Introduction

The United States has a long history of enacting "sexual psychopath" laws in response to sexual assaults against its citizens. The first such law was passed in 1937 in Michigan. Recent dramatic increases in the numbers of reported rapes and cases of child abuse and/or neglect (according to one report [1], over 500,000 of the former and almost 3 million of the latter annually) have produced nationwide efforts to protect potential victims through sex offender registration and community notification laws.

Two New Jersey cases involving the rape and murder of young girls contributed to the public outcry that led to the enactment of New Jersey's "Megan's Law" in 1994 and the federal version in 1996. The New Jersey law requires all individuals convicted of a sex offense--adult and juvenile, male and female--to register with their local police departments. After the offender has registered, the ease is reviewed by two representatives of the county prosecutor's office to determine which of three levels of community notification is most appropriate. Offenders considered to be at most serious risk for reoffense and most likely to inflict serious harm receive the highest level of notification (i.e., door-to-door notification of neighbors, as well as notification of school and community officials). Those offenders deemed a moderate risk are subject to police and school official notification, but not neighborhood notification. For offenders designated as low risk, only the police are notified.

The issues raised by these and similar policies are complex. On the one hand, few would question the need for community protection. At the same time, the privacy rights of released and potentially rehabilitated offenders cannot be ignored. Recognition of these conflicting values highlights the importance of defining and predicting risk as accurately as possible, while doing so within inevitable governmental budgetary constraints.

Toward this end, in 1995 New Jersey's Attorney General commissioned a panel of forensic experts to develop an objective measure for determining which level of notification each of the state's convicted sex offenders should receive.[2] The product of these meetings was the Registrant Risk Assessment Scale (RRAS), the focus of this paper. In devising the instrument, the panel considered a combination of legal and clinical factors, including the New Jersey statute governing the sentencing of sex offenders (N.J.S.A 2C:7-6) and the sex offender risk assessment research, which has generated a number of empirically-based risk assessment criteria. The scale was designed to give essentially equal weight to the seriousness of the probable offense should the offender recidivate and the likelihood that the offender would indeed recidivate. Although the RRAS is currently in use in New Jersey and has withstood numerous legal challenges,[3] it has not yet been subjected to extensive empirical analysis. The preliminary investigation reported here should be considered a first step toward a thorough evaluation of the scale.

The RRAS is a 13-item scale (see Appendix A)[4] designed to be scored by trained personnel with access to the offender's criminal case file. For each item, the respondent judges whether the offender's behavior qualifies as "low risk," "moderate risk," or "high risk." The items are divided roughly equally between "static" and "dynamic" predictors of risk. [5] The static variables--those that are not amenable to change--include details of the offender's criminal history such as degree of force used against the victim, degree of contact (e.g., penetration), age of the victim, the victim's relationship to the offender, the number of past sexual offenses, the duration of the offensive behavior, and any history of antisocial acts. The dynamic variables--those amenable to change over time--include length of time since the last offense (while at risk), response to sex-offender-specific treatment, substance abuse, therapeutic support, residential support, and employment/educational stability.

There is considerable research supporting the importance of the static factors noted above in predicting sexual offenses. McGrath, in his review of sex offender risk assessment and recidivism research, [6] notes that several studies have found all of the following to be among the factors positively associated with the likelihood of committing a new sexual offense: the use of force or violence, the victimization of someone who is not related to the offender, prior convictions for sex offenses, and deviant sexual arousal patterns. Similarly, Quinsey and his associates [7] identify psychopathy, measures of previous criminal behavior, and phallometric indexes of deviant sexual interests as predictors of sexual recidivism. In a follow-up study of child molesters, Prentky, Knight, and Lees [8] also report that paraphilias, the degree of sexual preoccupation with children, and the number of prior sexual offenses predict sexual recidivism. Finally, in their comprehensive meta-analysis of sex offender recidivism studies, Hanson and Bussiere [9] found that illegal deviant sexual history is a good predictor of sexual criminal recidivism. In contrast to the static factors identified here, Quinsey et al note that there is little research to date examining or documenting the effectiveness of dynamic predictors.

