PREDICTING ADULT OFFENDER RECIDIVISM: WHAT WORKS!

1996-07

By

Paul Gendreau, Claire Goggin & Tracy Little

University of New Brunswick

The views expressed are those of the authors and are not necessarily those of the Ministry of the Solicitor General of Canada.

This document is available in French. Ce rapport est disponible en français sous le titre: Les techniques efficaces de prévision de la récidive chez les délinquants adultes.

Public Works and Government Services Canada
Cat. No. JS4-1/1996-7E
ISBN: 0-662-25114-8

TABLE OF CONTENTS

Abstract

Executive Summary

Predicting Adult Offender Recidivism: WHAT WORKS!

Predictors of Recidivism

Actuarial Measures for Predicting Recidivism

METHOD AND PROCEDURE

Sample of Studies

Coding the Studies

Predictor Categories

Effect Size Calculation

Significance Testing

RESULTS

Predictor Domains: Category I

Predictor Domains: Category II

Actuarial Measures

DISCUSSION AND RECOMMENDATIONS

Predictor Domains

Actuarial Measures for Predicting Recidivism

Practical Suggestions for Practitioners

APPENDIX A

References

References in the Meta-Analysis

Abstract

Meta-analytic techniques were used to determine which predictor domains and actuarial assessment instruments were the best predictors of adult offender recidivism. One hundred and thirty-one studies produced 1141 correlations with recidivism. The strongest predictor domains were criminogenic needs, criminal history/history of antisocial behaviour, social achievement, age/gender/race and family factors. Less robust predictors included intellectual functioning, personal distress factors and socio-economic status in the family of origin. Dynamic predictor domains performed at least as well as the static domains. Recommendations for developing sound assessment practices in corrections were provided.

Executive Summary

Correctional policy makers and practitioners are faced with noticeable increases in prison populations, burgeoning probation caseloads, uncertain parole assessment guidelines, and the need to design more effective offender treatment programs. Resolution of these issues is indeed difficult. The literature on the assessment and prediction of criminal behaviour, however, should provide some useful answers in this regard.

Therefore, the purpose of this research project was to provide data on five basic questions regarding the assessment and prediction of criminal behaviour, (i.e., recidivism).

The questions were:

1. Which predictor domains predict recidivism and are some more potent than others?
2. Are dynamic predictors as a group inferior to static predictors in their ability to predict recidivism?
3. Are there differences amongst composite measures of risk prediction instruments and measures of antisocial personality in their ability to predict recidivism?
4. Are the most robust predictors of recidivism associated with different theories of criminal behaviour?
5. What practical guidelines are forthcoming from this research that will assist practitioners in making better assessments of criminal behaviour?

The necessary information was generated from a meta-analysis of the prediction of recidivism literature among adult offenders. One hundred and thirty-one studies were identified which produced 1141 correlations with recidivism.

The strongest predictors of recidivism were criminogenic need, criminal history/history of antisocial behaviour, social achievement, age/gender/race, and family factors. Weaker predictors included intellectual functioning, personal distress (i.e., anxiety, self-esteem), and social class of origin.

Dynamic predictors - those that measure change in the offender - predicted recidivism as well as static predictors such as age and criminal history.

Amongst the composite risk measures, the Level of Service Inventory (LSI-R) generated higher correlations with recidivism than did other risk measures (e.g., Salient Factor Score, Wisconsin) and measures of antisocial personality. In the case of measures of antisocial personality, the Psychopathy Check List (PCL-R) was superior to the MMPI Pd scale/Megargee system and various other antisocial personality scales. Differential association and social learning theories of criminal conduct were associated with the strongest predictors of recidivism.

Several practical guidelines emanated from this meta-analysis that will assist criminal justice practitioners in their goal of reducing prison overcrowding, managing probation and parole caseloads effectively, and designing better treatment programs.

1. The "ideal" assessment protocol should include, whenever possible, the following content areas.

Static Predictors

1. age
2. criminal history - both as an adult and juvenile and history of antisocial behaviour when the offender was a youth
3. family factors - criminality, rearing practices, and structure

Dynamic Predictors

4. antisocial personality
5. companions
6. criminogenic needs
7. interpersonal conflict
8. social achievement
9. substance abuse

2. The available risk instrument that is closest to ideal is the LSI-R. Revisions to existing measures such as the LSI-R or the development of new instruments, should incorporate items that tap into the offender's and his/her family's early antisocial history. The assessment of antisocial or psychopathic personality may be best left to a separate assessment protocol using the PCL-R.

3. The measurement of change - assessing offenders at different points in time - should be done routinely.

4. The magnitude of the prediction of recidivism by any of the predictor domains appears to be little affected by the choice of recidivism outcome employed. Reconviction should be used in any circumstance but, obviously, parole violation and re-incarceration should be gathered by parole and prison authorities respectively.

5. Any correctional agency that has the professional integrity to: (a) compare the ability of different assessment measures (e.g., LSI-R vs. PCL-R) to predict recidivism, (b) assess the usefulness of new techniques (e.g., Neutralization scale), (c) generate prediction data on promising predictor domains and distinct groups of offenders (e.g., violent offenders), will make a substantial contribution to our knowledge base, which, in turn, will benefit practice in the field.

Predicting Adult Offender Recidivism: What Works!

The efficient management of prisons, probation and parole service and the development and evaluation of treatment programs is contingent, in part, upon the adequacy of our knowledge concerning the predictors of criminal behaviour. There are two questions in this regard. That is, what risk factors are the most potent predictors of recidivism and which actuarial instruments are best suited to that assessment?

In the case of prisons, the United States is the leader in incarceration rates in the Western world, while Canada is usually ranked in the top three (Mauer, 1994; Staff, 1993). It is estimated that 4,000,000 offenders will be imprisoned in the United States by the year 2000 (DiIulio, 1991). The continuing "war on drugs" and the recent "three strikes and out" bill proposed by President Clinton and enthusiastically adopted by many U.S. states may well prove DiIulio's pessimistic projection to be accurate. In Canada, as compared to the U.S., there is less punitiveness in sentencing for most crimes (Lynch, 1993), but the trends for "getting tough" are emerging. Changes in juvenile justice legislation and judges' sentencing practices, and the tightening of parole guidelines are increasing Canada's incarceration rate perceptibly (Leschied & Gendreau, 1994; Moon, 1995).

Thus, it comes as no surprise that prison overcrowding will likely become worse. As a consequence, the ability of prison systems to manage themselves in a humane and cost-efficient manner is being seriously jeopardized. Two strategies have been suggested which will help alleviate the stresses in this matter. First, medium and maximum security institutions should be reserved only for those offenders who are identified as being among the highest risk to re-offend. Lower risk inmates should be transferred to minimum security prison placements or, more preferably, to community based correctional facilities. Secondly, at sentencing, lower risk offenders can be diverted to probation, thereby avoiding the use of prisons altogether.

Similarly, probation and parole caseloads have increased by approximately 160% in the U.S. during the last decade (Austin, 1995). These unprecedented increases have occurred at a time when some of the largest states (e.g., California, New York), are cutting staff (Mencimer, 1993). Probation officer caseloads have reached several hundred in some jurisdictions (e.g., California). One partial solution to probation's dilemma is to restrict officers' supervisory practices to offenders designated as high risk. Parole boards find themselves in a Catch 22 situation. They are under pressure to reduce prison overcrowding by granting more parole while, at the same time, responding to the public's and politicians' demands that high risk offenders remain incarcerated.

The effectiveness of offender treatment programs also depends on our knowledge of the predictors of recidivism (Gendreau et al., 1994). In this regard, Andrews and Bonta (1994) developed an elegant risk/need/responsivity theory of criminal conduct that links prediction with treatment. We will focus on the risk and need principles as they have the strongest implications for prediction.

The risk principle has two components. First, it states that treatment will be effective when treatment services are matched with the offender's risk of re-offending. A risk factor can be anything about the offender's past or present circumstances and behaviour that is predictive of criminal behaviour. Intensive services should be provided for higher risk offenders and minimal services for lower risk offenders. Mismatching level of service with offender risk has seldom shown reductions in offender recidivism, in fact, in some studies increases have been reported (Andrews et al., 1990a; Andrews et al., 1990b).

Secondly, Andrews and Bonta (1994) classified risk factors into two categories: static and dynamic. Static risk factors (i.e., age, previous convictions, early family factors) are aspects of the offender's past that are predictive of recidivism but are not subject to change. On the other hand, dynamic risk factors or needs reflect the present circumstances and behaviour of the offender, and, as such, are mutable. There are two types of offender needs: criminogenic and noncriminogenic. Examples of criminogenic needs are offenders' attitudes, cognitions, and behaviour regarding employment, education, peers, authority, substance abuse and interpersonal relationships that lead to conflict with the law. The importance of criminogenic needs rests in the promise that when treatment programs target criminogenic needs, reductions in offender criminal behaviour can reasonably be expected to occur (Andrews et al., 1990a).

The evaluation of treatment programs depends on the availability of accurate measures of risk. While randomized experiments are difficult to achieve in corrections agencies, quasi-experimental designs utilizing comparison groups can be readily created if offenders have been assessed as to their risk level. In support of the point is that the Andrews et al., (1990b) meta-analysis found the quality of the research design was a minor factor in assessing the effectiveness of services. They endorsed the use of evaluations that controlled for pretreatment risk levels if randomized experiments were not possible.

As noted at the outset, the achievement of the above goals is predicated on the assumption that practitioners and policy makers can arrive at a consensus as to what are the best predictors of recidivism and the most accurate prediction instruments available to assess recidivism based on a reading of the relevant criminal justice literature. We now assess the status of this assumption.

