An Analysis of the Argument That Clinicians
Under-predict Sexual Violence in Civil Commitment Cases
Richard Wollert, Ph.D.
Institute for the Study of Mandated Treatment
Running head: Sex Offender Commitment
Requests for reprints and
correspondence concerning this article should be sent to the author at the
Institute for the Study of Mandated Treatment,
This paper presents the results of an analysis that evaluated the assumptions and arguments set forth in a recent paper (Doren, 1998) which concluded that clinicians under-predict the risk that candidates for commitment as dangerous sex offenders will recidivate. This analysis indicated that such a conclusion was untenable because almost all of the assertions on which it was based might be disputed in one way or another. These flaws undermine the paper’s value as a source of secondary authority in hearings pertaining to the commitment of dangerous sex offenders. Nonetheless, its publication will probably be beneficial for professional practice in this arena because it raises several important issues that stand in need of resolution.
An Analysis of the Argument That Clinicians
Under-predict Sexual Violence In Civil Commitment Cases
A number of
states have now enacted legislation allowing for the post-prison civil
commitment of sex offenders on the grounds that they would probably recidivate
if they were to be released from custody (
Many articles that are relevant to the issue of sexual dangerousness have been published in response to this interest. One of these was authored by Doren (1998), who concluded that clinicians under-predict sexual violence because they classify the great majority of incarcerated sex offenders as likely nonrecidivists and thereby create a situation where their “false negative” error rate is much larger than their “false positive” error rate . This conclusion was based on an analysis which assumed that the maximum prediction rate for sexual violence ranged from .6% (in the State of Washington) to 12% (in Wisconsin) while the actual longterm recidivism rate for rapists and child molesters is much higher – from 39% for rapists to 52% for molesters. None of these premises, however, had ever been supported or even seriously articulated in the literature that had been previously published. It was therefore necessary for Doren, in the first four sections of his analysis, to explain why these assumptions were warranted. In a fifth section that concluded his paper he seemed to imply that the existence of high false negative rates meant that clinicians could classify more offenders as violent than is currently the case and that in the course of doing so they might be fulfilling such ethical obligations as their “duty to protect” (p. 111).
Doren’s article is significant because experts have cited it or submitted it as a basis for their opinions in many civil commitment cases. If all of its conclusions and arguments were widely accepted as accurate in such proceedings, a number of assumptions about recidivism patterns among sex offenders would be adopted. A climate would also be established where it would be almost untenable for any clinician to predict that a commitment candidate might fall at a recidivism risk justifying his release.
A critical review of Doren’s arguments and conclusions is justified in light of the significance of these possible effects. The remainder of this paper presents such a review. Each of the next five sections refers to one of the major sections of Doren’s paper and consists of a summary, critique, and focused evaluation of his assertions. The sixth and concluding section contains what I believe to be an appropriate overall evaluation of Doren’s position, followed by a discussion of its implications for further research and dialogue.
I. The Significance of Offender Types and Research Shortcomings for
Estimating the True Base Rate of Recidivism Among Dangerous Sex Offenders
Before the … true base rate can be discussed … the population being described must be …defined. The … set of sex offenders (who) are typically referred (for commitment) are extrafamilial child molesters … or rapists. The recidivism base rates found for these … offenders seem most pertinent to the commitment process.
All … studies …(on) recidivism …rates in previously convicted sex offenders share …shortcomings (that) …potentially … underestimate the true base rates. Reconviction rates represent a diluted measure of true reoffense rates. Data beyond reconviction rates show an increase in sexual recidivism in the 27%-47% range. The sexual predator laws … specify only that there be a certain probability that the person will recommit a … “sexually defined act”, not get … reconvicted.
The statutory language of these laws …(leaves) open the interpretation that the timeframe of relevance (for recidivism) is the offender’s lifetime. All … research potentially underestimates the … rate for sex offender recidivism because the studies end too soon.
The base rate of sexual recidivism for predatory sex offenders needs to be estimated in order to make any estimates of the extent to which clinicians under- or over-predict violence. In this section of his paper, Doren set forth three criteria for homing in on particularly valuable research findings for determining the “true” base rate of sexual recidivism for predatory offenders.
