Accepted for publication in Psychological Assessment

3 downloads 0 Views 221KB Size Report
conducted in the United States (e.g., Onifade et al., 2008), United Kingdom (e.g., Marshall, Egan,. English, & Jones, 2006), Canada (e.g., Schmidt, Campbell, ...
Accepted  for  publication  in  Psychological  Assessment   Running  head:  ASSESSING  YOUTH  OFFENDERS       Assessing  Youth  Offenders  in  a  Non-­‐Western  Context:     The  Predictive  Validity  of  the  YLS/CMI  Ratings       Chi  Meng  Chua,b,  Yirong  Leea,  Gerald  Zengb,  Grace  Yimc,  Chen  Yeh  Tanc,  Yaming  Angd,  Shannon   Chine,  &  Kala  Rubyc       aClinical  and  Forensic  Psychology  Branch,  Ministry  of  Social  and  Family  Development,  

Singapore   bCentre  for  Research  on  Rehabilitation  and  Protection,  Ministry  of  Social  and  Family  

Development,  Singapore     cProbation  Services  Branch,  Ministry  of  Social  and  Family  Development,  Singapore   dFreelance  researcher,  Singapore   eCommunity  Rehabilitation  and  Support  Service,  Singapore  Anglican  Community  Services,  

Singapore     Correspondence  should  be  addressed  to  Dr.  Chi  Meng  Chu  at:   Centre  for  Research  on  Rehabilitation  and  Protection   Ministry  of  Social  and  Family  Development   512  Thomson  Road,  MSF  Building,  12th  Floor,  Singapore  298136   Email:  [email protected]     It  is  the  policy  of  APA  to  own  the  copyrights  to  its  publications,  and  to  the  contributions  contained  therein,  in  order  to  protect  the   interests  of  the  Association,  its  authors,  and  their  employers,  and  at  the  same  time  to  facilitate  the  appropriate  reuse  of  this  material   by  others.  In  exercising  its  rights  under  U.S.  copyright  law,  APA  requires  authors  to  obtain  APA’s  permission  and/or  pay  APA  a  fee  to   republish  or  reuse  the  author(s)’  manuscript  under  certain  circumstances.    This  article  may  not  exactly  replicate  the  final  version   published   in   the   APA   journal.   It   is   not   the   copy   of   record.   To   access   the   final   version,   please   use   the   following   DOI:   http://dx.doi.org/10.1037/a0038670  

Assessing  Youth  Offenders   Authors’  Note   The  views  expressed  are  those  of  the  authors  and  do  not  represent  the  official  position  or   policies  of  the  Ministry  of  Social  and  Family  Development.       Acknowledgements   The  authors  would  like  to  thank  the  staff  of  Clinical  and  Forensic  Psychology  Branch  and   Probation  Services  Branch  of  the  Ministry  of  Social  and  Family  Development  for  their  support.   In  addition,  we  would  like  to  express  gratitude  to  Ms.  Jennifer  Teoh,  Ms.  Bernadette  Alexander,   Ms.  Aileen  Tan,  Mr.  James  Loh,  Mr.  Alvin  Koh,  Ms.  Amelia  Wong,  Mr.  Han  Siang  Lim,  Ms.  Cheryl   Tan,  Ms.  Amanda  Tan,  and  Ms.  Melissa  Yeo  for  their  facilitation  at  different  stages  of  the  project.       Declaration  of  interest:  None      

Assessing  Youth  Offenders   Abstract   Empirical  support  for  the  usage  of  the  Youth  Level  of  Service  measures  has  been  reported  in   studies  conducted  in  the  North  America,  United  Kingdom,  and  Australia.  Recent  meta-­‐analytic   studies  on  the  Youth  Level  of  Service/Case  Management  Inventory  (YLS/CMI)  have  revealed   that  the  measure  has  modest  to  moderate  predictive  validity  for  general  recidivism,  but  there   are  very  few  studies  on  the  predictive  validity  of  the  YLS/CMI  ratings  for  recidivism  in  non-­‐ Western  contexts.  This  study  examined  the  predictive  validity  of  the  YLS/CMI  2.0  ratings  for   general  recidivism  in  a  sample  of  3,264  youth  offenders  within  a  Singaporean  context  (Mfollow-­‐up  =   1,764.5  days;  SDfollow-­‐up  =  521.5).  Results  showed  that  the  YLS/CMI  2.0  overall  risk  ratings  and   total  scores  significantly  predicted  general  recidivism  for  both  male  and  female  youth  offenders.   Overall,  the  results  suggest  that  the  YLS/CMI  2.0  is  suited  for  assessing  youth  offenders  in  terms   of  their  risk  for  general  recidivism  within  a  non-­‐Western  context.    

      Keywords:  

Strengths,  Risk  Assessment,  Risk-­‐Needs-­‐Responsivity,  Predictive  Validity,   Protective  Factors  

                   

Assessing  Youth  Offenders   Assessing  Youth  Offenders  in  a  Non-­‐Western  Context:     The  Predictive  Validity  of  the  YLS/CMI  Ratings   Introduction   With  increasing  demand  for  resources  as  well  as  the  need  to  provide  evidence  and  to   ensure  accountability  for  the  efficacy  of  offender  rehabilitation,  using  offender  risk  assessment   measures  to  inform  classification  and  management  decisions  is  a  critical  component  of  any   offender  rehabilitation  practice.  As  such,  empirically  reliable,  valid,  and  culturally-­‐sensitive  risk   assessment  measures  that  systemically  assess  the  risk  and  needs  of  offenders  are  paramount  to   the  success  of  offender  rehabilitation  including  those  of  the  youth  offenders.   Risk-­‐Needs-­‐Responsivity  Framework  and  Risk  Assessment  Measures   According  to  the  Risk-­‐Needs-­‐Responsivity  (RNR)  framework  (Andrews  &  Bonta,  2010),   effective  offender  rehabilitation  requires  the  accurate  classification  of  the  offender’s  level  of   risk  and  needs.  With  accurate  identification  and  classification  of  risk  and  needs,  clinicians  can   make  informed  decisions  about  the  levels  of  supervision,  as  well  as  the  type  and  intensity  of  the   interventions  that  should  be  provided.  The  framework  also  states  that  intervention  should   target  those  criminogenic  needs  that  are  functionally  related  to  criminal  behavior.  Moreover,   the  RNR  framework  also  posits  that  the  style  and  mode  of  intervention  should  match  the   offender’s  abilities  and  learning  style.  RNR  principles  have  been  shown  to  be  important  in  both   offender  assessment  and  intervention  domains,  and  the  Level  of  Service  risk  assessment   measures  (including  the  Youth  Level  of  Service/Case  Management  Inventory  [YLS/CMI;  Hoge  &   Andrews,  2002,  2011])  are  the  most  widely  used  products  of  the  RNR  (Andrews,  Bonta,  &   Wormith,  2010).  In  the  past,  assessments  of  risk  and  needs  were  often  based  on  unstructured   clinical  judgments,  and  such  a  decision-­‐making  approach  was  criticized  as  being  inaccurate   (Ægisdóttir  et  al.,  2006;  Grove,  Zald,  Lebow,  Snitz,  &  Nelson,  2000;  Monahan,  1981).  However,   risk  assessment  practices  have  advanced  over  the  past  three  decades,  and  there  is  a  greater   reliance  on  risk  assessment  measures  that  are  structured  and  empirically  based.  These   structured  risk  assessment  measures  have  been  found  to  provide  valid  and  consistent  

