Notes on the development of Static-2002 2003-01

PDF Version (517 KB)

R. Karl Hanson
Department of the Solicitor General of Canada 

David Thornton
Sand Ridge Secure Treatment Center
Department of Health and Family Services, WI 

Author's Note

We would like to thank Ron Coles, Jean Proulx, Larry Motiuk, Marylee Stephenson, John Reddon, Lea Studer, Janice Marques, Roxanne Lieb and Lin Song for access to their original data sets.

Correspondence concerning this report can be addressed to either author.

R. Karl Hanson
Corrections Research
Department of the Solicitor General of Canada
340 Laurier Ave., West
Ottawa, Ontario
Canada
K1A 0P8

Telephone: (613) 991 2840   
Fax: (613) 990 8295
email: hansonk@ps-sp.gc.ca

David Thornton
Sand Ridge Secure Treatment Center
Department of Health and Family Services of Wisconsin
1111 North Road
Mauston, Wisconsin 53948
U.S.A.

Telephone: (608) 847 1744    Fax: (608) 847 1749
email: thorndm@dhfs.state.wi.us

ABSTRACT

This paper describes the development of a new risk scale for sexual offenders, Static-2002. Like Static-99 (Hanson & Thornton, 1999, 2000), Static-2002 is intended to be a widely applicable risk scale for the prediction of sexual recidivism, which can be coded using commonly available file information. Preliminary analyses (n = 2,169; k = 7) suggest that Static-2002 shows promise, but further research is needed before it can be used in applied contexts. This description of Static-2002 is presented with the hope that researchers can examine its utility and predictive accuracy in new samples.

NOTES ON THE DEVELOPMENT OF STATIC-2002

Why the need for a revision?

Static-99 (Hanson & Thornton, 2000) has been widely adopted as a measure of sex offence recidivism risk, with routine applications in jurisdictions as diverse as Sweden, Texas, and Taiwan. Static-99 is intended to assess the long-term potential for sexual and violent recidivism based on objective, easily obtainable information, such as official criminal history, victim characteristics, and age. The 10 items used in Static-99 were selected from the non-redundant items in two pre-existing risk scales: The Rapid Risk Assessment for Sex Offence Recidivism (RRASOR) (Hanson, 1997) and Thornton's Structured Anchored Clinical Judgement scale (SAC-J) (Grubin, 1998).

Replication studies have found that Static-99 shows levels of predictive accuracy similar to those found in the developmental samples (Barbaree, Seto, Langton, & Peacock, 2001; Nunes, Firestone, Bradford, Greenberg, & Broom, 2002; Sjöstedt, & Långström, 2001). Nevertheless, there are several reasons for revising the measure. The first is a desire for increased coherence and conceptual clarity. Scales that have been developed on a purely empirical basis can be effective in predicting recidivism. However, if evaluators are to considered variables external to an actuarial risk scale, then it is necessary to understand what the scale is already assessing.

Another goal was to improve the consistency of the scoring criteria. Static-99 was created by combining two different scales, each with their own coding rules. As well, Static-99 used the definitions of variables already coded in existing data sets. Consequently, those learning Static-99 are often confronted with unanticipated changes in the coding principles across items. Training would be facilitated and rater reliability improved by increasing the consistency of the coding rules.

Another concern with Static-99 is that certain (rare) cases result in counterintuitive scorings. For example, the score for an offender with a conviction for index non-sexual violence may decrease if he subsequently reoffends with a sexual offence.

It was also hoped that by adding new variables and refining the definitions of the existing variables, the predictive accuracy could increase. In the absence of a marked improvement in predictive accuracy, Static-2002 would still be an improvement over Static-99 if it retained the same levels of predictive accuracy while being simpler and easier to score.

The overall goals of the revision are the same as that of the original Static-99 (and RRASOR and SAC-J): namely, a widely applicable risk scale for the prediction of sexual recidivism that could be reliably scored from commonly available information. As with Static-99, it was hoped that the scale would show similar levels of predictive accuracy for institutional and community samples of rapists, child molesters and exhibitionists.

The emphasis in scale construction was on the prediction of sexual recidivism. Preliminary analyses were conducted on the current data sets in an attempt to create a separate scale for the prediction of any violent recidivism (including sexual) among sexual offenders. The resulting violence scale was sufficiently similar to the sexual recidivism scale that only the sexual recidivism scale was retained.

OVERALL STRATEGY FOR REVISION

Variables considered

The selection of variables for Static-2002 was guided by previous research on factors associated with sexual offence recidivism. Major sources of variables were the Hanson and Bussière (1998) meta-analysis as well as other empirically developed sex offender risk scales, such as the SORAG (Quinsey, Harris, Rice, & Cormier, 1998) and the MnSOST-R (Epperson, Kaul, & Hesselton, 1998). One important data set contained all sexual offenders released in British Columbia between 1980 and 1992 (n = 727; Hanson, Broom & Stephenson, 2001). The criminal history records in the Hanson et al. (2001) data set were deliberately coded in alternate ways allowing for empirical comparisons between coding methods (e.g., counting charges versus sentencing occasions). We also included a number of exploratory variables that were suggested by the constructs we were attempting to assess (e.g., age range of victims).

In keeping with the goals of Static-99, a central concern in variable selection was the ease of data collection. The only variables considered were the offender's age, his officially documented criminal history and the characteristics of the victims of sexual offences (age, relationship to offender, gender). The Static-99 item "Ever lived with a lover for two years" was deleted from Static-2002 because of evaluators' concerns that it was poorly documented in the offender records, and hard to validate in adversarial contexts. The resulting 22 individual variables were organized into the following five content areas: age at release, persistence of sex offending, deviant sexual interests, range of available victims, and general criminality (see Table 1).