His first criterion was that follow-up research on rapists and extrafamilial molesters should be studied while research on incesters and intrafamilial molesters should be screened out. This might seem like a reasonable focus because low rates of recidivism are usually found for incesters (Berlin, Hunt, Malin, Dyer, Lehne, & Dean, 1991) and they are rarely referred for commitment. Incarcerated incesters and intrafamilial molesters do, however, make up a very large proportion of the prison release class from which commitment candidates are selected. The pattern of errors that clinicians make in predicting recidivism for this group may also be different than their errors for rapists and molesters. Eliminating research on this group makes it impossible to evaluate whether or not this might be the case. This, in turn, limits the applicability of Doren’s conclusions.
His second criterion was that the results of studies that equate recidivism with rearrests or new charges for sex crimes should be weighed most heavily while studies that report recidivism in terms of reconvictions should be discounted. The adequacy of this criterion is also questionable. Although studies using broad criteria have generally reported larger recidivism estimates than studies based on reconvictions, these criteria may overestimate recidivism because convicted sex offenders are likely to be detained as suspects during the early phases of a sex offense investigation and then released. Contending that broad standards provide more accurate recidivism estimates than reconviction rates is also societally divisive because reconviction rates stem from a consensually accepted institution for trying evidence – that is, the criminal justice system.
His third criterion was that findings from studies with very long or lifetime follow-up periods should be regarded as highly informative while findings from shorter studies should be discounted. This standard runs counter to the view that an exclusive reliance on very long-term studies might also have shortcomings. The results of 5-year studies may be relatively easy to interpret, for example, because a large amount of the variance in 5-year studies is attributable to the risk condition of interest – that is, being a convicted sex offender. This would be more ambiguous for 20-year studies, where a substantial amount of variance might be attributable to other risk conditions (e.g., being a single male). It is also possible, in view of the high rate of sexual misconduct among non-criminal samples (Hall, 1990), that the sexual conviction or misconduct rates for nonrecidivating sex offenders in the late stages of a long-term study might approximate those for matched samples of men with no prior convictions for sex crimes. If this were the case, cumulative data from studies that went beyond this “parity point” might be regarded as misleading by the courts. This, in turn, might lead to a greater reliance on time-limited rather than open-ended follow-up studies.
Considering the assertions in this section in light of the foregoing critique, many disadvantages are associated with an approach that concentrates on studies of rapists and molesters, discounts reconviction data, and overlooks short-term follow-up research. Predictive expertise might be enhanced, however, if research efforts were focused on a) examining factors associated with overestimates of recidivism as well as underestimates; and b) determining the point of “diminishing marginal returns” for long-term studies.
II. The True Recidivism Base Rate for Extrafamilial Child Molesters
One study that stands out (in relation to the foregoing screening criteria is by) Prentky, Lee, Knight, and Cerce (1997). (They) investigated …115 extrafamilial child molesters over a 25-year period using the recidivism definition of “new sex offense charge” (as well as reconviction for a new sex offense). The definition of “new sexual charge” included only charges resulting from physical contact.
Prentky et al. … performed a … survival analysis to determine the failure rate of their sample. (They) documented a 52% failure rate for sexual re-offending …over 25-year(s).
Some portion of the people charged … may have been innocent (but) … this portion would likely be far smaller than the number… who are never caught. The 52% figure (therefore) represents the lowest approximation for extrafamilial child molester sexual recidivism.
Comparisons between (the) Prentky … findings and those from other relevant research will be made. The results described by Hanson, Scott and Steffy (1995) and Hanson, Steffy, and Gauthier (1993) … represent those that are best compared. The Hanson … research included a … similar sample …in size (n = 191). Hanson (used) a 31-year follow-up period … and survival analysis. 18.3% of the Hanson sample were incest offenders … compared to none in the (Prentky) research. The recidivism measure was … reconviction … versus … “new sex offense charge”. A sex offense recidivism rate of 35.1% (was obtained). (Recall that) the summary above (see the second paragraph of the first section) … demonstrated an increase of 27%-47% in recidivism base rates (when recidivism was equated with new charges rather than reconvictions). The Hanson … reconviction rate … translates to a closer approximation of actual sexual reoffense rates … by adding that range of increased recidivism to the 35.1% figure, resulting in a … range from 44.6%-51.6% (i.e., 27%-47% of 35.1%, added to 35.1%). (Also recall that) 18% of the Hanson sample were incest offenders, a group … known to have some of the lowest rates for sexual recidivism. Hanson’s … overall “translated” recidivism rate … would (thus) very likely be even higher than the 44.6%-51.6% range. The Hanson … results support and replicate a 52% “conservative approximation” rate for extrafamilial child molester … recidivism.