The present study explored the inter-relationships among the static RRAS items and examined whether and how these scores differed among convicted adult sex offenders sentenced in New Jersey to either probation (with or without a term of county jail incarceration of up to 364 days), state prison, or the Adult Diagnostic and Treatment Center (ADTC). Before sentencing, all of the state's adults convicted of serious sex offenses are required to undergo psychological evaluations at the ADTC to determine whether their offenses were committed in a "repetitive and compulsive" manner. If an individual is found to be a repetitive and compulsive sex offender, he is then eligible for an indeterminate term at the ADTC up to his statutory maximum, where he will receive inpatient sex offender treatment specific to his pattern of offending. Although some repetitive and compulsive sex offenders are given probation-those with less serious offenses-this is unusual, since these offenders are typically seen as the highest risk to the community.

At the conclusion of the psychological evaluation, if there is insufficient clinical evidence to determine that the offense was committed as the result of a repetitive and compulsive pattern, [10] the judge has the discretion to sentence the individual to either probation or state prison time, depending on the severity of the offense and the individual's perceived risk for committing a new offense. Based on this sentencing structure, the authors viewed these three groups as representing a continuum of reoffense risk. That is, those individuals sentenced to the ADTC were seen as "high risk" for re-offense, those sentenced to state prison were considered "moderate risk," and those sentenced to probation were deemed "low risk," at least at the time of their sentencing for that particular offense.

Participants

The authors examined the RRAS scores of three groups of convicted male sex offenders: 154 who were on probation, 245 who were assigned to the New Jersey state prison system, and 175 who were placed in the state's Adult Diagnostic and Treatment Center (ADTC) These 574 participants included all 338 adult male sex offenders registered in five New Jersey counties as of February 1,1997. [11]. Of the state prison group, 113 had been referred to the New Jersey State Parole Board Psychology Department for pre-parole evaluations between April, 1995 and February, 1997. Of the ADTC group, 123 had been released from the treatment center between 1976 and 1995 and were also part of the New Jersey Attorney General's Research Department study of RRAS inter-rater reliability. Juvenile and female sex offenders were not included due to differences in sentencing options for these groups of offenders. The five counties selected for data collection represent a cross-section of the state's population in terms of demographic characteristics, ranging from urban to suburban to rural.

For reasons of confidentiality, identifying demographic information about the offenders (e.g., age, marital status, etc.) was not disclosed by some data acquisition sources. Therefore, it is not possible to ascertain the presence or impact of differences among the placement groups on these variables. Other research and government data suggest that rapists (i.e., those who forcibly sexually assault adults, usually women) as a group tend to be similar to non-sex-offender prisoners in having assaultive, antisocial histories, while child molesters as a group are less antisocial. [12] Those assigned to centers like the ADTC typically have longer sex offense histories than those given probation or assigned to the general prison population.

Data Measures and Acquisition

A trained rater scored the RRAS for each offender based on information available in his case tile. Extensive interview data were available in the majority of the files; such material is a benefit in scoring, but no interviewing is specifically necessary for completing the RRAS. Since only one individual scored an offender's RRAS in the present study, inter-rater reliability estimates could not be obtained. However, members of the RRAS development committee trained all raters in RRAS scoring procedures.

Although the RRAS as currently used in New Jersey requires that each item be scored either 0, 1, or 3 on a scale of risk and then be given a multiplicative weighting of either 5, 3, 2, or 1 (refer to Appendix A), for the purposes of this exploratory investigation two modifications were made. First, whenever an item was scored as a "3" it was recoded as a "2" in order to create equal-interval scaling. Second, the multiplicative weightings were ignored so that each RRAS item had a score of 0 ("low risk"), 1 ("moderate risk"), or 2 ("high risk"). All of the analyses presented here, however, were also performed using the original 0-1-3 scoring; in all cases the results were essentially comparable.

The RRAS manual instructs the rater to omit scores where information in the record is insufficient. The number of cases with missing data on dynamic items (e.g., response to treatment, community support) precluded a thorough statistical analysis of all 13 RRAS items. Therefore, this study focuses on the static items assessed by the RRAS, for which all 574 participants in the study have been scored.

Design

The data analysis based on the static RRAS items included three components. First, an exploratory factor analysis examined the underlying factor structure of these seven items for the entire sample. Second, one-way analyses of variance were conducted to determine the differences among the three placement groups on the individual RRAS items. Third, discriminant analyses investigated how effectively these static RRAS items collectively discriminated among the sex offenders in regard to their assignment to either probation, state prison, or the ADTC.

Factor Structure of the Static RRAS Items

The intercorrelations among the static RRAS items are presented in Table 1. A principle component analysis of these seven items with a Varimax rotation yielded three orthogonal factors with eigenvalues of 2.20, 1.67, and 1.13 accounting for 31%, 24%, and 16% of the variance respectively.