Selecting and weighting variables

Scale construction was strongly grounded in the available data, but was not determined by it. There are any number of possible arrangements of the data that could yield similar levels of predictive accuracy. Other researchers following the same general procedure with the same data should identify the same content areas (and the same univariate relationships), but the specific cut-points and weightings could vary substantial (e.g., "age < 30" versus "age < 35").

The principles guiding the selection and weighting of items were simplicity and relevance. Whenever possible, decisions between alternate coding approaches were empirically based. When two alternate coding methods produced equivalent predictive accuracy, then the simpler coding method was selected. In general, simple weights work as well as more complex systems (Silver, Smith & Banks, 2000).

Table 1 Content areas and variables considered

Expected relationship with recidivism: + = weak (r = .10 to .15 range); ++ = moderate (r = .15 to .20 range); 0 = none; ? = unknown.

Scale construction started by examining the univariate relationships with the sexual recidivism. Variables without significant univariate relationships with sexual recidivism were excluded. The next step was to combine the significant variables within each content area. Given that the variables within the content areas were expected to be correlated (e.g., unrelated and stranger victims), there was no requirement that all the items within a content area independently contributed to the prediction of sexual recidivism. Instead, the variable could be retained if it resulted in any improvement in predictive accuracy. The inclusion of multiple indicators of similar constructs should raise the reliability of assessment.

Once the items within a content area were defined, multivariate analyses were conducted to determine whether the subscale added incrementally over the subscales already considered. Subscales that made unique contributions were weighted so that a unit increase in each subscale was associated with approximately the same amount of increase in recidivism risk as for the other subscales (odds ratios in logistic regression or Cox regression).

None of the data sets contained all of the variables under examination. Consequently, the analyses started with the data sets containing the most complete information, and then approximations would be created in other data sets. The analysis of the content areas was conducted with all the data available for that content area.

Several initial scales were tested and refined across subgroups (total, rapists, child molesters, extrafamilial child molesters, intrafamilial offenders). Slight differences were observed across groups (e.g., age was more important for rapists than extrafamilial child molesters), but none of the resulting scales showed substantially different levels of predictive accuracy. Consequently, the definitions/weightings used were those that appeared to be the most widely applicable.

Samples

An overview of the samples can be found in Table 2. All the offenders were released from institutions with the exception of the Manitoba Probation sample and about half of the offenders from the Washington sample, who received community sentences. Racial ethnicity was not recorded for most samples, but given the demographics of the provinces, states and countries from which they were selected, the offenders can be expected to be predominantly white. All offenders were adult males (18 years old or older at time of release).

All samples included information concerning the outcome variables of sexual and violent recidivism, and all but one study (Alberta Hospital Edmonton) included survival times. All studies had information concerning the predictor variables of age and prior sexual offences. The remaining predictor variables were missing in at least one study. No study was without missing data on at least three of the 22 variables.

Table 2 Study characteristics.

Note: EX = Extrafamilial child molesters; IN = Intrafamilial child molesters

Canadian Federal - Pacific Region (CS/RESORS Consulting, 1991; Hanson, et al., 2001). This study followed sexual offenders released in British Columbia between 1976 and 1992. The original aim of the study was to compare offenders who received mandatory community counselling (n = 401) and those released in earlier years without the benefit of this post-release program (n = 288). Offenders released in the 1983/84 fiscal year (n = 38) were removed from this sample to avoid overlap with the other Correctional Service of Canada cohort described below. Recidivism information was coded in 2000 from Royal Canadian Mounted Police records. Information was not available concerning relationship history and victim characteristics.

Canadian Federal Recidivism Study - 1983/1984 Releases (Bonta & Hanson, 1995a; see also Bonta & Hanson, 1995b). This study examined the 316 sexual offenders included in the complete sample of 3,180 federal offenders released by the Correctional Service of Canada (CSC) in the fiscal year 1983/1984. Sexual offenders were defined as those who were released following any sexual conviction. Recidivism information was collected in 1994 using national criminal history records maintained by the Royal Canadian Mounted Police (RCMP). Information was not available for victims, age at first sex arrest, age at first arrest for violence, and non-contact sex offences. Sexual sentencing occasions was approximated as 2/3 of convictions.

Canadian Federal 1991 to 1994 Releases (Motiuk, 1995; see also Motiuk & Brown, 1993; Motiuk & Brown, 1996). This study followed a group of sexual offenders released by CSC between 1991 and 1994. The offenders in this group were those who were reviewed in 1991 (see Motiuk & Porporino, 1993) while they were still incarcerated. Follow-up information was coded from 1994 RCMP records. Information was not available for breach and time free before index. Sexual sentencing occasions were approximated as 2/3 of convictions; non-contact sex offences were approximated by index convictions for exhibitionism; sentencing occasions for anything were approximated by twice prison sentences.

Millbrook Recidivism Study (Hanson, Steffy & Gauthier, 1993b; see also Hanson, Scott & Steffy, 1995; Hanson, Steffy & Gauthier, 1992, 1993a). This study collected long-term recidivism information (15-30 years) for child molesters released between 1958 and 1974 from Millbrook Correctional Centre, a maximum security provincial correctional facility located in Ontario, Canada. About half of the sample went through a brief treatment program. Recidivism information was coded from RCMP records in 1989 and 1991. Information was not available for breach, non-contact sex offences, time free before index, and age of first violent offence. Sentencing occasions for sex offences was approximated by 2/3 convictions; sentencing occasions for anything was approximated by 2/3 convictions for anything.

