Sexual Offender Treatment Efficacy Revisited

AbstractBSTRACT

Several authors have recently addressed current views of sexual offender treatment efficacy. Some maintain that offenders can gain from treatment while others argue that the vast majority cannot. Some researchers say that the field of sexual offender treatment is to new to be able to determine whether or not treatment works. This latter group notes that most studies in this field have not yet reached the point at which meta-analytic techniques can be applied; for this reason no definitive statements can be made about the utility of treatment. The present analysis examines the issues from a slightly different perspective. Data from a large group of studies are combined to identify patterns which can be examined later in more detail. More specifically, 79 sexual offender treatment outcome studies are reviewed, encompassing 10,988 subjects. Recidivism rates for treated versus untreated offenders are investigated according to age of offender, age of victim, offense type, type of treatment, location of treatment, decade of treatment, and length of follow-up. Each study is used as the unit of analysis, and studies are combined according to the number of treated versus untreated subjects who reoffended in each category. Clinical implications are drawn from these results.

Key Words: relapse prevention; sexual offender recidivism; sexual offender treatment; sexual offending

Introduction

Efforts to treat sexual offenders have proliferated over the last 50 years (Becker, 1990; Doshay, 1943; Frisbee & Dondis, 1965; Hanson, Steffy, & Gauthier, 1993). A national survey has reported that the number of sexual offender treatment programs has grown from 643 in 1986 to 1784 in 1994 (Freeman-Longo, Bird, Stevenson, & Fiske, 1995).

Concomitantly, public funding for sexual offender research and treatment has declined in the last decade. A poverty of research funds has hampered improved understanding of the effectiveness of various offender treatment interventions, perhaps due to the belief that no treatment is effective.

A variety of circumstances has contributed to the pessimistic climate surrounding the efficacy of sexual offender treatment. First, the field of sexual offender work has just emerged as a separate clinical specialty with its attendant idiosyncrasies of intervention and treatment. Second, treatment providers often have scarce resources to collect, analyze, and publish their work; when they do, their research may be preliminary in nature. Third, sexual offender treatment research frequently lacks the rigor of more established scientific fields. Some authors hold that imperfect designs result in findings that cannot be generalized, while others contend that withholding treatment to achieve a balanced design is unethical.

Meta-analysis: Help or Hindrance?

The technique of meta-analysis surfaced in psychology in the 1970s and 1980s. By the 1990s some authors argued that it was necessary to establish the validity of a construct (Sohn, 1995). The effect size statistic becomes the common denominator by which a number of studies can be compared and contrasted according to specific characteristics they share in common. This statistic also provides a weighting factor by which studies can be evaluated.

In 1989 Furby, Weinrott, and Blackshaw reviewed 55 sexual offender treatment studies, and found that no studies qualified for the meta-analytic procedure, as it requires randomized, matched treated and untreated groups. Their extensive examination of the literature revealed no compelling evidence that treated offenders' recidivism rates were lower than rates for untreated offenders.

More recently, Marques (1994) completed a study meeting the criteria for meta-analysis, and subsequently, Hall (1995) used those data and others to publish the first meta-analysis in the field. He reanalyzed recidivism data from 12 studies to ascertain effect sizes and found a slight overall treatment effect (r = .12). While Hall's study represents a significant step in this field, his N of 1313 subjects is only a small fraction of the data pool on sexual offender treatment.

Scientists continue to raise concerns about techniques such as meta-analysis. Festinger (1980) warned of the fallacy of premature efforts to subject a data set to advanced analytic techniques if the concepts are still somewhat vague. The quest for statistically significant findings may cause researchers to discard data that might otherwise validate the construct at hand. Seligman (1995) adds his caution that random assignment of subjects to treatment conditions may seem a rigorous research technique but may result in the wrong subject being given the wrong treatment; individuals may differ in the nature of treatment each needs. The standardization process on which meta-analysis is based may subvert the findings of treatment effectiveness. Social scientists remain divided about whether or not techniques designed to measure the effects of treatment outweigh other, less statistically rigorous approaches.

