.04	.01	.05	−.07	.03	−.01	.23*	—	.81**
11. PVO	−.01	−.06	.06	.01	.01	−.12	.01	.03	.17	.76**	—

p < .05.

p < .01.

We examined whether total scores on the BDHI and its subscales predicted sexual recidivism by calculating Cohen’s d effect sizes, odds ratios, and their respective 95% CIs (see Table 5). With the exception of posttreatment scores on the Assault and Verbal Hostility subscales, none of the pre- and posttreatment or simple difference scores significantly predicted sexual recidivism.

Table 5.

Predicting Sexual Recidivism From Pretreatment, Posttreatment, and Difference Total and Subscale Scores.

BDHI scale	Nonrecidivists			Recidivists			OR [95% CI]	d [95% CI]
BDHI scale	M (SD)	Range^a	n	M (SD)	Range	n	OR [95% CI]	d [95% CI]
Total score			94			26
Pretreatment	28.43 (12.08)	7-59		32.19 (9.72)	13-51		1.03 [0.99, 1.07]	0.32 [−0.11, 0.76]
Posttreatment	27.31 (12.70)	4-53		31.79 (12.19)	12-58		1.03 [0.99, 1.07]	0.36 [−0.08, 0.79]
Difference score	−1.12 (10.09)	(−31.5)-(+34)		−0.40 (14.51)	(−37)-(+22)		1.01 [0.97, 1.05]	0.06 [−0.37, 0.50]
Assault			94			26
Pretreatment	3.44 (2.47)	0-9		4.04 (2.29)	0-9		1.11 [0.93, 1.32]	0.25 [−0.19, 0.68]
Posttreatment	3.13 (2.59)	0-10		4.38 (2.61)	0-10		1.19* [1.01, 1.40]	0.48* [0.04, 0.92]
Difference score	−0.31 (2.05)	(−5)-(+6)		0.35 (3.20)	(−6)-(+10)		1.12 [0.94, 1.35]	0.28 [−0.15, 0.72]
Indirect hostility			94			26
Pretreatment	4.29 (2.28)	0-9		5.12 (2.02)	2-9		1.18 [0.97, 1.45]	0.37 [−0.06, 0.81]
Posttreatment	4.44 (2.39)	0-9		4.98 (2.36)	1-9		1.10 [0.92, 1.33]	0.23 [−0.21, 0.66]
Difference score	0.15 (1.97)	(−5)-(+5)		−0.13 (2.67)	(−6)-(+6)		0.94 [0.76, 1.15]	−0.13 [−0.57, 0.30]
Irritability			94			26
Pretreatment	4.44 (2.75)	0-11		5.00 (2.83)	0-9		1.08 [0.92, 1.26]	0.20 [−0.23, 0.64]
Posttreatment	4.34 (2.78)	0-11		5.29 (2.74)	1-10		1.13 [0.97, 1.33]	0.34 [−0.09, 0.78]
Difference score	−0.10 (2.81)	(−7)-(+7)		0.29 (3.25)	(−8)-(+6)		1.05 [0.90, 1.22]	[−0.30, 0.57]
Negativism			94			26
Pretreatment	2.07 (1.36)	0-5		2.42 (1.51)	0-5		1.20 [0.88, 1.63]	0.25 [−0.18, 0.69]
Posttreatment	2.11 (1.46)	0-5		2.31 (1.52)	0-5		1.10 [0.82, 1.47]	0.14 [−0.30, 0.57]
Difference score	0.04 (1.26)	(−3)-(+5)		−0.12 (1.81)	(−5)-(+2.5)		0.92 [0.67, 1.26]	−0.11 [−0.55, 0.32]
Resentment			94			26
Pretreatment	2.80 (2.01)	0-8		3.44 (1.99)	0-8		1.17 [0.94, 1.45]	0.32 [−0.12, 0.76]
Posttreatment	2.64 (2.24)	0-8		3.02 (2.22)	0-8		1.08 [0.89, 1.31]	0.17 [−0.26, 0.60]
Difference score	−0.16 (2.12)	(−7)-(+5)		−0.42 (2.39)	(−8)-(+3.5)		0.95 [0.78, 1.15]	−0.12 [−0.55, 0.32]
Suspicion			94			26
Pretreatment	4.81 (2.77)	0-10		5.12 (2.30)	2-10		1.05 [0.89, 1.23]	0.12 [−0.32, 0.55]
Posttreatment	4.14 (2.81)	0-10		3.92 (2.92)	0-9		0.97 [0.83, 1.14]	−0.08 [−0.51, 0.36]
Difference score	−0.66 (2.15)	(−6)-(+5)		−1.19 (2.25)	(−6)-(+3)		0.90 [0.73, 1.09]	−0.24 [−0.68, 0.19]
Verbal hostility			94			26
Pretreatment	6.40 (2.82)	1-12		7.06 (2.42)	3.5-13		1.09 [0.93, 1.28]	0.24 [−0.19, 0.68]
Posttreatment	6.51 (2.69)	1-13		7.88 (1.93)	5-11		1.24* [1.04, 1.49]	0.54* [0.10, 0.98]
Difference score	0.11 (2.65)	(−8)-(+7)		0.83 (2.74)	(−5)-(+7)		1.11 [0.94, 1.31]	0.27 [−0.07, 0.71]

Note. BDHI = Buss–Durkee Hostility Inventory; OR = odds ratio; CI = confidence interval.

