Você está na página 1de 22

The Dynamic Prediction of Criminal Recidivism: A Three-Wave Prospective Study Author(s): Shelley L. Brown, Michelle D. St.

Amand and Edward Zamble Source: Law and Human Behavior, Vol. 33, No. 1 (Feb., 2009), pp. 25-45 Published by: Springer Stable URL: http://www.jstor.org/stable/30219008 . Accessed: 20/05/2013 12:30
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Springer is collaborating with JSTOR to digitize, preserve and extend access to Law and Human Behavior.

http://www.jstor.org

This content downloaded from 111.68.111.198 on Mon, 20 May 2013 12:30:11 PM All use subject to JSTOR Terms and Conditions

Law HumBehav(2009) 33:25-45 DOI 10.1007/s 10979-008-9139-7

ORIGINAL ARTICLE

Prediction of Criminal The Dynamic Recidivism: A Three-wave Prospective Study


ShelleyL. Brown MichelleD. St. Amand Edward Zamble

Published online: 16 July 2008 American 41 oftheAmerican Association 2008 Psychology-Law Society/Division Psychological C

Abstract A three-wave, prospective panel design was used to assess theextent to whichstatic and dynamic risk factors couldpredict criminal recidivism in a sampleof 136 adultmale offenders releasedfrom Canadianfederal prisons. Static measureswere assessed only once, priorto release while dynamic measureswere assessed on three 1 month, and 3 months separateoccasions: pre-release, Recidivism was coded an averageof post-release. during 10.2-month (SD = 19.2). A seriesofCox follow-up period covarisurvival regression analyseswithtime-dependent ates and Receiver Characteristic (ROC) analyses Operator wereconducted the to assess predictive validity. Although combined static and time-dependent model dynamic (AUC = .89, CI = .81-.93) significantly (p < .01) outthepurestaticmodel (AUC = .81, CI = .73performed intervals did overlapto some extent. .87) the confidence for risk andmanagement Implications dynamic assessment are discussed.

recidivism risk Keywords Criminal . Dynamic assessment prediction . Multi-wave risk to recidivate Assessingan offender's upon release from functions of a prisonis one of the mostimportant correctional is theability organization. Equallyimportant, to accurately detectchangesin an offender's riskto recidivate suchthat an impeding failure can be prevented and To date,riskassessment has emphaeffectively managed. sized two types of risk factors:static and dynamic. such as criminal are staticriskfactors Although history considered constant and unchanging, thusnotamenable to in risk factors such as substance treatment,theory dynamic can change, abuseandcriminal attitudes and,consequently are amenableto treatment (Andrews& Bonta, 2006). scholars riskassessment now conNotably, contemporary risk factors a continuum from ceptualize along ranging fixation to extreme malleaoffence) complete (age at first maritalargument) (Hanson & bility(perceivedthreats, Harris, 2000; Quinsey, Jones, Book, & Barr, 2006; Wormith, 2005; Zamble& Quinsey, 1997). The fieldof forensic risk assessment has burgeoned Monahan how difsince (1981) highlighted considerably and psychologists to make ficultit is for psychiatrists of violent behavior. The last27 years accurate predictions of numerous assessment has witnessed the proliferation a of outto antisocial protocols designed predict variety in comes includinggeneralcriminalrecidivism prison populations(e.g., Level of Service/CaseManagement Bonta,& Wormith, 2004; (LS/CMI); Andrews, Inventory Statistical Information on RecidivismScale (SIR); Nufin forensicpsychiatric field, 1982), violent recidivism settings (e.g., Violence Risk AppraisalGuide (VRAG); Harris,Rice, & Quinsey, 1993), spousal assault (e.g.,
Springer

S. L. Brown M. D. St. Amand E. Zamble ofPsychology, ON, Queen'sUniversity, Kingston, Department Canada S. L. Brown Correctional Service of Canada, Ottawa, ON, Canada Present Address: S. L. Brown (0) ofPsychology, Carleton 1125Colonel University, Department Ottawa, ON, CanadaK1S 5B6 By Drive, @ carleton.ca e-mail: shelley_brown Present Address: M. D. St. Amand EastCoastForensic Dartmouth, NS, Psychiatric Hospital, Canada

This content downloaded from 111.68.111.198 on Mon, 20 May 2013 12:30:11 PM All use subject to JSTOR Terms and Conditions

26

Law HumBehav(2009) 33:25-45

increasethe probability of detecting (SARA) Guide; Kropp, in theory, Spousal AssaultRisk Assessment change, for dynamicvariablesthatare expectedto Hart,Webster,& Eaves, 1999), and sexual recidivism particularly (STATIC 99; Hanson & Thorton, 1999). Concomitantly, changerapidly. stateof the art statistical such as Receiver With the ofresearch andcolleagues approaches, exception byQuinsey Characteristic et hasbeen (ROC) analysis (Swets, 1986; (i.e.,Quinsey al.,2004,2006),sample Operator censoring that a serious limitation all ofthe Swets, Dawes, & Monahan,2000) have illustrated methodological characterizing riskassessment boast impressive multi-wave, studies.In fact,several of the contemporary protocols prospective ratesin therangeof 70-80% (Bonta, studieshave failedto indicatehow manysubjectswere predictive accuracy Harman, Hann, & Cormier, 1996; Gendreau, Goggin,& actuallyexcludedfromthe change analysisdue to the occurrence ofrecidivism tore-assessment. Evenmore Smith,2002; Harriset al., 2003; Kropp & Hart,2000). prior has advanced sinceMonahan's problematic is thelack of consensus howbestto Thus,thefield considerably regarding data.Although someexperts monograph. analyze change (e.g.,Andrews, To date,a considerable advocate the use of of research has accumure-test others Wormith, scores, Zamble) body lateddemonstrating such (e.g., Bonta,Gendreau, theability ofdynamic risk factors the Hanson,Motiuk)recommend as criminal criminal and creationof an actual change variablederivedfromthe attitudes, associates, employment, substanceabuse to predict adult criminalrecidivism existing data (e.g., -1 = change forthe worst;0 = no for the Little,& Goggin, change (Andrews& Bonta, 2006; Gendreau, (alwaysgoodoralwaysbad); or 1 = change the have vast of these studies and have for 1996). However, better). However, majority Quinsey colleagues opted an relied exclusivelyon 'single-wave' research designs. entirely different on traditional research approach relying research risk methods suchas within-subject measures Briefly, single-wave designsassess dynamic repeated designs factorsonly once, for example, priorto release. The usedin conjunction with that condesigns between-subject assessment or not trol results are thenused to predict whether for static risk(e.g.,ANCOVA's). criminalrecidivism occurredat some point duringthe theareaofdynamic riskassessment merits Accordingly, the ideal studywould follow-upperiod. Thus, froma statistical perspective, further exploration. Ultimately, research single-waveresearch designs have essentiallytreated employa multi-wave, longitudinal prospective factors as ifthey werestatic-immutable. threewaves) coupledwithappropriate dynamic Dynamic design(minimum risk prediction in studieshave rarelybeen multi-wave statistical thatnotonlyaddresssamplecensortechniques thatis, theyhave rarely or not ing but also readilyincorporate examined whether about risk information nature, recidivism can be successfully from thesystem- factors thatfluctuate over time.Cox regression survival predicted aticassessment andre-assessment ofdynamic covariates is one such information, analysis with time-dependent a prerequisite fortesting whether risk approach(Cox, 1972). Survivalanalysisin general, or not a dynamic is a factoris trulycriminogenic statistical thatestimates how long it takes to (Andrews& Bonta, 2006; technique et al., 1997). reachsomeevent(e.g.,criminal as well as the Douglas & Skeem,2005; Kraemer recidivism) While multi-wave studiesdo exist (see Andrews& rateof occurrence of thatevent.Thus,unliketraditional & Wormith, Robinson, 1984; Andrews 1984; Bonta,1996; statistical methods that of wheonlyaddressthequestion Hanson & Harris,1998; Motiuk,1991, 1999; Quinsey, ther ornotan individual willcommit a a newcrime during survival valuable also provides Book, & Skilling,2004; Quinsey,Coleman, Jones,& interval, specified analysis aboutwhena recidivistic Altrows, 1997; Quinseyet al., 2006) mosthave suffered information eventis mostlikely from or statistical shortcom- to occur. Additionally, survivalanalysishas the unique theoretical, methodological, ofbeingable to control fora variable ings. For example,apartfroma few notableexceptions advantage follow-up censoreddata (Hanson & Harris,1998; Hanson et al., 2007; Quinsey period and it also readily incorporates et al., 1997,2004,2006) thedynamic has not (Chung, risk literature & Witte, Schmidt, 1991). examined theroleof rapidly suchas riskfactors The current has severalobjectives. it will First, changing study moodand situational factors that whether or not the re-assessment of prospectriggers, playa prominent determine role in the coping-relapse model of criminal recidivism tively-rated riskcan improve dynamic accuracy predictive & half of the over and above static risk using statistically (Zamble Quinsey, 1997). Moreover, existing rigorous multi-wave studies haveusedrelatively smallsamplesizes approaches, Cox regression survival with namely, analysis that time-dependent covariates and ROC analysis.Second, it (e.g., 50-60) or have reliedon assessment protocols may not be sensitive enoughto detectchangein rapidly will use dynamicrisk measuresthat are theoretically riskfactors. As well,with theexception derived and sensitive to change.Third, it will investigate occurring dynamic ofresearch Motiuk and and whether or not risk factors by (1999) Quinsey colleagues, dynamic actually changeovera assessment and re-assessment schedule. existing studies have employed two-wave designs. systematic Lastly, Research that three or morewaves should the relative predictivepower of some of the most designs employ Springer

This content downloaded from 111.68.111.198 on Mon, 20 May 2013 12:30:11 PM All use subject to JSTOR Terms and Conditions

Law HumBehav(2009) 33:25-45

27

It is hypothesized that: the Statistical promisingstatic measures,specifically, on Recidivism Information Scale (SIR, Nuffield, 1982),the model will include the (1) The strongest prediction Checklist HareRevisedPsychopathy (PCL-R,Hare,2003), measurementof both static and prospectively and theChildhood Adolescent Taxon Scale (Harris, Rice, assessedand re-assessed risk. dynamic & Quinsey, 1994; Quinsey, Harris, Rice,& Cormier, 1998) The re-assessment of prospectively-rated (2) dynamic will also be examined. riskmeasured whiletheoffender is under community The coping-relapsemodel of criminal recidivism will improveprediction accuracyover supervision (Zamble & Quinsey,1997) will providethe theoretical and above thestrongest static modelestablished preforthestudy. The modelpositsthattherecidframework release. ivismprocessbeginswitha precipitating environmental of prospectively-rated (3) The re-assessment dynamic orin thelanguage ofHansonet al., (2007) an acute trigger riskmeasured whiletheoffender is under community Thisevent can be highly variable dynamic trigger. ranging will improveprediction supervision accuracyover from chronic lifestressors suchas marital discord, job loss, and above the strongest model established dynamic or financial stress to relatively mundane dailyhasslessuch pre-release. to deal withcrowded as having publictransportation sys(4) All of the dynamicmeasureswill change in the has occurred, the tems.Once the acute dynamic trigger who manage direction. Thatis, individuals expected will invokebothan acute cognitive and emoindividual to remain in thecommunity and not be returned to Individuals whoperceive tionalappraisalof thesituation. willevidence a noticeable declinein dynamic custody thesituation as threatening or problematic typically expeareas. problem riencenegativeemotions (e.g., hostility, anger,fear),an elevatedlevel of perceived (e.g., "I have no global stress control over my life") and, lastly, some awareness This METHOD theseverity oftheenvironmental regarding trigger(s). in turn results in an attempt to deal withthesituation, but most areineffective at coping with the Participants offenders giventhat not be What follows is a situation it will remedied. original about to be and thirty six male offenders emotions, worsening cycle of negative maladaptive cog- One hundred mediumor maximum-security andeventually theresumption ofcriminal conduct. releasedfrom minimum-, nitions, institutions locatedin Ontario, Canada participated The modelfurther whether or notan individual federal positsthat an in the ifthey will initially acute environmental selected to Offenders were experience trigger(s) study.' participate or problematic is consented to take part (consentrate: 56.4%) and were or perceivea situation as threatening mediatedthrough two subsetsof factors: to be releasedon either enduring per- scheduled discretionary (parole)or 45 days release(statutory sonalityor behavioralpatterns/static factorsand stable non-discretionary release)within of theinitial In Canada,offenders mechanisms. assessment. dynamic response pre-release Static factors arerelatively stable andinclude factors such sentenced to 2 yearsor moreare eligibleforday and full oftheNational ParoleBoard)after as criminal lifetraits and enduring (e.g., tempera- parole(at thediscretion history areindicative and one-third of their emotional Thesefactors of serving one-sixth sentence, ment, respecreactivity). an individual's toreact toandinterpret situations tively.In contrast, the law mandatesthatall offenders propensity can be operation- serving 2 years ormore after two-thirds in a maladaptive manner. This construct be released serving This excludes alized using measures such as Hare's Psychopathy of theirsentence(i.e., statutory release). Checklist-Revised indeterminant sentences (PCL-R; Hare, 2003) or the Statistical offenders serving (e.g., lifers)or on Recidivism scale (SIR, Nuffield, Information in prison offenders who are detained untilwarrant 1982). expiry In contrast, available responsemechanisms are more due to theirperceivedpropensity to re-offend violently offenders releasedon paroleare in nature, albeitnot as labile as environmental uponrelease.In general, dynamic lower risk best as less criminal are (e.g., history, completed programs) They triggers. conceptualized slow-changing releasecounterparts. The thantheir that behavior Lastly,offenders statutory patterns mayserveas treatment targets. ifthey in thestudy understood mechanism subsetincludes vari- wereonlyincluded stabledynamic English, response noreligible for ables such as coping ability,substanceabuse, criminal wereneither deportation, actively psychotic forat least6 months criminal associates,and social support. attitudes, Lastly, and wouldnotreachwarrant expiry and the theory proposesthatthe process is continuous interactive such that each response generatesa new 1 In sentenced to 2 years or more fallunder the Canada,offenders in another situ- purview resulting precipitating sequenceof events of thefederal whereas provincial governments government another and eventually, another whoreceive sentences less than 2 years. offenders ation, appraisal, response. manage
Springer