Predictors of Recidivism

There is little disagreement in the criminological literature about some of the predictors of adult offender recidivism such as age, gender, past criminal history, early family factors and criminal associates. There has been, however, considerable controversy and/or lack of interest in dynamic risk factors. There are three reasons for this. First, because of ideological concerns and the professional self-interest of significant segments of the professions of criminology and sociology, the import of individual differences (i.e., offender needs, abilities, attitudes, and personality styles) has been derided in some criminological literature (cf., Andrews & Wormith, 1989; Rowe & Osgood, 1984; Wilson & Herrnstein, 1985).

Secondly, some methodologists (e.g., Jones, 1996) have expressed scepticism about dynamic risk factors because of their supposed unreliability. Unlike their static counterparts, dynamic risk factors can change over time and their measurement involves some degree of subjectivity. Since elementary psychometric theory reminds us that unreliability in measurement necessarily leads to an underestimation of validity (Cronbach, 1990), this line of reasoning implies that, collectively, dynamic variables must be relatively weak predictors of criminal behaviour.

Thirdly, criminal justice professionals have been, by and large, seemingly oblivious to the possibility that assessment of criminogenic needs might enhance the prediction of criminal behaviour (Bonta, 1996; Gendreau & Ross, 1987). The widely used Wisconsin classification system (Baird, 1981) illustrates this point. This instrument contains a useful needs component, but Bonta (1996) found just two studies that reported on the predictive validity of these items. Furthermore, the emergence of the "new penology" (Feeley & Simon, 1992), which is concerned with managing large aggregates of offenders in a simplistic input-output business-like fashion, has further contributed to the lack of interest in dynamic variables.

This denial of the utility of dynamic risk factors, obviously, has serious ramifications for corrections professionals who are routinely required to reclassify offenders for prison transfers, parole/probation supervision, and treatment services. Simply put, reclassification is devalued if the measurement of change has little validity.

Three specific types of predictors have also been the subject of much debate. They are social class of origin, intelligence, and personal distress. Social class of origin (i.e., parents occupation, education), has been the bedrock variable used in support of sociological theories of crime that assert that criminal behaviour is determined largely by one's social location (cf., Andrews & Bonta, 1994). Tittle and Meier (1990; 1991) have challenged this view, showing social class of origin (SES) to be a very weak predictor of juvenile delinquency.

The notion that criminals are less intelligent than nonoffenders has been prevalent for decades (cf., Goddard, 1920). Over the years, a number of studies have demonstrated a correlation between intelligence and delinquency (Hirschi & Hindelang, 1977). Recently, with the publication of The Bell Curve (Herrnstein & Murray, 1994), arguably the strongest claim yet has been made that IQ is a particularly powerful predictor. Their conclusions have implications for the provision of treatment programs for offenders, since IQ, in their view, is considered to be largely immutable.

According to Andrews et al. (1990a) personal distress variables (e.g., low self-esteem, anxiety) are not risk factors and are, therefore, inappropriate targets for treatment. Their conclusions are in stark contrast to the practices of many therapists and programs that give priority to lowering offenders' anxiety level and raising their self-esteem. The genesis of this perspective is, most likely, a consequence of the training received in mental health theory and practise (e.g., psychodynamic theory, phenomenology), where treatment professionals initially gained experience before emigrating to corrections in the 1960s (Gendreau, 1996). The current widespread popularity of the recovery and self-help agendas (see Kaminer, 1992) lends further credibility to the notion that personal distress factors are suitable targets for intervention, a view which has, in our opinion, generalized to corrections where surveys of treatment programs have found that it is not uncommon for programs to attempt to alleviate offenders' personal distress (Gendreau et al., 1990; Hoge et al., 1993).

To date, reviews of the evidence concerning the predictors of recidivism have been limited in scope and have been narrative in nature - except for two reviews that employed meta-analytic procedures. One meta-analysis, however, was quite preliminary (Gendreau et al., 1992) while the other was restricted to twin and adoption studies that combined juvenile and adult samples (Walters, 1992).

Actuarial Measures for Predicting Recidivism

Bonta (1996) has categorized risk assessment measures within a developmental framework. First generation techniques are based on clinical intuition and professional judgement. There is a plethora of literature documenting the lack of validity of this approach (cf., Meehl, 1954), even amongst the most highly trained clinicians and scholars (Little & Schneidman, 1959). This perspective is still commonplace among corrections professionals (Clear & Gallagher, 1985).

Second generation assessments are actuarial in nature. They are based on standardized, objective risk prediction instruments, such as the Salient Factor Score (SFS) (Hoffman, 1983) that are based almost entirely on static criminal history items. These kinds of measures provide little direction for classification and treatment decisions because the fixed nature of the items does not allow for changes in the offenders behaviour to be reflected in subsequent re-testing.

Bonta's third generation consists of two types of instruments. One of them encompass risk prediction measures that include dynamic factors (e.g., Community Risk/Needs Management Scale, Motiuk, 1993; Level of Service Inventory (LSI-R), Andrews & Bonta, 1995; the Wisconsin, Baird, 1981) which assess a wide range of criminogenic needs. The second type includes personality test scales in the antisocial personality/sociopathy/psychopathy content area. While these scales (e.g., the MMPI Pd scale, the Psychopathy Checklist (PCL-R), Hare, 1991; the Socialization scale (Soc) of the California Personality Inventory (CPI), Gough, 1957) do contain static items, the majority are dynamic in nature.

Reviews of the risk measure literature have also been, with one exception (Simourd et al., 1991), narrative in nature. Their meta-analysis reported that the PCL-R and the Soc scale of the CPI were better predictors of recidivism than the MMPI Pd scale. Unfortunately, most of the studies available to the authors were postdictive.

A final comment concerns the fact that the validity of various theories of criminal behaviour relies, somewhat, on the prediction literature. Anomie/strain (Merton, 1957) and sub-cultural theories (Cohen, 1955; Matza, 1964) support SES and, to some extent, personal distress, as strong predictors. Contemporary reformulations of differential association, social learning, and control theories (Andrews & Bonta, 1994; LeBlanc et al., 1988; Widom & Toch, 1993) centre on antisocial peers, learned antisocial values, early criminogenic family factors, and personality dimensions (e.g., egocentricity). Strong biologically oriented theories base much of their credence on IQ and twin studies (see Herrnstein & Murray, 1994; Walters, 1992).

In summary, our review of the predictors of recidivism literature for adult offenders has indicated a need for a comprehensive, quantitative research synthesis (i.e., meta-analysis) of the major classes of the predictors of recidivism and of the available prediction instruments. The potential advantages of meta-analysis over narrative reviews have been summarized in detail elsewhere (Cooper & Hedges, 1994). It has become the review method of choice in many applied areas (e.g., Lipsey & Wilson, 1993) and has recently led to advances in knowledge in the correctional field (Andrews et al., 1990b; Bonta & Gendreau, 1990; Gendreau & Andrews, 1990; Lipsey, 1992; Walters, 1992).

The questions we address in this study are as follows:

1. Which predictor domains predict recidivism and are some more potent than others?
2. Are dynamic predictors as a group inferior to static predictors in their ability to predict recidivism?
3. Are there differences amongst composite measures of risk prediction instruments and measures of antisocial personality in their ability to predict recidivism?
4. Are the strongest predictors of recidivism associated with different theories of criminal behaviour?
5. What guidelines are forthcoming from the meta-analysis that will assist criminal justice professionals in making more accurate assessments of criminal behaviour?

METHOD AND PROCEDURE

Sample of Studies

A literature search for relevant studies published between January 1970 and June 1994 was conducted using the ancestry approach and library abstracting services. For a study to be included, the following criteria applied:

1. Data on the offender was collected prior to the recording of the criterion measures. A minimum follow-up period of six months was required. If a study reported more than one follow up period, data from the longest interval was used.
2. Treatment studies that directly attempted to change offender personality or behaviour were not included.
3. The criterion or outcome measure of recidivism had to be recorded when the offender was an adult (18 years or more).
4. The criterion or outcome measure had to have a no-recidivism category. Studies that used "more" vs. "less" crime categorizations were not used. The criterion measures were arrest, conviction, incarceration, parole violation or a combination thereof.
5. The study was also required to report statistical information that could be converted, using meta-analytic formulae (Rosenthal, 1991) into the common metric or effect size of Pearson r.

Coding the Studies

For each study the following information was recorded:

1. Coder characteristics: date, coder identity.
2. Study characteristics: published document, type of publication, funding source, multi-disciplinary authorship, judgement of senior author's knowledge of the area, gender of authors, affiliation of authors, geographic location of study, decade in which study was published.
3. Study sample characteristics: age, gender, race, urban/rural, socio-economic status, risk level, crime history, psychological make-up.
4. Study methodology: extreme groups design, attrition, follow-up length, type of outcome measure, sample size, statistical value.

The accuracy of coding was assessed using the index: agreement = # of agreements ÷ (# of agreements + # of disagreements) (Yeaton & Wortman, 1993). The second author coded all studies. The first author blindly coded a random sample of 30 studies. Percentage agreement scores for the two raters ranged from 85% to 98% across coding categories. Where disagreements occurred, the coding used was based on the first author's classification.

Predictor Categories

The predictors were initially sorted into 18 domains (Category I). Their coding criteria are detailed in Appendix A. Then, for the purposes of research synthesis, the 18 domains were collapsed into eight all-encompassing predictor domains (Category II): a) age/gender/race, b) criminal history, c) criminogenic needs, d) family factors, e) intellectual functioning, f) personal distress, g) SES, and h) social achievement.

Effect Size Calculation

Pearson product-moment correlation coefficients were produced for all predictors in each study that reported a numerical relationship with the criterion. When statistics other than Pearson r were presented, their conversion to r was undertaken using the appropriate statistical formulae (Rosenthal, 1991). Where a p value of greater than .05 was the only reported statistic, an r of .0 was assigned.