The controversy over standards of experimentation in the field of sexual offender treatment may be premature. Another task may be the investigation of what types of treatment seem to work, and in what circumstances. If trends emerge, they could then be pursued under more controlled experimental conditions. The immediate objective would be to determine if any sexual offender treatment is successful rather than to delineate its strengths and the conditions under which it exerts its effects.

Method

All relevant studies on sexual offender treatment from 1943 through 1996 were located using computer searches, reference lists, and contacts with authors. Some studies were eliminated from the resulting pool of 359 studies, particularly those with female (N = 10) and developmentally disabled offenders (N = 8). Studies with fewer than 10 subjects (N = 36) were not included because of the question of whether or not any trends seen would be replicated in larger subject pools. Additionally, studies involving biomedical treatment (N = 31 ) or physical castration (N = 2) were omitted due to the relative invasiveness of these techniques compared to others in this investigation. Studies whose subjects overlapped with those in other reports were excluded (N = 25), along with studies with unclear outcome data (N = 48). This latter category contained studies in which the recidivism rate was reported, but not the raw data from which the rate was derived. Studies which described treatment but presented no outcome data (N = 64) also were excluded. Literature reviews of sexual offender treatment were set aside (N = 41), together with 15 quantitative reviews of the literature, the data from these reviews were considered elsewhere. Only English language studies were included.

The resulting subset of data consisted of 79 studies with a total of 10,988 subjects. While this data set does not have the rigor of that of Hall's, the larger N here has the potential to reveal patterns not readily apparent in his data set with its relatively low base rates. The current subject pool does not include subjects who dropped out or were terminated during the course of treatment. Dropouts/noncompleters were excluded due to the lack of consistency with which data on these subjects were reported in the various studies; some studies limited their calculations to the rate of dropouts from the original pool of subjects, while others reported the recidivism rate for dropouts in their data pool. The elimination of the data on dropouts could have skewed the results. Studies such as that by Miner and Dwyer (1995) point to a differential effect that treatment may have on completers as opposed to dropouts.

A three-step procedure was used to combine data in the current analysis. First, each study was coded according to a standard set of categories: Recidivism Rate, Offender Age, Sexual Offender Type, Treatment Location, Treatment Length, Length of Follow Up, Decade of Treatment, and whether treatment was Voluntary or Mandatory. Second, the raw data were classified, category by category, across studies, e.g., recidivism rates for rapists and child molesters who received relapse prevention treatment. Third, recidivism was computed by tallying the subjects, across studies, who reoffended and dividing that number by the total number of offenders in that category.

Recidivism was defined, whenever possible, as the number of subjects who were rearrested for a new sexual offense. This definition was chosen because it is more conservative than just counting subjects who were convicted of a new sexual offense; many subjects may be found not guilty due to legal technicalities, regardless of guilt or innocence. By counting them as having reoffended if they were rearrested, the potential for positive treatment effect bias is reduced.

The three-step process permitted the compilation of data across studies even though a true meta-analysis was not done.

Procedure

A series of analyses was done comparing treated and untreated subjects. The data were analyzed according to offender subtypes: All Subjects, Juveniles, Rapists, Child Molesters, Exhibitionists, and Type Not Specified. In some analyses the Child Molester category was further differentiated into offenders with female or male victims and incest or nonincest perpetrators. Within each subcategory, the following six variables were examined: (a) the Decade in which treatment took place, (b) the Length of Follow-Up, (c) the Location of Treatment (prison or hospital or outpatient), (d) whether the mode of treatment was Relapse Prevention (RP) or not, (e) the Duration of Treatment, and (f) whether treatment was Voluntary or Mandatory. Categories (e) (Duration of Treatment) and (f) (Voluntary or Mandatory) were excluded from this report because the data did not discriminate treated from untreated subjects.