 

RRAS Items	Mean(SD)	DF	DC	AV	VS	NO	DO	HA
Degree of Force (DF)	.069(0.77)	--	.23	-.49	.40	.12	.03	.45
Degree of Contact (DC)	1.66(0.62)	.23	--	-.13	-.05	-.05	.06	.09
Age of Victim  (AV)	1.38(0.75)	-.49	-.13	--	-.38	.14	.27	-.23
Victim Selection (VS)	0.82(0.77)	.40	-.05	-.38	--	.36	-.02	.36
Number of Offenses/Victims (NO)	0.64(0.38)	.12	-.05	.14	.36	--	.50	.22
Duration of Offensive Behavior (DO)	0.80(0.94)	.03	.06	.27	-.02	.50	--	.07
History of Anti-Social Acts (HA)	1.00(0.81)	.45	.09	-.23	.36	.22	.07	--

Note: N = 574.RRAS = Registrant Risk Assessment Scale. All correlations with an absolute value of .09 or greater are significant at p<.05.

Table 2 presents the factor loadings. The items with the highest loadings on the first factor are Degree of Force, Victim Selection, History of Anti-Social Acts (all positive loadings), and Age of Victim (a negative loading). In combination, they describe an offender with a history of antisocial acts who engages in a forcible sexual assault on an adult stranger. The second factor is most clearly defined by two items, Number of Offenses/Victims and Duration of Offensive Behavior (both positive loadings), with a more moderate weighting on a third item, Age of Victim (also a positive loading). This cluster describes an offender with a history of repeated and extensive sexual contacts with more youthful victims. Finally, the third factor is almost fully defined by one item: Degree of Contact, which ranges from no contact whatsoever to penetration. It should be noted that a very substantial proportion of the sample (74%) was scored at the highest level on this item.

Factor Structure of Static RRAS Items Based On Principle Component Analysis With Varimax Rotation

 

		Format
RRAS Item	1	2	3
Degree of Force	.78	.01	-.30
Degree of Contact	.11	.02	-.92
Age of Victim	-.69	.45	.10
Victim Selection	.77	.14	.31
Number of Offenses/Victims	.27	.82	.18
Duration of Offensive Behavior	-.08	.85	-.17
History of Anti-Social Acts	.67	.20	-.09

Note: N = .RRAS = Registrant Risk Assessment Scale

Analysis of Variance and Post-Hoc Comparison of Means

One-way analyses of variance were conducted comparing the individual RRAS item scores of the offenders in the three different placement groups. Table 3 summarizes the results for the entire sample and for the Probation, State Prison, and ADTC groups separately. Each of the 7 static RRAS items produced a significant overall F, and Scheffe post-hoc comparison tests were therefore employed to determine which of the group means differed significantly in each case. An inspection of the table indicates that the offenders assigned to the ADTC had significantly higher scores than both the probationers and the state prisoners on Age of Victim (which translates into younger victims), Number of Offenses/Victims, and Duration of Offensive Behavior; and significantly higher scores than the probationers only (but not the state prisoners) on Degree of Force, Degree of Contact, and History of Anti-Social Acts. The state prisoners had significantly higher scores than both the probationers and the offenders assigned to the ADTC on Degree of Force and History of Antisocial Acts, and higher scores than the probationers only (but not the ADTC group) on Degree of Contact. Victim Selection, which assesses the degree to which the offender knew the victim, was the only static item on which there were no differences among the three groups.

Means, Standard Deviations, and Comparison Tests For the Three Placement Groups

					PR	PR	SP
		Mean (SD		F	vs	vs	vs
RRASItem	PR	SP	TC	Value	SP	TC	TC
Degree of Force	0.40(0.59)	0.93(0.78)	0.62(0.80)	25.49**	**	**	**
Degree of Contact	1.33(0.78)	1.81(0.48)	1.75(0.51)	34.14**	**	**	ns
Age of Victim	1.39(.60)	1.20(0.83)	1.62(0.68)	16.39**	*	**	**
Victim Selection	0.69(0.65)	0.87(0.81)	0.86((0.81)	3.04*	ns	ns	ns
Number of Offenses/ Victims	0.29(0.64)	0.47(0.71)	1.20(0.86)	73.57**	ns	**	**
Duration of Offensive Behavior	0.42(0.78)	0.62(0.90)	1.38(0.84)	60.26**	ns	**	**
History of Anti-Social Acts	0.63(0.68)	1.23(0.82)	1.02(0.80)	28.17**	**	**	*