The BDHI is ordinarily scored yes = 2/no = 1. However, in the current archival data set, single missing items and illegible responses were replaced by a score of 0.5. This is reflected in the ranges for each group in Table 5. Analyses wherein all half-scores were rounded down, as well as analyses in which all scores were rounded up were conducted. These adjustments did not alter the outcome of any analyses. Thus, the existing scores were retained.

p < .05.

The meta-analyzed norms for the functional populations, as well as modified Cutoff B (functional mean plus 1 standard deviation) and the standard error of difference (S_diff) for the RCI for the total and subscale scores are presented in Table 2. When these cutoff scores were applied to the current data, 75% of the sample (n = 90) had total scores that fell within the functional range at pretreatment (see Table 6). Of the participants with dysfunctional pretreatment total scores, approximately one fourth recovered by posttreatment (i.e., their scores fell within the functional range at posttreatment and change over treatment was reliable). Specifically, of the 30 offenders with dysfunctional total scores at pretreatment, 23.3% recovered and 3.3% improved. The remaining 73.3% did not evidence reliable change over the course of treatment. None of the offenders with dysfunctional pretreatment total scores had deteriorated by posttreatment. The corresponding proportions for the BDHI subscales are also reported in Table 6.

Table 6.

Pretreatment Status on the BDHI Total and Subscale Scores by Clinically Significant Change.

BDHI scale	Posttreatment (%)					Functional at pretreatment n (%)
BDHI scale	n	Deteriorated	Unchanged	Improved	Recovered	Functional at pretreatment n (%)
Total score	30 (25.0)	0 (0.0)	22 (73.3)	1 (3.3)	7 (23.3)	90 (75.0)
Assault	19 (15.8)	0 (0.0)	14 (73.7)	0 (0.0)	5 (26.3)	101 (84.2)
Indirect hostility	38 (31.7)	0 (0.0)	32 (84.2)	0 (0.0)	6 (15.8)	82 (68.3)
Irritability	32 (26.7)	0 (0.0)	22 (68.8)	0 (0.0)	10 (31.3)	88 (73.3)
Negativism	23 (19.2)	0 (0.0)	16 (69.6)	0 (0.0)	7 (30.4)	97 (80.8)
Resentment	46 (38.3)	5 (10.9)	21 (45.7)	2 (4.3)	18 (39.1)	74 (61.7)
Suspicion	58 (48.3)	0 (0.0)	41 (70.7)	0 (0.0)	17 (29.3)	62 (51.7)
Verbal hostility	30 (25.0)	0 (0.0)	25 (83.3)	0 (0.0)	5 (16.7)	90 (75.0)

Note. BDHI = Buss–Durkee Hostility Inventory.

To assess whether participants with dysfunctional pre- or posttreatment scores differed from participants with functional pre- or posttreatment scores on sexual recidivism, we calculated odds ratios and 95% CIs. Functional versus dysfunctional pre- and posttreatment scores did not significantly predict sexual recidivism, either in terms of the total or subscale scores (Table 7).

Table 7.

Sexual Recidivism by Functional Versus Dysfunctional BDHI Total and Subscale Scores at Pre- and Posttreatment (N = 120).

BDHI scale	Pre- or posttreatment	Functional	Dysfunctional	OR [95% CI]
Total score	Pretreatment	20.0% (18/90)	26.7% (8/30)	1.46 [0.56, 3.80]
Total score	Posttreatment	20.2% (18/89)	25.8% (8/31)	1.37 [0.53, 3.57]
Assault	Pretreatment	21.8% (22/101)	21.1% (4/19)	0.96 [0.29, 3.18]
Assault	Posttreatment	21.6% (22/102)	22.2% (4/18)	1.04 [0.31, 3.47]
Indirect hostility	Pretreatment	18.3% (15/82)	28.9% (11/38)	1.82 [0.74, 4.46]
Indirect hostility	Posttreatment	19.2% (15/78)	26.2% (11/42)	1.49 [0.61, 3.63]
Irritability	Pretreatment	18.2% (16/88)	31.3% (10/32)	2.05 [0.81, 5.15]
Irritability	Posttreatment	19.1% (17/89)	29.0% (9/31)	1.73 [0.68, 4.43]
Negativism	Pretreatment	19.6% (19/97)	30.4% (7/23)	1.80 [0.65, 4.98]
Negativism	Posttreatment	20.6% (20/97)	26.1% (6/23)	1.36 [0.47, 3.89]
Resentment	Pretreatment	20.3% (15/74)	23.9% (11/46)	1.24 [0.51, 2.99]
Resentment	Posttreatment	21.3% (17/80)	22.5% (9/40)	1.08 [0.43, 2.69]
Suspicion	Pretreatment	22.6% (14/62)	20.7% (12/58)	0.89 [0.37, 2.14]
Suspicion	Posttreatment	20.8% (16/77)	23.3% (10/43)	1.16 [0.47, 2.83]
Verbal hostility	Pretreatment	21.1% (19/90)	23.3% (7/30)	1.14 [0.42, 3.05]
Verbal hostility	Posttreatment	20.9% (18/86)	23.5% (8/34)	1.16 [0.45, 3.00]