Next, the obtained correlations were transformed using Fisher's table. Then, according to the procedures outlined by Hedges and Olkin (1985, p. 230-232), the statistic z±, representing the weighted estimation of Pearson r, was calculated for each predictor domain by dividing the sum of the weighted zrs per predictor domain by the sum of each predictor's sample size minus three across that domain.

In order to determine the practical utility of various predictors relative to each other, the common language (CL) effect size indicator (McGraw & Wong, 1992) was also employed. The CL measure is little affected by changes in base rates and selection ratios making it ideal for prediction studies (Rice & Harris, 1995a). The CL statistic converts an effect size into the probability that a predictor-criterion score sampled at random from the distribution of one predictor domain (e.g., criminogenic needs) will be greater than that sampled from another distribution (e.g., personal distress).

Significance Testing

To determine which of the predictor domains predicted criterion significantly different from zero, the mean z^± values for each domain were multiplied by the value of (N - 3k)^1/2, where N = the number of subjects per predictor domain and k = the number of predictors per domain (Hedges & Olkin, 1985).

One-way ANOVAs and the Student-Newman-Keuls (SNK) multiple comparison test were then applied to the mean r values of those domains which significantly predicted criterion better than zero in order to assess which domains differed significantly from each other.

Mindful of the debate regarding alternatives to the use of parametric methods as tests of significance in meta-analyses, the mean z^± values for significant predictor domains were also assessed using an analogue to the ANOVA's F-test, the goodness-of-fit statistic Q (Hedges & Olkin, 1985). Following that, post-hoc comparisons of the differences between mean z^± values of each pair of significant predictor domains were conducted using the z test (E. Marchand, personal communication, June 15, 1994).

Finally, one-way ANOVAs and the SNK test using Pearson r were employed to assess whether type of outcome criteria, length of follow-up, and study characteristics were related to effect size.

The CL statistic does not involve significance testing.

Unless otherwise specified, alpha was set at .05 2-tail for all significance tests.

RESULTS

One hundred and thirty-one studies were identified as suitable for the meta-analysis. These studies generated 1141 effect sizes with future criminal behaviour.

For those variables where at least 60% of the studies reported information on the study characteristics sampled, the results were as follows: a) 86% of the studies were published, 58% in journals, b) 73% of the senior authors had published in the area previously, 51% of the them were male, c) 44% and 54% of authors were based in an academic and government agency setting respectively, d) the studies were evenly distributed across the decades with the majority emanating from the United States and Canada, although Canadian studies contributed the majority (63%) of effect sizes, e) 95% of studies consisted of male or mixed samples, f) only 5% of studies employed an extreme groups design and g) 83% did not suffer subject attrition of more than 10% of their sample.

Predictor Domains: Category I

Table 1 presents the mean effect sizes for the 18 levels of Category I in conjunction with the number of effect sizes (k) and the total number of subjects associated with each predictor domain (N). The domains are grouped as follows: static (n = 10), dynamic (n = 7) and composite measures (n = 1).

Table 1

Mean effect sizes for predictor domains: Category I

Note. k = effect sizes per predictor domain; N = subjects per predictor domain; SD = standard deviation

^a F(16, 1001) = 5.59, p < .05.

¹2, 13 vs. 6, 8, 10, 15; 211 vs. 8, 10, 15; 312, 16 vs. 8, 15; ⁴1, 3, 5, 14, 17 vs. 15; Student-Newman-Keuls post-hoc comparison, p < .05.

*p < .05.

The following is an example of how to read Table 1. Across the 131 studies sampled, a quantitative relationship between the predictor age and recidivism was reported on 56 occasions involving a total of 61,312 subjects. The associated mean Pearson r for age with outcome was .15 (SD = .12), with younger age being positively correlated with poorer outcome. Mean z^±, the weighted estimation of Pearson r for age with outcome, was .11. Application of Hedges and Olkin's (1985) method for testing the significance of the mean z^± values confirmed age as a significant predictor of recidivism.

All predictor domains were significant predictors of recidivism. The largest mean r values were found for adult criminal history, antisocial personality, companions, and criminogenic needs. Risk scale measures, which contained information from several predictor domains, produced the highest mean r value with recidivism (.30).

The conclusions reached by the parametric (ANOVA, SNK) statistical analysis were virtually identical to those of the F-test analogue (Q, Z-test comparison). We report the results of the standard parametric analysis.

A one-way ANOVA applied to the mean r values (excluding composite risk scales) indicated there was a significant difference across the predictor domains [F(16, 1001) = 5.59]. A Student Newman - Keuls (SNK) multiple comparison test of the mean r values are specified in Table 1. Adult criminal history, and criminogenic needs produced the greatest frequency of significant differences. Each of these were significantly different from family structure, intellectual functioning, personal distress, and SES.

Predictor Domains: Category II

With the exception of the risk scales domain, the 17 predictor domains from Category I were collapsed into 8 groups. All predictor domains were significantly greater than 0 (see Table 2). There were significant differences among the eight predictor domains [F(7, 1010) = 10.00]. The SNK multiple comparison test of the mean r values revealed that the predictor domains criminal history and criminogenic needs were significantly greater than those of family factors, intellectual functioning, personal distress, SES.

Table 2

Mean effect sizes for predictor domains: Category II

Note. k = effect sizes per predictor domain; N = subjects per predictor domain; SD = standard deviation

^ccriminal history = adult plus pre-adult; ^dcriminogenic need factors = antisocial personality, companions, interpersonal conflict, criminogenic needs, and substance abuse.

^a F(7, 1010) = 10.00, p < .05; ^b F(1,1016) = 6.18, p < .05.

¹2, 6 vs. 3, 4, 5, 7; ²8 vs. 4, 5, 7; ³1, 3 vs. 4, 7; Student-Newman-Keuls post-hoc comparison,

p < .05.

*p < .05.

The eight predictor domains were classified into dynamic and static factors. The dynamic grouping consisted of criminogenic needs factors, personal distress, and social achievement. The mean r values for dynamic (.15) and static (.12) were significantly different [F(1, 1016) = 6.18].

The common language effect size indicator (CL) provided another approach to examining the relative usefulness of the 8 predictor domains from Table 2 as well as the static-dynamic comparison. The CL scores, summarized in Table 3, indicate the percentage of time that one of a pair of predictors produced larger correlations with outcome.

Table 3

Common language effect size indicators^a

^aCommon language effect size indicators for mean r values. Bracketed values favour vertical axis; unbracketed values favour horizontal axis.

^bAGR = age, gender, race; CH = criminal history/history of antisocial behaviour; CN = criminogenic need factors; F = family factors; I = intellectual functioning; PD = personal distress; SES = social class of origin; SA = social achievement.

Table 3 can be read in the following way. With regard to direction, unbracketed scores favour the horizontal axis predictor while bracketed scores favour the vertical axis predictor. For example, in comparing criminogenic needs (CN) with personal distress (PD), one can see that 73% of the time CN produced higher correlations with recidivism than did PD.

In the case of the static-dynamic comparison (Table 2) the CL score was 54% in favour of the dynamic predictor domain.

Actuarial Measures

Table 4 summarizes the mean effect sizes of the composite risk and personality scales with recidivism. All of the instruments predicted recidivism significantly different from zero. Amongst the risk scales, the LSI-R produced the highest correlation with recidivism (r = .35) but it was not significantly greater than the SFS, Wisconsin or the other risk scale domains [F(3, 119) = 1.52]. The Other domain consisted of SFS clones, that is, instruments containing about 5-10 items, almost all of which were static in nature.

Table 4

Mean effect sizes for Risk and AntiSocial Personality Scales

Antisocial Personality Scales^b

Note. k = effect sizes per predictor domain; N = subjects per predictor domain; SD = standard deviation.

^a F(3, 119) = 1.52, p < .05; ^b F(2,59) = 4.01, p < .05.

¹6 vs. 5, 7; Student-Newman-Keuls post-hoc comparison, p < .05.

*p < .05.

The LSI-R produced CL scores of 76% and 67% with the Wisconsin and SFS respectively when mean r was the dependent variable.

A comparison of the mean r values associated with the antisocial personality measures revealed a significant difference between measures [F(2, 59) = 4.01]. The SNK multiple comparison test reported that the PCL was a significantly better predictor than either the MMPI based measures or Other domain. CL analysis indicated that 83% of the time the PCL produced larger Pearson r correlations with recidivism than did the MMPI. The Social Achievement domain (accommodation, education, employment, and marital status) showed r's ranging from .15-.16.

Within the personal distress domain, 24 of 66 effect sizes tapped psychiatric symptomatology. The mean r (SD) was .00 (.17). Family factors did not include studies from the gene-crime relationship as Walters (1992) has already conducted a meta-analysis in this area. However, data from Tables 2 and 3 of Walter's (1992) study yielded a mean r and z^± of .08, indicating that genetic background was a significant predictor of recidivism.

Practitioners occasionally ask which official measure of recidivism is the most sensitive. Four outcomes - arrest, conviction, incarceration, and parole violation - were compared. Mean effect size values ranged from .13 to .19. The mean r values associated with incarceration were significantly greater than those of conviction or parole violation [F(3, 894) = 6.71]. In all comparisons, however, the CL scores were less than 60%.

DISCUSSION AND RECOMMENDATIONS

Prior to discussing the results it must be noted that the generalization of the results of any meta-analysis is limited by the nature of the studies examined.

Some valuable studies (e.g., Gendreau et al., 1979) could not be used because the researchers reported their results in formats (e.g., regression analyses) from which Pearson r's could not be calculated.

Little attempt was made to retrieve unpublished studies that were not immediately available. A common assumption is that one of the reasons that some studies are not published is because they may be lacking in methodological rigour which, in turn, affects the magnitude of effect sizes (see Lipsey & Wilson, 1993). Lipsey & Wilson's (1993) analysis applied to treatment studies but, so far, prediction studies have not shown similar results (Goggin & Gendreau, 1995).