 Note: RRAS = Registrant Risk Assessment Scale; PR = Probation (N of 154); SP = State Prison (N of 245); TC = Adult Diagnostic Treatment Center (N of 175); *p <.05; **p <.01; ns = p >.05

Discriminant Function Analyses

A discriminant analysis was performed using the seven static RRAS items as independent variables to predict the offenders' assignment to probation, state prison, or the ADTC. The -analysis was based on proportional placement corresponding to the actual distribution within the sample (i.e., 27% probation, 43% state prison, and 30% ADTC). With three classification groups, as many as two functions can be statistically significant discriminators. In this case, both functions provided significant discriminating power. Together they yielded a Wilk's lambda of .58 with an approximate F (14, 1130) = 24.91, p <.01. The first function had a canonical correlation (a summary measure of the degree of relation among the placement groups and the discriminant function) of .55 and accounted for 68% of the discriminating power of the two functions combined. The second discriminant function had a canonical correlation of .41 and accounted for the remaining 32% of the pair's discriminating power.

Table 4 presents the factor structure for each discriminant function and also each group's mean (or centroid) on that function. The factor structure can be interpreted in a manner similar to the loadings in a factor analysis. The group centroids for the first function suggest that the clearest separation occurs between offenders assigned to the ADTC and those assigned to either probation or state prison; the ADTC mean lies in the positive end of the function's continuum, while both of the other groups have negative means. Within this function, the two variables that relate most strongly to ADTC placement are Number of Offenses/Victims (.76) and Duration of Offensive Behavior (.69). The group centroids for the second function indicate that it most effectively discriminates between offenders sentenced to state prison and those given probation. Three static RRAS items were particularly useful in making this discrimination. The state prison group tended to score higher on Degree of Force (.65), Degree of Contact (.59), and History of Antisocial Acts (.63) than the probationers (as well as the ADTC group). Two variables--Age of Victim and Victim Selection (i.e., relationship to the offender)--had relatively low loadings on both functions.

Discriminant Function Analysis Using Group Placement (Probation, State Prison, or ADTC) as the Criterion

 

RRAS Item	Function 1	Function 2
Degree of Force	.09	.65
Degree of Contact	.35	.59
Age of Victim	.24	-.41
Victim Selection	.11	.17
Number of Offenses/Victims	.76	-.25
Duration of Offensive Behavior	.69	-.21
History of Anti-Social Acts	.21	.63
Group Means (Gr Centroids) For Each Function:
Probation	-.79	-.50
State Prison	-.14	.51
ADTC	.90	-.27

Note: N = 574; RRAS = Registrant Risk Assessment Scale; ADTC = Adult Diagnostic and Treatment Center

Cross-validation is an important follow-up step in evaluating the effectiveness of a set of discriminant functions. Since a discriminant analysis determines the best-fitting classification functions for a particular sample, the discriminatory power of these functions is likely to diminish when they are applied to a sample different from the one that generated them. To examine the extent of deterioration in this regard, the sample was divided with equal numbers of Probationers, State Prisoners, and ADTC offenders randomly assigned to one half-sample or the other. A discriminant analysis was then performed on one half-sample and the derived functions were used to classify the offenders in the second half-sample. There was a moderate decline in classification accuracy, with the overall "hit" rate dropping from 62.7% in the first half-sample to 56.4% in the cross-validation half-sample.

The predictive accuracy of the discriminant analysis can also be evaluated by more closely examining the classification matrix in Table 5. The rows present the sex offenders' actual placements and the columns present the offenders' predicted placements based on the discriminant functions. As the rows indicate, 49% of the "true" probationers, 64% of the "true" state prisoners, and 61% of the "true" ADTC offenders were correctly identified. These numbers reflect the sensitivity of the classification rules employed--that is, the probability that an individual of known legal disposition is accurately classified. As the columns reveal, 57% of the offenders classified as probationers were true probationers, 59% of the offenders classified as state prisoners were true state prisoners, and 60% of the men classified as ADTC offenders were in fact actual ADTC placements. This latter set of values reflects the positive predictive power of the classification rules--that is, the probability that an individual who scores at a given risk level has a corresponding legal disposition.