Institut Philippe Pinel (Montreal). (Proulx, Pellerin, McKibben, Aubut & Ouimet, 1995; see also Proulx, Pellerin, McKibben, Aubut & Ouimet, 1997; Pellerin et al., 1996). This study focused on sexual offenders treated at a maximum security psychiatric facility between 1978 and 1993. The Institut Philippe Pinel in Montreal provides long term (1-3 years) treatment for sexual offenders referred from both the mental health and correctional systems. Recidivism information was collected in 1994 from RCMP records. Information was not available concerning stranger victims, age at first sex offence, breach, non-contact sex offences, age at first any offence, time free before index, index non-sexual violence, age at first violent offence. Sentencing occasions for sex offences was approximated by ½ charges.

Alberta Hospital Edmonton - Phoenix Program. (Reddon, 1996; see also Studer, Reddon, Roper & Estrada, 1996). The sexual offenders in this study were drawn from those treated at the Phoenix (Alberta Hospital Edmonton) program between 1987 and 1994. The Phoenix program is an eclectic inpatient treatment program that receives many of its referrals from federal correctional facilities. Recidivism information was collected in 1995 using RCMP records. Information was not available concerning age at first sex offence, breach, non-contact sex offence, age at first arrest for anything, time free before index, and all variables concerning non-sexual violence. Sentencing occasions for sex offences were approximated by 2/3 convictions; sentencing occasions for anything were approximated by 2/3 convictions.

California's Sex Offender Treatment and Evaluation Project (SOTEP). (Marques & Day, 1996; see also Marques, Day, Nelson & West, 1993; Marques, Nelson, West & Day, 1994). The primary aim of this ongoing study is to examine the efficacy of treatment. The sample used in the current study included sexual offenders randomly assigned to treatment (n = 172), matched volunteer controls, treatment refusers, as well as a general sample of sexual offenders from the California correctional system (total sample of 1,137). Men who had offended only against their biological children were not included. Subjects were admitted to this study between 1985 and 1995; follow-up information was collected in 1995 based on local and national criminal records, as well as local police and probation reports. Information was not available concerning age at first sex offence, breach, non-contact sex offences, sentencing occasions for anything, age at first arrest for anything, time free prior to index, and all variables related to non-sexual violence.

Her Majesty's Prison Service (UK). (Thornton, 1997). This study provided a 16 year follow-up of all sexual offenders released from Her Majesty's (HM) Prison Service (England and Wales) in 1979 (n = 573). Recidivism information was based on Home Office records collected in 1995. Very few of the offenders in this sample would have received specialised sexual offender treatment. Information was missing concerning breach, time free prior to index, and juvenile arrests for sexual offences.

Washington. (Berliner, Schram, Miller & Milloy, 1995; Song & Lieb, 1995). This data set was created to evaluate Washington State's Special Sex Offender Sentencing Alternative (SSOSA), which allows judges to sentence sex offenders to community treatment. To be eligible for SSOSA, offenders must be facing their first felony conviction for sexual crimes other than first or second degree rape. The sample consisted of 287 offenders who received SSOSA and 300 who were statutorily eligible for SSOSA but did not receive it. The majority of the sample was White (85%). Offenders were convicted between January 1985 and June 1986, with follow-up data collected in December, 1990. Information was missing concerning breach, time free prior to index and all variables concerning non-sexual violence. Sentencing occasions for sex offences were approximated by 2/3 convictions; any non-contact sex offences was approximated by "index convictions were only exhibitionism or peeping"; sentencing occasions for anything were approximated by 2/3 convictions.

Manitoba Probation. (Hanson, 2002). This follow-up study was conducted as an evaluation of a risk scale used by probation officers in Manitoba, Canada. The 202 offenders were consecutive admissions to probation between May, 1997 and February, 1999. Recidivism information was collected in November, 2000, based on RCMP records. Unlike the RCMP records used in the other studies (which included only charges and convictions that went to court), the RCMP records for the Manitoba sample included unresolved charges and cases currently under police investigation. The demographic, victim and offence information was collected by the probation officers in the course of their normal duties. The criminal history variables needed for coding Static-99 were coded by trained research assistants (median r = .93; median Kappa = .87). Information was missing concerning male victims, breach, age at first arrest for anything, time free prior to index, and age at first arrest for violence. Sentencing occasions for sex offences was approximated by 2/3 convictions; any juvenile sexual offences was approximated by "sex offence committed prior to age 20".

Plan of analysis

There are two common approaches for analysing data from multiple data sets. One approach is to ignore differences between samples and combine the results into one big sample (combined approach). The other option identifies the relationships within each sample, then summarizes the results across samples (nested approach). The summary of nested effects can be accomplished through meta-analysis (Hedges & Olkin, 1985), or certain statistical procedures (e.g., Cox regression with each sample as a separate strata). The combined and the nested approaches often yield equivalent results, but when differences occurs, the nested/meta-analytic results are the more dependable. Combining samples can artificially increase effects (e.g., the sample with the largest proportion of high risk offenders is also the sample followed for the longest period of time) or decrease effects (e.g., the sample with the largest proportion of low risk offenders has the broadest criteria for recidivism). Given that difference results were observed for some variables using the combined versus the nested approaches, the nested approach was adopted as the primary method of data analysis.

RESULTS

In the combined sample, rapists were slightly higher risk to sexually recidivate (16%) than were child molesters (13%, p < .05), but the effect was largely due to the relatively low recidivism rates of intrafamilial child molesters (5%) compared to extrafamilial child molesters (18%). The rates for any violent recidivism (including sexual) were higher for rapists (34.0%) than extrafamilial child molesters (23.8%) and incest offenders (10.9%).