Another methodological point concerns one of the goals of meta-analysis. Hunter & Schmidt (1990) are interested in determining the maximum value that can be obtained in prediction if all variables were perfectly measured. Others insist that the goal of meta-analysis is to "teach us better what is, not what might some day be in the best of all possible worlds..." (Rosenthal, 1991, p. 25). We are of the latter view and did not attempt to statistically adjust for methodological artifacts, which may or may not have had an impact on the magnitude of the effect sizes obtained.

The data base was, regrettably, virtually silent on the prediction of recidivism among female offenders, minority groups, "white collar" offenders, and some important sample characteristics such as risk level and the psychological make-up of the subjects studied. Much of the effect size data on dynamic predictor domains came from Canada where there has been a strong emphasis on the assessment of individual differences (cf., Andrews & Bonta, 1994).

One should not assume that many of the correlations found in this meta-analysis (i.e., .10 - .30) are inconsequential. In fact, mean r values in this range can be indicative of substantial practical import (Hunter & Schmidt, 1990; Rosnow & Rosenthal, 1993). Indeed, the percentage improvement in predicting recidivism can equal the actual value of r (Rosenthal, 1991, p. 134), assuming base rates and selection ratios that are not in the extreme.

The fact that the data base consisted of just over 1,000 effect sizes involving almost 750,000 S's suggests that reasonable confidence can be placed in the results. Additional research, in our view, is not likely to change the direction or ordering of the results of the predictor domains to any marked degree.

The remainder of the discussion addresses the questions raised in the introduction.

Predictor Domains

The meta-analysis provided further confirmation of the narrative reviews which concluded that variables such as age, criminal history, companions, family factors, gender, social achievement, and substance abuse are significant and potent predictors of recidivism. On the other hand, it offered some important insights into several other predictor domains.

The time is long past when those offender risk factors that are dynamic in nature can be cavalierly ignored. Indeed, criminogenic needs produced higher correlations with recidivism (see Table 3) a much higher percentage of the time than did several other predictor domains. When considering all predictor domains, a statistically significant difference was found in favour of dynamic risk factors, but the CL effect size indicator was only 54%. Moreover, the two major static and dynamic categories, criminal history and criminogenic needs, were almost identical in predicting recidivism. While very few studies have assessed how well changes over time within dynamic factors predict recidivism, the data suggest that changes on criminogenic needs may produce strong correlations in that regard.

Early family factors and history of pre-adult antisocial behaviour are rarely included in adult offender risk prediction instruments. Fortuitously, a number of estimable studies (producing 103 effect sizes) were located that followed offenders from early years to adulthood. The combined family factors domain (Table 2) and pre-adult history of antisocial behaviour (Table 1) produced correlations of .12 and .13 with recidivism respectively, demonstrating once again that antisocial risk factors in childhood can have far reaching influence (e.g., Stattin & Magnusson, 1989).

Much controversy has focused upon how well personal distress, intelligence, and SES predict recidivism (Andrews & Bonta, 1994; Herrnstein & Murray, 1994; Tittle & Meier, 1990). From a treatment standpoint, the important result centred on the fact that personal distress turned out to be quite a weak predictor of recidivism. Moreover, one of the components of this domain, psychiatric symptomatology, which has characteristically been perceived as an important predictor of re-offending in the field of psychiatry (cf., Phillips et al., 1988), did not correlate (r = .00) with recidivism. This finding was based on few effect sizes; more research is needed to confirm this tentative result.

It would be reasonable, therefore, to assume that programs that insist on alleviating offenders' personal distress, as many do (Gendreau et al., 1994), will have little success in reducing offender recidivism. Meta-analyses of the offender treatment literature (e.g., Andrews et al., 1990b) are also supportive of this conclusion.

The studies in the meta-analysis that included measures of IQ were of the "traditional" sort (i.e., standard paper and pencil tests that measured linguistic and mathematical abilities). Granted, these sorts of IQ measures can produce modest correlations with criminal behaviour over long periods of time (Moffitt et al., 1994), it is generally agreed that this type of IQ assessment has reached its limits (Gardner, 1995). A much more productive strategy would be to focus on what is called practical or tacit intelligence, which is defined as the ability to learn and profit from experience, effectively monitoring one's own and other's feelings and needs, and solve the problems of everyday life (Gardner, 1983; Sternberg et al., in press).

This meta-analysis extended Tittle and Meier's (1990; 1991) pessimistic conclusions regarding the social class - crime link with delinquent samples to that of adult offenders. It is difficult to judge how social class theories will evolve in the future; for speculations on this matter see Andrews & Bonta (1994) and Tittle & Meier (1990). The most probable scenario is that social class theories will incorporate more psychological concepts (e.g., Agnew, 1992).

How well might the results from the meta-analysis generalize to specialized offender groups? Few violence prediction studies that predicted the occurrence of violence vs. no criminal activity were retrieved. Our reading of the literature indicates that the strongest predictors identified in this meta-analysis also apply to violent offenders (Harris et al., 1993; Reiss & Roth, 1993). As well, composite measures of general recidivism (i.e., LSI-R) correlate highly (r = .78) with measures intended to predict violence (i.e., PCL-R) (Loza & Simourd, 1994). One area where the predictors of violent offending may be quite different is in the area of impulsivity combined with overly hostile attributions of other peoples intent (Serin & Kuriychuk, 1994). Sex offenders present a somewhat different picture. At the risk of generalizing across such a complex group, there do appear to be a few predictors, centring on the offense itself, that are unique to this population (Hanson & Bussière, 1995). Moreover, factors that predict sexual violence are different than those that predict other types of violence (Bonta & Hanson, 1995).

In regard to theory development, the results from the meta-analysis are most supportive of recent advancements in differential association and social learning theories (see Andrews & Bonta, 1994, p. 104-124). These authors assert that, from the perspective of these theories, it is absolutely essential that criminogenic needs and antisocial associates be two of the strongest correlates of criminal conduct. Criminogenic needs establish the standards of conduct and generate the rationale for engaging in antisocial behaviour. Antisocial associates provide the opportunity for antisocial modelling to occur, govern the rewards and costs of such behaviour, and influence antisocial attitudes.

The less potent predictors in this meta-analysis (e.g., SES, personal distress, intellectual functioning) have traditionally been associated with the anomie/strain, subcultural, and biologically-oriented theories.

Actuarial Measures for Predicting Recidivism

Composite measures of risk, on average, produced substantially greater correlations with recidivism than antisocial personality scales. This is not surprising, because risk measures generally sample from a much wider variety of predictor domains than personality scales.

Amongst the former, the LSI-R produced higher correlations with recidivism than the SFS, the Wisconsin, or the Other category. While the mean differences among the four measures were not statistically significant, the CL effect size indicator provided a result of practical importance. The LSI-R produced larger correlations with recidivism than did the other three risk measures between 62% and 76% of the time.

The LSI-R, therefore, appears to be the current measure of choice. An impressive volume of studies confirming its predictive validity with recidivism and prison adjustment has been generated for a variety of offender populations (i.e., adults, juveniles, natives, females) (Andrews & Bonta, 1995).

In the area of antisocial personality assessment, a noteworthy finding was that Hare's (1991) PCL-R produced significantly greater correlations with recidivism than the widely used MMPI based systems. The PCL-R specializes in assessing the psychopathic dimension of antisocial personality. It is recommended by clinicians who are concerned with predicting violence (Harris et al., 1993).

Practical Suggestions for Practitioners

One advantage of a meta-analysis is that a thorough literature search uncovers some promising assessment approaches and measures that, upon further investigation, may prove to be highly useful for practitioners. Meta-analysis also encourages professional judgement where necessary (Light & Pillemer, 1984). Please be aware that these recommendations are based on very limited research and the first author's experience and clinical judgement. They are as follows:

1. Consider measures that assess hostility and/or aggression such as the Aggression Questionnaire (Buss & Perry, 1992). Gendreau et al., (1979a) reported moderate predictive validities on an earlier version of this test (i.e., Buss-Durkee, 1957). The fact that long-term follow-up studies of aggression in childhood correlated so well (r = .30 range) with criminal behaviour in later adulthood was impressive (Stattin & Magnusson, 1989).

2. Utilize measures that tap into selfish, narcissistic antisocial behaviour, criminal sentiments, and rationalizations for criminal behaviour (e.g., Walters & Chlumsky, 1993; Shields & Simourd, 1991; Shields & Whitehall, 1994).

3. Do not limit assessments of education and employment to just an offender' s past performance in this area. Measure current attitudes and performance regarding work and skill development. Substantial predictive validity could result (Gendreau et al., 1979a; Jenkins et al., 1977).

4. Pursue the usefulness of actuarial measures and predictor domains which zero in on offenders that specialize in one type of offense such as violence, paedophilia, or arson (Hare, 1991; Knight et al., 1994; Rice & Harris, 1994; Serin & Kuriychuk, 1994).

5. Collect information on bio-medical factors and relate these to future criminal behaviour. Data from this domain may produce small but helpful predictive validities of recidivism (cf., Raine, 1993). Prisons that have comprehensive medical screening protocols should have a wealth of information for future research.

6. Update the knowledge base about the predictive validities of commonly used psychological tests. There have to be thousands of CPI and MMPI protocols sitting in prison files that could be expediently analyzed to determine whether the time and effort involved in using these tests in corrections is worthwhile.

7. Whenever possible, compare and combine different types of assessment methods, such as structured interviews, actuarial measures, and social history information (see Andrews & Wormith, 1989; Gendreau et al., 1980; Rice & Harris, 1995b). Predictive validities are certain to increase.

In conclusion, the modest contribution from this meta-analysis has been to clarify which predictor variables and measures of risk will provide the most assistance to practitioners and policy makers to reach their objectives of reducing prison overcrowding, managing probation and parole caseloads effectively, and designing better treatment programs.