The positive predictive power demonstrates substantially greater accuracy than one would obtain by chance alone given the proportional representation of the three groups in the sample (27% probationers, 43% state prisoners, and 30% ADTC offenders). However, these base-rates are artifacts of how many offenders in each group were included in the study. Of more interest are the true base-rates of probation, state prison, and ADTC disposition for the New Jersey sex offender population; when the actual base rates in the New Jersey sex offender population are applied (56%, 34%, and 10% for probationers, state prisoners, and ADTC inmates respectively[13]), the positive predictive ratios are elevated even further for both state prisoners and ADTC offenders but diminished for probationers. [14]

Classification Matrix with Observed Placements in Rows and Predicted Placement in Columns

 

Actual Placement	% Correct	Probation	State Prison	ADTC	Total
Probation	49.35	76	61	17	154
State Prison	63.67	36	156	53	245
ADTC	61.14	20	48	107	175
Total	59.06	132	265	177	574

Note: ADTC = Adult Diagnostic and Treatment Center.

In New Jersey, the RRAS scores of hundreds of convicted sex offenders have been used to estimate the level of risk they pose and to determine the appropriate level of community notification. Although the current study examines only the "static" items of the RRAS, the findings shed light on the constellation of variables underlying these offenses, as well as clarifying the considerations that enter into the process by which law enforcement officials make correctional assignment decisions when evaluating this particularly troubling group of criminals. At the same time, the results point to the need for further analysis of the RRAS in order that it is utilized both effectively and appropriately.

As would be hoped for an instrument designed as a risk assessment scale for sexual recidivism, the two key orthogonal factors identified by the exploratory factor analysis of the static RRAS items parallel the findings of other researchers who have investigated predictors of sexual and non-sexual criminal activity. One RRAS factor is characterized by forcible sexual assault against an adult stranger by an offender with a history of antisocial acts. The other RRAS factor is defined by an extensive and long-standing preoccupation with sexually deviant behaviors as measured by number and duration of sexual offenses. A recent meta-analysis of sex offender recidivism studies, combining a large number of North American samples, uncovered a corresponding pattern: a sexual deviance factor was the best predictor of sexual recidivism, while a criminality factor predicted sexual recidivism less well but predicted non-sexual recidivism quite well. [15]

When comparing the mean static RRAS item scores for the probation, state prison, and ADTC samples in the present study, state prison sex offenders scored higher than the other two groups on criteria that load on the forcible assault factor. This is not surprising given that these are typically men whose most recent violent assault against an adult adds to an already extensive criminal history. As would also be expected, the ADTC sex offenders scored higher than either probationers or state prison sex offenders on criteria that load on the sexual deviance factor. Probationers scored lowest on all the RRAS items under consideration, consistent with the fact that these men have been allowed to remain in the community. [16]

Further documenting the significance of these group differences are the results of the discriminant function analysis. Employing weighted combinations of the static RRAS items, two factors were generated that correctly classified into the probation, state prison, or ADTC categories almost 60% of the offenders studied. These two discriminant functions were similar to the factors described above. The ADTC inmates were most effectively discriminated from the other two groups by a sexual deviance factor. This is as it should be, since ADTC sex offenders have been found "repetitive-compulsive sex offenders," [17] more likely to recidivate than other offenders. [18] State prisoners were best discriminated by a function combining force, antisocial history, and degree of contact (i.e., penetration) as well. In other words, many sex offenders incarcerated in the state prison system are rapists, men who forcibly sexually assault and penetrate their victims in a context of broader antisocial behavior (and although the loading is less strong, their victims tend to be older than those of other offender groups). The literature documents that rapists (in contrast to child molesters) are indeed more similar to a general criminal population than they are to other sex offenders. [19]

The empirical and theoretical issues surrounding risk classification merit further consideration here. First, how well did the discriminant analysis using the static RRAS items classify this particular sample of sex offenders? Under most circumstances, positive predictive power (PPP) is of most interest to the user of a scale. PPP is the probability that an individual who is classified in a certain group by the scale actually falls in that group. Although, for reasons previously noted, we have not computed formal operating characteristics for this subset of RRAS variables, it is evident that for high risk individuals-perhaps that group of most concern to the public-the RRAS discriminate functions did quite well. Although only 10% of the state's convicted sex offenders are considered sufficiently "high risk" to be incarcerated at the ADTC, roughly 60% of high-risk offenders were accurately classified by the RRAS discriminate function analysis. This represents a six-fold improvement over chance with regard to high-risk offenders. [20]

But despite the success of the static RRAS items in classifying sex offenders, there are limitations inherent in the criterion measure the authors' selected to measure risk. This preliminary investigation of the RRAS was not a longitudinal study that directly assessed rates of recidivism among a sample of sex offenders. Rather, the authors adopted a sensible but imperfect substitute for actual recidivism data, namely the legal disposition of each offender's case. That is, probationers were deemed "low risk," state prisoners were considered "moderate risk"' and ADTC offenders were identified as "high risk."