This content downloaded from 111.68.111.198 on Mon, 20 May 2013 12:30:11 PM All use subject to JSTOR Terms and Conditions

28

Law HumBehav(2009) 33:25-45

criterion was necessary from thedateof release.This final to ensurethatparticipants would be relatively easy to once released,giventhatthey contactin the community to a paroleofficer untilthey to report wouldbe required sentence reached theend of their (i.e., warrant expiry). and criminal For ethicalreasons, detaileddemographic could notbe collectedfortheparticiinformation history who declined to take part (43.6% of the original pants it was known thatindiHowever, subjectpool; 129/296). to agreetotakepart in thestudy if vidualsweremore likely thanmin(85.7%) rather theywerehousedin maximumimuminstitutions (54.0%), (53.7%) or mediumsecurity was notstatistically thedifference (x2 although significant, (2, N = 296)= 5.28, p < .10). Individualshoused in institutions have considminimumand mediumsecurity and actual freetime more freedom of movement erably maximumthantheir security counterparts. Consequently, it is plausible that they were less likely to volunteer to leave theircells because theyhad moreopportunities 61% of the outside of the researchproject.Similarly, decliners(78/129) statedthat theywere not interested becausethey did notexpectto have sufficient timeto take 10.9% of the after release.An additional partin thestudy declinersrefusedto even discuss the studywith the researchers in person,while another 5.4% indicated that releasetoeven weretoo stressed abouttheir they upcoming thinkabout talkingto someoneabout it. An additional consent but withdrew 4.7% initially agreedto participate before theinterviewers had an opportunity to conduct the interview. pre-release to notethat an additional 31 participants It is important criteria and had beeninterviewed mettheinclusion during the pre-release phase of the studyhowevertheywere excludedfromfurther analysisforvariousreasons(e.g., outsideof study parameters, changedreleasedestinations committedsuicide, unexpectedly deportedto another ordetained inprison, detained country temporarily pending revocation and chi-square hearing).A series of t-tests revealed that the31 excluded did not participants analyses in sentence differ from theremaining terms of sample age, marital releasetype, status, level, length, ethnicity, security or risklevel as measured theHare Psychopathy by either Checklist-Revised (PCL-R; Hare, 2003) or the Statistical on RecidivismScale-Revision1 (SIR-R1; Information were Nuffield, 1982). Lastly,19 of these31 individuals excludedbecause their releaseoutcome could notbe reliwhen the recidivism information was ably determined coded.2

The average age ofthesamplewas 33 whiletheaverage 4 years.Participants sentence lengthwas approximately were incarcerated for a varietyof criminaloffences murder, assault,sexual assault,robbery, including drug and property-related While54.4% of the crimes. offences, sample was releasedon parole,45.6% was releasedon release.Approximately two-thirds of thesample statutory one-third was either was Caucasian,whiletheremaining Black (15.4%), Asian (4.4%), Aboriginal (4.4%), or classified as Other ofthesamplewas single (7.4%). Two-thirds at thetimeof release. Measures is no single andvalidated measure there reliable Currently, of thecoping-relapse model.As a result, a combination of was used to and measures pre-existing newlydeveloped assesseach ofthefour ofthemodel:(1) static components (2) acute dynamic factors/enduring personality; triggers; (3) acute cognitive/emotional appraisals;and (4) stable mechanisms. It shouldbe notedthat the dynamic response distinction betweenacute and stable factors was made on theoretical purely grounds. StaticFactors/Enduring Personality The following fivestaticmeasures were assessed: age at time ofpre-release theStatistical Information on interview, 1 (SIR-R1;Nuffield, Scale-Revision Recidivism 1982)3;the Hare RevisedPsychopathy Checklist (PCL-R; Hare, 1991, Adolescent TaxonScale: Self-Report 2003); theChildhood Version(CATS-SR; Harriset al., 1994; Quinsey, Harris, The SIRmisconducts. Rice, & Cormier, 2006) and prison riskassessment toolcomprised R1 is an actuarial primarily +27 of static, criminal Scoresrange from variables. history to -30 withmorenegative scoresreflecting greater probabilities of general recidivism.The SIR-R1 and its excellent inter-rater (theSIR) have evidenced predecessor & Goldstone, (r = .96; Wormith 1984),moderreliability ateinternal (Chronbach's consistency alpha = .75; Nafekh & Motiuk, 2002) andmoderate (AUC's predictive validity between .71 and .74; Bontaet al., 1996; Serin& ranging Brown, 2001, UnpublishedManuscript).The internal

Footnote 2 continued torevoke these individuals orcanceltheir andreturn them suspension to thecommunity. Theother twoindividuals hadgoneunlawfully at information was coded. before therecidivism large(UAL) shortly Given that someUAL casesarenotnecessarily revoked, particularly 2 TheCorrectional are onlyUAL fora short of time, it was decided to period Service ofCanadahadsuspended theconditional if they exclude these cases from the sample. releaseof 17 ofthese individuals weretemporarily who,as a result, Service ofCanadamodified theSIR scale detained in either a provincial institution. Recidivism 3 In 1996theCorrectional jail or federal wascodedonSeptember 2001.Atthis the National Parole slightly to reflect in theoffender The revised 27th, time, changes population. theSIR-R1. Boardhadnotyet rendered a final decision interms ofwhether ornot scalewas named Springer

This content downloaded from 111.68.111.198 on Mon, 20 May 2013 12:30:11 PM All use subject to JSTOR Terms and Conditions

Law HumBehav(2009) 33:25-45

29

(Chronbach's consistency alpha = .76) and theinter-rater butwas recodedas follows:0, 1, 2, 3, or 4 or moremisconducts due to extreme As Table 1 illustrates skewness. correlation coefficient (intra-class (ICC) = .98) reliability of the SIR-R1 was also acceptablein the current was notan issue as noneof thePearsonr study. multicollinearity correlation coefficients exceeded+-.59. Inter-rater estimates werecalculated reliability using12% of thesample(thefirst author used official fileinformation of theinterviews to assess inter-rater AcuteDynamic and tape recordings Triggers forall of thevariables used in thestudy). reliability used measureof The following from The PCL-R is the most commonly sevensub-scales theProblem Survey 0 from 40 Checklist Brown & were used to Total scores to with Zamble, 1998a) (PSC; psychopathy. range higher ofpsychopathy. The PCLassess acutedynamic difficulscoresreflecting (1) marital/family higher degrees triggers: R is scoredon thebasis of a semi-structured interview and ties (sevenitemsdesigned to tape thequalityof intimate collateralfile review.The PCL-R has strong inter-rater partner and familysupport), difficulties (2) employment = = ICC N and items to measure .93, 1, (five 246) strong designed instability, reliability (averaged employment internalconsistency absenceofjob commitment/intrinsic barriers, (alpha = .84, N = 1, 246) (Hare, employment difficulties reviewshave demon- job motivation), 2003). Additionally, (3) accommodations (four meta-analytic thatthe PCL-R predicts bothgeneraland violent itemsmeasuring dissatisfacstrated neighborhood, criminogenic with meta-analytic effectsizes (r's) ranging tionwith and accommodation recidivism accommodation, instability), from .23 to .37 (Gendreau et al., 2002; Hemphill, Hare & financial difficulties items financial (4) (three measuring and poorfinancial skills),(5) pooruse management Wong,1998; Salekin,Rogers,& Sewell, 1996). The reli- stress unstructured leisuretime, of thePCL-R in thecurrent was also strong of time(fouritemsmeasuring ability study absenceof hobbies),(6) interpersonal conflict (fouritems (ICC = .82; alpha = .84). amountof conflict with peers, co-workers, of eight,childhoodand measuring The CATS-SR is comprised of antisocialbehavior(e.g., arrests family), and (7) physical and emotional health indicators adolescent (two items to age 16, schoolsuspension, alcoholabuse) (Harris measuring global mentalhealthand/or physicalhealth prior 0 to 16 withhigher problems withpsychiatric from failure to comply et al., 1994). Total scoresrangefrom ranging is rated scoresreflecting levels of childhood and adolescent medication to chronic back Each item on the pain). higher basis of a semi-structured interview and a collateral file Previous research has demonstrated that the antisociality. review. Each individual item was scored as follows: no of violentrecidivism CATS-SR is predictive (r = .18; it is reliable(alpha = .84; Folsom problemevident(score 0), problemsomewhatevident 2000) and that Nugent, evident & Atkinson, thenon-self ver- (score 1), or problem 2007). Additionally, (score2). report clearly the was PSC sion of theCATS has beenused in place of thePCL-R in Although developed based on prior theVRAG without theoverallpredictive accu- research (Zamble& Quinsey, 1997) thiswas thefirst study degrading touse thePSC. Consequently, thesubscales wereexamined et al., 2006). The racyoftheVRAG (AUC = .75; Quinsey in thecurrent extensively fortheir and adjusted CATS-SR demonstrated psychometric properties accepted reliability = = to For .78; (ICC prior primary analysis. study alpha .84). accordingly example,subThe finalstaticmeasure was defined as thenumber of scale itemswere deletedif theydemonstrated: (1) multiconducttoward collinearity (r's > .85); (2) poorinternal (e.g., consistency prison misconducts (e.g., disrespectful afteritem exclusion); (3) on staff/inmates, substance use,assaults staff, participation alpha improved considerably thelast12 months poorinter-rater inriots, incurred (ICC or kappaless than.65); or reliability hostage takings) during extreme skewness or kurtosis. As a result, were ofincarceration to the interview date. This (4) changes pre-release prior fromCorrectional Service made to fiveof theoriginal sevensubscales.Specifically, variable was coded directly the interpersonal conflict subscale was droppedentirely of Canada's OffenderManagement System (OMS) due to poorinter-rater thevariable from 0 to 17 (ICC = .96). Originally, ranged reliability (e.g.,kappaor ICC < .65
matrix for static Table 1 Correlation predictors Variable Age at release Information on Recidivism Statistical Scale-R1(SIR-R1) TaxonScale (CAT-SR) Adolescent Childhood Checklist-Revised (PCL-R) Psychopathy Prison misconducts
Note: ap <.05, bp <.001