Note. BDHI = Buss–Durkee Hostility Inventory. OR = odds ratio; CI = confidence interval. Values in parentheses indicate, respectively, the number recidivists and the total number of offenders in each cell.

Next, we examined whether clinically significant change predicted sexual recidivism (Table 8). As mentioned above, clinically significant change analyses were restricted to those offenders who had dysfunctional pretreatment scores (i.e., who evidenced treatment needs in the area of hostility). Groups were collapsed due to the small number of participants in some change categories. Offenders who evidenced positive change over treatment (i.e., recovered or improved) were compared with offenders who did not evidence reliable change (i.e., unchanged). With the exception of the Resentment subscale (five offenders deteriorated and none of these recidivated), no offenders with dysfunctional pretreatment scores deteriorated by posttreatment. Change in BDHI total or subscale scores was not significantly associated with sexual recidivism in any analyses (see Table 8). Statistical significance aside, the odds ratio indicated that offenders in the recovered/improved group had higher odds of sexual recidivism than offenders who remained unchanged. For dichotomous predictors, odds ratios of 1.4 (and the reciprocal, 0.71, if the relationship is in the opposite direction), 2.3 (0.43), and 3.7 (0.27) are considered, respectively, small, medium, and large effect sizes (using the Cox approximation to d, Equation 18 from Sánchez-Meca, Chacón-Moscoso, & Marín-Martínez, 2003).

Table 8.

Clinical Significant Change in BDHI Total and Subscale Scores and Sexual Recidivism.

	Posttreatment
BDHI scale	n	Recovered	Improved	Unchanged	Deteriorated	OR [95% CI]
Total score	30	57.1% (4/7)	0.0% (0/1)	18.2% (4/22)	0.0% (0/0)	0.22 [0.38, 1.29]
Assault	19	40.0% (2/5)	0.0% (0/0)	14.3% (2/14)	0.0% (0/0)	0.25 [0.02, 2.58]
Indirect hostility	38	50.0% (3/6)	0.0% (0/0)	25.0% (8/32)	0.0% (0/0)	0.33 [0.06, 2.00]
Irritability	32	30.0% (3/10)	0.0% (0/0)	31.8% (7/22)	0.0% (0/0)	1.09 [0.22, 5.52]
Negativism	23	42.9% (3/7)	0.0% (0/0)	25.0% (4/16)	0.0% (0/0)	0.44 [0.09, 2.90]
Resentment	46	33.3% (6/18)	0.0% (0/2)	23.8% (5/21)	0.0% (0/5)	0.56 [0.14, 2.18]
Suspicion	58	17.6% (3/17)	0.0% (0/0)	22.0% (9/41)	0.0% (0/0)	1.31 [0.31, 5.59]
Verbal hostility	30	40.0% (2/5)	0.0% (0/0)	20.0% (5/25)	0.0% (0/0)	0.38 [0.05, 2.88]

Note. ORs for collapsed groups recovered/improved vs. unchanged/deteriorated. BDHI = Buss–Durkee Hostility Inventory. OR = odds ratio; CI = confidence interval. Values in parentheses indicate, respectively, the number recidivists and the total number of offenders in each cell.

Discussion

The purpose of the current study was to examine whether hostility, as assessed by the BDHI and change on this measure predicted sexual recidivism. Higher posttreatment scores on the Assault and Verbal Hostility subscales significantly predicted sexual recidivism, both with medium effect sizes. The remaining subscales had small effect sizes in the expected direction (d > 0.20 for pre- or posttreatment scores, except for scores on the Suspicion subscale) but did not significantly predict sexual recidivism. We also found that the majority of offenders already had functional BDHI scores at pretreatment (i.e., the proportion of the sample with functional total or subscale scores ranged from 52% to 84%) and that most of those with dysfunctional pretreatment scores did not change reliably over the course of treatment (46%-84%). Offenders with dysfunctional scores had nonsignificantly higher rates of sexual recidivism than offenders with functional scores for the BDHI total score and for almost all subscale scores at pretreatment and posttreatment; these effect sizes were small.