In many instances researchers did not report data in all categories. This meant that the number of subjects differed from one analysis to the next depending on which information was or was not provided on any given factor.

The 79 studies provided little information on statistical significance; nevertheless, an attempt was made to indicate the relative strength of patterns in the data. Three criteria were formulated to distinguish stronger from weaker findings. First, a recidivism rate of less than 11% was proposed as a ceiling for a positive treatment outcome. This cutoff score was based on the common-sense notion that programs with 90% effectiveness would be acceptable to the general public and to researchers and clinicians in this field. The second strength criterion was an N of at least 100 subjects in each cell; cells with higher numbers had more validity than ones with only a few subjects. Third, a spread of at least 10% between treated and untreated subjects signaled yet another kind of strength; a 10% gap between treated and untreated subjects would suggest a clear-cut distinction between the two groups even though this difference did not prove that treatment was effective. Overall, a finding that fit just one of these three criteria indicated a pattern warranting future exploration, but findings that met all three were considered stronger.

Results

The 10,988 subjects were divided into the subtypes shown in Table 1. Treated subjects had lower recidivism rates than untreated subjects in all categories except for Type Not Specified. The lowest recidivism rate of 7.1% was for juveniles, a rate under 11%, thus meeting the first strength criterion. Each cell in this table except untreated exhibitionists met the second strength criterion, that of more than 100 subjects per cell. The categories of Child Molesters and Exhibitionists met the third criterion (a 10% difference between treated and untreated subjects), but the overall results did not.

Additional analyses were needed to investigate two specific issues:

	(1) any differential effect the location of treatment might have had and
	(2) whether newer treatment modalities had produced any amelioration of recidivism rates.

The following analyses involved a further examination of the data, addressing the use of RP, Treatment Location, and Offender Type. Table II displays a comparison of RP and other approaches. The decision was made to include studies in the RP category that were designated as such, as well as those that used cognitive behavioral techniques. Marshall and Anderson (1996) point out that definitions of RP need to be refined; in the interim, distinctions between these two techniques in individual studies could not be delineated clearly.

Table I. Subjects in the Data Pool

Subject Type		Recidivism Rate/Treated	Recidivism Rate/Untreated
Juveniles (N=1025)
	Rapists	5.8%(6/103)	N/A(a)
	Child molesters	2.1%(1/47)	N/A(a)
	Type unspecified	7.5%(66/875)	N/A(a)
Rapists (N=631)
	Juveniles	5.8%(6/103)	N/A(a)
	Adults	20.1%(79/393)	23.7%(32/135)
Child molesters (N=2184)
	Female victims	15.6%(35/225)	15.7%(28/178)
	Male victims	18.2%(46/253)	34.1%(28/82)
	Victims' gender mixed/not specified (N=1446)
	Incest	4.0%(16/397)	12.5%(4/32)
	Nonincest	1.7%(51/437)	32.0%(24/75)
	Incest/nonincest not specified (N=1243)

		Year 1	Year 3	Year 5	Years 5+
Juveniles
	Treated	5.2%(115)	21.6%(37)	-	6.8%(453)
	Untreated	-	-	-	-
Rapists
	Treated	-	19.8%(91)	22.6%(168)	29.3%(41)
	Untreated	-	19.4%(67)	27.9%(68)	-
Child molesters
	Female Treated	-	12.0%(51)(a)	19.8%(91)	-
	Female Untreated	-	13.3%(83)	17.9%(28)	-
	Male Treated	-	19.0%(158)(a)	14.5%(55)	-
	Incest Treated	-	6.9%(29)(a)	5.3%(57)	-
	Incest Untreated	-	18.2%(22)	0%(10)	-
	Nonincest Treated	18.1%(22)(b)	17.1%(41)	17.8%(146)	-
	Nonincest Untreated	-	41.9%(31)	25.0%(44)	-

Recidivism rates rose over time for juveniles, rapists and child molesters with female victims, while rates for child molesters with male victims and exhibitionists fell. In particular, the gap between treated and untreated child molesters with male victims appeared to grow the longer the subjects were followed in the community. This trend seems promising, with the caution that almost every cell has an N below 100.