Broadhurst and Maller (1992) studied … molesters and rapists over … 11 years. Recidivism … was (defined as) … reincarceration … following an offense of violence or sex. Recidivism … was found to be 34%. This percentage may be compared to Prentky’s (1997) 23% reconviction rate … or 30% new charge rate.
Radzinowicz (1957) studied … molesters over … four years. Sexual offense reconvictions were found in 11.3%. Prentky … documented a 12% reconviction rate over a four year “at risk” period.
Hanson and Bussiere (1996) performed a meta-analysis involving 61 different sex offender data bases … and found a 12.7% sex offense recidivism rate for 9,603 child molesters over an average period of four to five years. The near exactness of Hanson and Bussiere’s finding to the reconviction rate found by Prentky … during the first four … year period seems … supportive of the conclusion that the (rearrest) findings by Prentky … are in keeping with … other research involving shorter-term measures.
Some published studies … document far lower reconviction rates. These samples … should not be included in this type of investigation (because) subjects were placed directly on probation and thus likely to be viewed by sentencing courts as … a less serious risk … than subjects … incarcerated for their crimes.
The conclusion seems reasonable that Prentky’s (1997) 52% … rate is in keeping with the … molestation recidivism research. This … assumes that extrapolations must be made to make proper comparisons between the Prentky (1997) research and other relevant research. Extrafamilial child molester sexual recidivism should be thought of as having a conservative approximation of 52%.
As the foregoing
excerpts indicate, a single study by Prentky served
as the foundation for Doren’s conclusion that the rate of sexual recidivism for
extrafamilial child molesters should conservatively be set at 52%. Taking such a sweeping position on the basis
of a single study runs counter, however, to one of the most widely-practiced
beliefs in the social sciences – that is, that many studies are required to
advance a definitive conclusion about an issue because each individual study is
limited in one way or another.
Consistent with this view, which reflects the way the social sciences
have evolved, Prentky’s research needs to be
considered in light of
several limitations. The
validity of his survival analysis may have been undermined, for example, by any
of a number of shortcomings (Broadhurst & Maller, 1992; Greenhouse, Stangl,
& Bromberg, 1989; Harris & Moitra, 1978);
these threats were not, unfortunately, considered in the original article. It was also the case that Prentky’s
sample of molesters were highly likely to recidivate because they were
convicted of many more sex offenses (N=4.6) than a more typical sample
of incarcerated sex offenders, about 90% of whom have been convicted of only
one sex offense (Song & Lieb, 1995). Finally, it should be noted that statutory
sex offenses were included in Prentky’s definition of
recidivism. This may have been a serious
limitation because high recidivism estimates will be obtained in studies where
these crimes are counted, but the results of
studies may be discounted in commitment cases once it is understood that they
confound obviously violent sex offenses with more ambiguous crimes.
A valuable perspective for evaluating the
significance of Prentky’s article may be gained by
referring to some of the comments it contained.
In their introduction (p. 637), for example, the authors made it clear
that their primary purpose was to provide an overview of how variability in the
estimated rates of recidivism was affected by methodological variations. In later sections they acknowledged some of
the limitations of their research and
avoided the temptation
to endorse any recidivism rate
as definitive, observing instead that the “marked heterogeneity of sexual
offenders precludes automatic generalization of the rates reported here to
other samples” (p. 656). In light of the
restrained and relativistic interpretation that the Prentky
researchers placed on their findings, it is difficult to understand how Doren
could objectively claim that they serve as the basis for establishing something
so absolute as “the true recidivism base rate for
extrafamilial child molesters”. It is
also difficult to understand how, proceeding from an objective standpoint, he
could endorse Prentky’s highest recidivism estimate
without informing readers of the study’s purpose, its limitations, or the fact
that Prentky’s recidivism estimates for molesters
dropped into the 25% to 41% range when reconviction data were analyzed.
After introducing Prentky’s survival analysis, Doren cited two different sources of support for his position. One of these was Hanson’s 30-year survival analysis. Again, however, aspects of this research that held negative implications for Doren’s position were not discussed. A major unmentioned flaw, for example, was that the data for Hanson’s offenders who had died or reached the age of 70 without reoffending – those who probably had the longest trouble-free histories - were destroyed prior to being analyzed (Hanson, Scott, & Steffy, 1995). As a result the 35% recidivism rate that was obtained when the remaining data were analyzed is probably an overestimate and it is unlikely that a truly meaningful “translated recidivism rate” could be calculated on the basis of this figure. Furthermore, Doren’s opinion that the “translated recidivism rate” for the molesters in Hanson’s subjects would probably exceed 52% if the data for incesters were removed seems overly confident in light of the fact – which was not discussed – that Hanson found comparable recidivism rates for incesters, female molesters, and bisexual molesters.