APPENDIX A

CODING CRITERIA FOR PREDICTOR DOMAINS

Category I

Static Predictors

1. age: at time of data collection/assessment.
2. criminal history: adult - prior arrest, probation, jail, conviction, incarceration, prison misconducts.
3. history of antisocial behaviour: pre-adult - prior arrest, probation, jail, conviction, incarceration, alcohol/drug abuse, aggressive behaviour, conduct disorder, behaviour problems at home and school, delinquent friends.
4. family criminality: parents and/or siblings in trouble with the law.
5. family rearing practices: lack of supervision and affection, conflict, abuse.
6. family structure: separation from parents, broken home, foster parents.
7. gender.
8. intellectual functioning: WAIS/WISC, Raven, Porteous Q score, learning disabilities, reading level.
9. race: white vs. black/hispanic/native.
10. social class of origin (SES): socioeconomic status of parents (parental occupation, education or income).

Dynamic Predictors

11. antisocial personality/sociopathy/psychopathy scales: MMPI Pd, Megargee system, EPI-Psychoticism, CPI-Soc, PCL-R, DSM III personality disorders, any indices of egocentric thinking.
12. companions: identification/socialization with other offenders.
13. criminogenic needs: antisocial attitudes supportive of an antisocial lifestyle and behaviour regarding education, employment.
14. interpersonal conflict: family discord, conflict with significant others.
15. personal distress: anxiety, depression, neuroticism, low self-esteem, psychiatric symptomatology (i.e., psychotic episodes, schizophrenia, not guilty by reason of insanity, affective disorder), attempted suicide, personal inadequacy.
16. social achievement: marital status, level of education, employment history, income, address changes.
17. substance abuse: recent history of alcohol/drug abuse.

Composite Measures

18. LSI-R, SFS, Wisconsin, Other risk scales.

References

Agnew, R. (1992). Foundation for a general strain theory of crime and delinquency. Criminology, 30, 47-86.

Andrews, D.A. (1980). Some experimental investigations of the principles of differential association through deliberate manipulations of the structure of service systems. American Sociological Review, 45, 448-462.

Andrews, D.A., & Bonta, J. (1994). The Psychology of Criminal Conduct. Cincinnati, OH: Anderson.

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

Andrews, D.A., Bonta, J., & Hoge, R.D. (1990a). Classification for effective rehabilitation: Rediscovering psychology. Criminal Justice and Behavior, 17, 19-52.

Andrews, D.A., & Wormith, J.S. (1989). Personality and crime: Knowledge destruction and construction in criminology. Justice Quarterly, 6, 289-309.

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

Austin, J. (1995). Correctional options: An overview. Corrections Today, 57.

Baird, C.S. (1981). Probation and parole classification: The Wisconsin model. Corrections Today, 43, 36-41.

Bonta, J. (1996). Risk-needs assessment and treatment. In A. Harland (Ed.), Choosing correctional options that work: Defining the demand and evaluating the supply (pp. 18-32). Thousand Oaks, CA: Sage.

Bonta, J., & Gendreau, P. (1990). Reexamining the cruel and unusual punishment of prison life. Law and Human Behavior, 14, 347-372.

Bonta, J., & Hanson, R.K. (1995, August). Violent recidivism of men released from prison. Paper presented at the 103rd Annual Convention of the American Psychological Association, New York, NY.

Buss, A.H., & Durkee, A. (1957). An inventory for assessing different kinds of hostility. Journal of Consulting and Clinical Psychology, 21, 343-349.

Buss, A.H., & Perry, M. (1992). The aggression questionnaire. Journal of Personality and Social Psychology, 63, 452-459.

Clear, T.R., & Gallagher, K.W. (1985). Probation and parole supervision: A review of current classification. Crime and Delinquency, 31, 423-443.

Cohen, A.K. (1955). Delinquent boys: The culture of the gang. Glencoe, IL: Free Press.

Cohen, J. (1994). The earth is round (p .05). American Psychologist, 49, 997-1003.

Cooper, H., & Hedges, L.V. (1994). Handbook of research synthesis. New York: Russell Sage Foundation.

Cronbach, L.J. (1990). Essentials of Psychological Testing. New York: Harper Collins.

DiIulio, J.J. (1991). No escape: The future of American corrections. New York: Basic Books.

Feeley, M.M., & Simon, J. (1992). The new penology: Notes on the emerging strategy of corrections and its implications. Criminology, 30, 449-474.

Gardner, H. (1995). Cracking open the IQ box. The American Prospect, 20, 71-80.

Gardner, H. (1983). Frames of mind: The theory of multiple intelligence. New York: Basic Books.

Gendreau, P. (1996). Offender rehabilitation: What we know and what needs to be done. Criminal Justice and Behavior, 23, 144-161.

Gendreau, P. & Andrews, D.A. (1990). Tertiary prevention: What the meta-analysis of the offender treatment literature tells us about "what works". Canadian Journal of Criminology, 32, 173-184.

Gendreau, P., Andrews, D.A., Goggin, C., & Chanteloupe, F. (1992). The Development of Clinical and Policy Guidelines for the Prediction of Criminal Behaviour in Criminal Justice Settings. (Unpublished report, Ministry Secretariat). Ottawa: Solicitor General of Canada.

Gendreau, P., Cullen, F.T., & Bonta, J. (1994). Intensive rehabilitation supervision: The next generation in community corrections? Federal Probation, 58, 72-78.

Gendreau, P., Goggin, C., & Annis, H. (1990). Survey of existing substance abuse programs. Forum on Corrections Research, 2, 6-8.

Gendreau, P., Grant, B.A., & Leipciger, M. (1979). Self-esteem, incarceration and recidivism. Criminal Justice and Behavior, 6, 67-75.

Gendreau, P., Grant, B.A., Leipciger, M., & Collins, S. (1979). Norms and recidivism rates for the MMPI and selected experimental scales on a Canadian delinquent sample. Canadian Journal of Behavioural Science, 11, 21-31.

Gendreau, P., & Ross, R.R. (1987). Revivification of rehabilitation: Evidence from the 80's. Justice Quarterly, 4, 349-407.

Goddard, H.H. (1920). Human efficiency and levels of intelligence. Princeton, NH: Princeton University Press. Goggin, C.E., & Gendreau, P. (1995). Predicting psychiatric rehospitalization: A meta-analysis. Unpublished manuscript, University of New Brunswick, Saint John, New Brunswick.

Gottfredson, S.D., & Gottfredson, D.M. (1979). Screening for risk: A comparison of methods. Washington, DC: U.S. Government Printing Office.

Gough, H.G. (1957). Manual for the California Psychological Inventory. Palo Alto, Calif.: Consulting Psychologists Press.

Gough, H.G., Wenk, E.A., & Rozynko, V.V. (1965). Parole outcome as predicted from the CPI, the MMPI, and a Base Expectancy Table. Journal of Abnormal Psychology, 70, 432-441.

Hanson, R.K., & Bussière, M. (1996). Predictors of sexual offender recidivism: A meta-analysis. User Report 1996-04, Ottawa: Solicitor General of Canada.

Hare, R.D. (1991). The Revised Psychopathy Checklist. Toronto: Multi-Health Systems.

Harris, G.T., Rice, M.E., & Quinsey, V.L. (1993). Violent recidivism of mentally disordered offenders: The development of a statistical prediction instrument. Criminal Justice and Behavior, 20, 315-335.

Hedges, L.V., & Olkin, I. (1985). Statistical methods for meta-analysis. San Diego, CA: Academic Press.

Herrnstein, R.J., & Murray, C.A. (1994). The bell curve: Intelligence and class structure in American life. New York: Free Press.

Hirschi, T., & Hindelang, M.J. (1977). Intelligence and delinquency: A revisionist review. American Sociological Review, 42, 571-587.

Hoffman, P.B. (1983). Screening for risk: A revised Salient Factor Score (SFS 81). Journal of Criminal Justice, 11, 539-547.

Hoge, R.D., Leschied, A.W., & Andrews, D.A. (1993). An investigation of young offender services in the province of Ontario: A report of the repeat offender project. Toronto: Ontario Ministry of Community and Social Services.

Hunter, J.E., & Schmidt, F.L. (1990). Methods of meta-analysis: Correcting error and bias in research findings. Newbury Park, CA: Sage.

Jenkins, W.O., DeValera, E.K., & Muller, J.B. (1977). The Behavioral Evaluation Treatment and Analysis (BETA) system in the prediction of criminal and delinquent behaviour. Quarterly Journal of Corrections, 1, 44-50.

Jones, P.R. (1996). Risk prediction in criminal justice. In A. Harland (Ed.), Choosing correctional options that work: Defining the demand and evaluating the supply. Thousand Oaks, CA: Sage.

Kaminer, W. (1992). I'm Dysfunctional - You're Dysfunctional. New York: Addison-Wesley.

Knight, R.A., Prentky, R.A., & Cerce, D.D. (1994). The development, reliability, and validity of an inventory for the multidimensional assessment of sex and aggression. Criminal Justice and Behavior, 21, 72-94.

LeBlanc, M., Ouimet, M., & Tremblay, R.E., (1988). An integrative control theory of delinquent behaviour: A validation 1976-1985. Psychiatry, 51, 164-176.

Leschied, A.W., & Gendreau, P. (1994). Doing justice in Canada: YOA policies that can promote community safety. Canadian Journal of Criminology, 36, 291-303.

Lipsey, M.W. (1992). Juvenile delinquency treatment: A meta-analytic inquiry into the variability of effects. In T.D. Cook, H. Cooper, D.S. Cordray, H. Hartmann, L.V. Hedges, R.J. Light, T.A. Louis, & F. Mosteller (Eds.), Meta-analysis for explanation. New York: Russell Sage Foundation.