The most obvious drawback to this formulation is that many factors other than risk contribute to an offender's legal disposition. For instance, suppose an individual is truly high risk-that is, he has a strongly deviant sexual interest pattern and is therefore more likely than other sex offenders to relapse. Although in the "ideal" world this offender would be assigned to the ADTC, the actual outcome might be a negotiated plea of probation if the prosecutor's evidence is weak, the victim does not wish to testify, and the offender does not confess. This intrinsic error in the criterion measure cuts both ways in regard to evaluating the effectiveness of the RRAS. On the one hand, some of the misclassifications based on the RRAS may reflect actual errors in adjudication. On the other hand, some of the RRAS "hits" may not necessarily translate into accurate identifications of risk since there is an imperfect relationship between case disposition and true likelihood of recidivism. The extent to which these competing influences balance each other cannot be determined from the current study.

In short, the results presented may reveal as much about how different static factors influence adjudication decisions as they do about the actual risk levels posed by the offenders in question. For example, those convicted sex offenders with extensive histories of sexual deviance and preoccupation may be assigned to the ADTC, those with assaultive and antisocial histories absent significant evidence of sexual deviance may be sent to state prison' and those offenders who do not have extensive histories along either of these two dimensions may be given probation. An additional confounding consideration in this context is the degree to which risk is defined by both reoffense likelihood and reoffense seriousness. After all, exhibitionists have high recidivism rates, but their offenses are generally considered nuisance offenses. As a result, they are typically considered "low risk" and no community notification occurs. Some of these thorny issues simply cannot be decided by empirical investigations; rather, they require value judgments best left to the legislature and the courts.

The present study provides clear preliminary evidence for the utility of the static RRAS items, and suggests that the RRAS as a whole may prove to be a valuable addition to the arsenal of risk assessment and research instruments for evaluating potential fixture harm. Dynamic factors are undoubtedly important as well in arriving at appropriate community notification decisions, but the authors encountered significant data collection difficulties in these areas. The RRAS includes six items that might be considered dynamic measures: length of time since last offense (while at risk), response to sex-offender-specific treatment, substance abuse, therapeutic support, residential support, and employment/educational stability. Some of these, such as treatment response, are inappropriate to score for incarcerated state prisoners, who have little or no access to treatment, or for most probationers, who are similarly unlikely to be engaged in sex-offender-specific treatment. (In this regard, one recent meta-analysis found that those sex offenders who failed to complete treatment were indeed at higher risk to sexually reoffend [21]). In the same way, items assessing community adjustment can be legitimately scored only after the offender has been released.

As noted earlier, some authorities [22] have suggested that static factors generally have higher predictive value than do dynamic, clinical factors. But whether this remains true in the specific domain of sex offender risk assessment remains to be seen. There has not yet been a comparative study to determine what combination of weights or factors best predicts sexual recidivism. Similarly, the relationships among the growing family of sex-offender risk assessment instruments await future investigation. [23]

In principle, recidivism should be predicted most effectively by combining historical data and the enduring propensities of the individual (i.e., static factors) with more transient influences and precipitants (i.e., dynamic factors). [24] In some cases, static factors will merit greater weight than dynamic factors, and vice versa. For example, when determining whether an individual is of sufficient risk to warrant involuntary commitment as a sex offender, static factors are typically most salient. However, when determining whether to release an individual from such commitment, dynamic factors--such as progress in treatment--are typically given greater consideration. [25]

Turning again to the RRAS, one point is very clear. Despite the promising findings reported here, it is crucial that future research investigate how well this instrument predicts recidivism among sex offenders. That is, do those offenders who score highest on some combination of RRAS items indeed recidivate most frequently and thereby pose the greatest risk? Do those offenders who recidivate at a moderate rate obtain intermediate scores on some combination of RRAS items? And do sex offenders with the lowest scores on particular scale items in fact reoffend least frequently? A determination of this sort will obviously require longitudinal data so that the survival curves of different risk groups can be compared. On a less optimistic note, similar instruments developed in Minnesota and Canada tend to have only very modest positive correlations with recidivism. [26] Regardless, other research has shown that the risk predictions of clinical evaluators are more accurate when they systematically review a list of standard, empirically supported risk factors in making their judgments. [27] In this light, examination and inclusion of criteria such as those assessed by the RRAS can only improve clinical sex offender risk determinations.