Age
.33b

SIR-R1

CATS-SR

PCL-R

-.49b
-.21a

-.59b
-.57b .58b

-.31b

-.33b

.41b

.36b

Springer

This content downloaded from 111.68.111.198 on Mon, 20 May 2013 12:30:11 PM All use subject to JSTOR Terms and Conditions

30

Law Hum Behav (2009) 33:25-45

at anyone Wave). Six of theoriginal sevenmarital/family.86; Cohenet al., 1983; Fliegeet al., 2005). The alphasin itemsweredropped. mea- thecurrent werealso acceptable at all three wavesof Thus,thefinalmarital study support sureconsisted of one item:pooror absent marital .81-.82). support data collection: theemploya modified version of thePositive Affect (totalscorerange:0-1). Onlyone itemfrom Lastly, Negawith tive Affect ment scale was dropped dissatisfied Schedule & Tellegen, Clark, (i.e., (PANAS; Watson, job) leaving of four items(total 1988) was usedto assessbothnegative thefinal scale comprised andpositive affect. employment scorerange:0-8). Lastly, one item(i.e., no fixed to takeprevious criticisms address) The scale was modified (e.g., was deleted fromthe accommodation scale given that Nemanick& Munz, 1994) as well as past correctionsno offenders wereclassified as 'no fixed address' specific research intoaccount(Zamble & Quinsey, virtually 1997). scale was comprised of30 items andtwosubduringthe community phase of the study.Thus, the Theresultant of three scales: Negative AffectSchedule and Positive Affect resultant accommodation scale was comprised items(totalscorerange:0-6). No changesweremade to Schedule. The NegativeAffectSchedule contained16 theoriginal four itemleisurescale (totalscorerange:0-8) whilethe nervous) adjectives (e.g.,bored, anger, depressed, orthethree itemfinance scale (totalscorerange: The affect schedule contained 14 0-6). positive adjectives (e.g., final scales evidenced acceptable internalconsistency excited, at ease, peaceful).Total scorescan range proud, between 14 and 70 forthePositiveAffect with Schedule, (ICC (alpha ranges: .20-.84) and inter-rater reliability at each wave. Please note that the scores levels of .71-1.00) ranges: study higher reflecting higher reported positive low alpha values (e.g., .20) were associated with the emotions, while scoresfortheNegativeAffect Schedule in thecommunity can range finances andleisure sub-scales measured between 16 and 80, with scoresreflecting higher low internal consis- higher levels of reported emotions. the (waves 2 and 3). Despiteevidencing negative Although thefinancial andleisure measures wereretained due revised PANAS had notbeen used prior to thestudy, the tency, to their in demonstrated research. PANAS has demonstrated importance past original strong reliability con(alpha's = .86-.90; Watsonet al., 1988) and strong AcuteCognitive/Emotional r = .67) and discriminant Appraisal vergent (validitycoefficient: r= -.31) (Crawford& validity (validitycoefficient: Three measures wereusedtoassessthecognitive/emotionalHenry, from .84 2004). In thecurrent study, alpharanged of the model. The Perceived Problem to on the .90 depending wave. appraisal component study Index (PPI; Zamble, 1998) adapted fromZamble and checklist thatasks StableDynamic Mechanisms Quinsey(1997) is an interview-based Response to rateon a 7-point Likert scale theextent to respondents which he/she feelsworried ortroubled about15 potentially The finalcomponent of the model was operationalized areas (e.g., employment, constructs: substance abuse,supervision criminogenic money, drugs/alco- usingsix different the compliance, social support, criminal assohol). Total scoresare obtained by summing together copingability, attitudes. items(plausiblerange:15-105). Higherscores ciates,and criminal individual levelsofperceived Substance abuse and supervision represent higher criminogenic problems. compliance were this scale has not been it evimeasured two used subscales from the PSC described Although previously using in dencedexcellent at above. The supervision consistency thecurrent study internal compliancescale was originally eachwaveofdatacollection from .93 to.95). of fiveitemsbut thefollowing two items:(1) (alphasranged comprised The PerceivedStressScale (PSS; Cohen,Kamarck, & and(2) unrealistic releaseplansweredropped manipulative and/or Mermelstein, 1983) is a 14 itemself-report questionnaire due to poor inter-rater reliability poor internal thatmeasuresto whatextent individuals findtheirlives consistency. The resultant supervision compliancescale uncontrollable and overloaded. was of three items unpredictable, (totalscorerange:0-6) that Respondents comprised theextent towhich theoffender was unmotivated are askedto indicate how often have felta particular measured they thelast month Likert scale to comply with supervision way during usinga five-point requirements, experienced thatrangesfrom'never' (0) to 'fairly often'(4). Sample problems or failedto getting alongwithhis paroleofficer items include:'In thelastmonth, howoften haveyoubeen attend scheduled with his officer. The appointments parole that estimates were acceptableat each upsetbecauseof something happened unexpectedly?'. inter-rater reliability Total scores range from0 to 56 with higherscores wave of the study(ICC's .70-.80) as was the Wave 1 levels of perceived stress. the internal estimate the reflecting greater Although consistency (alpha = .70). However, PSS has not been previously administered to offender alphasweresomewhat low during thecommunity phaseof = it has demonstrated consis- thestudy (i.e., Wave 2 alpha .34; Wave 3 alpha = .49). samples acceptableinternal thisvariablewas retained due to its theoretandcriterion-relatedNonetheless, tency (alphas.83-.86) andconvergent in community coefficients: .61ical and practical validity samples(validity importance.
Springer

This content downloaded from 111.68.111.198 on Mon, 20 May 2013 12:30:11 PM All use subject to JSTOR Terms and Conditions

Law HumBehav(2009) 33:25-45

31

The original scale was comprised of six items substance the extent was experito whichthe offender measuring item encingan alcohol and/or drugproblem(individual scores: 0-no problem;1-problem somewhatevident; 2-problem clearlyevident).While all six items were retained fiveof thesix itemsrequired dichotomization (1 and 2's werecollapsedinto2's) giventhat less thanthree receiveda '2' rating any of thecomparticipants during of the munity phases study.However,even afterthis thevariable remained skewedat Waves highly adjustment 2 and3 with thevastmajority ofthesamplescoring 0, 1,or 2. Consequently, theresultant substance abusevariable was dichotomized 1-some of a evidence (0-no problem; substance abuse problem). estimates (.81Inter-reliability at each wave of thestudy. .93) wereacceptable At pre-release, was measured copingefficacy usingthe Situations Zamble, Questionnaire (CSQ; 1989) Coping while the Coping Interview (CI; Zamble & Porporino, assessments. 1988) was used for the community-based Both measuresassess how well an offender copes with situations potentially criminogenic alongtwo dimensions: thatameliorate benefits the problem) and (i.e., responses costs (i.e., responses thatworsenthe problem). The two measures of copingefficacy are similarin thattheyare administered andemploy thesamerating scheme. verbally differ in that the asks the to However, CSQ they participant describe howhe wouldreactto four hypothetical problem whiletheCI askstheparticipant how to describe scenarios, he is currently tworeallifeproblem situations copingwith Each response is rated (as identified by the interviewer). scales and averagecopingusingcost and benefit rating scoresare calculatedforeach situation. A final, efficacy score is obtainedthatranges aggregate coping efficacy from 1 to 20 withhigher scoresreflecting superior coping considers solutions, (e.g., generates ability multiple longtermconsequences, do not make the problem responses interworse). Both measureshave demonstrated strong rater studies (r's .80-.95) in previous (Zamble& reliability 1988) as well as thecurrent (ICC's ranPorporino, study, between .78 and .94). ged The Social Support Scheme (SSS; Brown& Zamble, that interview-based measure 1998b) is a semi-structured assessesthenumber sourcesof of perceived, high-quality informasupport(i.e., social, emotional,instrumental, within an individual's environment. a semitional) During structured interview are askedto identify how participants individuals of ten)promanydifferent (up to a maximum vide themwithsome formof social (e.g., leisuretime), instrumental (e.g., money, lodging),emotional (e.g., lisor informational tens,providesencouragement) support then license).The respondent (e.g., how to get a driver's rates(7-point Likert on the scale) each identified supporter fivequestions thatmeasurequalityof support: following

to youcomein contact with thisperson?", (1) "How often (2) "How muchdo you respect (3) How much him/her?", would he/she be upset if you started crime committing "How much do want to do what again?", (4) you he/she thinks areyouwith youshoulddo?", and (5) "How happy thesupport from this A totalSSS youaregetting person?". score is producedby multiplying the total numberof identified supporters (range 0-10) by the total quality items X 7 = maximum scoreof35). Total (5 rating support 0 scorescan thusrangefrom (no identified to supporters) 350, with higherscores representing superiorsupport. indicated that no modifications were Preliminary screening the SSS demonstrated required. Additionally, acceptable inter-rater the study(ICC's range: reliability throughout .67-.74). After theSSS had been administered, were participants askedwhether or notanySSS individuals werecriminally activeor had a criminal record. File information was also used to augment thisvariable.The original variablewas scored0-10 with10 representing individuals who identified 10 different criminalsupporters. Due to extreme skewness this variable was dichotomized as follows: 0-no criminalassociates, 1-at least one criminalassociate (kappa's: .87-.93). The construct of criminal attitudes was assessedusing two measures-the Criminal Self-EfficacyScale-15 (CSES-15; Brown, Zamble, & Nugent,1998a) and a modified version of theExpected Value of Crime slightly (EVC; Harris, 1975).The CSES-15 is a 15-item, Inventory that measures the true/false, self-report questionnaire extent to which a person believesthat he/she can execute a of violentand nonviolent crimessuccessfully as variety well as one's beliefin his/her criminal general expertise. "I items have much include: do not Sample experience intocars" and "If I was shotor stabbed I would breaking knowwhere to gethelpwithout The to thehospital". going CSES-15 rangesfrom0 to 15 withhigher scoresbeing associatedwithhigher degreesof criminal self-efficacy. the evidence that CSES-15 is interPreliminary suggests = consistent An earlier 20-item version nally (alpha .92). demonstrated (correlation acceptableconvergent validity with thePCL-R; r = .52) andcriterion (correlation validity with criminal r = .33) (Brown, & Nugent, Zamble, history; 1998b). A slightly modified version of the(EVC; Harris, 1975) was used to assess boththeanticipated positiveand negative consequences of crime. The measure was administered verbally.Individualswere firstasked to as different ofcrime generate many negative consequences as they couldup to a maximum often(e.g.,go tojail, lose Once the list was generated, the following two family). wereaskedabouteachconsequence: "On a scale questions of 1-10 where tenrepresents 'verybad', howbad wouldit
Springer

This content downloaded from 111.68.111.198 on Mon, 20 May 2013 12:30:11 PM All use subject to JSTOR Terms and Conditions