All of the individual variables in Table 1 were significantly related to sexual recidivism with the following exceptions: age youngest victim, age oldest victim, age range of victims, victims inside and outside the family, and index non-sexual violence. Consequently, these five variables were excluded from further analysis.

Some comments on the coding of specific variables is given below.

Age. Age was more strongly related to violent recidivism than sexual recidivism. As well, the age effect was stronger for rapists than child molesters. Nevertheless, there was a significant decline in all forms of recidivism with age (see Hanson, 2001). The coding selected (see Appendix A) was intermediate between the codings that appeared optimal for predicting sexual recidivism among child molesters (the lowest weights) and predicting violent recidivism in the total sample (the highest weights).

Persistence of sexual offending. Three variables were considered as indicators of the persistence of sexual offending: the number of prior sexual offences, the rate of sexual offending, and the age of first arrest for a sexual offence.

In order to choose between the various possible codings of prior sexual offences, the CSC Pacific data set (Hanson et al., 2001) was coded with four different definitions: arrests, charges, convictions, and sentencing occasions (for both index and prior offences). The number of index offences (whatever the definition) was not related to sexual offence recidivism. Prior sexual offences, however, consistently predicted recidivism, with the largest correlation being for sentencing occasions (r = .21, with r = .19 for the other three codings). Friendship, Thornton, Erikson and Beech (2001) found that in the UK records, sentencing occasions were more reliably recorded than were other indicators of prior offending. Consequently, sentencing occasions was selected because it was not only the most accurate, but also the simplest.

Examination of the cross-tabulation of prior sentencing occasions and recidivism suggested the following coding: 0 = 0; 1 = 1; 2, 3 = 2; 4 or more = 3. For those data sets using other definitions of prior sexual offences, the number of prior sentencing occasions was estimated using the ratios between convictions/charges/sentencing occasions observed in the CSC-Pacific data set: 1 sentencing occasion = 1.5 convictions or 2 charges.

The rate of sexual offending was created by dividing the total number of sentencing occasions for sexual offenses by the offender's age at release. Examination of the distribution suggested a useful break point of once every 15 years (.0667).

The age at first arrest for a sexual offence was strongly related to sexual recidivism. Given that this variable was correlated with both the offender's current age and the number of prior sexual offences, it did not independently contribute when these other variables were considered. Examination of the limited available data suggested that the existence of arrests both as a juvenile and as an adult captured enough meaningful variation to warrant inclusion in the Persistence subscale.

Deviant sexual interests. Three variables were considered as indicators of deviant sexual interests: any male victims, any non-contact sexual offences, and multiple young victims (defined as two or more victims less than age 12, at least one of which must be unrelated). All three variables predicted sexual recidivism, both individually and in combination (only one data set, however, contained all three variables - HM Prison Service). Consequently, all three variables were retained and given equal weight (0 = no, 1 = yes).

Range of victims/Relationship to Victims. Previous research has demonstrated that having unrelated victims and strangers as victims were both associated with increased risk for sexual recidivism. One explanation of these findings is that offenders who are willing to victimize stranger have a wider range of potential victims than those who offend only within the family. Consequently, it was hypothesized that having victims of widely different ages could also be a risk marker for sexual recidivism. This was not the case. In the current data sets, the age range of victims was unrelated to sexual recidivism. As well, the offenders who had victims both inside and outside the family (another indicator of range of available victims) were no different in their sexual recidivism rates than offenders who selected only unrelated victims.

Given that the only two variables retained from this section were Any Unrelated Victims and Any Stranger Victims, the section was renamed "Relationship to Victims". Both items independently contributed to the prediction of sexual recidivism (p < .001 using Cox regression). In the combined sample, the sexual recidivism rate was 6% for purely incest offenders, 13% for those who victimized acquaintances, and 21% for those who victimized strangers. One point was assigned to Any Unrelated Victims (0/1) and for Any Stranger Victims (0/1), resulting in a range of 0 to 2.

General criminality. The variables defining general criminality (breach, time free before index, age at first arrest, number of prior offences, prior non-sexual violence) were all intercorrelated (median r = .44) and factor analysis strongly suggested a single factor (eigenvalues of 2.64, .81, .68, .52, .35). Prior non-sexual violence was selected as the measure of non-sexual violence because it was a slightly stronger predictor of sexual recidivism than Any non-sexual violence. As previously mentioned, Index Non-Sexual Violence was not significantly related to sexual recidivism in the combined data set, and was excluded. Age at first arrest was also excluded because the relationships with sexual recidivism were inconsistent across data sets, and information concerning juvenile records was not well recorded.

The number of prior involvements with the criminal justice system was coded as follows: 0 = no prior charges for anything; 1 = any prior charges but less than 3 sentencing occasions; 2 = 3-13 prior sentencing occasions; 3 = 14 or more prior sentencing occasions. The other variables indicative of general criminality were as follows: Any Breach of Conditional Release (0 = none/1 = any), Any prior sentencing occasion for non-sexual violence (0 = none/1 = any), and less than 4 years at liberty prior to the index offence (0 = more than 4 years/1 = less than 4 years).

Combining and weighting subscales.

Sufficient information was available to compare the content areas in 6 of the 10 data sets (CSC - Pacific Region and CSC 83/84 did not have victim information, SOTEP lacked criminal history information, and Alberta Hospital Edmonton lacked survival dates). The six data sets had a total sample of 1,783 with 258 sexual recidivists.