Lipsey, M., & Wilson, D.B. (1993). The efficacy of psychological, educational, and behavioral treatment: Confirmation from meta-analysis. American Psychologist, 48, 1181-1209.

Little, K.B., & Shneidman, E.S. (1959). Congruencies among interpretations of psychological test and anamnestic data. Psychological Monographs: General and Applied, 73, 1-42.

Loza, W., & Simourd, D.J. (1994). Psychometric evaluation of the Level of Supervision Inventory (LSI-R) among male Canadian federal offenders. Criminal Justice and Behavior, 21, 468-480.

Lynch, J.P. (1993). A cross-national comparison of the length of custodial sentences for serious crimes. Justice Quarterly, 10, 639-660.

Matza, D. (1964). Delinquency and drift (2nd ed.). New York: Wiley.

Mauer, M. (1994). Americans behind bars: The international use of incarceration, 1992-1993. Washington, DC: Sentencing Project.

McGraw, K.O., & Wong, S.P. (1992). A common language effect size. Psychological Bulletin, 111, 361-365.

Meehl, P.E. (1954). Clinical versus statistical prediction. Minneapolis: University of Minnesota Press.

Mencimer, S. (1993). Righting sentences: Let's get smart about who should - and who shouldn't - be in jail. Washington Monthly, 5, 26-29.

Merton, R.K. (1957). Social theory and social structure. New York: Free Press.

Moffitt, T.E., Lynam, D.R., & Silva, P.A. (1994). Neuropsychological tests predicting persistent male delinquency. Criminology, 32, 277-300.

Moon, P. (1995, February 6). Prison crowding crisis building: Longer sentences, less early parole force inmates to share one-man cells. The Globe and Mail, pp. 1,6.

Motiuk, L.L. (1993). Where are we in our ability to assess risk? Forum on Corrections Research, 5, 14-19.

Motiuk, L.L., Bonta, J., & Andrews, D.A. (1986, June). Dynamic predictive criterion validity in offender assessment. Paper presented at the meeting of the Canadian Psychological Association, Ottawa, Ontario, Canada.

Phillips, M., Wolf, A., & Anderson, C. (1988). Psychiatry and the criminal justice system: Testing the myths. American Journal of Psychiatry, 145, 605-610.

Raine, A. (1993). The psychopathology of crime: Criminal behaviour as a clinical disorder. San Diego, CA: Academic Press, Inc.

Reiss, A.J., Jr., & Roth, J.A. (1993). Understanding and Preventing Violence. Washington, DC: National Academy Press.

Rice, M.E., & Harris, G.T. (1995a). Violent recidivism: Assessing predictive validity. Journal of Consulting and Clinical Psychology, 63, 737-748.

Rice, M.E., & Harris, G.T. (1995b). Cross-validation of actuarial predictions. (Research Report Vol. XII, No. 2). Penetanguishene, Ontario, Canada: Mental Health Centre.

Rice, M.E., & Harris, G.T. (1994). Predicting the recidivism of mentally disordered firesetters. (Research Report Vol. XI, No. 2). Penetanguishene, Ontario: Mental Health Centre.

Rosenthal, R. (1991). Meta-analytic procedures for social research. Newbury Park, CA: Sage.

Rosnow, R.L., & Rosenthal, R. (1993). Beginning behavioral research: A conceptual primer. New York: Macmillan Publishing Company.

Rowe, D.C., & Osgood, D.W. (1984). Heredity and sociological theories of delinquency: A reconsideration. American Sociological Review, 49, 526-540.

Schmidt, F.L. (1992). What do data really mean? American Psychologist, 47, 1173-1181.

Serin, R.C., & Kuriychuk, M. (1994). Social and cognitive processing deficits in violent offenders. International Journal of Law and Psychiatry, 17, 431-441.

Serin, R.C., Peters, R.D., & Barbaree, H.E. (1990). Predictors of psychopathy and release outcome in a criminal population. Psychological Assessment: A Journal of Consulting and Clinical Psychology, 2, 419-422.

Shields, I.W., & Simourd, D. (1991). Predicting predatory behavior in a population of incarcerated young offenders. Criminal Justice and Behavior, 18, 179-193.

Shields, I.W., & Whitehall, G.C. (1994). Neutralization and delinquency among teenagers. Criminal Justice and Behavior, 21, 223-235.

Simourd, D.J., Bonta, J., Andrews, D.A., & Hoge, R.D. (1991). Criterion validity of assessments of psychopathy: A meta-analysis. Unpublished manuscript, Carleton University, Ottawa, Ontario.

Staff. (1994, June 15). State legislatures moving toward tougher sentences. Criminal Justice Newsletter, 25, 1-2.

Staff. (1993). Basic facts about corrections in Canada. (Available from Correctional Services of Canada, Ottawa).

Stattin, H., & Magnusson, D. (1989). The role of early aggressive behaviour in the frequency, seriousness and types of later crime. Journal of Consulting and Clinical Psychology, 57, 710-718.

Sternberg, R.J., Wagner, R.K., Williams, W.M., & Horvath, J.A. (in press). Testing common sense. American Psychologist.

Tittle, C.R., & Meier, R.F. (1990). Specifying the SES/delinquency relationship. Criminology, 28, 271-299.

Tittle, C.R., & Meier, R.F. (1991). Specifying the SES/delinquency relationship by social characteristics of contexts. Journal of Research in Crime and Delinquency, 28, 430-455.

Walters, G.D. (1992). A meta-analysis of the gene-crime relationship. Criminology, 30, 595-613.

Walters, G. D., & Chlumsky, M. L. (1993). The lifestyle criminality screening form and antisocial personality disorder: Predicting release outcome in a state prison sample. Behavioral Sciences and the Law, 11, 111-115.

Widom, C.S., & Toch, H. (1993). The contribution of psychology to criminal justice education. Journal of Criminal Justice Education, 4, 251-272.

Wilson, J.Q., & Herrnstein, R.J. (1985). Crime and human nature. New York: Simon and Schuster.

Yeaton, W.H., & Wortman, P.M. (1993). On the reliability of meta-analytic reviews: The role of intercoder agreement. Evaluation Review, 17, 292-309.

References in the Meta-Analysis

Adams, D., & Fischer, J. (1976). The effects of prison residents' community contacts on recidivism rates. Corrective and Social Psychiatry and Journal of Behavioral Technology, Methods, and Therapy, 22, 21-27.

Andrews, D. A., & Friesen, W. (1987). Assessments of anticriminal plans and the prediction of criminal futures: A research note. Criminal Justice and Behavior, 14, 33-37.

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

Babst, D. V., Koval, M., & Neithercutt, M. G. (1972). Relationship of time served to parole outcome for different classifications of burglars based on males paroled in fifty jurisdictions in 1968 and 1969. Journal of Research in Crime and Delinquency, 9, 99-116.

Babst, D. V., Moseley, W. H., Schmeidler, J., Neithercutt, M. G., & Koval, M. (1976). Assessing the length of institutionalization in relation to parole outcome. Criminology, 14, 41-54.

Baker, L. A., Mack, W., Moffitt, T. E., & Mednick, S. (1989). Sex differences in property crime in a Danish adoption cohort. Behavior Genetics, 19, 355-370.

Barbaree, H. E., & Marshall, W. L. (1988). Deviant sexual arousal, offense history, and demographic variables as predictors of reoffense among child molesters. Behavioral Sciences and the Law, 6, 267-280.

Barnett, A., Blumstein, A., & Farrington, D. P. (1989). A prospective test of a criminal career model. Criminology, 27, 373-388.

Bayer, B. M., Bonta, J. L., & Motiuk, L. L. (1985). The PD subscales: An empirical evaluation. Journal of Clinical Psychology, 41, 780-788.

Bennett, L. A. (1974). Self-esteem and parole adjustment. Criminology, 12, 346-363.

Bieber, S. L., Pasewark, R. A., Bosten, K., & Steadman, H. J. (1988). Predicting criminal recidivism of insanity acquittees. International Journal of Law and Psychiatry, 11, 105-112.

Black, T., & Spinks, P. (1985). Predicting outcomes of mentally disordered and dangerous offenders. In D. P. Farrington & R. Tarling (Eds.), Prediction in criminology (pp. 174-192). Albany, NY: State University of New York Press.

Bogenberger, R. P., Pasewark, R. A., Gudeman, H., & Bieber, S. L. (1987). Follow-up of insanity acquittees in Hawaii. International Journal of Law and Psychiatry, 10, 283-295.

Bonta, J. (1989). Native inmates: Institutional response, risk, and needs. Canadian Journal of Criminology, 31, 49-61. Bonta, J., Lipinski, S. & Martin, M. (1992). The characteristics of aboriginal recidivists. Canadian Journal of Criminology, 38, 517-522.

Bonta, J., & Motiuk, L. L. (1985). Utilization of an interview-based classification instrument: A study of correctional halfway houses. Criminal Justice and Behavior, 12, 333-352.

Bonta, J., & Motiuk, L. L. (1987). The diversion of incarcerated offenders to correctional halfway houses. Journal of Research in Crime and Delinquency, 24, 302-323.

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

Broadhurst, R., & Maller, R. (1990). The recidivism of prisoners released for the first time: Reconsidering the effectiveness question. Australian and New Zealand Journal of Criminology, 23, 88-104.

Broadhurst, R. G., & Maller, R. A. (1991). Sex offending and recidivism. (Report No.3). Nedlands, Western Australia: The University of Western Australia, Crime Research Center.

Broadhurst, R., Maller, R., Maller, M., & Duffecy, J. (1988). Aboriginal and non-aboriginal recidivism in Western Australia: A failure rate analysis. Journal of Research in Crime and Delinquency, 25, 83-108.

Brown, L.D. (1978). The development of a parolee classification system using discriminant analysis. Journal of Research in Crime and Delinquency, 15, 92-108.

Buikhuisen, W., & Hoekstra, H. A. (1974). Factors related to recidivism. British Journal of Criminology, 14, 62-69.