32

Law HumBehav(2009) 33:25-45

be ifconsequence #1(e.g.,go tojail) actually happened?" and "On a scale of 1-10 where ten represents 'very #1would probable'whatare the odds thatconsequence occurif you did engagein crime?".Due to extreme kurtosisthe 10-point was scale into a threerating collapsed scale (i.e., values 1-3 were recodedto l's; pointrating values 4-7 were recodedto 2's and values 8, 9, and 10 were recodedto 3's). A total negativeconsequenceof crime scorewas then obtained the'badness' bymultiplying and 'probability' ratingsfor each consequence.These valueswerethen summed and divided bythetotalnumber of consequences. Total negativeconsequencesof crime scoresrangedfrom1 to 6 withhigher scoresreflecting a level of associated greater negative expected consequences withdoingcrime(ICC's range:.87-.96). A similar forthepositive was thenfollowed procedure consequencesof crime.In this case, the first question "How it if would be #1 asked, good consequence (e.g., score a lot of money) occurred?"whereasthe second questionasked, "What are the odds of consequence#1 kurtosis the 10 item occurring?". Again due to extreme scale was collapsed into a three-point ratingscale as above. Total expected of explained positive consequences crime scores ranged from 1 to 6 with higherscores a greater levelofexpected reflecting positive consequences of crime.In theory, the negativeEVC score shouldbe correlated withrecidivism (themoreperceived negatively of crimetheless likelyrecidivism negative consequences should occur) while the positiveEVC score shouldbe correlated withrecidivism (themoreperceived positively benefits of crime themorelikely shouldoccur). recidivism the original EVC measure(Harris,1975) demAlthough onstrated some evidenceof concurrent in thatit validity was correlated with criminal Harris didnot indices, history indices.However,the measuredemonreport reliability strated in thecurrent inter-rater acceptable reliability study (ICC's range:.77-.96). The Balanced Inventory of DesirableResponding-Version 6 (BIDR; Paulhus,1994) was also administered as a measure ofsocialdesirability. The BIDR is a 40-item, selfof two dimensions: selfreport questionnaire comprised and The BIDR has deception impression management. beenextensively researched andhas proven tobe a reliable and valid measure of (e.g., alpha = .86 amongprisoners) socialdesirability theBIDR was (Paulhus, 1998).Although assessed only once duringthe pre-release phase it was treated as a 'stabledynamic mechanism' variable response withinthe coping-relapse model. Consequently, it was includedin all multivariate analyses thatinvolvedthe individual mechanism subset(alpha = .83). response recidivism was coded Lastly, 'yes' if the offender's conditional release was officially revokedfor technical violations or abuse,criminal association) (e.g., substance
Springer

for new offences(chargesor convictions). Recidivism information was obtainedfromCorrectional Service of Canada's Offender Management System(OMS). Given thatCorrectional Serviceswas supervising all of theparwhen the recidivism data was coded the OMS ticipants systemcapturedall formsof recidivism rangingfrom technical violations crimes abuse)to minor (e.g.,substance (e.g., obstruct justice)to majorcrimes (e.g., murder). Procedure Data werecollected usinga three-wave prospective, panel 1 Wave occurred within a 45 window to design. day prior releasewhileWaves 2 and 3 occurred in thecommunity at 1-month and 3-month intervals, post-release respectively. Potential candidates were informed aboutthe purposeof thestudy andwerethen askedtoparticipate. Ifthey agreed, written consentwas obtainedand the interviewer proceeded withtheinterview of thestudy. Each preportion releaseassessment lastedapproximately 5-8 h. It included theinterview time (2-3 h), a filereview(1-2 h), scoring and of the (2-2.5 h) completion self-report questionnaires. The community assessments were considerably shorter, from2 to 3 h. Participants after were debriefed ranging each phase of the data collection.In total,six trained interviewers conducted thedatacollection. Twelvepercent of the pre-release interviews and of the post10% (16) releaseinterviews Wave 2, and 15 from Wave 3) (15 from were scoredby the first author to obtaininter-rater reliestimates. ability If the National Parole Board officially revokedan individual's conditional releaseprior to thecompletion of thecommunity his participation in thestudy assessments, was immediately terminated. were However,individuals retained in thestudy ifthey weresuspended (theprecursor to revocation) and temporarily detainedin custodybut were subsequently re-released rather thanrevoked.The amount oftime that an offender in temporary spent custody was recordedand added to his follow-up time in the This adjustment was madeto ensure that each community. offender would be assessed afterbeing 'at risk in the for1 and 3 months, community' respectively. Data Analysis A seriesof doublymultivariate measures analyrepeated univariate ses, single-group repeatedmeasuresanalyses, and pairwisecomparisons were used to assess changein Cox regression survivalanalysiswith dynamicfactors. covariates and ROC wereused to time-dependent analysis assess predictive accuracy. Cox regression survivalanalysisis a semi-parametric survivaltechniquethatgenerates regression parameters

This content downloaded from 111.68.111.198 on Mon, 20 May 2013 12:30:11 PM All use subject to JSTOR Terms and Conditions

Law HumBehav(2009) 33:25-45

33

thatapproximately 80% of thevariables had (Allison, emphasized using partiallikelihoodestimation procedures it allows the no morethan 25% oftheir valuesmissing at anyone wave. 2000; Cox, 1972). Like multiple regression, thehighest researcher to comparethe relativecontribution of mul- Further, ofmissing data(34%-42%) percentage of whether from the four variables or arose in thecomadministered tiple simultaneously, regardless self-reports not they are dichotomousor continuousin nature. munity. datawerereplaced with theoverall Missing sample meanforcontinuous variables data were Moreover,Cox regressionsurvival analysis can also (Time 1 missing score (X'Beta) thatrepresents replacedwiththe mean value corresponding to Time 1; generatea standardized the predictedsurvival function for each subject, or, Time 2 missing data werereplacedwiththemean value score thatrepresents the standardized one corresponding to Time 2, and Time 3 missing data were alternatively, variableor the best linearcombination of two replacedwiththe mean value corresponding to Time 3). predictor or more predictor variables includingtime-dependent However,the sample mode was used for dichotomous variables.It is analogousto the standardized A comparison of theresults withand conducted predicted variables. value obtained in regular linearregression mean that without substitution deletion was represents (pairwise approach the best linearcombination of predictor variables.The mean substitution tendedto used) revealedthatalthough X'Beta value can also be saved and treatedas a pre- reducethe magnitude of the univariate and multivariate dictor variable.Thus,it can be used in standard theoverall in theresults trend remained multiple effects, unchanged. to of variance No differences between regression generate proportions explained significant emerged study compland also in ROC analysisto generateArea Underthe eters and studydropoutsat Wave 2 on a numberof Curvevalues. and staticpredictor variables.However,at demographic Cox regressionreadily incorporates were significantly morelikelyto time-dependent Wave 3 study dropouts covariates thatfluctuate overtime)by incorpo- have been Black or Asian and also scoredsignificantly (variables the dynamic measurement scorethatwas takenin loweron thePCL-R. rating closest temporal proximity priorto the eventof interest intotheanalysis. forindi(i.e., revocation) Alternatively, Recidivism Rate viduals who were not revoked, the last dynamic measurement score (i.e., 3 monthassessment result)is revoked Overall,36.8% of thesample(50/136)was either used. Thus, although only one variableis used in the at large(n = 5) during thefollow(n = 45) or unlawfully theactualvaluevariesas a function analysis, ofthetiming that ranged from3 monthsto 19.2 months of theeventof interest. one of the up period This in turn addresses (M = 10.2, SD = 3.9). Approximately, 2 of therevocamost commonlimitations of past researchin this area, tions were for purelytechnical reasons(e.g., substance Individuals that failprior to renamely, samplecensoring. abuseviolation, curfew whiletheothers werefor violation) assessment can stillbe incorporated in theanalysis. Hownewcriminal of and/or convictions. The definition charges theanalysis does notdirectly ever,technically, incorporate recidivism usedinthecurrent was conditional release study changescoresper se. revocation. Violentrecidivism was not considered given ReceiverOperator Characteristic is a statistical techfor that only5% ofthesample(n = 7) wererevoked being used to assesspredictive (Swets niquefrequently accuracy with a violent offence(s). charged et al., 2000). The analysis AreaUnder theCurve generates between .50 and 1.0.Thesevalues (AUC) valuesthat range the of a recidivist AssessingChange in DynamicFactors: represent probability correctly selecting whenaskedto do so from Change AmongNonrecidivists a recidivist Within-Subject a pairofindividuals: anda non-recidivist. AUC valueswerecompared usingthe test of correlated ROC areas outlinedby Hanley and The first phase of the analysisfocusedon determining whether or notthedynamic measures for McNeil (1983): actually changed who werenotrevoked individuals thestudy during period Z= (A1- A2)/(SE +SE - 2rSE1SE2)1/2 (n = 86). Two doubly multivariate repeatedmeasures were The first was performed analysis analyses conducted.4 RESULTS and response subset on thecombined mechanisms triggers on theappraisals whilethesecondanalysis was performed MissingData data foreach variable of missing Overall,thepercentage from 0% to 18.3% at Time 1, 19.8% to 34.2% at ranged Time 2, and 25% to 41.6% at Time 3. It should be
was adopted theDV's were procedure giventhat 4 A multivariate correlated with oneanother as demonstrated byBartlett's significantly forbecausethe testof sphericity. Staticriskwas not controlled current was on within-subject emphasis change. Springer

This content downloaded from 111.68.111.198 on Mon, 20 May 2013 12:30:11 PM All use subject to JSTOR Terms and Conditions

34

Law Hum Behav (2009) 33:25-45

indicated thatthe comsubset."Wilk's Lamba criterion bined triggerand response mechanism subset was affectedby time, F (14, 86) = 11.14, significantly F (4, 86) = 8.72, < .0001 as was the appraisalsubset, p p < .0001. uniThese analyseswerefollowed by 18 single-group variate wereoriginally (as there analyses repeated-measures riskfactors) and a seriesofpair18 hypothesized dynamic were examined wise comparisons. violations Sphericity and were addressed usingMauchly'sTest of Sphericity method. As Table 2 demonusingtheGreenhouse-Geisser strates, (p < .05), within-subject significant changeswere in 13 of the 18 with the variables, following eight present variablesevidencing significant change at the p < .001 level: employment leisureproblems, problems, perceived global stress, perceivedproblemindex,negativeaffect, social support, coping ability, expectedpositiveconseand substance abuseproblems. Also,9 of quencesofcrime the 13 significant (p < .05) variablesyieldedsignificant leisure (see Table 3). Noteworthy, pairwisecomparisons of and problems expectedpositiveconsequences crime ina counter-intuitive both increased. direction; changed they financial difficulties, Additionally, employment problems, and perceivedproblemlevel actuallyincreasedbetween Time 1 andTime2 buteventually leveledoff byTime3.

univariate to (p < .05) were thenpermitted significance foruniquevariance in a stepwise Cox regression compete It should be notedthat theentry survival andexit analysis. the values were set at .10 due to interp conservatively correlations amongthevariables. all of thestatic variables withtheexcepInterestingly, tion of age were significantly relatedto survivaltime. all the staticfactors Consequently, (excluding age) comin for variance survival outcome peted unique usingCox The results of thisanalysisrevealed regression. stepwise thatonlythe SIR-R1 and prisonmisconducts made significant and unique contributions to the survivaltime. a standardmultiple regressionanalysis Furthermore, revealedthattheSIR-R1 scoreand thenumber of prison misconducts accounted for 28% ofthevariance in outcome ornotrevoked). thepredictive of (revoked Lastly, accuracy thebest static mismodel and prediction (SIR-R1 prison conducts) was high (AUC = .81). See Table 4 for a overview. complete The analysesdescribed for above were thenrepeated 1 eachoftheTime 1 dynamic variables. Time Specifically, univariate dynamicvariablesthat demonstrated significance (p < .05) in theCox regression forunique competed variancein one of two stepwise Cox regression survival The first Cox was restricted analyses. stepwise regression to appraisal subset variables while the second Cox Recidivism Prediction was restricted to the combinedtrigger and regression mechanism subset. response The next wave of analysesfocusedon identifying the eachofthe four variables evidenced Noteworthy, appraisal set of static and factors that a with survival time in the strongest dynamic optimally significant relationship expected successful releasesfrom failures direction index distinguished usingCox (p < .05). However, only perceived problem survival linear and andperceived entered thefinal regression analysis, multiple globalstress regression, stepwise equaROC analysis.Resultspertaining to the staticsubsetare tion. while 7 ofthe15response Similarly, mechanism/trigger an followed of information variables demonstrated a univariate first, presented by analysis (p < .05) significant, an of with survival time: followed availablepre-release (Time 1), relationship by analysis employment, unsupportive information' the 'time-dependent derivedfromthe pre- partner/single, social support,supervisioncompliance, release (Time 1) and post-release substance abuseandimpression (Time 2 and Time 3) coping ability, management assessments. between the onlytwoof thesevariables-substance abuse and impresLastly,a seriesof comparisons their in variousprediction models(e.g., 'best static'versus'best sion management retained contribution significant the Cox stepwiseregression. the relationship werealso conducted. Moreover, time-dependent) Five univariate Cox regression survival and survivaltime was analyseswere betweenimpression management conducted for each static variable assessedpre-release offenders who were more (i.e., counterintuitive. Specifically, scoreswere age at release, SIR-R1, PCL-R, CATS-SR, numberof likelyto exhibit highimpression management thatdemonstrated also more tosurvive thefollow-up See Next,riskfactors prisonmisconducts). likely during period. Table 5 fora complete overview the including correspond5 A seriesof on theTime 1 principal component analyses(performed of variance ing AUC values as well as the proportions data only) supported the uniquenessof the individualinfluences and accounted for in revocation outcome. accountedfor55% and 56% of the appraisalsubsets(e.g., one factor The next phase of the analysis addressedthe timevariancein each component, However,similarsupport respectively). was not renderedforthe available responsemechanismand trigger dependent riskfactors. Once again,each of the 18 timecomponents.Consequently,all multivariate analyses involvingthe risk via factors was first analyzed independently dependent available response mechanism and triggercomponent domains Cox regressionsurvival analysis with time-dependent conducted for these domains were subsequentlymerged while covariates to ascertain which variables wouldbe permitted multivariateanalyses for the individual influencesand appraisal subsetswere conductedindependently, as planned. into the Cox stepwiseregression. These analyses entry
Springer