Each of the five content areas independently predicted sexual recidivism (see Table 3). The independent contributions were tested using Cox regression with samples as strata (SPSS, Version 10, 1999). Cox regression is a form of survival analysis that controls for time at risk, as well as both categorical and ordinal covariates (see Cox & Oakes, 1984). By dividing the analysis by strata the results were equivalent to the nested/meta-analytic approach.

Table 3 Contribution of each subscale to the prediction of sexual recidivism (Cox regression).

** p < .01; ***p < .001.

The exponent of the regression coefficients is equivalent to an odds ratio. These exponents (e^B) can be interpreted as the proportion increase in recidivism rates for unit increase in the subscale, after controlling for the other content areas. The initial analysis found that the odds ratios for the Persistence of Sex Offending (1.19) and the General Criminality (1.20) subscales were somewhat smaller than for the other three subscales (range of 1.41 - 1.52). Consequently, these two subscales were recoded so that unit increases in all subscales resulted in similar increases in the sexual recidivism rate. For the Persistence subscale, the range was reduced to 0 - 3 (originally 0 - 5) and for the General Criminality subscale, the range was reduced to 0 - 3 (originally 0 - 6). The resulting exponents, shown in Table 3, suggested a 30% to 40% relative increase for each increase in the scale score. In the combined sample, for example, a score of 2 corresponded to a 4.8% recidivism rate and a score of 3 corresponded to a rate of 6.4% (6.5/4.8 = 1.35).

The final scale for Static-2002 is described in Appendix A.

Scale comparison

To estimate the predictive accuracy of Static-2002, it was compared to its subcomponents/earlier versions: namely, RRASOR (Hanson, 1997) and Static-99 (Hanson & Thornton, 2000; Phenix, Hanson & Thornton, 2000). Direct comparisons were not possible because none of the samples contained all the variables; consequently, comparisons were conducted using approximations for each scale. Static-99 was accepted if there were 3 or less missing items (out of 10). Static-2002 was accepted if there were 5 or less missing items (out of 13), provided that the missing items were not all of the same type (e.g., not all the criminal history items, or all the victim information). If an item was missing, it was scored "zero". RRASOR was scored only if all four variables could be estimated.

None of the scales were computed for the CSC-Pacific and CSC-83/84 samples because victim information was not available. Unfortunately, these were also the only two samples (n = 1,000) that included the variables "breach" and "time free prior to index". Consequently, these two variables were missing for all the Static-2002 scores. The following analyses should, therefore, be considered only as a rough guide to the relative predictive accuracy of the measures. In particular, the analyses would be expect to underestimate the predictive accuracy of Static-99 and Static-2002, while providing reasonable estimates of the accuracy of RRASOR.

Average values for the scales are presented in Table 4. All three measures were positively skewed, although the amount of skew was less for Static-2002 (.33, SD of skewness = .053) than for RRASOR (.80) and Static-99 (.59). 

Table 4 Average scores on RRASOR, Static-99, and Static-2002

For sexual recidivism, the predictive accuracy of Static-2002 was similar to that of Static-99 (ROC areas of .69 to .73 range; see Table 5). In the full sample, Static-2002 demonstrated an ROC area of .711 compared to an ROC area of .701 for Static-99 (Z = 1.14, p > .10, using Formula 3 from Hanley & McNeil, 1983). An advantage of Static-2002, however, was that it showed little variability across samples (Q = 5.65, p >.40) in comparison to Static-99 (Q = 14.08, p < .025). For Static-99, the lowest predictive accuracy was found for the two community samples (Washington and Manitoba, ROC areas of .59 for both), and the highest predictive accuracies were for the CSC 91/94 sample (.77) and the Alberta Hospital Edmonton treatment sample (.76). In contrast, the range for Static-2002 was from .65 (Millbrook) to .77 (CSC 91/94, Alberta Hospital Edmonton). Static-2002 predicted sexual recidivism as well for child molesters (.70) as rapists (.72).

Table 5 Sexual recidivism: Predictive accuracy of RRASOR, Static-99 and Static-2002.

*p < .05; **p < .01; ***p < .001
k = number of studies 

For any violent recidivism (see Table 6), the predictive accuracy of Static-2002 (.71) was greater than the predictive accuracy of Static-99 (.69; Z = 3.54, p < .01) and RRASOR (.62). For all three measures, the amount of variability across samples was no more than would be expected by chance. The average ROC areas for violent recidivism were somewhat higher for child molesters than rapists, but the differences were not significant for Static-2002 (Q = 0.37, p > .25), Static-99 (Q = 1.50, p > .20) or RRASOR (Q = 3.77, p < .10).

Table 6 Violent recidivism: Predictive accuracy of RRASOR, Static-99 and Static-2002.

⁺p < .10, all other Q statistics were associated with probabilities of > .10.

DISCUSSION

Static-2002 showed similar levels of predictive accuracy to Static-99 for the prediction of sexual recidivism. The initial analyses suggested that the new items could add valuable information, but, given that large amounts of missing data, further research is necessary to determine whether the new items/scoring result in a noticeable improvement in predictive accuracy.

The present study confirms previous results: Static-99 had slightly higher predictive accuracy than RRASOR, and the predictive accuracy of both scales was similar for rapists and child molesters. Such a finding is not surprising given that three of the seven samples had been previously used in the development of Static-99. The current study, however, includes new Static-99 data for four new data sets (n = 1,319) or 61% of the total sample (average predictive accuracy of .68 in these four data sets, range of .59 to .77).