Doren’s second source of support revolved around comparisons that showed that the reconviction rates found in the early years of the Prentky research were similar to the rates found in shorter studies by Broadhurst and Maller (1992), Radzinowicz (1957), and Hanson and Bussiere (1996). It was implied that these comparisons were useful for establishing that the figure generated by Prentky’s 25-year survival analysis of rearrest rates was generalizable to all incarcerated molesters. Doren did not, however, articulate a theory on how such different measures might be associated over a long period of time or set forth a convincing case that generalizations were justified because of similar sampling procedures and definitions of recidivism. His thesis was therefore largely speculative.
Towards the end
of this section, Doren acknowledged that other researchers had obtained
recidivism estimates that were considerably lower than 52%,
attempted to protect his position from the implications of these results by
pointing out that the subjects in these studies were
not considered dangerous enough to be incarcerated. The Washington State Institute for Public
Policy has reported, however, that about 13 % of all sex offenders incarcerated
in Washington over an 18 month period were rearrested for sexual offenses that
involved physical contact (Song & Lieb, 1995)
during a follow-up period that covered 5.7 years. Since 66% of those who were rearrested were
reconvicted (Lieb, 1997), the reconviction rate for
this sample was about 8.6%.
Considering the assertions in this section in light of the foregoing critique, a single study was inadequate to sustain Doren’s assertion that a “conservative approximation” of the “true base rate” of recidivism for molesters was 52%. The objectivity of his approach also seemed diminished by using rhetorical terms, rarely discussing inconsistent findings, and making speculative extrapolations. Our present knowledge base might be expanded, however, if researchers were to report more fully on the validity of their statistical techniques. Predictive expertise might also be enhanced if data were analyzed so that clearer distinctions might be made between violent and ambiguous sex offenses.
III. The True Recidivism Base Rate for Rapists
Figuring the true base rate for rapists … is a similar process to that … above. Again, the best research … is that by Prentky, Lee, Knight, and Cerce (1997). Rapists’ recidivism was measured as both a new sexual charge and a new conviction. The study employed a 25-year at-risk period. Using survival analysis, the … rate for 136 rapists as measured by a new … charge was 39%. The … reconviction rate … was 24%. The reconviction rate seriously underestimated the recidivism rate … by 62.5%.
Ideally, there would be another study … that involves … recidivism beyond reconviction. The Prentky … research appears … unique, so … reconviction rates must be used (for comparative purposes) instead. The only … long … study that …used a survival analysis (was) by Soothill & Gibbens (1978). (They) obtained … a sexual reconviction rate for rapists over the 22 year follow-up period of about 23%. The Prentky … finding of (a) 24% reconviction rate using a 25-year … period … is a close replication.
(The results of) shorter-term studies (Frisbie & Dondis, 1965; Radzinowicz, 1957; Rice, Harris, & Quinsey, 1990; Sturgeon & Marshall, 1980) compare favorably with Prentky’s … 11% reconviction rate over five years at risk and suggest that the Prentky … data represent the low side of the recidivism range. This conclusion is supported when the Prentky … results are compared to a (longer) study (Grunfield & Noreik, 1986) … and some studies … (that) employed less stringent measures of recidivism (than) … reconviction.
The conclusion seems reasonable that Prentky’s … 39% … recidivism rate is in keeping with (the) research. Given that Prentky … found reconviction rates to be frequently lower than … other studies, one may wonder if the (39%) new sexual charge rate found by these researchers was … low as well. (This) rate (also) assumes that all recidivating rapists get caught for their new crimes. Hence … it is concluded that rapist sexual recidivism should … have a conservative approximation … of about 39%.
This figure … appears to contradict the conclusion by some … that rapists have higher … rates than … molesters. This difference is easily explained by (recalling Prentky’s finding that) … rapists … show higher reconviction rates … (in) their first five years at risk (while) … molesters have higher (rates) when using either a less stringent definition of recidivism or longer follow-up periods.