Simple difference scores and clinically significant change yielded opposite nonsignificant results for the relationship between change in hostility and sexual recidivism. Greater reductions in scores on the Assault and Verbal Hostility subscales as measured by simple difference scores showed small associations (d > 0.20) with lower sexual recidivism, but neither these nor any of the other simple difference scores significantly predicted sexual recidivism. Clinically significant change on the BDHI total or subscale scores also did not significantly predict sexual recidivism (recovered/improved vs. unchanged/deteriorated), but offenders who evidenced significant treatment gains surprisingly had much higher odds of sexual recidivism compared with offenders who remained unchanged or deteriorated. This was true of total scores and most subscale scores. These unexpected findings of the clinically significant change analyses should be interpreted with caution because, in addition to being nonsignificant, there were very few offenders in the recovered/improved categories (5-20). However, it is worth noting that the finding that offenders who appear to make clinically significant treatment gains have the highest recidivism rates is not unique to the current study. For example, Wakeling, Beech, and Freemantle (2013) found that offenders who improved (but did not recover) by posttreatment had higher recidivism rates compared even with those offenders who deteriorated.

Our findings suggest that the Assault and Verbal Hostility subscales of the BDHI may be useful for predicting sexual recidivism. However, our findings do not clearly support the notion that reduced hostility is associated with reduced rates of sexual recidivism. This is inconsistent with the view that hostility is a dynamic risk factor for sexual recidivism. As noted in the introduction section, a dynamic risk factor must not only be predictive of recidivism and be changeable, but changes on that variable must be related to changes in risk for recidivism (Andrews & Bonta, 2010; Douglas & Skeem, 2005; Harris & Rice, 2003; Kraemer et al., 1997; Quinsey et al., 2006; Seto, 2008). The current study has demonstrated that scores on the BDHI can change and that scores on the Assault and Verbal Hostility subscales predict sexual recidivism. However, it remains to be clearly demonstrated if improvement on these subscales is associated with reduced risk of sexual recidivism.

Although our findings provide little support for the utility of the BDHI for assessing risk-relevant treatment targets and progress, they also do not conclusively rule out the possibility that hostility is a causal dynamic risk factor. For example, perhaps treatment-associated reductions in hostility do reduce the likelihood of sexual recidivism, but only for as long as that reduced level of hostility is maintained. It is unknown if reductions would have been maintained over the longtime period between treatment completion and opportunity for sexual recidivism. A fairer test would require more frequent assessments to capture the level of hostility just prior to recidivism (e.g., Harris & Rice, 2003). Furthermore, given the small sample size of the current study and the limited number of sexual recidivists, focusing on effect sizes rather than statistical significance may be more appropriate. Effect sizes in the current study were generally small by conventional standards, but were nevertheless often slightly larger than those found for hostility in Hanson and Morton-Bourgon’s (2004) meta-analysis, and those generally considered meaningful in the sexual offender recidivism literature (e.g., Hanson & Morton-Bourgon, 2005; Mann, Hanson, & Thornton, 2010). This was also the case for the simple difference scores on the Assault and Verbal Hostility subscales. Although most of our results fell below the threshold for statistical significance, the nonnegligible effect sizes suggest that it may be worthwhile to examine these subscales and similar constructs in future research with larger samples.

In addition to the small sample size noted above, other limitations warrant cautious interpretation of our findings. The absence of a control group of untreated offenders is problematic because it is not possible to determine whether the change observed was caused by participation in the treatment program or some other factor, such as the passage of time or repeated assessment. Examining sexual offenders in the current sample without regard for victim-type or other potentially important distinctions may also have been problematic. For example, Firestone et al. (2005) found that the BDHI predicted sexual recidivism for sexual offenders against children but not for rapists or offenders with both child and adult victims. Combining these groups as we did in the current study may have attenuated the observed association between hostility and sexual recidivism. The sample size in the current study was too small to permit an examination of these subgroups, but future studies may benefit from examining more homogeneous groups of sexual offenders.

Finally, the selection of meaningful cutoff scores is an ongoing challenge in research on clinical significant change with sexual offenders. The cutoff score and, therefore, this method of analyzing change are only as good as the norms on which they are based (Jacobson & Truax, 1991). Although we argue that our meta-analytic approach yielded more representative and appropriate norms and cutoff scores than typical approaches in research on sexual offenders (e.g., Nunes et al., 2011), the cutoff scores may still be far from ideal. The main challenges are the considerable overlap in scores between sexual offenders and nonsexual offenders (e.g., Fiqia, Lang, Plutchik, & Holden, 1987) and the lack of norms for samples independently identified as having dysfunctional levels of hostility.