Buikhuisen, W., & Meijs, B. W. G. P. (1983). A psychosocial approach to recidivism. In K. T. Van Dusen & S. A. Mednick (Eds.), Prospective studies of crime and delinquency (pp. 99-113). Boston, MA: Kluwer-Nijhoff.

Carlson, K. A. (1973). Some characteristics of recidivists in an Ontario institution for adult male first incarcerates. Canadian Journal of Criminology and Corrections, 15, 397-409.

Carroll, J. S., Wiener, R. L., Coates, D., Galegher, J., & Alibrio, J. J. (1982). Evaluation, diagnosis, and predicting in parole decision making. Law and Society Review, 17, 199-228.

Cloninger, C. R., & Guze, S. B. (1973). Psychiatric disorders and criminal recidivism. Archives of General Psychiatry, 29, 266-269.

Donnelly, H. C., & Bala, G. H. (1991). 1985 releases: Five year post release follow-up. Albany, NY: Department of Correctional Services.

Eron, L. D., & Huesmann, L. R. (1984). The relation of prosocial behaviour to the development of aggression and psychopathology. Aggressive Behavior, 10, 201-211.

Farrington, D. (1983). Offending from 10-25 years of age. In K. Van Dusen & S. Mednick (Eds.), Prospective studies of crime and delinquency (pp. 17-33). Boston, MA: Kluwer-Nijhoff.

Farrington, D. P., Biron, L., & Leblanc, M. (1982). Personality and delinquency in London and Montreal. In J. Gunn & D. P. Farrington (Eds.), Abnormal offenders, delinquency and the criminal justice system (pp. 153-195). New York: Wiley.

Farrington, D. P., Gundry, G., & West, D. J. (1975). The familial transmission of criminality. Medicine, Science, and the Law, 15, 177-86.

Feder, L. (1991). A comparison of the community adjustment of mentally ill offenders with those from the general prison population: An 18-month follow up. Law and Human Behavior, 15, 477-493.

Gabrielli, W. F., & Mednick, S. A. (1983). Genetic correlates of criminal behaviour. American Behavioral Scientist, 27, 59-74.

Gendreau, P., Grant, B.A., & Leipciger, M. (1979) Self-esteem, incarceration and recidivism. Criminal Justice and Behavior, 6, 67-75.

Gendreau, P., Madden, P., & Leipciger, M. (1979). Norms and recidivism for first incarcerates: Implications for programming. Canadian Journal of Criminology, 21, 416-441.

Gibbens, T. C. N. (1984). Borstal boys after 25 years. British Journal of Criminology, 24, 51-62.

Gordon, A. M. (1983). Drugs and delinquency: A ten year follow-up of drug clinic patients. British Journal of Psychiatry, 142, 169-173.

Gottfredson, D. M., Wilkins, L. T., & Hoffman, P. B. (1978). Guidelines for parole and sentencing: A policy control method. Lexington, MA: DC Heath.

Gottfredson, M. R., & Gottfredson, D. R. (1980). Decision making in criminal justice: Toward the rational exercise of discretion. Cambridge, MA: Ballinger.

Greenberg, W. M., Shah, P. J., & Seide, M. (1993). Recidivism on an acute psychiatric forensic service. Hospital and Community Psychiatry, 44, 583-585.

Grygier, T., Blum, F., & Porebski, R. (1971). Decision and outcome: studies in parole prediction. Canadian Journal of Criminology and Corrections, 13, 133-146.

Gunn, J., Robertson, G., Dell, S., & Way, C. (1978). Psychiatric aspects of imprisonment. London: Academic Press.

Hamparian, D. (1987). How well can we predict for juveniles? Juvenile delinquency and adult crime. In F. N. Dutile & C. H. Foust (Eds.), The prediction of criminal violence (pp. 169-179). Chicago, IL: Charles C. Thomas.

Hann, R. G., & Harman, W. G. (1988). Release risk prediction: Testing the Nuffield scoring system for native and female inmates. Ottawa: Ministry of the Solicitor General of Canada.

Hann, R. G., & Harman, W. G. (1992). Predicting general release risk for Canadian penitentiary inmates. User Report # 1992-07, Ottawa: Solicitor General of Canada.

Hare, R. D., McPherson, L. M., & Forth, A. (1988). Male psychopaths and their criminal careers. Journal of Consulting and Clinical Psychology, 56, 710-714.

Hart, S. D., Kropp, P. R., & Hare, R. D. (1988). Performance of male psychopaths following conditional release from prison. Journal of Consulting and Clinical Psychology, 56, 227-232.

Hodgins, S. (1983). A follow-up study of persons found incompetent to stand trial and/or not guilty by reason of insanity in Quebec. International Journal of Law and Psychiatry, 6, 399-411.

Hodgins, S. (1987). Men found unfit to stand trial and/or not guilty by reason of insanity: Recidivism. Canadian Journal of Criminology, 29, 51-70.

Hoffman, P. B. (1983). Screening for risk: A revised Salient Factor Score (SF81). Journal of Criminal Justice, 11, 539-547.

Hoffman, P. B., & Beck, J. L. (1974). Parole decision-making: A salient factor score. Journal of Criminal Justice, 33, 195-206.

Hoffman, P. B., & Beck, J. L. (1976). Salient Factor Score validation: A 1972 release cohort. Journal of Criminal Justice, 4, 69-76.

Hoffman, P. B., & Beck, J. L. (1980). Revalidating the Salient Factor Score: A research note. Journal of Criminal Justice, 8, 185-188.

Hoffman, P. B., & Beck, J. L. (1984). Burnout-age at release from prison and recidivism. Journal of Criminal Justice, 12, 617-623.

Hoffman, P. B., & Beck, J. L. (1985). Recidivism among released federal prisoners: Salient Factor Score and five-year follow-up. Criminal Justice and Behavior, 12, 501-507.

Hoffman, P. B., Stone-Meierhoefer, B., & Beck, J. L. (1978). Salient factor score and release behaviour: Three validation samples. Law and Human Behavior, 2, 47-62.

Holcomb, W. R., & Ahr, P. R. (1988). Arrest rates among young adult psychiatric patients treated in inpatient and outpatient settings. Hospital and Community Psychiatry, 39, 52-57.

Holland, T. R., Holt, N., Levi, M., & Beckett, G. E. (1983). Comparison and combination of clinical and statistical predictions of recidivism among adult offenders. Journal of Applied Psychology, 68, 203-211.

Huesmann, R. L., Eron, L. D., Lefkowitz, M. M., & Walder, L. O. (1984). Stability of aggression over time and generations. Developmental Psychology, 20, 1120-1134.

Janes, C. L., Hesselbrock, V. M., Myers, D. G., & Peniman, J. H. (1979). Problem boys in young adulthood: Teacher ratings and twelve-year follow-up. Journal of Youth and Adolescence, 8, 453-472.

Jenkins, W. O., DeValera, E. K., & Muller, J. B. (1977). The Behavioral Evaluation Treatment and Analysis (BETA) system in the prediction of criminal and delinquent behaviour. Quarterly Journal of Corrections, 1, 44-50.

Kassebaum, G., Ward, D., & Wilner, D. (1971). Prison treatment and parole survival: An empirical assessment. Toronto: Wiley.

Klein, S. P., & Caggiano, M. N. (1986). The prevalence, predictability, and policy implications of recidivism. Santa Monica, CA: Rand Corporation.

Kolvin, I., Miller, F. J. W., Fleeting, M., & Kolvin, P. A. (1988). Social and parenting factors affecting criminal-offence rates. British Journal of Psychiatry, 152, 80-90.

Lambert, L. R., & Madden, P. G. (1976). The adult female offender: The road from institution to community life. Canadian Journal of Criminology and Corrections, 18, 3-15.

Lane, D. A. (1987). Personality and antisocial behaviour: A long-term study. Personality and Individual Differences, 8, 799-806.

LeBlanc, M., & Fréchette, M. (1989). Male criminal activity from childhood through youth: Multilevel and developmental perspectives. New York: Springer-Verlag.

LeBlanc, M., Vallières, E., & McDuff, P. (1993). The prediction of males' adolescent and adult offending from school experience. Canadian Journal of Criminology, 25, 459-478.

Liberton, M., Silverman, M., & Blount, W. R. (1992). Predicting probation success for the first-time offender. International Journal of Offender Therapy and Comparative Criminology, 36, 335-347.

Little, G. I., & Robinson, K. D. (1989). Relationship of DUI recidivism to moral reasoning, sensation seeking, and MacAndrew alcoholism scores. Psychological Reports, 65, 1171-1174.

Loney, J., Whaley-Klahn, M. A., Kosier, T., & Conboy, J. (1983). Hyperactive boys and their brothers at 21: Predictors of aggressive and antisocial outcomes. In K. T. Van Dusen & S. A. Mednick (Eds.), Prospective studies of crime and delinquency (pp. 181-205). Boston, MA: Kluwer-Nijhoff.

MacNaughton-Smith, P. (1976). Permission to be slightly free. Ottawa: Law Reform Commission of Canada.

McCord, J. (1982). A longitudinal view of the relationship between paternal absence and crimes. In J. Gunn & D. Farrington (Eds.), Abnormal offenders, delinquency, and the criminal justice system (pp. 113-128). Chichester: Wiley.

McCord, J. (1983). A forty year perspective on effects of child abuse and neglect. Child Abuse and Neglect, 7, 265-270.

McGarvey, B., Gabrielli, W. F., Bentler, P. M., & Mednick, S. A. (1981). Rearing social class, education, and criminality: A multiple indicator model. Journal of Abnormal Psychology, 90, 354-364.

McGurk, B. J., Bolton, N., & Smith, M. (1978). Some psychological, educational, and criminological variables related to recidivism in delinquent boys. British Journal of Social and Clinical Psychology, 17, 251-54.