This content downloaded from 111.68.111.198 on Mon, 20 May 2013 12:30:11 PM All use subject to JSTOR Terms and Conditions

Law HumBehav(2009) 33:25-45 Table 2 Within successes: measures results subject-change among repeated measure Dynamic set Trigger Single/unsupportive partner Employment problems Accommodation problems Financial problems Leisure problems Health problems set Appraisal Perceived globalstress Perceived index problem affect Negative Positive affect mechanism set Response social Strong support Criminal associates Positive coping ability Criminal self-efficacy valueofcrime negative Expected valueofcrime Expected positive Poorsupervision compliance Substance abuseproblems Managementa Impression Time1 (N = 86) M (SD) Time2 (N = 86) M (SD) Time3 (N = 86) M (SD) F

35

0.5 (0.5)
2.1 (2.1) 2.1 (1.3) 1.3 (1.3)

0.6(0.5)
2.6 (2.1) 1.8 (1.6) 1.8 (1.3) 4.6 (1.6) 0.5 (0.8) 18.0(6.1) 27.9 (8.6) 27.6 (7.6) 47.0 (6.5) 146.2(52.3) 0.3 (0.5) 10.9(3.9) 4.9 (3.7) 3.6 (1.4) 2.0 (0.8) 0.7 (0.9)

0.5(0.5)
1.6 (1.8) 2.0 (1.5) 1.4 (1.1) 4.7 (1.6) 0.4 (0.7) 17.5(4.6) 24.8 (7.1) 25.2 (5.3) 48.9 (5.0) 161.8(40.3) 0.2 (0.4) 12.0(3.8) 4.6 (3.3) 3.9 (1.3) 1.9 (0.8) 0.9 (1.0)

10.51"* 1.05 3.34* 33.16** 0.09 10.02** 7.60** 17.94** 1.58 12.75** 3.74* 9.13** 0.31 3.74* 12.87** 0.83

3.01"

3.1 (2.3) 0.5 (0.8) 20.5 (6.5) 24.1 (9.4) 31.0 (8.8) 47.8 (8.0) 136.9(50.6) 0.4 (0.5) 10.1 (2.4) 4.9 (3.5) 3.4 (1.6) 1.5 (0.7) 0.8 (1.1)

0.3(0.4)

0.1(0.3)

0.1(0.3)

10.25**

1 month Note:Time 1 refers to information to release, Time2 andTime3 refer to information and3 months after gathered prior gathered release, respectively a couldnotbe assessed that datawerenotcollected at Time2 and3 Within-subject change given
* ** < .001 p < .05, p

illustrated thatthe following13 variablesevidenceda modelwas comprised of two variables from the dynamic survival time:single/ combinedresponsemechanism and trigger subset:subsignificant relationship (p < .05) with unsupportive, employment problems,perceived global stance abuse and impressionmanagementand two stress, level, problem negativeaffect, positive variablesfromthe appraisalsubset:perceivedproblem perceived affect,strongsocial support,positive coping ability, level and perceivedglobal stress.Lastly,the strongest criminal of five self-efficacy, expectednegativevalue of crime, Time-dependent dynamicmodel was comprised com- variablesfrom the combined expectedpositivevalue of crime,poor supervision response mechanism/trigger in Table 6, 7 of subset:employment andsubstance abuse.As illustrated social support, pliance, problems, single/unthe 13 variables thatevidenced at theunivar- supportive significance partner, expected positive consequencesof andtwovariables from theappraisal subset: iate level also displayed the stepwise crime significance during negative and perceived level. regressionphase. These variables included substance affect problem social support, At this stage, the relativepower of each prediction abuse, employment difficulties, single/unsupportive partner,expected positive consequences, model was assessed using ROC analysis and standard and perceivedproblemlevel. It is also As Table 7 illustrates, as hypothesized negativeaffect, multiple regression. that the between the model included both staticand noteworthy relationship expected positive strongest prediction of crimeand recidivism was counterintui- prospectively rateddynamic risk(AUC = .89, CI = .81consequences tive; individualswho generateda greaternumberof .93; R2 = .43, p < .0001). The inclusion of dynamic thosere-assessed in thecommunity positive consequences of crime were less likely to variables, particularly recidivate. enhanced (see Table 7). greatly predictive accuracy In sum,thestrongest static modelwas comprised of the As Table 8 illustrates, a number of significant differmisconducts whilethestrongest Time 1 modelsemerged, SIR-R1andprison encesbetween thevarious however, only
Springer

This content downloaded from 111.68.111.198 on Mon, 20 May 2013 12:30:11 PM All use subject to JSTOR Terms and Conditions

36 Table 3 Within successes: pairwise subject-change among comparisons measure Dynamic set Trigger Single/unsupportive partner Employment problems Accommodation problems Financial problems Leisure problems Health problems set Appraisal Perceived globalstress Perceived level problem affect Negative Positive affect mechanism set Response social Strong support Criminal associates Positive coping ability Criminal self-efficacy valueofCrime Expected negative valueofCrime Expected positive Poorsupervision compliance abuseproblems Substance Time 1 versus Time2 (N = 86) Da -0.1 -0.5 0.2 -0.4 -1.5* 0.0 2.4* -3.8* 3.4* 0.7 -9.3 0.1 -0.8 -0.1 -0.2 -0.5* 0.0 0.2*

Law HumBehav(2009) 33:25-45

Time2 versus Time3 (N = 86) D 0.1 1.1" -0.2 0.3 -0.1 0.0 0.5 3.2* 5.8* -1.5 -15.6 0.2 -1.1 0.3 -0.3 0.6 -0.1 0.0

Time1 versus Time3 (N = 86) D 0.0 0.6 0.1 -0.1 -1.6* 0.0 3.0* -0.6 2.4* -0.8 -24.9* 0.1 -1.9 0.2 -0.5 -0.4* -0.1 0.2*

1 month Note:Time 1 refers to information to release, Time2 andTime3 refer to information and 3 months after gathered prior gathered release, respectively a D = Difference score * p < .05 (Bonferonni adjustment applied)

modelincluded bothstatic and time level. Spe- pre-release two differences reachedthe .001 significance predictive risk the combined static and 1 measures. assessors should cifically, pre-release dynamic dynamic Accordingly, to incorporate both staticas well as dynamic model significantly the SIR-R1 model, continue outperformed A more intotherisk detailed p < .001 as did the combinedstaticand time-dependent information appraisal process. is now provided reviewof theresults followed model. dynamic by a disand directions for cussionof thelimitations, implications future research. DISCUSSION Static Factors In 1995,Andrews ofthe "Overall, proclaimed, exploration a validities of assessments of changeremains As expected, all of thestatic measures predictive exceptage signifiand the issue for the revocation at the univariate level. issue is, general major perhaps, major cantly predicted of theory and practice in thepsychology of Interestingly factors--the SIR-RI however, development onlytwostatic crime"(p. 54). Accordingly, thecurrent the andprison misconducts entered thefinal solution. study explored stepwise of theSIR-R1 to predict revocation is predictive validityof change data using a three-wave, The ability general male with research et 136 offenders released consistent al., 1996; Cormier, (Bonta prospective designinvolving prior from Canadianprisons. theobservation that thenumber of prison 1997). Similarly, As predicted, to add significant, the greatest level of predictive was theonlystatic measure accuracy misconducts was achieved when both static and time-dependent incremental variance above andbeyond theSIR-R1 is also In measures were included. to the with consistent 2000; Palmer, dynamic regards preprevious findings (Nugent, theTime 1 dynamic it is necessary releaseinformation, modelwas gen- 1997). At thisjuncture to emphasizethat misconducts was operationalized as a static variable of predicting prison to thestatic modelin terms erally equivalent revocation. onceagain, themost accurate (i.e., number of prisonmisconducts incurred However, general duringthe
Springer

This content downloaded from 111.68.111.198 on Mon, 20 May 2013 12:30:11 PM All use subject to JSTOR Terms and Conditions

Law HumBehav(2009) 33:25-45 Table 4 Beststatic subset: cox stepwise andreceiver characteristic (ROC) results regression operator Static measure -2 Log La Bb (with variable(s)) 418.64 -0.09 0.02 34.40*** 407.75 -0.09 0.02 34.40*** 0.26 0.08 9.75** -8.3 29.5 .99 .39 -171.56 53.68 .05** -8.6 .99 -171.56 .23*** SE b X2c

37

% change in Incremental AUC (CI)h Be % change r2(X'Betag) in hazard rate hazard rate (unstandardized)d (standardized)f .78 (.70-.85) .81 (.73-.87)

Step1 SIR-Ri' Step2 SIR-R1 Prison misconducts

R2 = .28. No other Note:N = 136.FinalR2= .28. Adjusted variables metthe.10 significance levelfor entry into themodel a = 454.10;-2 Log L = -2 multiplied -2 Log L (without variable) value bythelog likelihood b Unstandardized thebaseline function b. It represents thedegree to that survival increases ordecreases as a function of a unit inthe change variable
d

2 = ScoreStatistic Thisvaluerepresents the percentage change inthe hazard rate for eachone-unit increase inthe variable while all other variables inthe holding model constant e Standardized B Thisvaluerepresents the inthe hazard rate for eachincrease ofonestandard deviation inthe while variable all other percentage change holding in themodel variables constant

is a standardized X1Beta score that thepredicted survival function for eachsubject oralternatively, thebestlinear of combination represents variables. It is analogous to thestandardized value obtained in regular linear thebestlinear predictor predicted regression representing combination ofpredictor variables theCurve; CI = 95% confidence intervals h AUC = Areaunder SSIR-R1 = Statistical Information on Recidivism Scale-Revised ** p < .01,***p < .001
g

12 monthperiodpriorto release). Futureresearchthat variablesdid not evidencesignificant change:accommothis variable as dation health criminal affect, conceptualizes dynamic (e.g., pre-release: problems, problems, positive has the number of prisonmisconducts been declining or self-efficacy and supervision two compliance.Further, sinceadmission; has thenumber of unexpected results and Leisure-related increasing post-release: emerged. problems interviews or issued warrants been the extent to which individuals consethe disciplinary suspension perceived or increasing sincerelease?)could enhance declining pre- quences of crimein a positivelightactuallyincreased dictive evenfurther. sucha concrete during thefirst 3 months of releaseamongthesuccessful accuracy Additionally, measureof behavior are discussed further in the changewould be relatively easy to cases. These specific findings record andwouldrequire minimal sections. subjective interpretation proceeding on the part of decision-makers providedthatobjective criteria defining'significant change' are established DynamicFactors: StrongEmpiricalSupport a priori. factors that demonstrated Dynamic significant changeand entered thefinal wereconsidered to have solution Change in DynamicFactors stepwise evidencedstrong In sum,seven of the empirical support. As predicted, the vast majority of the dynamicfactors original 18 time-dependent dynamicfactorsmet these evidencedsignificant change in the expecteddirection criteria. They includedtwo variablesfromthe trigger thestudy. model: employment and cases, subset of the coping-relapse during Specifically amongthesuccessful 11 of the 18 dynamic variables and single/unsupportive changedsignificantly partner;two variables from the in thehypothesized and appraisalsubsetof the model: negativeaffect and perfinancial, direction---employment, three the substance abuse difficulties the variables from ceived and decreased, level; expected problem lastly, of crimeincreased, criminal asso- model's responsemechanismsubset: substanceabuse, negative consequences and expectedpositiveconsequencesof ciation, negativeaffect,perceived problemlevel and social support, lessened,and copingability, per- crime. perceived global stress All of thesefindings were consistent withthe copingceived social support,and maritalsupportimproved. to thestated thefollowing relapsemodelof criminal recidivism However, (Zamble& Quinsey, contrary hypothesis,
Springer