Assessing  Youth  Offenders   assessments  of  risk  for  future  offending  behavior  and  potential  intervention  needs  (Hoge,   2002),  and  the  YLS/CMI  represents  one  of  the  most  widely  used,  structured  risk  assessment   measures  for  assessing  the  risk  of  general  recidivism  and  criminogenic  needs  in  youth   offenders.     The  YLS/CMI  is  comprised  of  static  and  dynamic  risk  factors  that  are  associated  with   reoffending;  in  particular,  static  risk  factors  are  variables  that  are  not  amenable  to  change   through  planned  intervention  over  time.  Hence,  these  static  factors  are  unlikely  to  be   ameliorated  for  purpose  of  managing  or  reducing  the  risk  of  future  offending  over  time   (Douglas  &  Skeen,  2005).  In  contrast,  dynamic  risk  factors  for  reoffending  (also  known  as   criminogenic  needs)  fluctuate  with  time  and  circumstances.  The  dynamic  risk  factors  can  be   changed  as  a  result  of  deliberate  intervention  (Webster,  Douglas,  Belfrage,  &  Link,  2000),  and   hence  reducing  the  overall  level  of  risk  for  future  offending.  The  YLS/CMI  assists  the   practitioners  to  reliably  assess  risk  factors  and  criminogenic  needs,  and  provide  much  needed   information  to  manage  risk  and  target  the  relevant  areas  for  intervention.  In  addition  to  these   risk  factors,  the  YLS/CMI  has  an  accompanying  case  management  component  to  guide  the   practitioners  to  make  recommendations  and  also  account  for  other  factors  that  may  affect  the   successful  management  of  the  youth  offender.     Furthermore,  the  YLS/CMI  also  allows  for  the  assessment  of  strengths  in  the  youth   offenders,  and  such  an  assessment  can  provide  essential  information  to  risk  assessment  and   intervention  planning.  Recent  studies  have  emphasized  on  the  utility  of  protective  factors  for   mitigating  the  level  of  risk  that  an  individual  poses,  and  also  possibly  reducing  the  likelihood  of   recidivism  (e.g.,  Hartman,  Turner,  Daigle,  Exum,  &  Cullen,  2009;  Lodewijks,  de  Ruiter,  &   Doreleijers,  2010).  However,  there  are  few  studies  on  the  predictive  utility  of  strengths  or   protective  factors  in  youth  risk  assessment  measures,  namely  the  Structured  Assessment  of   Violence  Risk  in  Youth  (SAVRY;  e.g.,  Lodewijks,  de  Ruiter,  &  Doreleijers,  2010;  Rennie  &  Dolan,   2010),  and  no  published  study  has  systematically  examined  the  YLS/CMI  strength  ratings  yet.    

Assessing  Youth  Offenders   Predictive  Validity  of  the  Ratings  for  YLS  Measures  across  Various  Countries   Empirical  support  for  the  usage  of  the  YLS  measures  have  been  reported  in  studies   conducted  in  the  United  States  (e.g.,  Onifade  et  al.,  2008),  United  Kingdom  (e.g.,  Marshall,  Egan,   English,  &  Jones,  2006),  Canada  (e.g.,  Schmidt,  Campbell,  &  Houlding,  2011),  Australia  (e.g.,   McGrath  &  Thompson,  2012),  Japan  (Takahashi,  Mori,  &  Kroner,  2013),  and  Singapore  (Chu,  Ng,   Fong,  &  Teoh,  2012).  Recent  meta-­‐analytic  studies  on  the  YLS  measures  have  revealed  that  its   ratings  have  modest  to  moderate  predictive  validity  for  general  recidivism.  In  particular,   Schwalbe  (2007)  found  a  mean  weighted  area  under  curve  (AUC)  of  .641  based  on  a  review  of   11  YLS  studies.  As  a  general  rule  for  practice,  AUCs  greater  than  .54,  .63,  and  .71,  as  well  as   correlation  coefficients  (r)  that  are  greater  than  .10,  .24,  and  .37,  are  regarded  as  small,   moderate,  and  large  effects  respectively  (Rice  &  Harris,  2005).  In  an  overlapping  but  larger   sample  of  19  studies,  the  mean-­‐weighted  correlation  between  YLS  total  scores  and  general   recidivism  was  .32  (Olver,  Stockdale,  &  Wormith,  2009).  In  their  meta-­‐analysis,  Olver  et  al.  also   showed  that  the  ratings  of  the  YLS  measures  had  lower  predictive  validity  for  general   recidivism  when  they  were  used  in  other  western  contexts  outside  of  Canada  (mean-­‐weighted   correlation  of  .26  vs.  .35).  Olver  and  colleagues  suggested  that  “‘international’  differences   contributed  to  the  variability  across  studies”  (p.  348).     Further  examination  of  such  variability  across  contexts  with  regard  to  the  predictive   validity  of  the  ratings  for  the  Level  of  Service  risk  assessment  measures  revealed  that  the   location  effect  is  a  function  of  the  authors’  allegiance  (i.e.,  being  a  Canadian,  which  reflects  the   integrity  of  the  assessment),  and  the  reliability  of  the  outcome  measure(s)  (Andrews  et  al.,   2011).  This  finding  is  also  reflected  in  other  risk  assessment  measures  that  are  developed  in   Canada  (see  Harris,  Rice,  &  Quinsey,  2009;  Olver  et  al.,  2009,  Yang,  Wong,  &  Coid,  2010).   Moreover,  the  generalizability  of  the  original  development  sample  to  the  samples  used  in   subsequent  studies  does  affect  the  predictive  validity.  In  other  words,  there  may  be  a  decrease   in  true  predictive  validity  of  the  ratings  for  a  risk  assessment  measure  as  “it  transverses   national,  hence  legal,  boundaries”  (Andrews  et  al.,  2011,  p.  426).  Differences  in  legislation,  

Assessing  Youth  Offenders   definitions  of  outcomes,  interpretation  of  the  criteria  for  rating  of  the  risk  assessment  measure   will  affect  its  predictive  validity  when  used  in  different  contexts.     To  the  best  of  the  authors’  knowledge,  there  are  only  two  published  studies  on  the   predictive  validity  of  the  YLS/CMI  ratings  for  various  recidivistic  outcomes  within  non-­‐Western   contexts  (Chu  et  al.,  2012;  Takahashi  et  al.,  2013).  It  was  noted  that  Chu  et  al.’s  study  (N  =  104)   was  restricted  to  youth  who  sexually  offended  in  a  Singaporean  context;  notably,  the  YLS/CMI   (total  score)  was  useful  in  predicting  nonsexual  recidivism  for  youth  who  sexually  offended   (AUC  =  .65),  but  was  limited  in  predicting  sexual  recidivism  (AUC  =  .29).  On  the  other  hand,   Takahashi  et  al.  (N  =  389)  found  the  YLS/CMI  ratings  had  wide-­‐ranging  predictive  validity  for   nonviolent,  violent,  and  general  recidivism  for  both  community-­‐based  and  institutionalized   juvenile  offenders  (AUCs  =  .50  to  .87);  in  particular,  shorter  follow-­‐up  periods  and  assessments   for  community-­‐based  juvenile  offenders  resulted  to  higher  predictive  validity  for  recidivistic   outcomes,  especially  in  terms  of  nonviolent  and  general  recidivism.  Notwithstanding  these  two   studies,  there  is  generally  a  dearth  of  published  studies  on  the  predictive  validity  of  the  ratings   for  the  YLS/CMI  subscales  within  non-­‐Western  contexts.   Application  of  the  RNR  Framework  and  YLS/CMI  in  Singapore   Singapore  is  an  independent  island-­‐state  in  South  East  Asia  with  a  total  population  of   5.4  million  (Singapore  Department  of  Statistics,  2013).  Pertaining  to  crime  statistics,  youth   arrests  accounted  for  about  10%  of  all  arrests  in  Singapore  (Singapore  Police  Force,  2013).   Many  statutes  in  Singapore  are  based  on  English  common  law  (e.g.,  the  Criminal  Procedure   Code,  2012),  but  there  are  some  statutes  that  are  based  on  legislation  from  other  jurisdictions;   for  example,  the  Penal  Code  (2008)  is  based  on  the  Indian  Penal  Code,  which  was  (nonetheless)   first  formulated  by  the  English  in  1800s.  As  such,  there  are  similarities  in  the  way  that  offenses   are  defined  in  Singapore  when  compared  with  the  abovementioned  countries,  but  the  exact   language  of  the  laws  might  vary  somewhat.  In  particular,  cultures  and  societies  often  define   what  attitudes  and  behaviors  are  considered  “normal”  and  “deviant.”  Although  there  is  some   agreement  across  cultures  about  what  constitutes  offending  behavior,  the  development  of  

Assessing  Youth  Offenders   deviant  attitudes  and  behaviors  can  differ  due  to  cultural  norms,  gender  roles,  morals,  religion,   taboos,  and  expectations  (e.g.,  Bhugra,  Popelyuk,  &  McMullen,  2010;  Lahlah,  Van  der  Knaap,   Bogaers,  &  Lens,  2013).   It  is  possible  that  the  motivation,  risk  factors,  and  pathways  for  offending  may  differ   cross-­‐culturally  due  to  cross-­‐cultural  differences  as  to  how  individuals  cope,  self-­‐regulate,  or   even  report  crime.  Therefore,  it  may  be  necessary  to  examine  the  empirical  evidence  whenever   there  is  any  adaptation  of  assessment  and  intervention  frameworks  (that  are  developed  for  the   Western  contexts)  into  non-­‐Western  contexts,  as  well  as  the  accompanying  measures  (e.g.,   measures  might  need  new  norms  or  cut-­‐offs  that  would  suit  the  new  context  due  to  the   aforementioned  factors).   In  the  early  2000s,  there  was  a  collective  move  by  the  youth  and  adult  correctional   services  toward  using  structured  and  empirically-­‐informed  approach  with  regard  to  assessing   youth  offenders’  risk  and  needs.  The  RNR  framework  was  introduced  in  Singapore  to  provide  a   theoretical  and  empirical-­‐based  approach  to  conduct  offender  assessment  and  rehabilitation.   Importantly,  the  Youth  Level  of  Service/Case  Management  Inventory  (YLS/CMI  and   subsequently  the  YLS/CMI  2.0)  was  chosen  as  the  primary  risk  assessment  measure  to  assess   the  risk  and  needs  of  youth  offenders  (Chua,  Chu,  Yim,  Chong,  &  Teoh,  2014).  Within  the   Singaporean  context,  YLS/CMI  2.0’s  coding  criteria  and  cut-­‐offs  for  risk  categories  were   modified  and  developed  following  its  introduction  in  2011,  respectively,  for  assessing  the  local   youth  offenders  and  in  accordance  with  the  local  legislation  and  procedures.   Present  Study    