Criteria	Low Risk	0	Moderate Risk	1	High Risk	3	Comments	Total
Serious-ness of offenses x 5
1. Degree of Force	No Physical force; no threats		Threat; minor physical force		Violent; use of weapon; significant victim harm
2. Degree of Contact	No contact; fondling		Fondling under clothing		Penetration
3. Age of victim	18 or over		13-17		Under 13
							Subtotal:
Offense history x 3
4. Victim selection	Household /family member		Acquaintance		Stranger
5. Number of offenses/ Victims	Fist known offense/ victim		Two known offenses/ victims		Three or more offenses/ victims
6. duration of offensive behavior	Less than 1 year		1 to 2 years		Over 2 years
7. Length of time since last offense	5 or more years		More than 1 but less than 5 years		1 year or less
8. History of antisocial acts	No history		Limited history		Extensive history
							Subtotal:
Characteristics of offender x 2
9. Response to treatment	Good progress		Limited progress		Prior unsuccessful treatment or no progress in current treatment
10. Substance abuse	No history of abuse		In remission		Not in remission
							Subtotal:
Community support x 1
11. Therapeutic support	Current/ continued involve-ment in therapy		Intermittent		No involvement
12. Residential support	Supportive/supervised setting; appropriate location		Stable & appropriate location but no external support system		Problematic location and/or unstable; isolated
13. Employment/educational stability	Stable and appropriate		Intermittent but appropriate		Inappropriate or none
							Subtotal:
							Total:

Low range: 0-36; Moderate range: 37-73; High range: 74-111

1. Prentky, R.A. (1994). Introduction: The assessment and treatment of sex offenders, Criminal Justice and

Behavior, 21, 6-9.

2. The third author, Philip H. Witt, was a member of this panel.

3. See Witt, P. H., DelRusso, J., Oppenheim, J. & Ferguson, G. (1996). Sex offender risk assessment and the law, Journal of Psychiatry and Law, 24, 343-377.

4. The manual for the RRAS can be obtained from Deputy Attorney General Jessica Oppenheim, Department of Law and Public Safety, Division of Criminal Justice, Hughes Justice Complex, Trenton, NJ 08625.

5. Quinsey, V. L., Lalumiere, M. L., Rice, M. E. & Harris, G. T. (1995). Predicting sexual offenses. In J. C. Campbell (ed.) (pp. 114-137).

6. McGrath, R. J. (1991). Sex offender risk assessment and disposition planning: A review of empirical and clinical findings, International Journal of Offender Therapy and Comparative Criminology, 35, 328-350.

7. Rice, M.E., Harris, G.T. & Quinsey (1990). A follow-up of rapists assessed in a maximum security psychiatric facility, Journal of interpersonal violence, 5, 435448; 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; Quinsey et al., supra note 5.

8. Prentky, R. A., Knight, R. A. & Lee, A. F. 5. (1997). Risk factors associated with recidivism among extrafamilial child molesters, Journal of Consulting and Clinical Psvchology, 65, 141-149.

9. Hanson, R. K. & Bussiere, M. T. 1996, Predictors of sexual offender recidivism: A meta-analysis. Published by Canada Public Works and Government Services, ISBN 0-662-24790-6.

10. The burden of proof for a repetitive and compulsive finding is on the State, at a level of preponderence of the evidence. See Witt, P.H. and Frank, M. (1988). Psychological evaluations under the New Jersey Sex Offender Act, A New Jersey Trial Lawyer, 2, 37-43.

11. Ferguson, G.E. (1997). An investigation into the validity of the registrant risk assessment scale as a legal tool and clinical instrument. Dissertation Abstracts International, AAT 9809070.