Given the limitations of the current study and the still frequent use of the BDHI and similar measures (e.g., the Aggression Questionnaire; Buss & Perry, 1992), additional research investigating the practical utility of these scales for the assessment of meaningful treatment change and risk of sexual recidivism is needed. To this end, norms gathered through our meta-analysis could prove useful and may provide a general guide to the establishment of norms for other measures. If our null findings were convincingly replicated, the BDHI would not be helpful for assessing risk-relevant treatment targets. However, scores on the subscales of the BDHI that pertain to overt behaviours, such as Assault and Verbal Hostility, may nevertheless have some utility in recidivism prediction.

Footnotes

Acknowledgements

We thank Kelly Babchishin for assistance with the meta-analysis.

Authors’ Note

The views expressed are those of the authors and do not necessarily represent the views of the Correctional Service of Canada.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported in part by grants from the Social Sciences and Humanities Research Council of Canada and funding from Carleton University.

Notes

References

Andrews

D. A.

Bonta

(2010). The psychology of criminal conduct (5th ed.). Cincinnati, OH: Anderson Publishing.

Babchishin

K. M.

Pettersen

Nunes

K. L.

Cortoni

(2011, June). Meta-analysis of functional and dysfunctional norms for measures of treatment targets for violent offender programs. In Cortoni

(Chair), Reducing recidivism among violent offenders: The effectiveness of treatment. Symposium conducted at the 2nd North American Correctional and Criminal Justice Psychology Conference, Toronto, Ontario, Canada.

Barnett

G. D.

Wakeling

H. C.

Mandeville-Norden

Rakestrow

(2013). How useful are psychometric scores in predicting recidivism for treated sex offenders? International Journal of Offender Therapy and Comparative Criminology, 56, 420-446. doi:10.1080/1068316X.2011.607820

Beech

A. R.

(1998). A psychometric typology of child abusers. International Journal of Offender Therapy and Comparative Criminology, 42, 319-339.

Beech

A. R.

Fisher

Beckett

(1999). STEP3: An evaluation of the prison sex offender treatment programme (Home Office Occasional Paper). London, England: Home Office.

Beech

A. R.

Ford

(2006). The relationship between risk, deviance, treatment outcome and sexual reconviction in a sample of child sexual abusers completing residential treatment for their offending. Psychology, Crime & Law, 12, 685-701. doi:10.1080/10683160600558493

Beggs

S. M.

Grace

R. C.

(2011). Treatment gain for sexual offenders against children predicts reduced recidivism: A comparative validity study. Journal of Consulting and Clinical Psychology, 79, 182-192. doi:10.1037/a0022900

Berkowitz

Jaffee

Troccoli

B. T.

(2001). On the correction of feeling induced judgmental biases. In Forgas

J. P.

(Ed.), Feeling and thinking: The role of affect in social cognition (pp. 131-152). Cambridge, UK: Cambridge University Press.

Borenstein

Hedges

Higgins

Rothstein

H. R.

(2005). Comprehensive Meta-Analysis (Version 2). Englewood, NJ: Biostat.

10.

Buss

A. H.

Durkee

(1957). An inventory for assessing different kinds of hostility. Journal of Consulting Psychology, 21, 343-349.

11.

Buss

A. H.

Perry

(1992). The Aggression Questionnaire. Journal of Personality and Social Psychology, 63, 452-459.

12.

Cohen

(1988). Statistical power analysis for the behavioral sciences (2nd ed.). New Jersey, NJ: Lawrence Erlbaum Associates.

13.

Douglas

K. S.

Skeem

J. L.

(2005). Violence risk assessment: Getting specific about being dynamic. Psychology, Public Policy, and Law, 11, 347-383. doi:10.1037/1076-8971.11.3.347

14.

Fiqia

N. A.

Lang

R. A.

Plutchik

Holden

(1987). Personality differences between sex and violent offenders. International Journal of Offender Therapy and Comparative Criminology, 31, 211-226.

15.

Firestone

Nunes

K. L.

Moulden

Broom

Bradford

J. M.

(2005). Hostility and recidivism in sexual offenders. Archives of Sexual Behavior, 34, 277-283. doi:10.1007/s10508-005-3116-8

16.

Fossati

Barratt

E. S.

Carretta

Leonardi

Grazioli

Maffei

(2004). Predicting borderline and antisocial personality disorder features in nonclinical subjects using measures of impulsivity and aggressiveness. Psychiatry Research, 125, 161-170. doi:10.1016/j.psychres.2003.12.001

17.

Gannon

T. A.

Collie

R. M.

Ward

Thakker

(2008). Rape: Psychopathology, theory and treatment. Clinical Psychology Review, 28, 982-1008. doi:10.1016/j.cpr.2008.02.005

18.