Considering  that  there  is  currently  limited  empirical  knowledge  pertaining  to  the  

YLS/CMI  2.0’s  predictive  validity  for  recidivistic  outcomes  within  non-­‐Western  contexts,  the   present  study  sought  to  examine  the  predictive  validity  of  the  YLS/CMI  2.0  total  scores  and   overall  risk  ratings  for  general  recidivism  in  Singapore  using  a  large  sample  of  youth  offenders.   In  addition,  the  present  study  sought  to  examine  the  predictive  validity  of  the  YLS/CMI  2.0  

Assessing  Youth  Offenders   subscale  ratings  for  general  recidivism,  as  well  as  the  association  between  general  recidivism   and  (1)  the  strength  ratings,  and  (b)  other  needs/special  considerations.   Method   Source  Sample   The  sample  consisted  of  3,264  youth  (aged  12  –  18  years)  who  were  convicted  of   criminal  offenses.  They  were  referred  to  the  Probation  Services  Branch  of  the  Ministry  of  Social   and  Family  Development  (Singapore)  between  January  2004  and  December  2008,  and  were   placed  on  community  supervision  following  their  court  sentencing.  The  mean  age  of  these  youth   at  referral  to  the  Probation  Services  Branch  was  15.42  years  (Mdn  =  15.00;  SD  =  1.19),  and  the   large  majority  of  the  youth  were  males  (90.4%,  2,951/3,264).  Slightly  more  than  half  of  the   youth  were  Chinese  (53.6%,  1,749/3,264);  31.9%  were  Malay  (1,042/3,264),  9.3%  (303/3,264)   were  Indian,  and  5.2%  (170/3,264)  were  of  other  ethnicity.  The  current  sample  included  96.9%   (3,264/3,370)  of  the  youth  offenders  who  were  placed  on  community  supervision  during  this   period;  the  remaining  could  not  be  coded  as  a  result  of  missing  information  or  file  retrieval   difficulties.  In  terms  of  offense  characteristics,  the  mean  number  of  index  offenses1  committed   was  2.61  (SD  =  2.82,  range  =  1  –  40);  31.6%  (1,030/3264)  had  committed  index  violent   offense(s),  2.1%  (69/3,264)  had  committed  index  sexual  offense(s),  and  78.6%  (2,564/3,264)   had  committed  index  nonviolent  nonsexual  offense(s).  A  small  minority  of  the  sample  (1.9%,   63/3,264)  had  a  prior  offense  history2  as  indicated  on  criminal  records.   Measures     YLS/CMI  2.0  (Hoge  &  Andrews,  2011).  The  YLS/CMI  2.0  is  a  structured  assessment   instrument  designed  to  facilitate  the  effective  intervention  and  rehabilitation  of  youth  who  have   committed  criminal  offenses  (aged  12  –  18  years)  by  assessing  their  risk  level,  criminogenic   needs,  and  strengths.  It  consists  of  42  items  (scored  as  either  Present  or  Absent)  that  are  divided   into  eight  subscales  (Prior  or  Current  Offenses/Dispositions,  Family  Circumstances/Parenting,  

                                                                                                                1  Index  offense(s)  refers  to  the  offense(s)  that  the  youth  was  charged  with  and  convicted  of,  when  they  first  came  into  

contact  with  the  juvenile  justice  system  during  this  period  (i.e.,  2004  to  2008).   2  Prior  offense  history  refers  to  the  youth’s  history  of  convictions,  as  indicated  in  official  records.  The  prior  offense   history  did  not  include  those  offenses  that  the  youth  were  not  charged  with  as  a  result  of  diversionary  procedures.  

Assessing  Youth  Offenders   Education/Employment,  Peer  Relations,  Substance  Abuse,  Leisure/Recreation,   Personality/Behavior,  and  Attitudes/Orientation).  The  item  scores  (i.e.,  the  number  of  indicated   risk  factors/needs)  can  be  aggregated  to  obtain  a  total  risk/needs  score.  In  addition  to  the  eight   subscales,  the  YLS/CMI  2.0  also  consists  of  items  that  pertain  to  noncriminogenic  needs  and   responsivity  factors,  which  can  be  rated  as  present  or  absent.  It  should  be  noted  that  the   YLS/CMI  2.0  coding  descriptions  for  some  items  were  adapted,  with  consultation  from   Professor  Robert  Hoge,  to  suit  the  Singaporean  context.  For  example,  for  the  items  in  Prior  or   Current  Offenses/Dispositions  subscale,  the  term  “convictions”  was  localized  to  follow  legal   terminology  in  Singapore.  Similarly,  for  the  items  in  Family  Circumstances/Parenting,   Education/Employment  subscale,  school  and  work  contexts  were  localized  to  include  the   learning  centers  in  youth  correctional  institutions  and  also  compulsory  military  service  for  male   youth.  Localized  examples  were  also  included  in  the  item  descriptions  for  the  various  subscales   to  assist  with  the  ratings.   With  regard  to  the  determination  of  the  cut-­‐off  scores,  the  distributions  of  scores  from   the  original  normative  samples  were  considered  (as  advised  by  Professor  Hoge).  In  addition,   the  probability  of  future  recidivism  for  individuals  with  particular  scores,  as  well  as  the   specificity  and  sensitivity  of  the  scores  were  also  taken  into  account  in  order  to  fine-­‐tune  the   cut-­‐offs  for  the  Singaporean  male  and  female  youth  offender  samples.  The  cut-­‐off  scores  of  the   risk  bins  for  the  male  youth  offenders  under  community  supervision  in  Singapore  are:  0  to  10   (Low),  11  to  19  (Moderate),  20  to  26  (High),  and  27  to  42  (Very  High).  On  the  other  hand,  the   cut-­‐off  scores  of  the  risk  bins  for  the  female  youth  offenders  on  community  supervision  in   Singapore  are:  0  to  12  (Low),  13  to  19  (Moderate),  and  20  to  42  (High).  There  is  no  Very  High   risk  bin  for  the  female  youth  offenders  on  community  supervision.  As  the  proportion  of  the   youth  offenders  who  were  assessed  as  High  and  Very-­‐High  risk  were  small  in  this  sample   (4.0%),  we  have  decided  to  group  these  two  categories  together  for  purpose  of  analyses.   Although  the  predictive  validity  of  the  Overall  Risk  Rating  was  examined  in  this  study,  it  should   be  noted  that  the  raters  had  not  utilized  any  professional  override  to  change  the  Overall  Risk  

Assessing  Youth  Offenders   Rating.   Procedure   The  approval  for  the  current  research  study  was  obtained  from  the  Ministry  of  Social   and  Family  Development  before  the  commencement  of  the  study.  For  the  purpose  of  this  study,   two  psychologists,  one  probation  officer,  as  well  as  five  research  assistants  had  conducted   clinical  file  reviews  between  January  2011  and  September  2012.  They  had  completed  coding  for   the  YLS/CMI  2.0  based  on  file  information  available  at  the  time  of  the  initial  assessment  at  the   presentencing  stage;  information  available  subsequent  to  the  presentencing  stage  were  not   considered  for  coding  purposes  to  minimize  criterion  contamination.  As  such,  this  is  a   prospective  study  (i.e.,  using  data  from  the  time  of  the  index  offense  to  predict  events  that   occurred  after  the  index  offense)  with  the  data  coded  retrospectively.  These  raters  had  attended   a  3-­‐day  training  program  conducted  by  accredited  trainers;  the  training  program  involved   lectures,  discussions,  case  studies  and  scoring  practices,  as  well  as  a  test.  The  clinical  files  that   were  obtained  from  Probation  Services  Branch  contained  (a)  psychological  reports  prepared  by   psychologists  at  CFPB;  (b)  presentence  reports  prepared  by  probation  officers;  (c)  charge   sheets;  (d)  statement  of  facts;  (e)  any  previous  assessment  and  treatment  reports;  as  well  as  (f)   school  reports.       Psychological  and  presentence  reports  contain  specific  information  pertaining  to   several  key  areas  of  assessment  (i.e.,  personal,  family,  psychiatric,  and  criminal  offending   histories  as  well  as  the  current  offending  behaviors  and  risk  management  issues).  Specifically,   these  areas  yielded  valuable  information  about  the  youth’s  upbringing,  interaction  with  peers   and  authorities,  general  and  academic  functioning,  values,  family  and  school  environments,  as   well  as  information  relating  to  the  youth’s  offending,  treatment,  management/supervision,  and   responsivity  issues.  To  examine  the  inter-­‐rater  reliability  for  the  ratings  pertaining  to  the   YLS/CMI  2.0,  the  raters  had  separately  coded  a  randomly  selected  sample  of  31  files,  and  the   intraclass  correlation  coefficients  for  single  rater  (using  absolute  agreement  definition;  ICCs)   were  .63  (good)  for  the  YLS/CMI  2.0  total  score  (see  Cicchetti,  1994,  for  a  classification  of  ICCs).  