Even though the level of predictive accuracy was similar, Static-2002 has potential advantages over Static-99. Static-2002 predicted any violent recidivism as well as it predicted sexual recidivism, and predicted violent recidivism better than Static-99. Static-2002 was less variable than Static-99 across settings. As well, the content areas of Static-2002 have increased conceptual clarity compared to Static-99.

Conceptual clarity is important for evaluators who wish to consider whether external factors should be used to adjust the risk level suggested by an actuarial tool. Researchers may be able to improve the predictive accuracy of Static-2002 on purely empirical grounds by including other variables, such as those related to childhood behaviour problems (e.g., SORAG, Quinsey et al., 1998), or lack of cooperation with supervision (e.g., Hanson & Harris, 2000). The greatest advances in risk assessment, however, will come from accurate explanations for recidivism. Research is required to established the construct validity of the content areas of Static-2002, and, more importantly, identify the causes of recidivism. Prior offending is a good predictor variable, but it is poor theoretical variable. Events that happened twenty years ago cannot compel future crimes. Instead, researchers need to identify the current life situations and the enduring characteristics that predispose certain individuals to offend sexually, whether they be memories, habits, values, cognitive deficits, or unfortunate life choices.

Although Static-2002 shows promise, further research is necessary before it can be used in applied contexts. Specifically, the results need to be replicated with independent data sets. Given that the major utility of static risk scales is to judge long-term recidivism potential, the replication studies should include a follow-up period of at least 10 years (15 years preferably). It is likely that the replication studies would need to include a combined sample size of over 1000 sexual offenders (200 to 400 in each study) before it is possible to associate specific risk levels to specific ranges of scores.

As with Static-99, Static-2002 was not intended to provide a comprehensive assessment of recidivism risk. There are many factors associated with sexual recidivism, such as sexual preoccupations and access to victims (Hanson & Harris, 2000), that evaluators may want to consider. How best to combine static, historical factors with dynamic (changeable) risk factors remains an important topic for future research.

REFERENCES

Barbaree, H. E., Seto, M. C., Langton, C., & Peacock, E. (2001). Evaluating the predictive accuracy of six risk assessment instruments for adult sex offenders. Criminal Justice and Behavior, 28, 490-521.

Berliner, L., Schram, D., Miller, L. L., & Milloy, C. D. (1995). A sentencing alternative for sex offenders: A study of decision making and recidivism. Journal of Interpersonal Violence, 10, 487-502.

Bonta, J., & Hanson, R. K. (1995a). [10-year recidivism of Canadian federal offenders]. Unpublished raw data.

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

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

CS/RESORS Consulting. (1991). An evaluation of community sex offender programs in the Pacific Region. Report presented to the Correctional Service of Canada Regional Headquarters (Contract no. 21803-0-A602/01-XSB). Vancouver, B.C.: Author.

Epperson, D. L., Kaul, J. D., & Hesselton, D. (1998). Minnesota Sex Offender Screening Tool - Revised (MnSOST-R): Development, performance, and recommended risk level cut scores. Iowa State University/Minnesota Department of Corrections.

Friendship, C., Thornton, D., Erikson, M. & Beech, A.R. (2001). Reconviction: A Critique and Comparison of Two Main Data Sources in England and Wales. Legal and Criminological Psychology, 6, 121-129.

Grubin, D. (1998). Sex offending against children: Understanding the risk. Police Research Series Paper 99. London, UK: Home Office.

Hanley, J. A., & McNeil, B. J. (1983). A method of comparing the areas under Receiver Operating Characteristic curves derived from the same cases. Radiology, 148, 839-843.

Hanson, R. K. (1997). The development of a brief actuarial risk scale for sexual offense recidivism. (User Report 97-04). Ottawa: Department of the Solicitor General of Canada.

Hanson, R. K. (2001). Age and sexual recidivism: A comparison of rapists and child molesters. User Report 2001-01. Ottawa: Department of the Solicitor General of Canada.

Hanson, R. K. (2002). Evaluation of Manitoba's Secondary Risk Assessment. Unpublished manuscript.

Hanson, R. K., Broom, I., & Stephenson, M. (2001). [Updated recidivism information on sexual offenders in the CSC Pacific Region]. Unpublished raw data.

Hanson, R. K., & Bussière, M. T. (1998). Predicting relapse: A meta-analysis of sexual offender recidivism studies. Journal of Consulting and Clinical Psychology, 66 (2), 348-362.

Hanson, R. K., & Harris, A. J. R. (2000). Where should we intervene? Dynamic predictors of sex offense recidivism. Criminal Justice and Behavior, 27, 6-35.

Hanson, R. K., Scott, H., & Steffy, R. A. (1995). A comparison of child molesters and non-sexual criminals: Risk predictors and long-term recidivism. Journal of Research in Crime and Delinquency, 32(3), 325-337.

Hanson, R. K., Steffy, R. A., & Gauthier, R. (1992). Long-term follow-up of child molesters: Risk prediction and treatment outcome. (User Report No. 1992-02.) Ottawa: Corrections Branch, Ministry of the Solicitor General of Canada.

Hanson, R. K., Steffy, R. A., & Gauthier, R. (1993a). Long-term recidivism of child molesters. Journal of Consulting and Clinical Psychology, 61, 646-652.

Hanson, R. K., Steffy, R. A., & Gauthier, R. (1993b). [Long-term recidivism of child molesters]. Unpublished raw data.

Hanson, R. K., & Thornton, D. (1999). Static-99: Improving actuarial risk assessments for sex offenders. User Report 99-02. Ottawa: Department of the Solicitor General of Canada.