Gunn

Gristwood

(1975). Use of the Buss-Durkee Hostility Inventory among British prisoners. Journal of Consulting and Clinical Psychology, 43, 590.

19.

Hall

G. C. N.

Hirschman

(1991). Toward a theory of sexual aggression: A quadripartite model. Journal of Consulting and Clinical Psychology, 59, 662-669.

20.

Hanson

R. K.

Bussière

M. T.

(1998). Predicting relapse: A meta-analysis of sexual offender recidivism studies. Journal of Consulting and Clinical Psychology, 66, 348-362.

21.

Hanson

R. K.

Morton-Bourgon

K. E.

(2004). Predictors of sexual recidivism: An updated meta-analysis (Report No. 2004-02). Ottawa, Ontario: Public Safety Canada.

22.

Hanson

R. K.

Morton-Bourgon

K. E.

(2005). The characteristics of persistent sexual offenders: A meta-analysis of recidivism studies. Journal of Consulting and Clinical Psychology, 73, 1154-1163. doi:10.1037/0022-006X.73.6

23.

Hanson

R. K.

Thornton

(2003). Notes on the development of Static-2002 (Report No. 2003-01). Retrieved from http://www.publicsafety.gc.ca/cnt/rsrcs/pblctns/nts-dvlpmnt-sttc/nts-dvlpmnt-sttc-eng.pdf

24.

Harris

G. T.

Rice

M. E.

(2003). Actuarial assessment of risk among sex offenders. Annals of the New York Academy of Sciences, 989(1), 198-210. doi:10.1111/j.1749-6632.2003.tb07306.x

25.

Hatzitaskos

P. K.

Soldatos

C. R.

Sakkas

P. N.

Stefanis

C. N.

(1997). Discriminating borderline from antisocial personality disorder in male patients based on psychopathology patterns and type of hostility. Journal of Nervous and Mental Disease, 185, 442-446.

26.

Hedges

L. V.

Vevea

J. L.

(1998). Fixed- and random-effects models in meta-analysis. Psychological Methods, 3, 486-504.

27.

Helmus

Thornton

Hanson

K. R.

Babchishin

K. M.

(2012). Improving the predictive accuracy of the Static-99 and the Static-2002 with older sex offenders: Revised age weights. Sexual Abuse: A Journal of Research and Treatment, 24(1), 64-101. doi:10.1177/1079063211409951

28.

Holland

T. R.

Levi

Beckett

G. E.

(1983). Ethnicity, criminality, and the Buss-Durkee Hostility Inventory. Journal of Personality Assessment, 47, 375-379.

29.

Howells

Day

Wright

(2004). Affect, emotions and sex offending. Psychology, Crime & Law, 10, 179-195. doi:10.1080/10683160310001609988

30.

Jacobson

N. S.

Follette

W. C.

Revenstorf

(1984). Psychotherapy outcome research: Methods for reporting variability and evaluating clinical significance. Behavior Therapy, 15, 336-352.

31.

Jacobson

N. S.

Truax

(1991). Clinical significance: A statistical approach to defining meaningful change in psychotherapy research. Journal of Consulting and Clinical Psychology, 59, 12-19.

32.

Keeling

J. A.

Rose

J. L.

Beech

A. R.

(2006). A comparison of the application of the self-regulation model of the relapse process for mainstream and special needs sexual offenders. Sexual Abuse: A Journal of Research and Treatment, 18(4), 373-382. doi:10.1080/14789940600658293

33.

Keeling

J. A.

Rose

J. L.

Beech

A. R.

(2007). Comparing sexual offender treatment efficacy: Mainstream sexual offenders and sexual offenders with special needs. Journal of Intellectual & Developmental Disability, 32, 117-124. doi:10.1080/13668250701402767

34.

Kingston

D. A.

Firestone

Wexler

Bradford

J. M.

(2008). Factors associated with recidivism among intrafamilial child molesters. Journal of Sexual Aggression, 14, 3-18. doi:10.1080/13552600802074924

35.

Knight

R. A.

Prentky

R. A.

(1990). Classifying sexual offenders: The development and corroboration of taxonomic models. In Marshall

W. L.

Laws

D. R.

Barbaree

H. E.

(Eds.), The handbook of sexual assault: Issues, theories, and treatment of the offender (pp. 23-52). New York, NY: Plenum Press.

36.

Kraemer

H. C.

Kazdin

A. E.

Offord

D. R.

Kessler

R. C.

Jensen

P. S.

Kupler

D. J.

(1997). Coming to terms with the terms of risk. Archives of General Psychiatry, 54, 337-343. doi:10.1001/archpsyc.1997.01830160065009

37.

Langton

C. M.

Marshall

W. L.

(2001). Cognition in rapists: Theoretical patterns by typological breakdown. Aggression and Violent Behavior, 6, 499-518. doi:10.1016/S1359-1789(00)00029-X

38.