Assessing  Youth  Offenders   The  ICCs  (single  rater,  absolute  agreement  definition)  for  the  eight  subscales  were  .43  (fair)  for   Prior  or  Current  Offenses/Dispositions,  .50  (fair)  for  Family  Circumstances/Parenting,  .60   (good)  for  Education/Employment,  .50  (fair)  for  Peer  Relations,  .49  (fair)  for  Substance  Abuse,   .45  (fair)  for  Leisure/Recreation,  .55  (fair)  for  Personality/Behavior,  and  .48  (fair)  for   Attitudes/Orientation.   In  terms  of  the  definition  of  the  recidivistic  outcome,  general  recidivism  refers  to  any   conviction  of  sexual  (e.g.,  indecent  exposure,  molestation,  peeping,  rape,  and  sodomy),  violent   (e.g.,  physical  assault,  rioting,  murder,  and  robbery),  nonviolent  nonsexual  (e.g.,  theft,  fraud,   burglary,  drug  use,  and  drug  trafficking)  offenses  that  were  committed  following  the  initial   court  order,  breaches  of  court  orders,  or  any  combination  of  the  aforementioned  outcomes.   Official  recidivism  data  were  only  obtained  following  the  completion  of  coding  of  all  other   variables  and  the  cut-­‐off  date  for  the  recidivism  data  was  20  April  2011.   Statistical  Analyses   The  sample  was  characterized  using  descriptive  statistics,  with  categorical  data   reported  as  numbers  and  percentages,  and  continuous  data  presented  in  relation  to  the  mean   and  standard  deviation.  Histograms  of  the  continuous  data  were  plotted  to  check  for  skewed   distributions.  Chi-­‐square  tests  of  association  were  computed  for  categorical  data,  and   correlational  analyses  were  also  conducted  to  examine  the  relationship  between  continuous   data  and  the  recidivistic  outcome  (dichotomous  data).  Receiver  Operating  Characteristics  (ROC)   analyses  were  conducted  to  examine  the  predictive  validity  of  the  YLS/CMI  2.0  total  scores,  and   Cox  regression  analyses  were  also  conducted  to  examine  whether  the  YLS/CMI  2.0  overall  risk   ratings  were  predictive  of  recidivistic  outcomes  whilst  accounting  for  differences  in  follow-­‐up   period.  Benjamini  and  Hochberg  False  Discovery  Rate  (FDR)  corrections  were  conducted  to   control  for  Type  I  error  that  may  arise  from  computing  multiple  comparisons;  specifically,  it  is  a   less  conservative  but  more  powerful  statistical  approach  than  Bonferroni-­‐type  adjustments   (Benjamini  &  Hochberg,  1995).  Effect  sizes  were  also  computed  to  demonstrate  the  strength  of   the  associations  between  variables.  Analyses  were  conducted  using  SPSS  version  19.  

Assessing  Youth  Offenders   Results   Recidivism  Data    

The  mean  follow-­‐up  period  was  1,764.5  days  (Mdn  =  1,762.5,  SD  =  521.5,  range  =  840  –  

2,666).  With  regard  to  the  recidivism  rates,  37.6%  (1,228/3,264)  of  the  current  sample  was   convicted  of  new  offense(s)  during  the  follow-­‐up  period;  33.5%  (1,095/3,264)  of  the  sample   were  convicted  of  new  nonviolent  nonsexual  offenses,  10.3%  (336/3,264)  violent  offenses,  and   0.5%  (17/3,264)  sexual  offenses.   YLS/CMI  2.0  Total  Score  and  Subscales     Table  1  shows  the  means  and  standard  deviations,  as  well  as  the  correlation  to  general   recidivism  (i.e.,  any  type  of  reoffense  and/or  breach  of  court  orders)  for  the  subscale  scores  of   the  YLS/CMI  2.0.  The  mean  total  score  of  the  YLS/CMI  2.0  for  the  overall  sample  was  11.78  (SD   =  4.10,  range  =  1  –  26),  and  the  correlation  between  the  YLS/CMI  2.0  total  score  and  general   recidivism  was  .24,  p  <  .001.  All  the  YLS/CMI  2.0  subscale  scores  were  also  significantly   correlated  to  general  recidivism  even  after  FDR  corrections.  Table  2  shows  the  breakdown  of   risk  categories  for  each  subscale.  In  terms  of  predicting  general  recidivism  in  the  overall   sample,  as  well  as  the  male,  and  female  subsamples,  the  AUCs  for  the  YLS/CMI  2.0  total  score   were  .64  (95%  Confidence  Interval  [95%  CI]  =  .62  -­‐  .66,  SE  =  0.01,  p  <  .001),  .65  (95%   Confidence  Interval  [95%  CI]  =  .63  -­‐  .67,  SE  =  0.01,  p  <  .001),  and  .67  (95%  Confidence  Interval   [95%  CI]  =  .60  -­‐  .73,  SE  =  0.03,  p  <  .001)  respectively.     Further,  Cox  regression  analyses  revealed  that  the  YLS/CMI  2.0  Overall  Risk  Ratings  of   Low,  Moderate,  and  High  were  significantly  different  from  each  other  in  the  overall  sample   (Hazard  RatioModerate-­‐Low  [HR]  =  2.09,  95%  CI  =  1.84  –  2.38,  p  <  .001;  HRHigh-­‐Low  =  3.61,  95%  CI  =   2.82  –  4.60,  p  <  .001;  HRHigh-­‐Moderate  =  1.72,  95%  CI  =  1.37  –  2.17,  p  <  .001),  as  well  as  male   (HRModerate-­‐Low  =  2.09,  95%  CI  =  1.83  –  2.38,  p  <  .001;  HRHigh-­‐Low  =  3.62,  95%  CI  =  2.79  –  4.71,  p  <   .001;  HRHigh-­‐Moderate  =  1.74,  95%  CI  =  1.36  –  2.22,  p  <  .001)  and  female  subsamples  (HRModerate-­‐Low  =   2.17,  95%  CI  =  1.35  –  3.39,  p  =  .001;  HRHigh-­‐Low  =  4.18,  95%  CI  =  2.04  –  8.54,  p  <  .001;  HRHigh-­‐ Moderate  =  1.92,  95%  CI  =  1.01  –  3.66,  p  <  .05).  