As in the second
section, Doren relied on his interpretation of Prentky’s
study as the source of his conclusion in this section that the rate of sexual
recidivism for rapists should conservatively be set at 39%. Many of the limitations discussed in the
previous section are therefore relevant to the present section (e.g., the
importance of multiple studies in the social sciences, threats to the use of
survival analysis, confounding violent and ambiguous sex offenses, disregarding
Prentky’s purpose and cautionary warnings, proposing
a “true” base rate). Although further
discussion of these points is unnecessary, a couple of facts should be
summarized for the purpose of evaluating the generalizability of Prentky’s research on rapists. Like his molesters, Prentky’s
rapists were highly likely to recidivate because they had been convicted of
many more sex offenses (N=3.5) than a more typical sample of
incarcerated sex offenders. The
recidivism estimates he obtained for rapists also dropped markedly, from 39%
into the 15% to 24% range, when reconviction data were analyzed. Doren
alluded to the last percentage,
but sought to discount
its negative implications for his position by claiming – without citing any
external evidence in support of his claim - that “the reconviction rate seriously
underestimated the recidivism rate”. As
the critique in the first section suggested, an objective approach would
preclude assuming that one estimate is inaccurate simply because it yields a
lower recidivism rate than another.
After introducing Prentky’s survival analysis, Doren cited two different sources of support for his position. One of these was Soothill & Gibbens’ (1978) long-term survival analysis. Comparing the results of Soothill and Gibbens’ study with Prentky’s findings on rapists was inappropriate, however, because Prentky classified rapists as “men whose sexual offenses were on victims who were 16 years of age or older” (p. 638). Soothill and Gibbens, in contrast, classified rapists as being offenders who had victimized girls “under 13” (p. 269). It would have been more appropriate to compare Soothill and Gibbens’ findings with those for Prentky’s molesters, who were “men whose victims were under the age of 16” (p.638). Such a comparison would have revealed a finding at odds with the reasoning in Doren’s second section – that is, that only 23% of Soothill & Gibben’s subjects were convicted within 22 years while the 25 year conviction rate for Prentky’s subjects was 41%.
Doren’s second source of support revolved around comparisons that showed that the reconviction and rearrest rates found in the earlier years of the Prentky research were similar to the rates found in nine shorter studies. As in the second section, however, Doren did not explain how these results insured the generalizability of Prentky’s 25-year survival analysis of rearrest rates to incarcerated sex offenders.
the end of this section, Doren acknowledged that Prentky’s
recidivism rate for rapists was lower than that for molesters, which seemed to
contradict the conclusions of other researchers that rapists are more likely to
recidivate. Then he
attempted to support
his position by claiming
that rapists recidivated more often in the short run
while molesters recidivated at a higher rate in the long run. He
did not , however, offer any evidence other
than Prentky’s results to substantiate this claim. He
also did not consider the possibility that the molesters in Prentky’s
sample recidivated more often than the rapists because they had a more
extensive history of offenses, which has repeatedly been shown to correlate
with recidivism (Hanson & Bussiere, 1997).
As in the second section, a single study was inadequate in the third section to sustain Doren’s assertion that a “conservative approximation” of the “true base rate” of recidivism for rapists was 52%. The remaining portions of the evaluation for Doren’s second section are also relevant for the evaluation of his third section.
IV. The “Over-Prediction” of Violence
One … criticism of clinicians’ predictions of …violence is the …bias towards its over-prediction. Such a perspective … has assumed that … violence was a … rare event. This can no longer be assumed … when the violence in question is sexual recidivism.
state’s referral rate (for the filing of commitment petitions) reflects the rate at which predictions for
likely sexual recidivism are made (by clinicians). The percentage of potentially
eligible sex offenders referred for possible commitment ranges … from about
These numbers allow a determination of potential false negative rates … and false positive rates … for two extreme possibilities… (that) will demonstrate the degree to which over-predictions of sexual predation can occur.
At one extreme (suppose) that all referred offenders … (would have been) sexually violent (if released). Between 27% and 51.4% (i.e., 39%-12%, and 52%-0.6%) of the actual recidivists would not have been predicted as likely to exhibit future sexual predation. The over-prediction of recidvism would equal zero while the under-prediction of violence would be very great.
(At) the other extreme (suppose) that … every … prediction of … sexual predation were wrong. The false negative rate would still be … greater than the false positive rate.
There is (consequently) a very significant under-prediction of sexual predation when it comes to … sexual predator laws. This pattern … (will likely persist) for a long time. Only … when a state refers a percentage of sex offenders for possible commitment equal to … the known recidivism base rates will the concept of … over-prediction … become accurate.
