FEB-21-2008 16:12 From:PSEPC CORRECT 6139908295 To:206 587 5088 1 2 3 4 5 6 7 8 9 STATE OF ?WASHINGTON FRANKLIN COUNTY SUP}:RIOR COURT In re the Detention of: WILLIAM DAVENPORT Res ondent_ NO. 99-2-50349-2 DECLARATION OF DR. R. KARL HANSON 10 11 12 13 14 15 1. I am DR. R KARL HANSON. I am a registered psychologist in Ontario, Canada. I am a Senior Research Ofticer with Public Safdy Canada.o and Adjunct Professor in the Department of Psychology at Carleton University, Ottawa, Canada. 1 am a fellow of the Canadian Psychological Association and received the 2002 Sign.ificant Achievement Award from the Association for the Treatment of Sexual Abusers. My CV is attached. 2. I am the developer of Static-99, an actuarial risk tool widely used in the evaluation 1.6 17 18 19 of recidivism risk of adult male sexual offenders. I have carefully reviewed the evidence on the validity of risk assessments for sexual offendeTS (Hanson, RK., & Morton-Bourgon. K.E. [2007] The accuracy of recidivism risk assessnumls for sexual offenders: A mer.a-ana~ysis_ / 20 21 Corrections User Report No. 2007-01 Ottawa: Public Safety Canada). As well) 1 have 22 23 24 examined the relationship between age at release and recidivism among sexual oilcllders (Hanson.o R.K. (2006). Docs Static-99 predict recidivism among older sexual offenders?"' Sexual Abuse: A. Journal of Research and Treatment, 18, 343-355. Hanson, R. K. (2002). I A "meta-analysis" is, simply put, an analysis ofal1alyses_ Tn this meta-.malysis, we reviewed 577 findings from 79 different samples_ 25 26 DECLARATION OF KARL HANSON AlTORNliY (jI.?NI.!AAL'S 01'1'1(;1:: CrilTlinill Jusli(;c Division gOO Fourih Iwo:mll:, Sllil~ Seanle, WA 98104 (206) 464-6430 20()() FEB-21-2008 16:12 From:PSEPC CORRECT 6139908295 To:206 587 5088 1 "Recidivism and age: .Follow-up data on 4,673 sexual offenders." Journal oj'In.terpersonaL Violence, 17, 1046-1062. 2 3 4 3. I am very familiar with Dr. Wollert's writings on the evaluation of rccidivi~m risk for sexual offendt.'Ts. I have previously provlded el'itique~ of his testimony in civil commitment cases- T have had occasion to review Dr. Wollerl's testimony an.d affidavits in specific civil commitment C~lses. On several occas,ions, I have written responses indicating that he is 5 6 7 8 9 mi:mpplying my research or misstating the data4. T have reviewed Dr. Wollert':; declarations (dated November 23, 2007 and January 16, 2008) submitted in this case, pages 76 to 96 in the Motion to Set Aside Judgment by R.J Thompson dated approximately January 15, 2008, as well as various manuscripts authored by Dr. Wollert that purport to provide the scienti1ic basis for the assertions contained in his Declaration. Specifically, T have reviewed the following: "Poor Diagnostic Reliability, The Null-Bayes Logic Model, And Their .1mpJications For Sexually Violent Predator Evaluations," [Draft]; "Validation of a Bayesian Method For Assessing Sexual Recidivism Risk" (presented at AP A Conf. August 2007); and "Low Base Rates Limit Expert Certainty When Current Actuarials Are Used to Identify Sexually Violent Predators," Psychology, Public Policy and Law 2006, Vol. 12. No. a, 56?85. 10 11 12 13 14 15 16 17 18 19 s- DT. Wollert's testimony contains some statements that are true by definitlon (e.g., Bayes' theorem), some statements that are consistent with widely held professional opinion, some statements that are debatable, and some statements that arc demonstrably false. These false statements include both misrepresentations of facts as well as misrepresentations of statistics and research methods. Furthennore, Dr, WoHert does not aid the audience in judging the weight to be given his various assertions because he fails to distinguish between peer supported professional opin.ion and his own speculations_ 20 21 22 23 24 25 26 DECLARATION OF KARL HANSON 2 ATIORNEY GENERAL'~ OFFICE Criminal Justice Division 800 FOIll'th ~ S\,;lo: 2000 Seattle. WA 98104 "v,=",,,.. (206) 464?64:10 FEB-21-2008 16:12 From:PSEPC CORRECT 6139908295 To:206 587 5088 I 2 3 6. Dr. Wollert's most serious error is that the method he proposed to mathematically adjust actuarial risk predictions [Ising factors extemal to the actuarial scheme is incorrect. Specifically, he argues that Bayes' theorem can be used to adjust Static-99 rigk predictions based on the offender'S age at the time of release. Such adjustments only make sense when age is lUlfelated to Static-99 scores. When the offender'S age is already correlated with Static99-a5 it in fact is~- adjustments such as those made by Dr. Wollert can eithcr overestimate or undcrcstllnate the recidivism rates. The adjustments made by Dr. Wollcrt in tIus case have the effect of underestimating recidivism rates. 