Assessing  Youth  Offenders   YLS/CMI  Strength  Ratings  and  Other  Needs/Special  Considerations   The  mean  number  of  strengths  was  0.29  (Mdn  =  0;  SD  =  0.62;  range  =  0  to  4);  the   majority  of  the  sample  was  not  rated  as  having  strengths  (78.9%,  2,576/3,264),  15%   (489/3,264)  had  one  strength,  4.7%  (152/3,264)  had  two,  1.4%  (45/3,264)  had  three,  and   0.2%  (2/3,264)  had  four.  There  was  also  a  significant  correlation  between  the  number  of   strengths  and  general  recidivism,  roverall  =  -­‐.14,  p  <  .001.  Examining  gender  differences,  the   correlation  between  the  number  of  strengths  and  general  recidivism  was  significant  for  the   male  subsample  (rmale  =  -­‐.15,  p  <  .001)  but  not  for  the  female  subsample  (rfemale  =  -­‐.02,  ns).  Table   3  shows  the  other  needs/special  considerations  that  are  significantly  associated  with   recidivism.  For  the  male  subsample,  11  variables  were  significantly  associated  with  general   recidivism  after  FDR  correction.  In  contrast,  none  were  significantly  associated  with  general   recidivism  for  the  female  subsample  after  FDR  correction.   Discussion   Predictive  Validity  of  the  YLS/CMI  2.0  Ratings  in  a  Non-­‐Western  Context   Apropos  of  overall  predictive  validity  of  the  YLS/CMI  2.0  ratings,  we  noted  that  ROC  and   Cox  regression  analyses  revealed  that  the  YLS/CMI  2.0  total  scores  and  overall  risk  ratings  were   predictive  of  general  recidivism  for  male  and  female  youth  offenders.  The  YLS/CMI  2.0  risk   categories  (using  Singapore  norms)  had  also  differentiated  the  Low-­‐,  Moderate-­‐,  and  High-­‐risk   groups,  with  each  group  having  significant  differences  in  terms  of  time  to  reoffense  -­‐  this   suggests  that  the  local  norms  have  sufficient  validity.  In  addition,  ROC  and  correlational   analyses,  with  a  mean  follow-­‐up  of  almost  5  years,  showed  that  the  YLS/CMI  2.0  total  scores   were  moderately  predictive  of  general  recidivism  (AUCoverall  =  .64,  roverall  =  .24;  AUCmale  =  .65,  rmale   =  .24;  AUCfemale  =  .67,  rfemale  =  .26).  These  indices  were  consistent  and  comparable  with  the   results  from  recent  meta-­‐analyses  examining  the  predictive  validity  of  the  YLS/CMI  2.0  ratings   from  non-­‐Canadian  jurisdictions  (Olver  et  al.,  2009;  Schwalbe,  2007),  but  were  significantly   lower  than  those  predictive  validity  indices  for  the  Japanese  community  subsample  in   Takahashi  et  al.’s  (2013)  study  (AUC  =  .76).  Despite  the  significant  predictive  utility  of  the  

Assessing  Youth  Offenders   YLS/CMI  2.0  ratings,  there  is  still  a  fair  amount  of  the  variance  in  the  recidivism  rate  between   the  offenders  that  is  unexplained.  The  unexplained  variance  may  be,  in  part,  explained  by   differences  in  intraindividual-­‐,  environmental-­‐  and  system-­‐level  variables  (e.g.,  developmental   changes  across  the  follow-­‐up  period,  differences  between  actual  and  reported  crimes  as  well  as   neighborhood  crime  rates  and  socioeconomic  factors)(Olver  et  al.,  2009;  Onifade,  Petersen,   Bynum,  &  Davidson,  2011).  Nevertheless,  this  study  has  yielded  a  set  of  variables  that  are   significantly  predictive  of  general  recidivism  in  the  Singaporean  context  over  a  substantial   follow-­‐up  period.   The  mean  total  score  of  this  sample  was  similar  to  the  mean  from  the  Japanese  study   (Takahashi  et  al.,  2013),  but  it  was  generally  lower  than  those  figures  reported  in  published   studies  from  Western  contexts  (e.g.,  Marshall  et  al.  2006;  McGrath  &  Thompson,  2012;  Olver,   Stockdale,  &  Wong,  2012;  Welsh  et  al.,  2008).  Apart  from  the  domains  of   Education/Employment  as  well  as  Leisure/Recreation,  the  means  of  the  subscales  were   generally  lower  than  those  taken  from  the  published  Western  studies.  However,  the  means  of   the  subscales  were  higher  than  those  from  the  Japanese  study  except  for  the  domains  of  Prior  or   Current  Offense/Disposition  and  Personality/Behavior.  Thus,  it  appears  that  the  observed   differences  are  likely  functions  of  the  characteristics  of  the  comparison  groups  (e.g.,  some   included  high-­‐risk,  institutionalized  youth  offenders,  and  those  with  mental  health  issues),   cross-­‐boundary  differences  (e.g.,  a  relatively  less  serious  youth  crime  situation  in  Singapore  as   compared  to  other  Western  contexts,  and  tough  laws  to  combat  substance  abuse),  and   differences  in  coding  criteria  of  the  YLS/CMI  (which  could  be  a  reflection  of  the  differences  in   the  legal  environments).  However,  the  scores  for  the  Prior  or  Current  Offense/Disposition   appeared  very  low  even  when  compared  to  the  mean  from  the  Japanese  study  –  the  low  prior   offense  rate  might  have  impacted  this,  and  that  range  restriction  of  scores  might  have  affected   the  validity  findings.  Overall,  this  suggests  that  a  reexamination  of  the  coding  criteria  may  be   necessary  to  obtain  more  variability  in  the  subscale  score,  which  may  in  turn  increase  the  utility   of  this  subscale  for  predicting  further  offenses.  

Assessing  Youth  Offenders   Pertaining  to  gender  differences,  the  present  study  showed  that  the  female  youth   offenders  were  rated  higher  on  the  YLS/CMI  2.0  total  score,  as  well  as  most  of  the  other   subscales  (apart  from  Prior  or  Current  Offenses/Dispositions)  as  compared  to  their  male   counterparts.  This  pattern  of  results  is  somewhat  similar  to  several  Western  studies  on  the   YLS/CMI  (e.g.,  Olver  et  al.,  2012;  Schmidt  et  al.,  2011;  Thompson  &  McGrath,  2012),  but  is   different  from  others  (e.g.,  Jung  &  Rawana,  1999;  Marshall  et  al.,  2006).  With  regard  to  the   YLS/CMI  2.0  subscales,  the  results  of  this  study  showed  that  eight  and  four  subscales  were   univariately  associated  with  general  recidivism  for  the  male  and  female  subsamples   respectively.  It  is  clear  that  this  finding  whereby  different  criminogenic  needs  are  predictive  of   recidivism  relates  the  Need  principle  in  the  RNR  framework,  and  that  targeting  these  needs  will   reduce  the  propensity  for  criminal  reoffending.  On  the  other  hand,  the  findings  on  gender   differences  relate  to  the  Responsivity  principle.     Compared  to  their  male  counterparts,  it  seems  that  the  female  youth  offenders  in   Singapore  have  a  higher  level  of  criminogenic  needs  (as  measured  on  the  YLS/CMI  2.0)  when   they  enter  the  juvenile  justice  system.  Possibilities  for  such  an  observation  could  be:  (a)  there   are  different  pathways  for  offending  across  gender,  which  would  be  linked  to  differences  in  risk   factors  for  offending  behavior;  and/or  (b)  many  “gender-­‐neutral”  assessment  measures  that   were  developed  for  males  had  limited  relevance  for  females,  thus  some  risk  factors  and  needs   that  are  most  relevant  to  female  offenders  might  have  been  omitted  or  ignored  (e.g.,  Taylor  &   Blanchette,  2009;  Salisbury  &  Van  Voorhis,  2009;  Van  Voorhis,  Wright,  Salisbury,  &  Bauman,   2010).  Such  possibilities  could  be  areas  of  future  empirical  inquiry  within  the  Singaporean   context,  and  there  appears  to  be  a  need  to  feature  gender-­‐responsive  risk  factors  in  risk   assessment  measures  too.  In  addition,  there  are  some  grounds  to  explore  gender-­‐responsive   interventions  for  youth  offenders  considering  the  findings  on  criminogenic  needs  in  the  present   study  (see  Table  3).   Notwithstanding  the  gender  differences,  we  hypothesized  that  the  other  domains  exert   their  influence  on  the  perpetuation  of  criminal  conduct  through  the  abovementioned  significant  

Assessing  Youth  Offenders   risk  factors  within  our  local  context,  but  this  is  an  empirical  question  that  needs  to  be  further   explored  using  structural  equation  modeling.  Although  the  static  domain  (i.e.,  Prior  or  Current   Offenses/Dispositions)  remained  as  a  significant  predictor  of  general  recidivism  in  this  study   (over  the  long  term),  it  is  clear  that  other  dynamic  variables  have  a  part  to  play  in  the  risk   assessment  of  youth  offenders.     In  terms  of  YLS/CMI  strength  ratings  and  other  needs/special  considerations  (i.e.,   responsivity  factors  and  other  needs),  we  believe  that  this  is  the  only  published  study  on  the   YLS/CMI  (to  the  best  of  the  authors’  knowledge)  that  have  examined  the  utility  of  these   components  in  addition  to  the  YLS/CMI  subscales.  The  number  of  strengths  in  the  male   subsample  had  an  inverse  univariate  relationship  with  general  recidivism  suggesting  that   protective  factors  that  buffered  the  youth  offenders  against  further  criminal  offending  behavior   albeit  a  relatively  small  effect  size.  However,  this  relationship  was  not  found  in  the  female   subsample  –  suggesting  that  there  could  be  different  protective  factors  across  gender.  Perhaps   strengths  also  may  interact  with  risk  factors  when  used  to  predict  (non)recidivism,  but  the   actual  processes  that  may  not  have  been  systematically  examined  here.  For  example,  protective   and  risk  factors  were  found  to  predict  desistance  within  3  years,  but  desistance  over  a  longer   period  was  negatively  predicted  by  only  risk  factors  (Stouthamer-­‐Loeber  et  al.,  2008),   suggesting  that  protective  factors  have  a  shorter  “shelf  life”  in  terms  of  predicting   (non)recidivism.  More  research  is  clearly  needed  in  this  area.     In  the  extant  literature  on  the  YLS/CMI,  very  little  empirical  work  has  been  published  on   the  other  needs/special  considerations.  One  recent  study  examined  whether  risk  and  need   assessment  is  linked  to  the  case  management  of  youth  offenders  as  well  as  whether  adherence   to  RNR  principles  in  case  management  is  related  to  recidivism  (Luong  &  Wormith,  2011),  but   this  study  did  not  examine  the  utility  of  the  other  needs/special  considerations.  The  present   study  suggests  that  other  needs/special  considerations  (e.g.,  chronic  history  of  family   criminality,  and  family  financial/accommodation  problems,  diagnosis  of  Conduct   Disorder/Oppositional  Defiance  Disorder,  gang  involvement,  and  history  of  running  away)  