Lee

J. K. P.

Pattison

Jackson

H. J.

Ward

(2001). The general, common, and specific features of psychopathology for different types of paraphilias. Criminal Justice and Behavior, 28, 227-256. doi:10.1177/0093854801028002005

39.

Lothstein

L. M.

Jones

(1978). Discriminating violent individuals by means of various psychological tests. Journal of Personality Assessment, 42, 237-243.

40.

Maiuro

R. D.

Cahn

T. S.

Vitaliano

P. P.

Wagner

B. C.

Zegree

J. B.

(1988). Anger, hostility, and depression in domestically violent versus generally assaultive men and nonviolent control subjects. Journal of Consulting and Clinical Psychology, 56, 17-23. doi:10.1037/0022-006X.56.1.17

41.

Malamuth

N. M.

(1998). An evolutionary-based model integrating research on the characteristics of sexually coercive men. In Adair

J. G.

Belanger

Dion

K. L.

(Eds.), Advances in psychological science (Vol. 1, pp. 151-184). Hove, UK: Psychology Press.

42.

Malamuth

N. M.

Sockloskie

R. J.

Koss

M. P.

Tanaka

J. S.

(1991). Characteristics of aggressors against women: Testing a model using a national sample of college students. Journal of Consulting and Clinical Psychology, 59, 670-681.

43.

Mandeville-Norden

Beech

A. R.

(2009). Development of a psychometric typology of child molesters: Implications for treatment. Journal of Interpersonal Violence, 24, 307-325. doi:10.1177/0886260508316479

44.

Mandeville-Norden

Beech

A. R.

Hayes

(2008). Examining the effectiveness of a UK community-based sexual offender treatment programme for child abusers. Psychology, Crime & Law, 14, 493-512. doi:10.1080/10683160801948907

45.

Mann

R. E.

Beech

A. R.

(2003). Cognitive distortions, schemas, and implicit theories. In Ward

Laws

D. R.

Hudson

S. M.

(Eds.), Sexual deviance: Issues and controversies (pp. 135-153). Thousand Oaks, CA: SAGE.

46.

Mann

R. E.

Hanson

R. K.

Thornton

(2010). Assessing risk for sexual recidivism: Some proposals on the nature of psychologically meaningful risk factors. Sexual Abuse: A Journal of Research and Treatment, 22(2), 191-217. doi:10.1177/1079063210366039

47.

Mann

R. E.

Hollin

C. R.

(2007). Sexual offenders’ explanations for their offending. Journal of Sexual Aggression, 13, 3-9. doi:10.1080/13552600701365621

48.

Mann

R. E.

Hollin

C. R.

(2010). Self-reported schemas in sexual offenders. Journal of Forensic Psychiatry & Psychology, 21, 834-851. doi:10.1080/14789949.2010.511240

49.

Marshall

W. L.

Hambley

L.-S.

(1996). Intimacy and loneliness, and their relationship to rape myth acceptance and hostility toward women among rapists. Journal of Interpersonal Violence, 11, 586-592.

50.

Marshall

W. L.

Moulden

(2001). Hostility toward women and victim empathy in rapists. Sexual Abuse: A Journal of Research and Treatment, 13(4), 249-255. doi:10.1177/107906320101300403

51.

Marziano

Ward

Beech

A. R.

Pattison

(2006). Identification of five fundamental implicit theories underlying cognitive distortions in child abusers: A preliminary study. Psychology, Crime & Law, 12, 97-105. doi:10.1080/10683160500056887

52.

Milner

R. J.

Webster

S. D.

(2005). Identifying schemas in child molesters, rapists, and violent offenders. Sexual Abuse: A Journal of Research and Treatment, 17(4), 425-439. doi:10.1007/s11194-005-8053-5

53.

Moss

H. B.

(1989). Psychopathy, aggression, and family history in male veteran substance abuse patients: A factor analytic study. Addictive Behaviors, 14, 565-570.

54.

Mossman

(2013). Evaluating risk assessments using receiver operating characteristic analysis: Rationale, advantages, insights, and limitations. Behavioral Science & the Law, 31, 23-39. doi:10.1002/bsl.2050

55.

Nunes

K. L.

Babchishin

K. M.

Cortoni

(2011). Measuring treatment change in sex offenders: Clinical and statistical significance. Criminal Justice and Behavior, 38, 157-173. doi:10.1177/0093854810391054

56.

Nunes

K. L.

Pettersen

Hermann

Looman

Spape

(2014). Does change on the RAPE and MOLEST scales predict recidivism? Sexual Abuse: A Journal of Research and Treatment. Advance online publication. doi:10.1177/1079063214540725

57.

Paris

Zweig-Frank

Bond

Guzder

(1996). Defense styles, hostility, and psychological risk factors in male patients with personality disorders. Journal of Nervous and Mental Disease, 181, 153-158.