7. The available research has demonstrated that age is related to Static-99 scores: specifi,cally, the Static-99 scores of older offenders are lower than tile scores of younger offenders. TItis finding does not mean that older offcndcrs-merely because they are olderare less likely to reoffcnd than younger offenders with comparable Static-99 scores. Rathcr, it indicates that the older offenders were different from the YOlUlger offenders even when they were younger, that is) they are not just the younger offenders grown up. 8. "Base rate" is defined as the frequency with which a given event occurs within a given popu]ation_ In the context of sexual offender risk assessment, the base rate is simply lhe probability that the typical sex offender will reoffend, or the recidivism. ra.te of the average member a class or group. Dr. Wollert, based on data I presented, llses age to estimate new (lower) recidivism rates for older sexual offenders who have previously been assessed as high risk to reoffend. Dr. WoUert's met110d of making these adjustments uses Bayes' theorem" Bayes' theorem includes separate measures of thc a) recidivism base rate and b) the discriminative properties of the scale. This theorem is true only under certain conditions, and it is not true in the data presented by Dr. Wollert. The theorem is false when the variables used to adjust the recidivism base rates are correlated with scores on the assessment measure. Specifically, Dr. Wollert's application of Bayes' theorem is false because age at release is 4 5 6 7 8 9 10 tt 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 DECLARATTON OF KARL HANSON 3 ATTORNEY GENERAL'S OFFICE Criminal JlI~1ICI:' lJivlsion 800 FOIRI04 (206) 464-6430 FEB-21-2008 16:12 From:PSEPC CORRECT 6139908295 To:206 587 5088 '1 2006, at 353. Dr. Wolled's appJication of an "age-sensitive version of Bayes's Theorem" to that data docs not constitute such an iodepE:ndcnt replication. Tn esscnce, Dr. Wollert is 2 3 4 proposing a new actuarial risk tool without doing the necessary research to establish its predictive validity. More troubling is that he appears to be relying on my research to suggest that Tagrce with his analysis, when in fact :r disagree with it. 12. 5 6 The effect of age on sexual recidivism risk has not been resolved in the 7 scientific community. Some studies find lower recidivism rates for older offenders, whereas other stLldics do not. Tn his testimony, Dr. Wollert reports only those fIndil1g~ in which older sexual offenders were lower risk for recidivism. Curiously, he reports on thc same samples rcpeatcdlyas though (hey were independent replications (e.g_, Hanson 2005 and 2006 are the same subjects; the Warkworth data set, analyzed by Drs. Barbaree and Langton, is mentioned sevcral tiJ.ues). He does not mcntion the studies in which age at release had no incremental validity onee age at index offence was considered (Rice & HalTis, 2007), nOT 8 9 10 11 12 13 14 the studies of offenders very similar to those fo'Und in civil commitment settings (i.e., high risk sexual offenders commilLed for treatment) in which age at release was, unrelated to recidivism (Thornton & Knight, 2007). 15 16 17 18 13. Even if advanced age has some relationship to reduced recidivism (a position I believe to be true), it is merely one factor among many that could illfiuence recidivism risk beyond that measured by Static-99. Research has demonstrated a number of factors external to Static~99 19 20 21 22 23 24 add incremental validity, inc1udhlg measures of sexual deviance, lifestyle impuJsivity, and interpersonal deficits (Beech et at, 2002; Hanson et at, 2007; Thornton, 2002). In many data sets, the association between age and recidivism arc relatively small compared to the effects of other risk factors (Thomton & Knight, 2007). 