This content downloaded from 111.68.111.198 on Mon, 20 May 2013 12:30:11 PM All use subject to JSTOR Terms and Conditions

38 Table 5 Time1 dynamic subset: cox stepwise andROC results regression Dynamic measure and responses Triggers Step1 Substance abuse Step2 Substance abuse Impression management Appraisal Step1 Perceived index problem Step2 Perceived index problem Perceived stress global -2 Log La bb SE b (with variable(s))
X2c

Law HumBehav(2009) 33:25-45

% change % change Incre-mental AUC (CI)h Be in hazard in hazard rate rate r2(X1Betag) (unstandardized)d (standardized)f .65 91.8 .20*** .20*** .77 (.69-.85)

426.85 420.97

1.63 .35 21.24***

5.1

1.43 .36 26.31*** -.11 .05 30.81*** 437.51 0.05 0.01 17.81*** 432.64 0.04 0.01 19.02*** 0.05 0.02 4.93*

4.2' -10.2

.72 .44

105.4 55.3

.03* .68 (.59-.77)

5.4 4.1 5.3

.47 .38 .33

60.0 46.2

.*** .70 (.61-.78) .02*

subset: FinalR2= .23.Adjusted R2= .22.Appraisal subset: FinalR2= .13.Adjusted N = 136.Trigger andresponse Note: R2= .11.No other variables metthe.10 significance levelfor intothemodel entry a = 454.10;-2 Log L = -2 multiplied -2 Log L (without value variable) bythelog likelihood b Unstandardized b. It represents thedegree to that thebaseline survival function increases ordecreases as a function of a unit change in the variable
c

2 = Score Statistic

e Standardized B
f
g

Thisvaluerepresents thepercentage change in thehazard rateforeachone-unit increase in thevariable whileholding all other variables constant

Thisvalue the inthe hazard rate for eachincrease ofonestandard inthe while deviation variable all other represents percentage change holding variables constant thepredicted survival for X'Beta is a standardized score that function eachsubject oralternatively, thebestlinear combination of represents variables predictor h AUC = Areaunder theCurve; CI = 95% confidence intervals SGiventhat this this valuerepresents relative risk rather than % change inhazard rate with substance variable is dichotomous (i.e.,individuals than with no substance abuseproblems) abuseproblems are5 times more to be revoked individuals likely SGiventhat this variable was dichotomous AUC valuescouldnotbe calculated
* p < .05, *** p < .001

the that one; thefinding 1997) orwith pastresearch, except perceived positive consequences of crime actually increased cases during each succesamongthesuccessful sive wave of data collection was unanticipated. Some of theresearch interviewers feltthatthe 'crime-free' particiwere more than their pants likely supposedly 'criminallyactive' counterparts to speak freelyabout the positive of crimewhilethe 'criminally-active' indiconsequences vidualsweremoredefensive and moreaptto emphatically 'crimeneverpays'. Alternatively, it is equally proclaim, that in individuals, plausible prosocial assumably working a conventional job earninga conventional salary are awareofthebenefits ofcrime, fast andeasy acutely namely Thishypothesis money accompanied byless responsibility. is highly and requires further validation. speculative It is also noteworthy thatthe social support variable it didnot retained itspredictive thefactthat powerdespite discriminate betweencriminal and pro-socialsupporters.
Springer

itis important to emphasize that at each waveof However, thestudy, to criminal theratioofprosocial supporters supare porters alwaysexceededfourto one; thustheresults consistent withsocial learning (Akers,1998) that theory criminal behavior is a learned andthat the postulates process of the deviant behavior results from the strength directly and probability of its reinforcement amount, frequency others. through DynamicFactors: Moderate EmpiricalSupport Perceived global stress, poorcoping,and perceived negaof crimewerecategorized tiveconsequences as moderate of revocation predictors giventhattheychangedsignififirst 3 months of the studyand they the cantlyduring intheunivariate, notthe revocation albeit predicted general multivariate the finding thata global analyses.Although stress measure is predictive of conditional release failure

This content downloaded from 111.68.111.198 on Mon, 20 May 2013 12:30:11 PM All use subject to JSTOR Terms and Conditions

Law HumBehav(2009) 33:25-45

39

(CI)h (.69-.80) AUC .82 (.74-.89) .83 (.76-.90) .84 (.78-.91) .85 (.61-.80) (.63-.81) .71 .73

r2

(X1Betag) Incremental

.25***

.05**

.04**

.02**

.01ns

.16**

rate 58.4 change 111.7 75.1 % hazard in(standardized)f 64.9 55.3 59.9 58.4 59.9 47.7 41.9 63.2 59.6 58.4 44.8 40.5 101.4 85.9

Be

.75

.56 .46

.50 .44 .47

.46 .39 .47 .35

.49 .37 .47 .46 .34

.70

.62

rate 6.3' 5.2' change 25.8 hazard % in (unstandardized)d 3.4' 25.0 -0.9 3.1' 2.01 21.9 -0.9 3.3' 2.5 20.4 -0.8 9.9 -37.1 8.1

Z2c

10.68*** 12.24*** 12.39*** 8.05** 10.95*** 8.11** 39.32*** 3.30' 29.41*** 6.80**

12.85*** 8.05** 3.45t 50.67*** 5.31"

45.93*** 48.41***

results b SE ROC and Bb regression

.29

.30 .07

.35 .07 .00

.34 .07 .00 .39

.33 .07 .00 .39 .20

.01

.02

1.84

1.6 .23

.22 1.22 -.01

1.13 0.20 0.71 -0.01

1.20 0.19 0.91 -.01 -0.46

0.09

0.08

La stepwise Log cox variable(s)) 415.09 403.23 -2 (with subset:

394.34

390.63

385.21

420.31 415.24

dynamic

partner

partner

abuse abuse abuse abuse abuse affect affect responses consequences-crime support support support measure Time-dependent and 6 1 1 3 2 5 2 4 Substance Substance Single/unsupportive Substance Social Substance Social Single/unsupportive Substance Social Employment Employment Positive Negative Negative Employment Employment Step Step Step Step Step Step Step Table Dynamic Trigger Appraisal

Springer

This content downloaded from 111.68.111.198 on Mon, 20 May 2013 12:30:11 PM All use subject to JSTOR Terms and Conditions

40

Law HumBehav(2009) 33:25-45

(CI)h AUC

entry for level

than

r2 significance .02t .10 (X'Betag) Incremental the met model rate

revoked be constant to model likely

the more in constant variables times 2

either weak or no empirfactors exhibited Eightdynamic ical support. four variables from thetrigger Theyincluded one leisureand health, subset:accommodation, finances, and variablefromthe appraisalsubset:positiveaffect, three variables from the response mechanismsubset: X2c and crimicriminal self-efficacy, supervision compliance, 5.54* each rather one increase increases ifany nal associates. Variables wereplacedinthiscategory likelihood of for Appraisal the were risk one of criteria met: (1) significant log following b .34. and the was absentduring thecourseof thestudy change one-unit the function .02 = SE increase variable was in no wayrelated to conditional releasefailure function by relative R2 each an calculated(2) significant butit did nottranslate changewas evident for be for survival intopredictive or (3) significant power, changewas absent survival not rate but re-test information enhanced rate 0.04 Adjusted Bb predictive validity. multiplied represents Thereare severalpossibleexplanations to accountfor .37. -2 baseline could = hazardhazard in thosedynamic did notevidence thestability factors that = intervals predicted values the L health R2 the thethe significant change(i.e., accommodation problems, values that in La in Log these criminal and affect, positive self-efficacy, problems, to Final .001 in it have -2 Log First, supervision compliance). retrospect, may AUC variable(s)) < confidence changechange p -2 been unrealistic to expect significant change in accomrepresents subset: degree 95% (with or criminal modation financial selfdifficulties, *** patterns, 454.10; the that = dichotomous = the first 3 months of release. Not were only efficacy during .01, CI partner) are dichotomous themajority oftheoffenders toreside at a halfway < percentage required response percentage score is p house duringthe first6 monthsof release but most the the and represents ** Curve; index variable) little theearlystages to have very It money appeared during variables supportive .05, of release. b. a the that variable Similarly, expectation global change a under < Trigger B standardized statistic in occur over the course of attitude would p problem (without naturally represents these this represents a with * 136. L Area in of an active treatment intervention 3 months the absence is measure continued that that = model Score .10, was in hindsight, value value = 6 naive.In regards to health, the giventhat Log N = < the Perceived -2 Unstandardized the of was not from a majority sample suffering psychiatric p X'Beta AUC Standardized X2 This This Given Given into Table Dynamic a b c d e f g h i individuals Note: j t disorder or illnessany real changesthatwere occurring
Springer

variables are predictor the variable of variables in other the change 41.9 all hazard in other % in (standardized)f individuals No variables change holding combination .17. single = other unit Be .35 R2 while all linear a (i.e., of best rate holding variable the Adjusted function the rate hazard .18. a while in in = as change 4.1 R2 hazard % in variable alternatively, change (unstandardized)d deviation Final or decreases % the value or in subset: subjectthan standard

mainstream criminological perspecappearsto contradict tives (e.g., Personal-Interpersonal-Community (PIC-R) Andrews & Bonta,2006) thatrejectpersonal Perspective, withthe coping-relapse distress variablesit is consistent as well as generalstrain model of criminal recidivism was not theory (Agnew,1992). Surprisingly, copingability thestrongest subset ofdynamic predictors, although among it did evidencesignificant direcchangein the predicted tion. Moreover, those changes when examined of any other variable predictedgeneral independently who werenot revocation. individuals Lastly,as predicted an increased revokedduring the studyperiodexhibited associatedwith awarenessof the negativeconsequences crime with each successive wave of data collection. the observed Moreover,when examined individually, revocation. themaggeneral Although changespredicted ofthechangewas notstrong nitude to permit entry enough intothefinal multivariate it demonstrate that does solution, can change in a relatively short of criminal attitudes period is adopted. timeif theappropriate measure strategy

Weakor no Empirical Factors: Dynamic Support

This content downloaded from 111.68.111.198 on Mon, 20 May 2013 12:30:11 PM All use subject to JSTOR Terms and Conditions

Law HumBehav(2009) 33:25-45 Table 7 Comparison ofprediction models: andROC results regression Prediction model reason Revocation-any R2(X'Beta) SIR-Rlc Static (time1) Dynamic Static anddynamic (time1) Timedependent dynamic Static andtime dependent .23*** .27*** .26*** .34*** .38*** .43*** R2(X1Beta) Adjusted .22 .27 .25 .33 .37 .43 R2changea (X' Beta) .05** .03** .11*** .15*** .20***

41

AUC (CI)b (X1Beta) .78 (.70-.85) .81 (.73-.87) .80 (.72-.87) .85 (.78-.91) .85 (.77-.91) .89 (.81-.93)

Note:a Represents theincremental increase in R2 achieved when is compared eachmodel to theSIR-R1model b AUC (CI) = Areaunder theCurve; 95% confidence intervals C on Recidivism SIR-R1= Statistical Information Scale
** *** p < .01, p < .001

Table 8 ROC results:pairwisecomparisons betweenprediction problems was coded based on whether or nottheoffender models to have leisure-related In thecommuexpected problems. evidenceforleisure was based on whether z value Modelcomparison: revocation nity, problems SIR-R1versus static SIR-R1versus (Time1) dynamic SIR-R1versus static anddynamic (Time1) SIR-R1versus time dependent SIR-R1versus static andtime dependent Static versus (Time1) dynamic static anddynamic Static versus (Time1) Static versus time dependent Static versus static andtime dependent Time1 dynamic versus time dependent -1.67" -0.56 -2.99*** -1.52 -3.17"*** 0.11 -2.22** 0.89 -2.53** 1.16

Note:a Thebeststatic model for theprediction ofnewoffences was oftheSIR-R1i solely comprised * p < .05,** p < .001 p < .01,***

of the individuals witha history of amongthe minority psychiatric problemswould have been masked by the ofnon-psychiatric cases. Similarly, overwhelming majority ofphysical health themajority that problems werereported wereof a chronic rather thanacutenature suchas liveror heartdisease, back problems, or arthritis. Thus, change wouldnotbe expected. thefinding that Lastly, supervision did notchangeduring thecourseof thestudy compliance was surprising. was once again due to Perhapstheresult measurement problems. leisureproblems over Although changedsignificantly in a counterthecourseof thestudy, thechanges occurred leisureproblems deteriointuitive direction, specifically, of rated duringthe first3 months release among the successfulcases. It is possible that the unanticipated been a meachangein leisure problems mayhave simply Priorto release, evidence for leisure surement artifact.

ornottheoffender was actually difficulties in experiencing thisarea. Consequently, it is possiblethatthepre-release and post-release measures of leisureactivity wereactually two distinct thus invalidating the constructs, measuring changeanalysis. While the number of criminalassociatesdiminished thesuccessful cases during thefirst 3 months ofthe among theobserved intopredicstudy; changesdid nottranslate tivepower.This finding contradicts previous single-wave outcome studies that have found that prediction repeatedly criminal is one of thebestpredictors association of criminal recidivism (e.g., Andrews& Bonta, 2006; Goggin, & Gray1998). This finding Gendreau, mayhave in part, been due to the direct measurement "Are you approach, in contact with that has a criminal anyone currently should 'no' a question that an automatic record?", generate most offenders arerequired to abideby response giventhat a non-association clause.Regardless, thisfinding does not, and shouldnot,void previous research Andrews & (e.g., that has demonstrated a relaBonta,2006) clearly strong tionship between criminal associates and criminal behavior. It is also worth that was noting impression management and significantly relatedto recidivism. Thus, negatively offenders who scoredhigher on impression management While thisfinding is somewere morelikelyto survive. it does confirm similar that whatcounter-intuitive results in other been observed correctional have recently samples 2005; Mills,Loza, & Kroner, 2003). (e.g.,Mills & Kroner, who More research is neededto ascertain whyoffenders are less likelyto score highon impression management reoffend.