Hanson, R. K., & Thornton, D. (2000). Improving risk assessments for sex offenders: A comparison of three actuarial scales. Law and Human Behavior, 24(1), 119-136.

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

Marques, J. K., & Day, D. M. (1996). [SOTEP follow-up data for 1995]. Unpublished raw data.

Marques, J. K., Day, D. M., Nelson, C., & West, M. A. (1993). Effects of cognitive-behavioral treatment on sex offenders' recidivism: Preliminary results of a longitudinal study. Criminal Justice and Behavior, 21, 28-54.

Marques, J. K., Nelson, C., West, M. A., & Day, D. M. (1994). The relationship between treatment goals and recidivism among child molesters. Behaviour Research and Therapy, 32, 577-588.

Motiuk, L. L. (1995). [Sex offender recidivism information for case load and new releases: March, 1991 to July, 1994]. Unpublished raw data.

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

Motiuk, L. L., & Brown, S. L. (1996). Factors related to recidivism among released federal sex offenders. (Research Report No. R-49). Ottawa, Canada: Correctional Service of Canada.

Motiuk, L. L., & Porporino, F. J. (1993). An examination of sex offender case histories in federal corrections. (Research Report No. R-30). Ottawa, Canada: Correctional Service of Canada.

Nunes, K. L., Firestone, P., Bradford, J. M., Greenberg, D. M., & Broom, I. (2002). A comparison of modified versions of the Static-99 and the Sex Offender Risk Appraisal Guide (SORAG). Sexual Abuse: A Journal of Research and Treatment, 14, 253-269.

Pellerin, B., Proulx, J., Ouimet, M., Paradis, Y., McKibben, A., & Aubut, J. (1996). Étude de la récidive post-traitement chez des agresseurs sexuels judiciarisés. Criminologie, 29, 85-108.

Phenix, A., Hanson, R. K., & Thornton, D. (2000). Coding rules for the Static-99. Corrections Research: Manuals and Forms. Ottawa: Department of the Solicitor General of Canada.

Proulx, J., Pellerin, B., McKibben, A., Aubut, J., & Ouimet, M. (1997). Static and dynamic predictors of recidivism in sexual offenders. Sexual Abuse: A Journal of Research and Treatment, 9, 7-28.

Proulx, J., Pellerin, B., McKibben, A., Aubut, J., & Ouimet, M. (1995). [Static and dynamic predictors of recidivism in sexual aggressors]. Unpublished raw data.

Quinsey, V. L., Harris, G. T., Rice, M. E., & Cormier, C. A. (1998). Violent offenders: appraising and managing risk. Washington, DC: American Psychological Association.

Reddon, J. R. (1996). [Phoenix Program for Sex Offender Treatment: An evaluation update with recidivism data obtained in September, 1995]. Unpublished raw data.

Silver, E., Smith, W. R., & Banks, S. (2000). Constructing actuarial devices for predicting recidivism: A comparison of methods. Criminal Justice and Behavior, 27 (6), 733-764.

Sjöstedt, G., & Långström, N. (2001). Actuarial assessment of sex offender recidivism risk: A cross validation of the RRASOR and the Static-99 in Sweden. Law and Human Behaviour, 25, 629-645.

Song, L., & Lieb, R. (1995). Washington State sex offenders: Overview of recidivism studies. Olympia, WA: Washington State Institute for Public Policy.

SPSS Inc. (1999). SPSS advanced models 10.0. Chicago: SPSS Inc.

Studer, L. H., Reddon, J. R., Roper, V., & Estrada, L. (1996). Phoenix: An inpatient treatment program for sex offenders. Journal of Offender Rehabilitation, 23, 91-97.

Thornton, D. (1997). [A 16-year follow-up of 563 sexual offenders released from HM Prison Service in 1979.] Unpublished raw data.

APPENDIX A

Coding rules for Static-2002

The following should be considered only as a general overview of the coding rules of Static-2002. Detailed coding rules are currently being developed. Researchers wishing to use Static-2002 should contact either Karl Hanson or David Thornton for the most recent version of the coding rules before proceeding. Although the items may appear simple, previous experience with Static-99 suggests that careful attention to the coding rules is necessary to obtain adequate reliability.

Static-2002 is intended for offenders with an index sexual offence who have an opportunity to reoffend by virtue of receiving a community sentence or by release from a custodial sentence. They must be males, age 18 or older, who have committed a sexual offence against an identifiable victim.

Age at release (0 - 3)

18 to 24.9 = 3

25 to 34.9 = 2

35 to 49.9 = 1

50 or older = 0

Use the age of the offender at the time that the risk assessment is to be used.

Persistence of sexual offending (0 - 3)

This section includes three variables: a) prior sentencing occasions for sexual offences (0 - 3), b) arrests for sexual offences as both an adult and a juvenile (0/1), and c) rate of sexual offences (0/1). Code each of the individual variables, then recode the subscale as follows:

Prior sentencing occasions for sexual offences

0 = 0

1 = 1

2, 3 = 2

4 or more = 3

A sentencing occasion is when the offender attends court, admits to the offence or is found guilty, and receiving some form of sanction (fine, prison, conditional sentence). Offenders may be convicted of more than one offence at the same sentencing occasion. In England, an official caution count as one sentencing occasion. Count both adult and juvenile sentencing occasions.

Offenders may go to court and receive more than one sentence for a single crime spree. In this case, all the convictions related to the same crime spree count as one sentencing occasion. For two sentencing occasions to be considered distinct, the offender must have committed a crime and been sanctioned for it prior to committing the second crime (and being sanctioned for it). When the offender is convicted for a crime that was committed prior to his previous conviction, the new conviction is considered pseudo-recidivism and is not counted separately.