Assessing  Youth  Offenders   variables  were  found  to  be  associated  with  general  recidivism  in  youth  offenders.  Some  of  these   variables  have  already  been  shown  to  be  associated  with  delinquent  or  offending  behavior   within  the  Singaporean  context  (Ang  &  Huan,  2008;  Chu,  Daffern,  Thomas,  &  Lim,  2012);   therefore,  the  CMI  component  allows  the  practitioners  to  consider  the  relevant  these  variables   that  may  otherwise  affect  the  assessment  of  risk,  as  well  as  the  offender  rehabilitation  process   (as  responsivity  issues).  Similar  to  the  YLS/CMI  strength  ratings,  more  research  needs  to  be   conducted  to  understand  the  utility  of  these  variables  better.     Limitations  and  Future  Directions   Firstly,  we  relied  on  the  electronic  data  and  archival  file  data  for  coding  of  the  risk   assessment  measures  and  recidivism  follow-­‐up,  hence  there  would  inevitably  be  an   underestimate  of  the  reoffending  due  to  the  further  offenses  not  having  been  detected.   Likewise,  the  retrospective  methodology  employed  in  the  present  study  would  have  also   underestimated  the  presence  of  risk  factors,  criminogenic  needs,  responsivity  factors,  as  well  as   strengths.  Secondly,  some  of  the  dynamic  risk  factors  and  protective  factors  might  have  lower   predictive  validity  over  the  longer  term,  so  it  is  possible  that  the  predictive  accuracy  of  the   YLS/CMI  2.0  ratings  for  general  recidivism  might  have  been  affected  given  the  developmental   changes  that  the  youth  offenders  undergo  and  long-­‐term  follow-­‐up  in  this  study.  Thirdly,  the   predictive  validity  might  also  have  been  affected  by  the  inter-­‐rater  reliability  of  ratings  for  the   subscales;  although  they  were  not  classified  as  Poor,  predictive  accuracy  would  most  likely  be   improved  with  more  reliable  ratings.  This  is  especially  important  in  high-­‐stakes  situations,  such   as  those  relating  to  making  decisions  regarding  the  level  of  supervision  or  incarceration.   Fourthly,  it  is  unclear  whether  some  of  the  other  needs/special  considerations  are  related  to  the   various  subscales  and  whether  they  should  be  equally  weighted  when  we  consider  these   variables  in  determining  risk  (which  is  probably  made  more  difficult  with  dichotomous   responses  [yes/no]);  these  issues  need  to  be  explored  further.  Finally,  like  much  risk   assessment  research,  the  predictive  validity  of  the  measure  might  be  artificially  lowered  by  the   probation  officers’  or  psychologists’  identification  and  diffusion  of  potential  criminal  offending  

Assessing  Youth  Offenders   behavior  that  might  be  exhibited  during  the  individuals’  court  orders  via  psychological  (e.g.,   counseling  or  relaxation),  increased  supervision,  and/or  social  (e.g.,  social  or  sporting   activities)  interventions.  As  such,  it  is  possible  that  the  predictive  accuracy  of  the  risk   assessment  instrument  might  be  attenuated.     Future  research  on  youth  risk  assessment  measures  should  employ  prospective  and   repeated  measures  designs,  in  which  the  risk  assessments  are  based  on  interviews  as  well  as   information  that  is  available  in  archival  records.  Moreover,  it  is  beneficial  to  examine  the  short-­‐   and  long-­‐term  validity  of  the  measure  and  its  components  (see  e.g.,  Chu,  Thomas,  Ogloff,  &   Daffern,  2013),  and  to  further  map  the  relationships  between  risk  factors,  strengths,  as  well  as   other  needs/special  considerations.  Furthermore,  it  will  be  useful  to  examine  how  the  usage  of   the  YLS/CMI  2.0  in  case  management  of  youth  offenders  can  contribute  to  improved  outcomes.  

Assessing  Youth  Offenders   References   Ægisdóttir,   S.,   White,   M.   J.,   Spengler,   P.   M.,   Maugherman,   A.   A.,   Anderson,   L.   A.,   Cook,   R.   S.,   …   Rush,   J.   D.   (2006).   The   meta-­‐analysis   of   clinical   judgment   project:   Fifty-­‐six   years   of   accumulated   research   on   clinical   versus   statistical   prediction.   The   Counseling   Psychologist,  34,  341-­‐382.   Andrews,  D.  A.,  &  Bonta,  J.  (2010).  The  psychology  of  criminal  conduct  (5th  ed.).  Cincinnati,  OH:   Anderson  Publishing.   Andrews,  D.  A.,  Bonta,  J.,  &  Wormith,  S.  J.  (2010).  The  Level  of  Service  (LS)  assessment  of  adults   and  older  adolescents.  In  R.  K.  Otto  &  K.  Douglas  (Eds.),  Handbook  of  violence  risk   assessment  (pp.  199-­‐225).  New  York,  NY:  Routledge.   Andrews,   D.   A.,   Bonta,   J.,   Wormith,   J.   S.,   Guzzo,   L.,   Brews,   A.,   Rettinger,   J.,   &   Rowe,   R.   (2011).   Sources   of   variability   in   estimates   of   predictive   validity:   A   specification   with   level   of   service  general  risk  and  need.  Criminal  Justice  and  Behavior,  38,  413-­‐432.   Ang,   R.,   &   Huan,   V.   S.   (2008).   Predictors   of   recidivism   for   adolescent   offenders   in   a   Singapore   sample.  Criminal  Justice  and  Behavior,  35,  895-­‐905.   Benjamini,  Y.,  &  Hochberg,  Y.  (1995).  Controlling  the  false  discovery  rate:  A  practical  and   powerful  approach  to  multiple  testing.  Journal  of  the  Royal  Statistical  Society:  Series  B   (Methodological),  57,  289-­‐300.   Bhugra,  D.,  Popelyuk,  D.,  &  McMullen,  I.  (2010).  Paraphilias  across  cultures:  Contexts  and   controversies.  Journal  of  Sex  Research,  47,  242-­‐256.   Chu,  C.  M.,  Daffern,  M.,  Thomas,  S.  D.  M.,  &  Lim,  J.  Y.  (2012).  Violence  risk  and  gang  affiliation  in   youth  offenders:  A  recidivism  study.  Psychology,  Crime  and  Law,  18,  299-­‐315.     Chu,  C.  M.,  Ng,  K.,  Fong,  J.,  &  Teoh,  J.  (2012).  Youth  who  sexually  offended:  The  predictive   validity  of  the  ERASOR,  J-­‐SOAP-­‐II,  and  YLS/CMI  in  a  non-­‐Western  context.  Sexual  Abuse:   A  Journal  of  Research  and  Treatment,  24,  153-­‐174.  