Springer

This content downloaded from 111.68.111.198 on Mon, 20 May 2013 12:30:11 PM All use subject to JSTOR Terms and Conditions

42

Law HumBehav(2009) 33:25-45

EmpiricalValidationof the DynamicRisk Factor: the LitmusTest Rethinking a gold standard forthestatistical analysisof Establishing to recidivism is one of the most data in relation change As the formidable obstaclesfacingcorrectional research. thereappearsto be no conliterature reviewillustrated of whatconstitutes themostappropriate sensusin terms method foranalyzing data and recidivism. Howchange involvecreating artificial ever,two commonapproaches or usinghierarchical changevariables regression approascores add incremental variancein ches (i.e., do re-test scores).These approaexplainedoutcomeabove pre-test ches have likelybeen adoptedbecause theyare literally consistent withearlierdefinitions thatdescribedynamic riskfactors of change as, "....ones in whichassessments a level of criterion re-tests) (or validity predictive possess of pretests" that is incremental to thecriterion validity (p. et al., 1990). Moreover, thesemethods, 31, Andrews specan be readily hierarchical cifically regression, appliedto two-wave panel designs. Recall that two-wave panel the majority of previousmultidesignshave dominated wave research. In contrast, thisstudy tooktheposition thattheempirical litmus orgoldstandard for whether ornota test testing risk factor is dynamic truly dynamic requires predicted three analyses.These includestandard typesof statistical survival within-subject change analyses,Cox regression with andtime-dependent time-invariate covariates analysis a riskfactor was interand ROC analysis.Furthermore, as if it over changed significantly timeas preted dynamic evidencedby the within-subject change resultsand, if abouthow thevariable information incorporating changed overtimeactually This later increased predictive accuracy. criterionwas demonstrated empiricallythroughCox survival and ROC analysis. regression analysis with Althoughthis approachmay seem inconsistent contraditional definitions and methods it is nonetheless, withmodem day definitions of the dynamic risk sistent factor.Current statethatin orderfor a risk definitions to be deemeddynamic it mustbe shownthatprefactor is improved dictive information accuracy by incorporating abouthow thefactor & changesovertime(e.g., Andrews Bonta, 2006; Douglas & Skeem, 2005; Kraemeret al., 1997). The Cox regression approachalso has several additionaladvantages.Not only can it incorporate an indefinite number of assessment waves butit readily prevents information about samplecensoring byincorporating cases thatfail beforethe first scheduledre-assessment. Additionally, within-subject analyses (as well as more formsof change analysessuch as growth sophisticated curveanalysis) and Cox regression can analyze procedures tinkerwith the change withouthaving to artificially
Springer

scale of a variable. For themoment, we strongly original as beingan ideal litmus testforthe our approach support empiricalvalidationof dynamicrisk factors.However further and statistical of thismatter conceptual exploration is required. StudyImplications themajority of risk Overall,the study supports retaining thecoping-relapse modelof criminal recidfactors within ivism. However,in the absence of structural equation the studycan neither nor refute the modeling, support interactional and bi-directional predicted relationships thedomains and subsequently, amongthevariables. among theextent to whichthe 'acutedynamic Additionally, trigmechanisms' domains gers' and 'stabledynamic response requires additional comprise truly distinctconstructs examination. betweenthese two Perhapsthe distinction is unnecessary who have notyet constructs foroffenders desisted orachieved a certain leveloflifestyle For stability. theseindividuals theprocesshas either or alreadystarted neverreallyended,thusrending theconcept of a 'trigger' that irrelevant. thiswouldindicate variables Theoretically classifiedas 'triggers'should be re-classified currently as 'response mechanisms'. Once an offender has perhaps then achieveda certain level of lifestyle stability perhaps of triggers couldbe re-introduced. theconcept Operationally, this would mean that duringthe early stages of would be to release,the goal of community supervision Once this stabilizetheoffender's lifestyle. objectivewas would be re-directed and met, supervisionstrategies focusedon identifying of earlywarning signsindicative destabilization. The study also demonstrated that information self-report betweensuccan play a valuablerole in discriminating of interest are variables cesses and failures. Particularly derived from an offender's selfperception such exclusively levelof support, as perceived level and perceived problem future assessment perceivedglobal stress.Consequently, shouldconsider moreweight to selfprocedures affording information of this nature. reported StudyLimitations the studygenerated resultsthere Although encouraging were variouslimitations such as attrition, missingdata, andreliance on newandunvalidated covariates, premature measures. 20% of the samplerefused to Approximately, takepartin the study once releasedintothe community. a greater of thesample(about35%) Moreover, percentage refused to complete theself-report administered questions in thecommunity. An analysisof thedata indicated that thestudy didnotdiffer from overall, dropouts substantially

This content downloaded from 111.68.111.198 on Mon, 20 May 2013 12:30:11 PM All use subject to JSTOR Terms and Conditions

Law HumBehav(2009) 33:25-45

43

thestudy on a number ofkeyvariables at Wave completers 2. However, theWave 3 dropouts weresignificantly more and of Asian or Aboriginal likelyto be non-psychopathic decent.Whileit is difficult to ascertain whythiswas the case in reference to ethnicity we surmise thatnon-psyweremorelikely to dropoutbecausethey chopaths simply had less timeto meetwitha researcher giventhatthey tendedto be busy livingprosociallives-balancing full timejobs withfamilies and baseballleagues.In contrast, tended tobe unemployed with lotsoffree time psychopaths on their from a research hands,consequently perspective data was theywere easierto trackdown.While missing also problematic, a comparison of the analysiswithand without meansubstitution revealed that thesamevariables as significant of whether or emerged regardless predictors notmeansubstitution was employed. Thus,whilemissing data was troublesome, the mannerin which it was addressed did notalterthestudy findings. The premature covariate is a phenomenon that emerges whenthedynamic variable of interest or exerts its changes influence on the outcomebeforethe researcher has an to measureit (Menard,1991; Plewis, 1985). opportunity thislimitation to thecurrent Although posed a real threat and mayhave accounted forsomeof thenonsignifstudy icantfindings, it is necessary to emphasize thatthestudy was stillable to report significant changeacrossa number of key variables. Moreover, changewas linkedto conditionalreleasefailure. Another limitation ofthestudy was theuse of a number ofnew,unstandardized measures. these measures Although were strongly groundedin theoryand based on past research nonetheless wereempirically untested at the they of the As a it was difficult to beginning study. consequence, ascertain whether or nottheabsenceof significant findings forcertain variables was genuine or simply a function of faultymeasurement given the low internal particularly estimates for variablessuch as leisureand consistency financial management. recidivism was measured basedsolelyon official Lastly, information the actual recidivism rate. thereby deflating thestudy was unableto lookat thepredictors of Moreover, violent recidivism due to low base rateissues. DirectionforFuture Researchand Conclusions

neededto determine whether or nota genuine distinction existsbetween stableand acutedynamic factors. actually The issue of how best to assess real-world dynamic factors to releasefrom a secureenvironment prior requires additional research.Future studies should investigate methods forreliably behavior in prison that best assessing real life situations(e.g., how does the approximates offender timein prison? Has he acquired spendleisure any debtswhileinside?How has his employment performance been during incarceration?). Future research shouldalso explorewhether or notthe relevance ofcertain factors does in factchangeas dynamic a function of timeand population as others have argued (e.g., Kraemeret al., 1997). For example,determining whatdynamic factors are mostrelevant the early during phasesof releaseversusthosethatdo notbecomeimportantuntilthe offender's life has stabilizedwould yield substantial operational gains. In conclusion, the study demonstrates thatthe systematicre-assessment of dynamic riskcan enhance predictive themostpromising models However, accuracy. prediction will be those that incorporate static as well as timeinformation. dependent dynamic
Thisresearch was supported Acknowledgements by Correctional Service ofCanadaandSocialSciences Humanities Research Council LouiseNoce,Dianne (SSHRC). We thank TanyaRugge, Joy Irving, Julie Perkins andSebastien Gratton for their assistance in Necovski, datacollection andoffender We also thank Motiuk, tracking. Larry Director Research Correctional Service of Canada General, Branch, for this research. Thisarticle is basedon Shelley Brown's facilitating doctoral dissertation andportions oftheresearch were ina published non-referred Forumon Corrections Research government journal, We wouldalso like to thank the reviewers fortheir insightful comments on this manuscript.

24-27. (2002), 14(1),

REFERENCES
R. (1992).Foundation for a general strain ofcrime and Agnew, theory 1992.tb01093.x.
30, 47-87. doi:10.1111/j.1745-9125. delinquency.Criminology,

A general Akers,R. L. (1998). Social learningand social structure: theoryof crime and deviance. Boston: NorthernUniversity

Press.

Allison, P. D. (2000). Survival analysis using the SAS system:A

conduct Ohio:Anderson (4thed.). Cincinnati, Publishing. A number of avenuesforfuture research from the emerged D. A., Bonta, for Andrews, J.,& Hoge,R. D. (1990).Classification concerns measurement refinestudy.Firstand foremost, effective rehabilitation: JusRediscovering psychology. Criminal ment. Althoughthe study incorporated several new tice andBehavior, doi:10.1177/0093854890017001004. 17,19-52. D. A.,Bonta, & Wormith, J.S. (2004).Levelofservice/ measures thatwereheavily in theory and sensi- Andrews, J., grounded case management LS/CMI manual. Toronto, Ontario, inventory: tive to change, certain measures (e.g., SSS) require Canada: MultiHealth Systems. refinement to operational prior implementation. Similarly, Andrews, D. A., & Robinson, D. (1984). The levelof supervision that measures are ensuring actuallycapable of detecting Second report.Ottawa: Community Division of the inventory: Ontario of Correctional Services. attention. further research is Ministry changerequires Additionally, Springer

Andrews,D. A., & Bonta, J. (2006). The psychologyof criminal

North Carolina: SAS Institute Inc. (4th ed.).Cary, practical guide.