12. Department of Justice, Bureau of Justice Statistics, Sex offenses and sex offenders, NCJ-163392, 1997.

13. New Jersey Administrative Office of the Courts. (1997). Judgement of conviction database, courtesy of James

Dunnemann, Research Analyst; these base-rates are remarkably similar to those for the U.S. as a whole: "On a given day there are approximately 234,000 [sex] offenders convicted of rape or sexual assault under the care, custody, or control of corrections agencies; nearly 60% of these sex offenders are under conditional supervision in the community." U.S. Department of Justice, Bureau of Justice Statistics, supra note 12, at 15.

14. We have not computed actual operating characteristics for two reasons. First, there is no "gold standard" of diagnosis against which to compare the subset of RRAS criteria we are examining. In traditional operating characteristic analysis, the ability of an instrument to recognize a diagnosis (of for example a disease or mental disorder) is compared to the standard method of making that diagnosis; there is no analogous diagnosis of risk that could be used as a comparison for the RRAS. Second, we are not examining the entire RRAS. Due to the difficulty in collecting data for the dynamic RRAS criteria, this study is only examining a subset of RRAS criteria, the static factors. Hence, the full instrument is not being examined. Nonetheless, one can see that particularly for high-risk individuals, a positive predictive power of 0.6 is a substantial improvement over the ADTC base-rate of 0.1, which is how often one could expect to be correct in high risk classification by chance.

15. Hanson, R. K. & Bussiere, M. T. supra note 9 note at 10:

Overall, the strongest predictors of sexual offense recidivism were factors related to sexual deviance. Sex offenders were more likely to recidivate if they had deviant sexual interests, had committed a variety of sexual crimes, had begun offending sexually at an early age, or had targeted boys, strangers, or unrelated victims. Sexual interest in children as measured by phallometric testing was the single strongest predictor of sexual offense recidivism. After sexual deviance, the next most important predictors were general criminological factors, such as any prior offenses, age, and antisocial personality disorder. These factors mark a dimension common to many criminal populations that has been variously referred to as "low self-control", psychopathy, or lifestyle instability. There is extensive research linking general criminological factors to non-sexual recidivism among both sexual and non-sexual offender populations. Although criminal lifestyle was, in itself, only moderately related to sexual offense recidivism, there is some evidence that the combination of deviant sexual preferences and psychopathy places offenders at particularly high risk for committing sexual crimes. [citations omitted]

16. Although the foregoing analysis is based only on static factors, scored 0, 1, 2, the study did examine group means for the full RRAS, including dynamic factors and scored and weighted in the original manner. The full RRAS group means are 38, 53, and 60 for probationers, state prison sex offenders, and ADTC sex offenders, exactly in the sequence one would expect. Additionally, a small sample (N 10) of civilly committed sex offenders was available; their full RRAS mean was 62.

17. To be found repetitive-compulsive, the sex offender's offense must be part of a broader pattern of deviant sexual behavior, and the sex offender must exhibit loss of sexual control. See Witt & Frank, supra note 10.

18. Hanson, R. K. & Bussiere, M. T., supra note 9, for general support of breadth and chronicity of sexual deviance as predictor of recidivism.

19. Hudson & Seto. Rape: Psychopathology and theory. In D. Richard Laws & William O'Donohue (eds.) Sexual Deviance: Theory, Assessment, and Treatment (332-355), New York: Guilford Press, 1997; U.S. Department of Justice, Bureau of Justice Statistics, supra note 12;

20. Admittedly, this is an exploratory study, and much further work needs to be done to examine survival curves of the three risk groups before concluding that such an improvement over chance can be sustained

21. Hanson & Bussiere, supra note 9.

22. For example, Quinsey et al supra note 5.

23. Some may object to the use of the term "weights," given that some instruments, such as Robert Prentky's Adult Sex Offender Risk Assessment Schedule, give all items equal weights. However, by varying the number of items in each category, such as sexual deviance or response to treatment, one can readily give that factor more weight.

24. See discussion in Hanson, R. K. (1998, in press). What do we know about sex offender risk assessment? Journal of Psvchology, Public Policy, and Law.

25. Hanson, R. K. (1998, in press), id.

26. See discussion in Hanson, supra note 9.

27. Dix, G.E. (1976). Differential processing of abnormal sex offenders. Journal of Criminal Law, Criminology, and Police Science, 67, 233-243; Hanson & Bussiere, supra note 9; Smith, W. R. & Monastersky, C. (1986). Assessing juvenile sex offenders' risk for reoffending. Criminal Justice and Behahavior, 13, 115-140; Hanson, R. K. (1998, in press), supra note 24.