Assessing  Youth  Offenders   Chu,  C.  M.,  Thomas,  S.  D.  M.,  Ogloff,  J.  R.  P.,  &  Daffern,  M.  (2013).  The  short-­‐  to  medium-­‐term   predictive  accuracy  of  static  and  dynamic  risk  assessment  measures  in  a  secure  forensic   hospital.  Assessment,  20,  230-­‐242.   Chua,  J.  R.  Z.  H.,  Chu,  C.  M.,  Yim,  G.,  Chong,  D.,  &  Teoh,  J.  (2014).  Implementation  of  the  Risk-­‐ Need-­‐Responsivity  framework  within  the  juvenile  justice  agencies  in  Singapore.   Psychiatry,  Psychology  and  Law,  21,  877-­‐889.     Cicchetti,  D.  V.  (1994).  Guidelines,  criteria,  and  rules  of  thumb  for  evaluating  normed  and   standardized  assessment  instruments  in  psychology.  Psychological  Assessment,  6,  284-­‐ 290.   Criminal  Procedure  Code,  68  Singapore  Statutes  (2012).   Douglas,  K.  S.,  &  Skeem,  J.  L.  (2005).  Violence  risk  assessment:  Getting  specific  about  being   dynamic.  Psychology,  Public  Policy,  and  Law,  11,  347-­‐383.   Grove,  W.  M.,  Zald,  D.  H.,  Lebow,  B.  S.,  Snitz,  B.  E.,  &  Nelson,  C.  (2000).  Clinical  versus  mechanical   prediction:  A  meta-­‐analysis.  Psychological  Assessment,  12,  19-­‐30.   Harris,  G.  T.,  Rice,  M.  E.,  &  Quinsey,  V.  L.  (2009).  Allegiance  or  fidelity:  A  clarifying  reply.  Clinical   Psychology:  Science  and  Practice,  17,  82-­‐89.   Hartman,   J.   L.,   Turner,   M.   G.,   Daigle,   L.   E,   Exum,   M.   L.,   &   Cullen,   F.   T.   (2009).   Exploring   gender   differences   in   protective   factors:   Implications   for   understanding   resiliency.   International  Journal  of  Offender  Therapy  and  Comparative  Criminology,  53,  249-­‐277.   Hoge,  R.  D.  (2002).    Standardized  instruments  for  assessing  risk  and  need  in  youthful  offenders.     Criminal  Justice  and  Behavior,  29,  380-­‐396.   Hoge,  R.  D.,  &  Andrews,  D.  A.  (2011).  Youth  Level  of  Service/Case  Management  Inventory  2.0   (YLS/CMI  2.0):  User’s  manual.  Toronto,  Ontario,  Canada:  Multi-­‐Health  Systems.   Hoge,  R.  D.,  &    Andrews,  D.    A.  (2002).  Youth  Level  of  Service  Case  Management  Inventory   (YLS/CMI):  User's  manual.  Toronto,  Ontario,  Canada:  Multi-­‐Health  Systems.   Jung,  S.,  &  Rawana,  E.  P.  (1999).  Risk  and  need  assessment  of  juvenile  offenders.  Criminal  Justice   and  Behavior,  26,  69-­‐88.  

Assessing  Youth  Offenders   Lahlah,  E.,  Van  der  Knaap,  L.  M.,  Bogaerts,  S.,  &  Lens,  K.  M.  E.  (2013).  Making  men  out  of  boys?   Ethnic  differences  in  juvenile  violent  offending  and  role  of  gender  role  orientations.   Journal  of  Cross-­‐Cultural  Psychology,  44,  1321-­‐1338.   Lodewijks,  D.  P.  B.,  de  Ruiter,  C.,  &  Doreleijers,  T.  A.  H.  (2010).  The  impact  of  protective  factors   in  desistance  from  violent  reoffending:  A  study  in  three  samples  of  adolescent  offenders.   Journal  of  Interpersonal  Violence,  25,  568-­‐587.   Luong,  D.,  &  Wormith,  J.  S.  (2011).  Risk/need  assessment  to  probation  practice  and  its  impact   on  recidivism  of  young  offenders.  Criminal  Justice  and  Behavior,  38,  1177-­‐1199.   Marshall,  J.,  Egan,  V.,  English,  M.,  &  Jones,  R.  M.  (2006).  The  relative  validity  of  psychopathy   versus  risk/needs-­‐based  assessments  in  the  prediction  of  adolescent  offending   behavior.  Legal  and  Criminological  Psychology,  11,  197-­‐210.   McGrath,  A.,  &  Thompson,  A.  P.  (2012).  The  relative  predictive  validity  of  the  static  and  dynamic   domain  scores  in  the  risk-­‐need  assessment  of  juvenile  offenders.  Criminal  Justice  and   Behavior,  39,  250-­‐263.   Monahan,  J.  (1981).  Prediction  of  violent  behavior:  An  assessment  of  clinical  techniques.  Beverly   Hills,  CA:  Sage.   Olver,   M.   E.,   Stockdale,   K.   C.,   &   Wong,   S.   C.   P.   (2012).   Short   and   long-­‐term   prediction   of   recidivism  using  the  Youth  Level  of  Service/Case  Management  Inventory  in  a  sample  of   serious  young  offenders.  Law  and  Human  Behavior,  36,  331-­‐344.   Olver,  M.  E.,  Stockdale,  K.  C.,  &  Wormith,  J.  S.  (2009).  Risk  assessment  with  young  offenders:  A   meta-­‐analysis  of  three  assessment  measures.  Criminal  Justice  and  Behavior,  36,  329-­‐353.   Onifade,  E.,  Davidson,  W.,  Campbell,  C.,  Turke,  G.,  Malinowski,  J.,  &  Turner,  K.  (2008).  Predicting   recidivism  in  probationers  with  the  Youth  Level  of  Service/Case  Management  Inventory   (YLS/CMI).  Criminal  Justice  and  Behavior,  35,  474-­‐483.   Onifade,  E.,  Petersen,  J.,  Bynum,  T.  S.,  &  Davidson,  W.  S.  (2011).  Multilevel  recidivism  prediction:   Incorporating  neighborhood  socioeconomic  ecology  in  juvenile  justice  risk  assessment.   Criminal  Justice  and  Behavior,  38,  840-­‐853.  

Assessing  Youth  Offenders   Penal  Code,  224  Singapore  Statutes  (2008).   Rennie,  C.,  &  Dolan,  M.  (2010).  The  significance  of  protective  factors  in  the  assessment  of  risk.   Criminal  Behaviour  and  Mental  Health,  20,  8-­‐22.   Rice,  M.  E.,  &  Harris,  G.  T.  (2005).  Comparing  effect  sizes  in  follow-­‐up  studies:  ROC  area,  Cohen’s   d  and  r.  Law  and  Human  Behavior,  29,  615-­‐620.   Salisbury,  E.  J.,  &  Van  Voorhis,  P.  (2009).  Gendered  pathways:  An  empirical  investigation  of   women  probationers’  paths  to  incarceration.  Criminal  Justice  and  Behavior,  36,  541-­‐ 566.   Schmidt,   F.,   Campbell,   M.   A.,   &   Houlding,   C.   (2011).   Comparative   analyses   of   the   YLS/CMI,   SAVRY,  and  PCL:YV  in  adolescent  offenders:  A  10-­‐year  follow-­‐up  into  adulthood.  Youth   Violence  and  Juvenile  Justice,  9,  23-­‐42.   Schwalbe,   C.   S.   (2007).   Risk   assessment   for   juvenile   justice:   A   meta-­‐analysis.   Law   &   Human   Behavior,  31,  449-­‐462.   Singapore  Department  of  Statistics.  (2013).  Yearbook  of  statistics  2013.  Retrieved  on  January  5,   2014,  from:  www.singstat.gov.sg/statistics/browse_by_theme/population.html.   Singapore  Police  Force.  (2013).  Annual  crime  brief  2012.  Retrieved  on  July  29,  2013,  from:   http://www.spf.gov.sg/stats/stats2012_improvements.htm.   Stouthamer-­‐Loeber,  M.,  Loeber,  R.,  Stallings,  R.,  &  Lacourse,  E.  (2008).  Desistance  and   persistence  in  offending.  In  R.    Loeber,  D.  P.,  Farrington,  M.  Stouthamer-­‐Loeber,  &  H.  R.   White,  Violence  and  serious  theft:  Risk  and  promotive  factors  from  childhood  to  early   adulthood  (pp.  269-­‐308).  Mahwah,  NJ:  Erlbaum.   Takahashi,  M.,  Mori,  T.,  &  Kroner,  D.  G.  (2013).  A  cross-­‐validation  of  the  Youth  Level  of   Service/Case  Management  Inventory  (YLS/CMI)  among  Japanese  juvenile  offenders.   Law  and  Human  Behavior,  37,  389-­‐400.     Taylor,  K.,  &  Blanchette,  K.  (2009).  The  women  are  not  wrong:  It  is  the  approach  that  is   debatable.  Criminology  and  Public  Policy,  8,  221-­‐229.  

Assessing  Youth  Offenders   Van  Voorhis,  P.,  Wright,  E.  M.,  Salisbury,  E.,  &  Bauman,  A.  (2010).  Women’s  risk  factor  and  their   contributions  to  existing  risk/needs  assessment:  The  current  status  of  a  gender-­‐ responsive  supplement.  Criminal  Justice  and  Behavior,  37,  261-­‐288.   Webster,  C.  D.,  Douglas,  K.  S.,  Belfrage,  H.,  &  Link,  B.  (2000).  Capturing  change:  An  approach  to   managing  violence  and  improving  mental  health.  In  S.  Hodgins  &  R.  Müller-­‐Isberner   (Eds.),  Violence  among  the  mentally  ill  (pp.  119-­‐144).  Dordrecht,  The  Netherlands:   Kluwer/Academic.   Welsh,  J.  L.,  Schmidt,  F.,  McKinnon,  L.,  Chattha,  H.  K.,  &  Meyers,  J.  R.  (2008).  A  comparative  study   of  adolescent  risk  assessment  instruments.  Assessment,  15,  104-­‐115.   Yang,  M.,  Wong,  S.  C.  P.,  &  Coid,  J.  (2010).  The  effect  of  violence  prediction:  A  meta-­‐analytic   comparison  of  nine  risk  assessment  tools.  Psychological  Bulletin,  136,  740-­‐767.  