58.

Pawlak

A. E.

(1994). Factors associated with sexual aggression among rapists and non-offenders (Doctoral dissertation). Carleton University, Ottawa, Ontario, Canada.

59.

Polaschek

D. L. L.

Ward

(2002). The implicit theories of potential rapists: What our questionnaires tell us. Aggression and Violent Behavior, 7, 385-406. doi:10.1016/S1359-1789(01)00063-5

60.

Quinsey

V. L.

Harris

G. T.

Rice

M. E.

Cormier

C. A.

(2006). Violent offenders: Appraising and managing risk (2nd ed.). Washington, DC: American Psychological Association.

61.

Ramírez

J. M.

Andreau

J. M.

(2006). Aggression, and some related psychological constructs (anger, hostility, and impulsivity) some comments from a research project. Neuroscience & Biobehavioral Reviews, 30, 276-291. doi:10.1016/j.neubiorev.2005.04.015

62.

Reinehr

R. C.

Swartz

J. D.

Dudley

H. K.

(1984). Ethnic differences in the measurement of hostility in forensic patients. Revista Interamericana de Psicología, 18, 53-64.

63.

Rice

M. E.

Chaplin

T. C.

Harris

G. T.

Coutts

(1994). Empathy for the victim and sexual arousal among rapists and nonrapists. Journal of Interpersonal Violence, 9, 435-449.

64.

Rice

M. E.

Harris

G. T.

(2005). Comparing effect sizes in follow-up studies: ROC Area, Cohen’s d, and r. Law and Human Behavior, 29, 615-620. doi:10.1007/s10979-005-6832-7

65.

Rodriguez

M. C.

Maeda

(2006). Meta-analysis of coefficient alpha. Psychological Methods, 11, 306-322.

66.

Sánchez-Meca

Chacón-Moscoso

Marín-Martínez

(2003). Effect-size indices for dichotomized outcomes in meta-analysis. Psychological Methods, 8, 448-467. doi:10.1037/1082-989X.8.4.448

67.

Seto

M. C.

(2008). Pedophilia and sexual offending against children: Theory, assessment, and intervention. Washington, DC: American Psychological Association.

68.

Sinha

B. K.

Watson

(2006). Hostility and personality disorder. Imagination, Cognition and Personality, 25, 45-57.

69.

Syvertson

K. L.

Romney

D. M.

(1985). A further attempt to differentiate violent from non-violent offenders by means of a battery of psychological tests. Canadian Journal of Behavioural Science, 17, 87-92.

70.

Thornton

(2002). Constructing and testing a framework for dynamic risk assessment. Sexual Abuse: A Journal of Research and Treatment, 14, 139-153. doi:10.1177/107906320201400205

71.

Valliant

P. M.

De Wit

Bowes

(2004). Cognitive and personality factors associated with assaultive and domestic offenders. Psychological Reports, 94, 1180-1184. doi:10.1177/0886260508314931

72.

Vassar

Hale

(2009). Reliability reporting across studies using the Buss-Durkee Hostility Inventory. Journal of Interpersonal Violence, 24(1), 20-37. doi:10.1177/0886260508314931

73.

Wakeling

H. C.

Barnett

(2011). Measuring grievance thinking in sexual offenders: The revised Dissipation Rumination Scale. Journal of Sexual Aggression, 17, 273-289. doi:10.1080/13552600.2010.495422

74.

Wakeling

H. C.

Beech

A. R.

Freemantle

(2013). Investigating treatment change and its relationship to recidivism in a sample of 3773 sex offenders in the UK. Psychology, Crime & Law, 19, 233-252. doi:10.1080/1068316X.2011.626413

75.

Wakeling

H. C.

Freemantle

Beech

A. R.

Elliott

I. A.

(2011). Identifying predictors of recidivism in a large sample of United Kingdom sexual offenders: A prognostic model. Psychological Services, 8, 307-318. doi:10.1037/a0025516

76.

Ward

(2000). Sexual offenders’ cognitive distortions as implicit theories. Aggression and Violent Behavior, 5, 491-507. doi:10.1016/S1359-1789(98)00036-6

77.

Ward

Hudson

S. M.

Johnston

Marshall

W. L.

(1997). Cognitive distortions in sex offenders: An integrative review. Clinical Psychology Review, 17, 479-507.

78.

Ward

Keenan

(1999). Child molesters’ implicit theories. Journal of Interpersonal Violence, 14, 821-838.

79.

Whitaker

D. J.

Hanson

K. R.

Baker

C. K.

McMahon

P. M.

Ryan

. . . Rice

(2008). Risk factors for the perpetration of child sexual abuse: A review and meta-analysis. Child Abuse & Neglect, 32, 529-548. doi:10.1016/j.chiabu.2007.08.005