Offences are considered as sexual based on the behaviour, not the name. For example, Assault plead down from Sexual Assault would still count as sexual if the motivation for the original offence was sexual. As with Static-99, Static-2002 is intended for offenders with at least one identifiable victim. It does not apply to purely prostitution offenders, although prostitution offences would count as sexual offences if the offender had a record for any Category A offences (see Static-99 coding rules; Phenix, Hanson & Thornton, 2000).

Arrest for sexual offences as juvenile

Arrest prior to age 18 and conviction after age 18 = 1

No arrest for a sexual offence prior to age 18 = 0

This items is only scored for offenders who have a conviction for a sexual offence as an adult. It is not scored for juvenile offenders who are released after age 18. The adult conviction must be for different behaviour than that which lead to the arrest as a juvenile.

Rate of sexual offending

Less than one sentencing occasion every 15 years = 0

One or more sentencing occasions every 15 years = 1

This is calculated by summing all the sentencing occasions for sexual offences (including index) and dividing that number by the offender's age at release. For example, if the offender was 29 years old and have one prior sexual offence, his rate would be (1 + 1)/29 = .06897, and he would receive the point for "high rate".

Deviant sexual interests (0 - 3)

This section contains three items: a) any convictions for non-contact sexual offences (0/1), b) any male victims (0/1), and c) young, unrelated victims (0/1). The three items are summed to make a total score that ranges from 0 to 3 (no recode).

Non-contact sexual offences

Any conviction for a non-contact sexual offence = 1

None = 0

This is the same as the Static-99 item. Count only convictions for obscene phone calls, internet luring, possession of child pornography, exhibitionism, and voyeurism. Do not count prostitution related offence or other offences without an identifiable human victim.

Any male victims

Any male victims = 1

None = 0

Same as Static-99.

Young, unrelated victims (0, 1)

Two or more victims less than age 12,
one of which must be unrelated = 1.

Otherwise = 0

This item is intended as a proxy for sexual interest in children. The offender receives the point if he has two separate victims who have yet to obtain their 12^th birthday. One of the victims must be unrelated. If all the victims are related, or if there is only one victim less than age 12, the offender gets a score of zero.

Relationship to victims of sexual offences (0 - 2)

This sections contains two items: Any unrelated victims (0/1) and any stranger victims (0/1). Stranger victims are also unrelated. The items are summed to get a subscale score that can range from 0 to 2.

Any unrelated victims

Any unrelated victims = 1

None = 0

Related general means "too close to marry", but it also includes 1^st cousins, nephews and spouses. This is the same item as in Static-99 (see Phenix et al., 2000).

Any stranger victims

Any stranger victims = 1

None = 0

The relationship is consider as one between strangers if the offender and victim did not know each other 24 prior to the sexual offence.

This is the same item as in Static-99 (see Phenix et al., 2000).

General Criminality (0 - 3)

This subscale contains four items: a) sentencing occasions for anything (0 - 3), b) any breach of conditional release (0/1), c) less than 4 year free prior to conviction for index offence (0/1), and ) any prior conviction for non-sexual violence. The raw score is recoded as follows:

Arrest/Sentencing occasions for anything

No prior arrest for anything = 0

Any arrest, but less than 3 sentencing occasions for anything = 1

3 - 13 sentencing occasions = 2

14 or more sentencing occasions = 3

Do not count the index offence(s). Count sexual and non-sexual sentencing occasions. Do not count very minor offences for which it would be impossible to go to jail or to receive a community sentence (e.g., drinking under age, speeding).

Breach

Any breach of conditional release = 1

None = 0

This item is scored "1 = yes" if the offender has ever been charged with an offence related to failure to conform to the demands of bail, parole, probation or community supervision (e.g., failure to comply with probation, parole revoked). It is also scored if the offender is arrested for a new offence while on a community supervision order (including bail, parole, probation).

Years Free Prior to Index

Less than 36 month free prior to committing the sexual offence that resulted in the index conviction, or

Less than 48 months free prior to conviction for index sex offence = 1,

More than 48 months free prior to index conviction and

More than 36 months free prior to committing the sexual offence that resulted in the index conviction = 0

Calculate the number of months between the date of release into the community from the previous involvement with the law, and the date of the most recent sex offence (the index sex offence). For offenders who received custodial sentences, the release into the community would be the release date from prison. For offenders sentenced to the community, it would be the date of sentence. Arrest and charges count as involvements with the law, even if the offender was not convicted.

Two dates are associated with the index sexual offence: a) the date the offence occurred, and b) the date the offender was convicted. The offender gets one point if either the time span between release and reconviction was less than 48 months, or the time span between release and sexual reoffending was less than 36 months.

Count juvenile offences for offenders who were less than 22 years of age at the time of conviction for the index sexual offense.

Prior violent non-sexual convictions

Any prior conviction for non-sexual violence = 1

No prior convictions = 0

Violent non-sexual offences are those that involve confrontation with the victim, such as robbery, assault, and murder, as well as arson and threatening. In contrast to Static-99, count only violent offences that were not sexually motivated. If an offender attempts a sexual assault, but is only convicted of Assault, the assault would count as a sexual offence and would not count as non-sexual violence. The criminal event that results in the conviction for non-sexual violence must be different from the criminal event that resulted in a conviction for sexual offending. The conviction for non-sexual violence must occur prior to the index sexual offence. Do not count convictions for non-sexual violence that are included in the index cluster of offences (i.e., the crime spree that included the index sex offence).

Date modified:

2015-12-08