14. In summary, Tbelieve that it is the role of evaluators to estimate recidivism rates 25 26 based on the average recidivism rates for members of the class that the offender most closeJy DECLARATION OF KARL HANSON 5 A11'()RNEY (JEN~RAL'S OFFICE Crirnin(,1 Jusli(;~ Division ROO Fourth Avenue, Suite 1()()() S~;IItI~. IN" 98104 (20t:;) 464"64JO FEB-21-2008 16:13 From:PSEPC CORRECT 6139908295 To:206 587 5088 1 resembles. The estimates of the recidivism rates ShOllld be those observed in actual recjdiv1~'TI'l 2 3 4 5 studies, and not those generated by arithmetic manipulations based on incorrect assumptions, such as those propos,ed by Dr, Wollen. .For purposes of Statie-99, those estimates should be detennined by reference to Appendix Six 1n the Static-99 scoring manual (Harris et aI., 2003, Static-99 coding rules: Revised 2003. Ottawa: Department of the Solicitor General of Canada). The table appearing in Hanson 2006 is not in.tended to be a "new" or "updated" Static-99 table as Dr. Wollert suggests when he refers to "Hanson's (2006) updated Static-99 table (t=.97)." Wollert 2007 at 2. 6 7 8 9 10 I declare under penalty ofpetjury that the foregoing is true and accurate to the best of my knowledge. DATED this II 12 13 14 15 -:;2./ ~day of Februury, 2008, at Ottawa, Canada. ~/I:;- 16 17 18 19 20 21 22 23 24 25 26 DECLARA nON OF KARL HANSON 6 J\TWRNEY GENERAI.'~ OFFICE Criminal JUSUl:ro:: DIviSion 800 FOUI1h Avenue, SuiL~ Se.mlro::. w.... 911 J04 2000 . (206) 4()4.6430 The following example demonstrates how Bayes' theorem can produce meaningless numbers when the criteria used to create the new base rate is correlated with the test results. All the numbers were obtained by re-arranging Table 5 from Hanson & Thornton (1999). In the Static-99 development samples, there were 129 offenders with a score of six or more. Ofthese 129 offenders, 39% (n = 50) reoffended sexually within 5 years and 79 did not. This 39% recidivism rate is the best estimate ofProb (recidJ score of6+). We will now show how Bayes's theorem produces a completely different (false) number when the base rate is changed in a manner that is correlated with Static-99 scores. In the 1086 offenders in the Static-99 development samples, 18% (n = 195) sexually reoffended after 5 years, and 82% did not. Of the 195 offenders who reoffended, 50 received a score of6+, yielding a sensitivity, Prob (6+Jrecid), of 501195 = 25.6%. Of the 891 offenders who did not reoffend, 79 received a score of6+, yielding Prob (6+Jnon-recid) of8.9%. When the above numbers of applied to Bayes' theorem: Prob (recidJ6+) = (.256)(.18)/[(.256)(.18) + .089(.82)] = .39 The probability .39 is as it should be; the sensitivity and specificity are correct and match to that sample. Now consider creating a subsample with a higher base rate. This higher base rate was created by restricting the sample to only offenders with a score of 4 or more. Of the 519 offenders with a score of 4 or more, there is an observed recidivism rates of 25.4%. When this new base rate is used in Bayes' theorem with the original sensitivity and specificity, the results are incorrect because the sensitivity and specificity has changed in reality, but not in the formula. Prob (recidJ6+) = (.256)(.254)/[(.256)(.254) + .089(.746)] = .49. By restricting the subsample to only high risk offenders, Bayes' theorem implies that the five year sexual recidivism rate for the 6+ group increases from .39 to .49, which is absurd. In these calculations, the observed recidivism rate for the 6+ group has remained unchanged at 39%. Only the reference group has changed. By carefully selecting reference groups, it is possible to change the distribution of scores and thereby increase or decrease Bayesian probabilities as desired. Such distortions are only observed when the overall sensitivity and specificity are used. When the sensitive and specificity for the subsample are used, Bayes' theorem provides correct predictions. To use Bayes' theorem, it is necessary to establish that the factors used to change the base rate are either unrelated to the relative recidivism risk of individuals, or are uncorre1ated with the actuarial measure. An appropriate use of Bayes' theorem, for example, would be to adjust 5 year predictions to 10 year predictions based on increased recidivism rates at 10 years. It is inappropriate to use Bayesian adjustments when the new base rate is created by using a factor that is correlated with the recidivism risk measure, such as age.