In this section Doren claimed that the clinician-predicted rate of sexually violent recidivism (PSVR) for any state with a commitment law is equivalent to the percentage of all incarcerated sex offenders who are referred for commitment consideration. Then he reasoned that the false negative rate may be determined for molesters and rapists, respectively, by subtracting the PSVR from the “true” rates of sexually violent recidivism (TSVR) he identified in the second and third sections. Under-prediction of violence, according to this method, was the remainder left after a small PSVR was subtracted from a large TSVR. This conceptualization might be scrutinized in another venue, but it is unnecessary to pursue this direction here because Doren’s application of it was fatally flawed in two respects. First, as the critiques of his second and third sections have indicated, there is no compelling reason to assume that the TSVR’s proposed by Doren are accurate. Second, in order to generate unbiased estimates for molesters and rapists, the PSVR that is inserted in Doren’s “under-prediction” equation should refer only to molesters and rapists. The one he used, however, was based on the rates for incesters as well as molesters and rapists. Since the commitment referral rate for incesters is very low, which Doren cited as a rationale in the first section for focusing on rapists and molesters, the PSVR he relied upon consistently underestimated the appropriate figure. All of his estimates were inflated as a result.
Taken together, the flaws described in the critique of this section point to the conclusion that an adequate foundation was not established for making any estimates about the under-prediction of violence by clinicians.
The recidivism base rates relevant to the sexual predator laws are within the mid-range of probability. This result was … translated into a demonstration that all … states … have significantly under-predicted … sexual predation in the implementation of their sex offender commitment laws. This … is (at odds with) the concern (that) clinicians over-estimate … violence.
The main implications of the above findings is (that they enhance) clinicians’ ability to be accurate in making predictions of … sexual predation … (by) using the more effective actuarial methods instead of purely clinical assessment procedures.
(Previous) research (is limited because it) has employed the outcome measure of … sexual offense reconviction within the first five years post-incarceration. There are two implications of this shortcoming (for the practice of assessment). The first … is that if an individual has … (many) risk factors … then the prediction might … be stated that the person will likely be reconvicted of a sexual offense within his first five years or so post-incarceration.
Just because an offender does not (however) show the risk factors associated with …being reconvicted within a … short period of time does not mean that he will not (eventually) be a recidivist. (The second implication is that) clinicians should therefore be far more reluctant to make definitive predictive statements about non-recidivism than making definitive statements about recidivism.
In the past, the usual
ethical argument concerned whether or not clinicians should even be … making
predictions of dangerousness (because of) the problems with predicting low frequency events and the
over-prediction of violence. The
likelihood that some (sex) offenders
will recidivate can (however) be
predicted with significant accuracy. Maybe … our ethical responsibilities related to the “duty to
protect” (need to be reconsidered) …
when we can scientifically conclude that a significant set of offenders will
likely be dangerous again.
critiques of the preceding sections indicate that there is very little reason
to believe that Doren’s conclusions are accurate. An extensive analysis of the implications he
discussed in relation to his conclusions therefore seems unwarranted.
At the end
of presenting his fourth section,
however, he seemed to imply vaguely that it would be constructive
for states to attempt to equalize under-prediction and
over-prediction by increasing the percentage of incarcerated sex offenders who
are referred for commitment. At the end
of his fifth section he
advanced a similar view, suggesting that clinicians who increase their
commitment recommendations might regard this as somehow being consistent with
carrying out an ethical guideline known as the “duty to protect”.
merits some comment because the greatest contribution that clinicians might
make to the legal system would be to figure out how to improve the accuracy of
their predictions. Within the context of
this superordinate goal, it doesn’t seem reasonable to invest a lot of effort
into equating predictive error rates.
Increasing the wrongful detention rate – which is the same thing as
increasing the false positive rate - would also probably undermine the
credibility of existing commitment laws.
The reason for this is that
suggestion might be placed in some
perspective by noting that all clinicians who accept dangerous sex offender
cases need to be cognizant of their ethical responsibilities. Accurate information on these issues is
therefore always valuable. Doren did
not, however, cite any sources that might be reviewed to obtain further
information about the concept of “duty to protect”. This term was also not found in a search of
reference materials that are commonly consulted for the purpose of clarifying
ethical principles (American Psychological Association, 1992; Association for
the Treatment of Sex Abusers, 1997; Keith-Spiegel and Koocher,
1985). Most likely, Doren confused
ethical and legal concepts, because clinicians may be sued for malpractice if
they are negligent in discharging whatever duties to protect are
legally-mandated (Beck, 1998). What is
the best way of discharging these duties?