Assessing  Youth  Offenders   Table  1.  Means,  standard  deviations,  and  correlations  to  general  recidivism  for  the  YLS/CMI  2.0  total  and   subscale  scores.               YLS/CMI   M   SD    rrecidivism   p             Overall  Sample  (N  =  3,264)           YLS/CMI  Total  Score   11.78   4.10   .24   <  .001a        Prior  or  Current  Offenses/Dispositions   0.18   0.41   .14   <  .001a        Family  Circumstances/Parenting   2.11   1.17   .21   <  .001a        Education/Employment   2.02   1.54   .18   <  .001a        Peer  Relations   3.06   1.12   .05   .006a        Substance  Abuse   0.18   0.53   .07   <  .001a        Leisure/Recreation   2.21   0.96   .10   <  .001a        Personality/Behavior   0.85   0.98   .11   <  .001a        Attitudes/Orientation   1.18   0.86   .15   <  .001a             Male  Subsample  (n  =2,951)           YLS/CMI  Total  Score   11.63   4.05   .25   <  .001a        Prior  or  Current  Offenses/Dispositions   0.19   0.41   .14   <  .001a        Family  Circumstances/Parenting   2.07   1.16   .21   <  .001a        Education/Employment   2.00   1.56   .19   <  .001a        Peer  Relations   3.01   1.12   .06   .002a        Substance  Abuse   0.16   0.49   .07   <  .001a        Leisure/Recreation   2.20   0.94   .10   <  .001a        Personality/Behavior   0.84   0.98   .11   <  .001a        Attitudes/Orientation   1.17   0.84   .14   <  .001a             Female  Subsample  (n  =  313)           YLS/CMI  Total  Score   13.21   4.29   .26   <  .001a        Prior  or  Current  Offenses/Dispositions   0.12   0.33   .19   .001a        Family  Circumstances/Parenting   2.46   1.27   .24   <  .001a        Education/Employment   2.27   1.32   .14   .016a        Peer  Relations   3.57   0.97   .06   ns        Substance  Abuse   0.36   0.83   .10   ns        Leisure/Recreation   2.25   1.14   .10   ns        Personality/Behavior   0.92   0.97   .11   .047        Attitudes/Orientation   1.27   1.01   .23   <  .001a             Note:  a  denotes  that  the  difference  was  statistically  significant  (p  <  .05)  after  FDR  correction.  

Assessing  Youth  Offenders    Table  2.  The  breakdown  of  risk  categories  for  YLS/CMI  2.0  overall  risk  rating  and  subscales     Risk  Categories   YLS/CMI   Low   Moderate   High*           Overall  Sample  (N  =  3,264)         Overall  Risk  Rating   1,325  (40.6%)   1,809  (55.4%)   130  (4.0%)        Prior  or  Current  Offenses/Dispositions   2,688  (82.4%)   573  (17.6%)   3  (<  0.1%)        Family  Circumstances/Parenting   2,102  (64.4%)   1,086  (33.3%)   76  (2.3%)        Education/Employment   580  (17.8%)   2,104  (64.5%)   580  (17.8%)        Peer  Relations   162  (5.0%)   1,378  (42.2%)   1,724  (52.8%)        Substance  Abuse   2,858  (87.6%)   370  (11.3%)   36  (1.1%)        Leisure/Recreation   250  (7.7%)   472  (14.5%)   2,542  (77.9%)        Personality/Behavior   1,528  (46.8%)   1,730  (53.0%)   6  (0.2%)        Attitudes/Orientation   574  (17.6%)   2,640  (80.9%)   50  (1.5%)           Male  Subsample  (n  =  2,951)         Overall  Risk  Rating   1,199  (40.6%)   1,642  (55.6%)   110  (3.7%)        Prior  or  Current  Offenses/Dispositions   2,414  (81.8%)   534  (18.1%)   3  (0.1%)        Family  Circumstances/Parenting   1,941  (65.8%)   951  (32.2%)   59  (2.0%)        Education/Employment   551  (18.7%)   1,872  (63.4%)   528  (17.9%)        Peer  Relations   148  (5.0%)   1,334  (45.2%)   1,469  (49.8%)        Substance  Abuse   2,613  (88.5%)   317  (10.7%)   21  (0.7%)        Leisure/Recreation   198  (6.7%)   456  (15.5%)   2,297  (77.8%)        Personality/Behavior   1,397  (47.3%)   1,548  (52.5%)   6  (0.2%)        Attitudes/Orientation   501  (17.0%)   2,406  (81.5%)   44  (1.5%)           Female  Subsample  (n  =  313)         Overall  Risk  Rating   126  (40.3%)   167  (53.4%)   20  (6.4%)        Prior  or  Current  Offenses/Dispositions   274  (87.5%)   39  (12.5%)   0  (0%)        Family  Circumstances/Parenting   161  (51.4%)   135  (63.4%)   17  (5.4%)        Education/Employment   29  (9.3%)   232  (74.1%)   52  (16.6%)        Peer  Relations   14  (4.5%)   44  (14.1%)   255  (81.5%)        Substance  Abuse   245  (78.3%)   53  (16.9%)   15  (4.8%)        Leisure/Recreation   52  (16.6%)   16  (5.1%)   245  (78.3%)        Personality/Behavior   131  (41.9%)   182  (58.1%)   0  (0%)        Attitudes/Orientation   73  (23.3%)   234  (74.8%)   6  (1.9%)           *  denotes  that  the  High  and  Very  High  risk  groups  were  combined  for  analyses  purposes  due  to  small   proportions  in  relation  to  other  subgroups.                                          

Assessing  Youth  Offenders   Table  3.  Significant  associations  between  other  needs/special  considerations  and  general  recidivism   (split  by  gender;  chi  square  analyses).     Recidivist   Nonrecidivist           Other  Needs/Special  Considerations   n  (%)      n  (%)    p    φ             Male  Subsample  (n  =  2,951)           Chronic  History  of  Offenses    (Family)   275/1,133  (24.3)   269/1,818  (14.8)   <  .001a   .12   Marital  Conflict  (Family)   99/1,133  (8.7)   116/1,818  (6.4)   .017   .04   Financial/Accommodation  Problems  (Family)   258/1,133  (22.8)   286/1,818  (15.7)   <  .001a   .09   Abusive  Father  (Family)   52/1,133  (4.6)   38/1,818  (2.1)   <  .001a   .07   Other  Problems  (Family)   55/1,133  (4.9)   59/1,818  (3.2)   .027   .04   Diagnosis  of  Conduct  Disorder/Oppositional   14/1,133  (1.2)   8/1,818  (0.4)   .015   .05   Defiant  Disorder   Financial/Accommodation  Problems   31/1,133  (2.7)   30/1,818  (1.7)   .044   .04   Gang  Involvement  (Ever  Been  Involved)   388/1,133  (34.2)   462/1,818  (25.4)   <  .001a   .10   History  of  Assault  on  Authority  Figures   15/1,133  (1.3)   6/1,818  (0.3)   .002a   .06   History  of  Bullying   149/1,133  (13.2)   120/1,818  (6.6)   <  .001a   .11   History  of  Running  Away   148/1,133  (13.1)   96/1,818  (5.3)   <  .001a   .14     History  of  Sexual/Physical  Assault   318/1,133  (23.4)   373/1,818  (20.5)   <  .001a    .09   Peers  Outside  Age  Range   166/1,133  (14.7)   204/1,818  (11.2)   .006a   .05   Poor  Problem-­‐solving  Skills   393/1,133  (34.7)   518/1,818  (28.5)   <  .001a   .07     Victim  of  Physical/Sexual  Abuse   18/1,133  (1.6)   14/1,818  (0.8)   .037   .04   Witness  of  Domestic  Violence   20/1,133  (1.8)   12/1,818  (0.7)   .005a   .05                     Female  Subsample  (n  =  313)                 History  of  Running  Away   30/95  (31.6)   38/218  (17.4)   .005   .16     Victim  of  Physical/Sexual  Abuse   10/95  (10.5)   9/218  (4.1)   .029   .12                   Note:  a  denotes  that  the  difference  was  statistically  significant  (p  <  .05)  after  FDR  correction.