This content downloaded from 111.68.111.198 on Mon, 20 May 2013 12:30:11 PM All use subject to JSTOR Terms and Conditions

44

Law HumBehav(2009) 33:25-45

D. A., & Wormith, J.S. (1984). Criminal sentiments and R. K., Harris, A. J.R., Scott, T. L., & Helmus, L. (2007). Andrews, Hanson, BranchUser Report. criminal behaviour. Ottawa: therisk oncommunity Programs Assessing ofsexual offenders supervision: Solicitor General Canada. Thedynamic Canada:Public Ottawa, supervision project. Safety Canada. assessment and treatment. In A. T. Bonta,J. (1996). Risk-needs Hartland R. K.,& Thornton, D. (1999).Static 99: Improving actuarial (Ed.), Choosingcorrectional optionsthat work: Hanson, thedemand and evaluating thesupply risk assessments Ottawa: Defining (pp. 18-32). (User 1999-02). forsexoffenders report Thousand Oaks:Sage Publications. ofSolicitor General ofCanada. Department W. G.,Hann, R. G.,& Cormier, R. B. (1996).The Checklist-Revised. Bonta, J., Harman, Hare, R. D. (1991). The Hare Psychopathy ofrecidivism sentenced offenders: A ON: Multi-Health Toronto, prediction among federally Systems. re-validation ofthe SIR scale.Canadian Journal Checklist-Revised (PCL-R) ofCriminology, Hare,R. D. (2003). Hare Psychopathy 38, 61-79. Canada:Multi-Health Ontario, (2nded.). Toronto, Systems. S. L., & Zamble,E. (1998a). Problem checklist. Harris, A. R. (1975). Imprisonment and the expectedvalue of Brown, survey Test.Queen'sUniversity, Ontario. criminal choice. American Review, 40, 71-87.doi: Kingston, Unpublished Sociological 10.2307/2094448. scheme. Brown,S. L., & Zamble,E. (1998b). Social support Test.Queen'sUniversity, Ontario. G. T., Rice, M. E., & Quinsey, V. L. (1993). Violent Harris, Unpublished Kingston, S. L., Zamble,E., & Nugent, P. (1998a). Criminal recidivism ofmentally disordered offenders: Thedevelopment of Brown, SelfTest.Queen'sUnia statistical instrument. Criminal Justice and Behavefficacy Scale-15(CSES-15). Unpublished prediction ON. ior,20, 315-335. versity, Kingston, S. L., Zamble,E., & Nugent, P. (1998b). Criminal G. T.,Rice,M. E., & Quinsey, V. L. (1994).Psychopathy as a Brown, self- Harris, scale-20(CSES-20). Unpublished Test.Queen's Unitaxon: Evidence that area discrete class.Journal efficacy of psychopaths ON. and Clinical Psychology, 62, 387-397. doi: versity, Kingston, Consulting 10.1037/0022-006X.62.2.387. C. F., Schmidt, A. D. (1991).Survival A P., Witte, Chung, analysis: Journal G. T., Rice, M. E., Quinsey, V. L., Lalumibre, M. L., Boer, 7, 59-98. doi: Harris, D., survey. of Quantitative Criminology, 10.1007/BF01083132. & Lang,C. (2003). A multi-site of actuarial risk comparison R. (1983).A global measure of instruments forsex offenders. Cohen, S., Kamarck, T.,Mermelstein, Assessment, 15, Psychological 413-425.doi:10.1037/1040-3590.15.3.413. stress. Journal 24, perceived of Healthand Social Behaviour, 385-396.doi:10.2307/2136404. J.F., Hare,R. D., & Wong,S. (1998). Psychopathy and Hemphill, B. (1997).Yes SIR! A stable risk tool.Forum on recidivism: A review. Cormier, 3, prediction Legaland Criminological Psychology, 139-170. Corrections Research, 9(1), 1-7. H. C., Kazdin, A. E., Offord, D. R., Kessler, R. C., Jensen, models and life-tables discus- Kraemer, Cox, D. R. (1972). Regression (with P. S., & Kupfer, D. J.(1997).Coming toterms with theterms of B34, 187-220. sion).Journal ofRoyalStatistical Society, J.R., & Henry, J.D. (2004). The Positive andNegative risk. Archives Crawford, 54, 337-343. ofGeneral Psychiatry, Affect Schedule(PANAS): Construct measurement Kropp,P. R., & Hart,S. D. (2000). The Spousal AssaultRisk validity, and normative data in a largenon-clinical Assessment andvalidity inadult male (SARA)guide: properties sample. Reliability offenders. Law and Human Behavior,24, 101-118. doi: British Journalof Clinical Psychology, 43, 245-265. doi: 10.1348/0144665031752934. 10.1023/A: 1005430904495. J. L. (2005). Violenceriskassessment: Kropp, P. R.,Hart S. D., Webster C. D., & Eaves,D. (1999).Manual Douglas,K. S., & Skeem, about Public assault risk assessment Getting specific being dynamic. Psychology, Policy, forthespousal guide(3rded.).Toronto, andLaw,11,347-383.doi:10.1037/1076-8971.11.3.347. Canada:Multi-Health Systems. R. D., in thesocial sciences: P., Walter, Menard, Fliege,H., Rose, M., Arck, O. B., Kocalevent, S. (1991). Quantitative applications B. F. (2005).ThePerceived Stress research. Park:Sage Publications. Weber, C., Klapp, QuestionLongitudinal Newbury naire Validation andreference valuesfrom Mills, J. F., & Kroner, D. G. (2005). An investigation intothe (PSQ) reconsidered: different clinicaland healthy adult samples.Psychosomatic between desirable andoffender relationship socially responding Medicine, 67,78-88.doi:10.1097/01.psy.0000151491.80178.78. self-report.Psychological Services, 2, 70-80. doi: 10.1037/1541-1559.2.1.70. & Atkinson, J.L. (2007).Thegeneralizability oftheLSIFolsom, J., R and the CAT to the prediction of recidivism in female Mills,J.F., Loza, W., & Kroner, D. G. (2003). Predictive validity offenders. Criminal Justice andBehaviour, doi: 34, 1044-1056. socialdesirability: Evidence fortherobustness of selfdespite 10.1177/0093854807300097. offenders. Criminal Behaviour andMental Health, report among P. (2002).Is the PCL-Rreally the 13, 140-150. doi:10.1002/cbm.536. Gendreau, P.,Goggin, C., & Smith, measure ofoffender risk? A lesson inknowledge Monahan, J. (1981). The clinicalprediction behaviour. "unparalleled" of violent accumulation. Criminal Justice and Behaviour, 29, 397-426. DC: Government Office. Washington, Printing doi:10.1177/0093854802029004004. L. L. (1991). Antecedents and consequences Motiuk, of prison C. (1996).A meta-analysis ofthe A systematic assessmentand re-assessment Gendreau, P.,Little, T., & Goggin, adjustment: of adult offender recidivism: What works! CriminolDoctoral Carleton UniverDissertation, predictors approach. Unpublished Ontario. 34, 575-607.doi:10.111 Ottawa, ogy, l1/j.1745-9125.1996.tb01220.x. sity, L. L. (1999). Assessment G. (1998).Associates andsocial Motiuk, methods in corrections. In P. M. C., Gendreau, P., & Gray, Goggin, interaction. Forum on Corrections Research, Harris to results: correc10(3),24-27. (Ed.), Research community Effective & McNeil, B.J. ofcomparing the areas tions:Proceedings (1983).A method Hanley, J.A. of the 1995 and 1996 conferences of the under Receiver Characteristic curves derived from the International Corrections Association Operating (ICCA) (pp. Community samecases.Radiology, MD: American Correctional Association. 148,839-843. Lanham, 171-190). R. K., & Harris, A. J. R. (1998). Dynamic of L. (2002). The Statistical on Hanson, Nafekh, M., & Motiuk, predictors Information 1 (SIR-RI)scale: A psychometric sexual recidivism Ottawa: of Recidivism Revised examina1998-01). (UserReport Department theSolicitor General ofCanada. tion ON: Correctional Service (Research Ottawa, R-126). Report ofCanada. K. R., & Harris, A. J.R. (2000). Where do we intervene? Hanson, ofsexual offense recidivism. Criminal Justice Nemanick, R. C., & Munz,D. C. (1994). Measuring thepoles of Dynamic predictors andBehavior, 27,6-35.doi:10.1177/0093854800027001002. andpositive moodusing thePositive Affect negative Negative Springer

This content downloaded from 111.68.111.198 on Mon, 20 May 2013 12:30:11 PM All use subject to JSTOR Terms and Conditions

Law HumBehav(2009) 33:25-45

45

V. L., Jones, K. N. (2006).The B. G., Book,A. S., & Barr, Affect Schedule AndActivation Deactivation Check- Quinsey, Adjective list.Psychological ofantisocial behavior forensic 74, 195-199. Reports, dynamic prediction among psychiatric A prospective field Journal P. M. (2000). The use of detention Factors ofInterpersonal legislation: patients: study. Nugent, doi:10.1177/0886260506294238. detention decisions and recidivism Violence, 21, 1539-1565. amonghigh-risk affecting R. T., Rogers, and inOntario. Doctoral R., & Sewell,K. W. (1996). A review Dissertation, Salekin, offenders federal Unpublished of thePsychopathy Checklist and Psychopathy ON. Queen'sUniversity, meta-analysis Kingston, in Canada: Research Checklist-Revised: Predictive of Dangerousness. ClinJ. (1982). Parole decision-making Nuffield, Validity ical Psychology: Science andPractice, from 340 toward decision 3, 203-215. (Available guidelines publicsafety, Laurier Canada:KiA OP9). for West, Ottawa, Ontario, Serin,R. C., & Brown,S.L. (2001). Empirical guidelines W. T. (1997).Paroleprediction current clinicians: thepractice of offender riskassessment. Palmer, Enhancing using psychological A longitudinal criterion and event and behavioural predictors: Unpublished publication. J.A. (1986).Indices ofdiscrimination ordiagnostic DoctoralDissertation, analysis. Queen's Swets, accuracy: history Unpublished TheirROCs and implied models.Psychological ON. Bulletin, 99, Kingston: University, 100-117.doi:10.1037/0033-2909.99.1.100. D. L. (1994). Balancedinventory Paulhus, ofdesirable responding. J. (2000). Psychological J.A., Dawes,R. M., & Monahan, manual 6. Unpublished Reference forBIDR version Manuscript, Swets, science canimprove decisions. Science ofBritish BC. Columbia, Psychological diagnostic University in thePublicInterest, 1, 1-26.doi:10.1111/1529-1006.001. Paulhus,D. L. (1998). Paulhus DeceptionScales (PDS): The user'smanual. Watson, L. A., & Tellegen, A. (1988).Development and balancedinventory D., Clark, of desirable responding-7 validation ofbrief measures ofpositive andnegative affect: The ON: Multi-Health Inc. Toronto, Systems and Social Psychology, PANAS scales.Journal I. (1985).Analyzing Measurement and explanation Plewis, ofPersonality change: data.Chichester, 54, 1063-1070. doi:10.1037/0022-3514.54.6.1063. using longitudinal England: Wiley. offender assessment: Issues and of V. L., Book,A., & Skilling, T. A. (2004).A follow-up Wormith, S. (2005). Re-assessing Quinsey, deinstitutionalized menwith intellectual disabilities andhistories theirsolutions in criminal justice. Journalof Community inIntellectual ofantisocial behaviour. Journal Research Corrections, XIV(3),6-8, 17-30. ofApplied J.S., & Goldstone, C. S. (1984).Theclinical andstatistical doi:10.1111/j.1468-3148.2004.00216.x. Wormith, Disabilities, 17,243-253. V. L., Coleman,G., Jones, I. (1997). ofrecidivism. Criminal Justice andBehavior, B., & Altrows, 11,3Quinsey, prediction 34. doi:10.1177/0093854884011001001. Proximal antecedents of eloping andreoffending among superE. (1989).Coping situation Test. visedmentally disordered offenders. Journal questionnaire. of Interpersonal Zamble, Unpublished ON. Violence, 12,794-813.doi:10.1177/088626097012006002. Queen'sUniversity, Kingston, index.Unpublished Test. V. L., Harris, G. T., Rice,M. E, & Cormier, C. A. (1998). Zamble,E. (1998). Perceived problem Quinsey, ON. and managing risk. Violent Queen'sUniversity, Kingston, offenders: Appraising Washington, F. J. (1988). Coping,behaviour, and Association. DC: American Zamble,E., & Porporino, Psychological inprison inmates. NewYork:Springer-Verlag. V. L., Harris, G. T., Rice,M. E., & Cormier, C. A. (2006). adaptation Quinsey, V. L. (1997). The processof criminal Violent and managing risk(2nd ed.). Zamble,E., & Quinsey, offenders: Appraising recidivism. Press. DC: American Association. Cambridge: Cambridge University Washington, Psychological

Springer

This content downloaded from 111.68.111.198 on Mon, 20 May 2013 12:30:11 PM All use subject to JSTOR Terms and Conditions

Você também pode gostar