DISCUSSION

Under specific conditions, less than 11% of treated offenders reviewed here reoffended. This finding suggests a number of implications involving both the discipline of sexual offender treatment, and research on offender treatment.

Many treated sexual offenders had recidivism rates below 11%; this suggests that most offenders may not need the extreme measure of permanent institutionalization, provided that they abide by conditions tailored to their individual offender profiles. Once treated, offenders may be monitored in the community for extended periods at a fraction of the cost of more restrictive or invasive measures. Marshall and Anderson (1996) recently proposed that some offenders can be treated without the protracted RP technique. If so, treatment could be shortened for some offenders at considerable cost savings.

Juveniles responded well to treatment; the demonstrated efficacy of juvenile offender treatment programs is a strong argument for their continued existence. While many other types of offenders seemed to respond to treatment, child molesters with male victims remain an enigma. This population may need more aggressive interventions. Further investigations need to address this issue.

Early-research grouped sexual offenders into homogeneous categories. Clustered together, research data yielded almost no treatment effects. When subjects are subdivided by type, the efficacy of treatment for at least some of them becomes more apparent. Prentky and Knight's (1991) classification system for rapists indicates that these offenders are heterogeneous. Future research should focus on which treatment works with whom, rather than debating whether or not to treat.

The technique of profile analysis might facilitate the examination of whether or not specific kinds of sexual offenders respond to treatment differently. Few studies have investigated how to treat exhibitionists and child molesters with male victims; most of the analyses contained here had an N of less than 100 per cell for these two subtypes. Research concerning treatment efforts with individuals in these two subgroups might produce additional insight into their long-term prognosis. Profile analysis also might be used to discover effective interventions with learning disabled or cognitively impaired offenders, and for rapists with, and without, a prior history of criminal behaviors. In addition, results from treatment in prison should continue to be distinguished from results in hospital setting to highlight differences.

An analysis such as this generates many methodological issues. The choice of recidivism index poses one such dilemma. In this study recidivism was defined as a rearrest for a sexual offense, whenever possible. On the one hand, this definition yields higher, and perhaps more accurate, recidivism rates than reconviction data. On the other hand, praictitioners comment that treated offenders may be more likely to be questioned about unsolved sexual crimes than their untreated counterparts.' While their rearrest rates maybe higher, the rate at which their charges are found unsubstantiated may be higher as well. Is there a rearrest bias against treated offenders, and, if so, should this affect the recidivism index chosen? Research exploring these issues has not, as yet, been done.

The external validity of analyses such as this one would increase if data collection and reporting procedures were standardized. The current study assumed that all RP treatment is the same, but Marshall and Anderson (1996) point to the lack of consistency with which RP is used. In addition, many programs rely on a combination of interventions. While a broad brush approach may be helpful, it limits the extent to which the utility of one particular technique can be isolated for examination.

Inaccurate or insufficient data within individual studies considered for this analysis may have spawned methodological errors. Data from certain studies could be used only marginally or not at all in an effort to preserve the integrity of' this analysis. For example, some authors reported recidivism percents without providing the raw data upon which the analyses were based; these studies could not be included in the current analysis. If guidelines could be formulated for the recording of research data, then this potential selection bias could be reduced. The work of Hanson et al. (1997) represents an initial step in establishing such guidelines.

Definitions of sexual offender types and treatment modalities could be standardized as well. This practice would increase the extent to which generalizations could be made across offender populations, Ultimately, proactive planning and centralized funding may be necessary to institute standardized research and treatment protocols across research sites (Hanson, 1997). In this way, meta-analytic studies could establish with more certainty which aspects of offender treatment work and which do not. With larger N's, a more detailed examination could be done about issues such as length of follow-up as a critical factor in any analysis of recidivism.