McWilliams, W. (1975). Sentencing and recidivism: An analysis by personality type. British Journal of Social Work, 5, 311-324.

Mednick, B. R., Baker, R. L., & Carothers, L. E. (1990). Patterns of family instability and crime: The association of timing of the family's disruption with subsequent adolescent and young adult criminality. Journal of Youth and Adolescence, 19, 201-219.

Mednick, B., Reznick, C., Hocevar, D., & Baker, R. (1987). Long term effects of parental divorce on young adult male crime. Journal of Youth and Adolescence, 16, 31-45.

Minor, K. I., & Hartmann, D. J. (1992). An evaluation of the Kalamazoo probation enhancement program. Federal Probation, 56, 30-35.

Mitchell, S., & Rosa, P. (1981). Boyhood behaviour problems as precursors of criminality: A fifteen year follow-up study. Journal of Child Psychology and Psychiatry, 22, 19-33.

Moffit, T. E., (1987). Parental mental disorder and offspring criminal behaviour: An adoption study. Psychiatry, 50, 346-360.

Moffitt, T. E., Gabrielli, W. F., Mednick, S. A., & Schulsinger, F. (1981). Socioeconomic status, IQ, and delinquency. Journal of Abnormal Psychology, 90, 152-156.

Motiuk, L.L. (1991). Antecedents and consequences of prison adjustment: A systematic assessment and reassessment approach. Unpublished doctoral dissertation, Carleton University, Ottawa, Ontario.

Motiuk, L. L., & Bonta, J. M. (1991, June). Prediction and matching in corrections: An examination of the risk principle in case classification. Paper presented at the Canadian Psychological Association Annual Convention, Calgary, AB, Canada.

Motiuk, L. L., Bonta, J., & Andrews, D. A. (1986). Classification in correctional halfway houses: The relative and incremental predictive criterion validities of the Megargee-MMPI and LSI-R systems. Criminal Justice and Behavior, 13, 33-46.

Motiuk, L. L., & Brown, S. L. (1993). The validity of offender needs identification and analysis in community correction (Report No. R-34). Ottawa, Canada: Research and Statistics Branch, Correctional Service of Canada.

Motiuk, L. L., & Brown, S. L. (1993). Survival time until suspension for sex offenders on conditional release (Report No. R-31). Ottawa: Research and Statistics Branch, Correctional Service Canada.

Motiuk, L. L., & Porporino, F. J. (1988). Offender risk/needs assessment: A study of conditional releases (Report No. R-01). Ottawa: Correctional Service of Canada.

Motiuk, L. L., & Porporino, F. J. (1989). Field test of the community risk/needs management scale: A study of offenders on case load (Report No. R-06). Ottawa: Research Branch, Correctional Service Canada.

O'Donnell, C. R., & Stanley, K. G. (1974). An adult furlough center: Correlates of parole success. Journal of Community Psychology, 2, 83-85.

Pallone, N. J., & Hennessey, J. J. (1977). Some correlates of recidivism among misdemeanants and minor felons: A 12-month follow-up. Journal of Social Psychology, 101, 321-322.

Pallone, N. J., & Hennessey, J. J. (1977). Empirical derivation of a scale for recidivism proneness among parolees: A multivariate model. Offender Rehabilitation, 2, 95-110.

Payne, C., McCabe, S., & Walker, N. (1974). Predicting offender-patients' reconvictions. British Journal of Psychiatry, 125, 60-64.

Petersén, K. G. I., Matousek, M., Mednick, S. A., Volavka, J., & Pollack, V. (1982). EEG antecedents of thievery. Acta Psychiatrica Scandinavica, 65, 331-38.

Petersilia, J., & Turner, S. (1987). Guideline-based justice: Prediction and racial minorities. In D. M. Gottfredson & M. Tonry (Eds.), Prediction and classification: Criminal justice decision making (pp. 151-177). Chicago, IL: University of Chicago Press.

Platt, J. J., & Labate, C. (1976). Recidivism in youth heroin offenders and characteristics of parole behaviour and environment. International Journal of Addictions, 11, 651-657.

Porporino, F. J., Zamble, E., & Higginbottom, S. (in press). Assessing models of predicting risk for criminal recidivism. Journal of Criminal Justice.

Putnins, A. L. (1982). The Esyenck Personality Questionnaires and delinquency prediction. Personality and Individual Differences, 3, 339-340.

Quinsey, V. L., Pruesse, M., & Fernley, R. (1975). Oak Ridge patients: Prerelease characteristics and postrelease adjustment. Journal of Psychiatry and Law, 3, 63-77.

Rice, M. E., & Harris, G. T. (1991). A comparison of criminal recidivism among schizophrenic and nonschizophrenic offenders. International Journal of Law and Psychiatry, 15, 397-408.

Rice, M. E., & Harris, G. T. (1993). Psychopathy, schizophrenia, alcohol abuse and violent recidivism. (Research Report Vol. X, No.3). Penetanguishene, ONT, Canada: Mental Health Centre.

Rice, M. E., Harris, G. T., Lang, C., & Bell, V. (1990).Criminological and psychiatric predictors of recidivism among male insanity acquittees. (Research Report Vol. VII, No. 4). Penetanguishene, ONT, Canada: Mental Health Centre.

Rice, M. E., Quinsey, V. L., & Harris, G. T. (1991). Sexual recidivism among child molesters released from a maximum security psychiatric institution. Journal of Consulting and Clinical Psychology, 59, 381-386.

Robins, L. N., & Ratcliffe, K. S. (1980). Childhood conduct disorders and later arrest. In L. N. Robins, P. J. Clayton, & J. K. Wing (Eds.), The social consequences of psychiatric illness (pp. 248-263). New York: Brunner/Mazel.

Robinson, D., & Porporino, F. J. (1989). Validation of an adult offender classification system for Newfoundland and Labrador (Report No. R.04). Ottawa: Correctional Service Canada.

Rogers, S. (1981). Factors related to recidivism among adult probationers in Ontario. Toronto: Ontario Ministry of Correctional Services.

Roundtree, G. A., Edwards, D. W., & Parker, J. B. (1984). A study of the personal characteristics of probationers as related to recidivism. Journal of Offender Counselling, Services, and Rehabilitation, 8, 53-61.

Sampson, A. (1974). Post-prison success prediction. Criminology, 12, 155-173.

Serin, R. C. (1990). Recidivism prediction: A comparison of the Psychopathy Checklist and actuarial risk scales. Ottawa: Correctional Service Canada.

Serin, R. C., & Barbaree, H. E. (1993). Decision issues in risk assessment. Federal Probation, 5, 22-25.

Serin, R. C., Peters, R. D., & Barbaree, H. E. (1990). Predictors of psychopathy and release outcome in a criminal population. Psychological Assessment: A Journal of Consulting and Clinical Psychology, 2, 419-422.

Shannon, L. W. (1985). Risk assessment vs. real prediction: The prediction problem and public trust. Journal of Quantitative Criminology, 1, 159-185.

Stattin, H., & Magnusson, D. (1989). The role of early aggressive behaviour in the frequency, seriousness and types of later crime. Journal of Consulting and Clinical Psychology, 57, 710-718.

Steadman, H. J. (1973). Follow-up on Baxstrom patients returned to hospitals for the criminally insane. American Journal of Psychiatry, 130, 317-319.

Steadman, H. J., & Cocozza, J. J. (1974). Careers of the criminally insane: Excessive social control of deviance. Lexington, MA: D.C. Heath and Company.

Steadman, H. J., Cocozza, J. J., & Melick, M. E. (1978). Explaining the increased arrest rate among mental patients: The changing clientele of state hospitals. American Journal of Psychiatry, 135, 816-820.

Tarling, R., & Perry, J. A. (1985). Statistical methods in criminological prediction. In D.P. Farrington & R. Tarling (Eds.), Prediction in criminology (pp 150-173). Albany, NY: State University of New York Press.

Tennent, G., & Way, C. (1984). The English special hospital- A 12-17 year follow up study: A comparison of violent and nonviolent re-offenders and non-offenders. Medicine, Science, and the Law, 24, 81-91.

Thornberry, T. P., & Jacoby, J. E. (1979). The criminally insane: A community follow-up of mentally ill offenders (pp. 178-198). Chicago, IL: University of Chicago Press.

Wallander, J. L. (1988). The relationship between attention problems in childhood and antisocial behaviour eight years later. Journal of Child Psychology and Psychiatry, 29, 53-61.

Werner, E., & Palmer, T. (1976). Psychological characteristics of successful and unsuccessful parolees: Implications of heteroscedastic and non-linear relationships. Journal of Research in Crime and Delinquency, 13, 165-178.

West, D. J., & Farrington, D. P. (1977). The delinquent way of life. London: Heinemann.

Wheeler, G. R., & Hissong, R. V. (1990). Transferability of probation risk-assessment instruments: A case for caution. Evaluation and Program Planning, 13, 399-406.

Widom, C. S. (1989). The cycle of violence. Science, 244, 160-166.

Wilbanks, W. L. (1985). Predicting failure on parole In D. P. Farrington & R. Tarling (Eds.), Prediction in criminology (pp. 78-94). Albany, NY: State University of New York Press.

Wormith, J. S., & Goldstone, C. S. (1984). The clinical and statistical prediction of recidivism. Criminal Justice and Behavior, 11, 3-34.

Wormith, J. S., & Ruhl, M. (1986). Preventive detention in Canada. Journal of Interpersonal Violence, 1, 399-430.

Wright, K. N., Clear, T. R., & Dickson, P. (1984). Universal applicability of probation risk-assessment instruments. Criminology, 22, 113-134.

Zamble, E., & Porporino, F. (1990). Coping, imprisonment, and rehabilitation: Some data and their implications. Criminal Justice and Behavior, 17, 53-70.