By completing a “careful clinical assessment” (Beck,
1998). Since such assessments are
statutorily required in all sex offender commitment cases, the duty to protect
is also invariably discharged in these cases.
Given this framework, it would seem misleading to imply that
the duty to protect in sex offender commitment cases might be satisfied by a
different route, such as by making more commitment recommendations.
Adoption of any policy for equating predictive error rates should be carefully scrutinized. Increasing the sensitivity of expert witnesses to the ethical guidelines that bear on their activities is an important goal, but achieving this goal requires a more accurate and specific consideration of principles than that provided by Doren.
This paper has presented a critical analysis of the assumptions and arguments that led Doren to conclude that clinicians under-predict sexual violence. Evaluating his major points in detail, it was apparent that nearly all of them might be disputed in one way or another. In some sections of his presentation, for example, assumptions were adopted that were either unjustified or no more compelling than alternative assumptions that were never considered. In others, methodological limitations of studies that were cited as being supportive of his thesis were not discussed. In still others, extrapolations were made that went far beyond the data. Throughout the paper, it would have been appropriate to consider other findings that were inconsistent with the arguments being proposed but such findings were under-emphasized.
author’s conclusion s or
th e implications of his conclusions.
Even if Doren’s reasoning were more compelling, the scope of his article would still be incomplete because he confined himself to estimating the degree to which violence is under-predicted when the only custody options being considered are total confinement or release. In doing so he neglected to consider whether violence would continue to be under-predicted if one of the custody options were to release dangerous sex offenders (DSO’s) to the community on the condition that they abide by a set of restrictions that might dictate the place of their residence and the terms of treatment, parole, and participation in a monitoring program. This is an important oversight because 1) a number of conditions have been found to be effective for reducing the risk of sexual recidivism (Federoff, Wisner-Carlson, Dean, & Berlin, 1992; Hall, 1995a; Hall, 1995b; Kruesi, Fine, Valladares, Phillips, & Rapoport, 1992; Land, 1995); 2) for each DSO, some clinical assessment needs to be made at some point as to whether the imposition of these or other conditions might significantly reduce his recidivism risk (Covington, 1997); 3) commitment statutes require that DSO’s must be conditionally released when it may be expected that their recidivism risk will be reduced below the commitment standard by adhering to a set of alternatives that are less restrictive than total confinement (Covington, 1997). Doren therefore did not portray the predictive tasks associated with dangerous sex offender cases as being as complex as they actually are.
In spite of its many shortcomings, Doren’s paper makes a number of contributions that are worthwhile to mention. It raises several questions, for example, that bear further discussion. These questions include how recidivism should be defined in dangerous sex offense cases and the length of time that should be covered by predictions of sexual dangerousness. It also points up the need for targeting future research efforts on empirical and methodological issues that are specifically important for the resolution of these types of cases. Regarding empirical research, it would seem particularly useful to collect data on 1) the sex crime rates of sex offenders and matched samples of men never convicted of a sex offense; 2) the rate with which sex offenders recidivate by perpetrating sex crimes that are of central importance for commitment statutes (e.g., forcible rape and molestation) versus their perpetration of other types of sex crimes (e.g., public indecency and statutory offenses); and 3) conditions and interventions that reduce the risk of sexual recidivism. Regarding methodological guidelines, there is a need for authoritative reviews that describe different techniques for calculating recidivism estimates, compare the advantages and disadvantages of using them for research on dangerous sex offenders, and set forth guidelines as to when they might be used and what caveats need to be mentioned in reporting their results. Finally, the publication of Doren’s position has stimulated a vigorous exchange of views. However this dialogue ends, its resolution will almost certainly enhance the level of professional expertise in an arena that is at a very early stage of development.
American Psychological Association. (1992). Ethical principles of psychologists and code of conduct. American Psychologist, 47, 1597-1611.
Association for the Treatment of Sex Abusers. (1997). Ethical
standards and principles for the management of sexual abusers.
Beck, J.C. (1998). Legal and ethical duties of the clinician treating a patient who is liable to be impulsively violent. Behavioral Sciences and the Law, 16, 375-389.
Broadhurst, R.G. & Maller, R. (1992). The recidivism of sex offenders in the western Australian prison population. British Journal of Criminology, 32, 54-80.