The above analysis combined data across studies while permitting a rudimentary assessment of the strength of each finding. When viewed in this way, the data suggest that the field may have progressed beyond the skepticism triggered by Furby et al. (1989). Data from multiple studies suggest that treatment may lower recidivism rates, at least for some sexual offenders.

CONCLUSIONS

The efficacy of sexual offender treatment currently looms large in public discourse. The question of how best to treat may be as complex as the reasons people offend. Even though research remains in the formative stages, what has been learned so far has practical utility. A variety of treated sexual offenders reoffend at rates below 11%. This finding, suggests that some effective components of the treatment process may have been identified. Practitioners working with offenders should master standard curricula explicating these treatment tools, so they can apply them in a uniform and consistent manner. Future research could then enhance what is already known about how to treat sexual offenders.

Should offender treatment be abandoned until its efficacy is incontrovertibly established? While this course may be tempting from a scientific perspective, the public safety ramifications of withholding even relatively ineffective treatment from dangerous offenders cannot be risked.

REFERENCES (a)

(a) Extended reference list not included.

Subject type	Treated Recidivism Rate	Untreated Recidivism Rate
Juveniles ((N=1025)	7.1% (73/1025)	N/A(a)
Rapists (N=528)(b)	20.1%(79/393)	23.7%(32/135)
Child molesters (N=2137)(c)	14.4%(241/1676)	25.8%(119/461)
Exhibitionists (N=331)	19.7%(61/310)	57.1%(as/21)
Types not specified (N=6967)	13.1%(786/5979)	12.0%(119/988)
Totals (N=10,988)	13.0%(1240/9383)	18.0%(282/1605)
Note. Does not include dropouts. (a) No data available in this category. (b) Does not include 103 juveniles. (c) Does not include 47 juveniles.

Type of Offender	Relapse prevention(a)	Group / behavioral / other	Unspecified	Untreated
Juveniles (N=1025)	9.8%(10/102)	6.8%(63/923)	N/A(b)	N/A(b)
Rapists (N=528)(c)	8.3%(6/72)	N/A(b)	22.7%(73/321)	23.7%(32/135)
Child molesters (N = 2137)(d)	8.1%(18/221)	18.3%(96/524)	13.6%(127/931)	25.8%(119/461)
Exhibitionists (N=331)	0%(0/13)	20.5%(61/297)	N/A(b)	57.1%(12/21)
Type not specified (N=6967)	5.6%(17/305)	14.3%(699/4872)	8.7%(70/802)	12.0%(119/988)
Totals (N=10,988)	7.2%(51/713)	13.9%(919/6616)	13.1%(270/2054)	17.6%(282/1605)
Note. Does not include dropouts. (a) Percentage recidivism. (b) No data available in this category. (c) Does not include 103 juveniles. (d) Does not include 47 juveniles.

Type of Offender	Outpatient(a)	Prison	Hospital	Unspecified or mixed	Untreated
Juveniles (N=1025)	6.3% (13/205)	6.9% (10/144)	8.5% (19/223)	6.8% (31/453)	N/A (b)
Rapists (N=528)(c)	N/A (b)	14.1% (14/99)	18.0% (16/89)	23.9% (49/205)	23.7% (32/135)
Child molesters (N=2137)(d)	13.9% (66/475)	21.4% (52/243)	14.2% (55/388)	11.9% (68/570)	25.1% (119/461)
Exhibitionists (N=331)	19.0% (50/263)	N/A (b)	N/A (b)	23.4% (11/47)	57.1% (12/21)
Type not specified (N=6967)	8.2% (51/620)	7.7% (133/1734)	17.5% (520/2968)	12.5% (82/657)	12.0% (119/988)
Totals (N-10,988)	11.5%(180/1563)	9.4% (209/2220)	16.6%( 610/3668)	12.5% (241/1932)	17.68% (282/1605)
Note. Does not include dropouts. (a) Percentage recidivism. (b) No data available in this category. (c) Does not include 103 juveniles. (d) Does not include 47 juveniles.