Doren, D. (1998). Recidivism base rates, predictions of sex offender recidivism, and the “sexual predator” commitment laws. Behavioral Sciences and the Law, 16, 97-114.
Wisner-Carlson, R., Dean, S., &
Greenhouse, J.B., Stangl, D., & Bromberg, J. (1989). An introduction to survival analysis: Statistical methods for analysis of clinical trial data. Journal of Consulting and Clinical Psychology, 57, 536-544.
Hall, G. (1990). Prediction of sexual aggression. Clinical Psychology Review, 10, 229-245.
Hall, G. C. N. (1995a). Sexual offender recidivism revisited: A meta-analysis of recent treatment studies. Journal of Consulting and Clinical Psychology, 63, 802-809.
Hall, G. (1995b). The preliminary development of theory-based community treatment for sexual offenders. Professional Psychology: Research and Practice, 26, 478-483.
development of a brief actuarial risk scale for sexual offense recidivism.
Hanson, R.K. & Bussiere, M. (1998). Predicting relapse: A meta-analysis of sexual offender recidivism studies. Journal of Consulting and Clinical Psychology, 66, 348-362.
Hanson, R.K., Scott, H., & Steffy, R. (1995). A comparison of child molesters and nonsexual criminals: Risk predictors and long-term recidivism. Journal of Research in Crime and Delinquency, 32, 325-337.
Hanson, R.K., Steffy, R., & Gauthier, R. (1993). Long-term recidivism of child molesters. Journal of Consulting and Clinical Psychology, 61, 646-652.
Harris, C. M. & Moitra, S. (1978). Improved statistical techniques for the measurement of recidivism. Journal of Research in Crime and Delinquency, 15, 194-213.
P. & Koocher, G. (1985). Ethics in psychology.
Kruesi, M., Fine, S., Valladares, M., Phillips, R., and Rapoport, J. Paraphilias (1992). A double-blind crossover comparison of clomipramine versus desipramine. Archives of Sexual Behavior, 21, 587-593.
options for sex offenders. In B.
Schwartz & H. Cellini (Eds.), The sex
offender (pp. 18/1-18/7).
Lieb, R. (1997). Findings
from the Community Protection Research Project: A Chartbook.
Prentky, R.A., Knight, R.., Lee, A., & Cerce, D. (1995). Predictive validity of lifestyle impulsivity for rapists. Criminal Justice and Behavior, 22 (2), 106-128.
Prentky, R., Lee, A., Knight, R., & Cerce, D. (1997). Recidivism rates among child molesters and rapists: A methodological analysis. Law and Human Behavior, 21, 635-659.
Song, L. & Lieb, R. (1995).
Soothill, K. & Gibbens, T. (1978). Recidivism of sexual offenders: A re-appraisal. British Journal of Criminology, 18, 267-276.
Wollert, Institute for the Study of Mandated Treatment. Correspondence concerning this article should
be sent to the author at the Institute for the Study of Mandated Treatment,
 A brief digression might clarify how these error rates are related to the concepts of under-prediction and over-prediction as they are used by Doren. In a typical prediction paradigm, incarcerated subjects who are classified as likely recidivists are called “positives”. Likely non-recidivists are called “negatives”. Positives who actually recidivate after their release are called “true” positives while positives who don’t are “false” positives. Negatives who don’t recidivate are “true” negatives while negatives who do are “false” negatives. Each subject in a prediction study is therefore assigned to one of four categories as a true positive (TP), true negative (TN), false positive (FP), or false negative (FN). The rate of accurate predictions is the sum of those in the TP and TN categories divided by the total number of subjects for whom predictions were made (N). The overall error rate is the sum of FP plus FN divided by N. Errors are biased towards over-prediction when FP/N is greater than FN/N. They are biased towards under-prediction when FN/N is greater than FP/N. Therefore, Doren’s conclusion that clinicians under-predict violence is another way of saying that their false negative error rate exceeds their false positive error rate.
 Several conventions were adopted for the purpose of presenting Doren’s assertions. For the sake of accuracy, the most important sections of his text were excerpted and presented in italics. Ellipses were used to signify portions of text that were omitted. Nonitalicized additional text, enclosed in parentheses, was inserted where it was believed that this would serve a clarifying function. Some of the articles cited by Doren also had apparently not been published when his manuscript was accepted, but are now in print. For these articles the author modified Doren’s citations to reflect this change in status.