Risk Assessment Research Findings

1
Risk AssessmentResearch Findings

• Dr Sarah Clarke
• Ahimsa (Safer Families) Ltd
• Plymouth, Devon Tel [phone]

2
Research findings - methodology

• There are three dominant approaches
• 1. Unstructured clinical assessment
• Largely or solely based on clinical constructs
  (as if violence propensity will somehow manifest
  itself at interview - I cannot imagine that this
  is the sort of man who would be capable of ..).
  
• No rules no transparency they ignore base
  rates prone to bias and illusory (rather than
  empirical) correlations about risk.
• Results consistently poor and rarely better than
  chance (background factors far more predictive
  than clinical impressions - interview poor means
  of data collection - self-report notoriously
  unreliable).
• Conroy and Murrie 2007, Andrews et al
  2008, Howitt 2006, Quinsey et al 2006, Bancroft
  and Silverman 2002

3
Research findings - methodology

• 2. Actuarial risk assessment
• Employs a statistical prediction derived from a
  finite list of factors that have been empirically
  established to be positively correlated with
  future violence within a given population.
• They typically include factors such as age,
  criminal history, substance use history, prior
  assaults, severity of injury (e.g. VRAG, DVRAG,
  PCL-R).
• Consistently outperform unaided clinical
  assessments
• BUT based almost exclusively upon criminal
  justice population and does not predict when,
  under what circumstances, against whom, and with
  what severity or frequency an individual will act
  violently nor does it predict other behaviours
  that are likely to also to cause harm.
• Quinsey et al 2006, Conroy and Murrie 2007,
  Maden 2007, Litwack 2001

4
Research findings - methodology

• 3. Structured clinical risk assessment
• Draws on the science of actuarial approaches but
  attempts to take advantage of clinical insights.
• The assessment approach is structured to ensure
  all crucial empirically-based risk factors are
  considered in every case but it allows the
  assessor freedom to examine factors that may be
  unique to the individual concerned (e.g. SARA,
  RAF).
• The anchor for the assessment is still background
  information, usually most reliably derived from
  the court bundle, criminal records etc risk
  assessment is essentially an investigative task.
• See Maden 2007, Conroy and Murrie 2007, Mulvey
  and Lidz 1995.

5
Research findings Beware of untested assumptions

• Some of the factors commonly assumed to increase
  risk are not supported by empirical research.
• For example, denial is a commonly cited risk
  factor for violence recidivism, yet the research
  evidence for this is weak we need to
  discriminate between liars and
  self-deceivers.
• Similarly, pregnancy on the part of the victim
  can lead to a reduction in risk, as well as an
  increase (or business as usual) we need to
  know about the mans attitude to the pregnancy.
• See Maruna and Mann 2006, Henning and Holdford
  2006, review by Jasinki 2004 and Jasinki and
  Kanfor 2001.

6
Research findings the limitations of RA
technology

• The importance of base rates is commonly ignored
• Base rates (the prevalence or frequency of a
  behaviour in a given population) indicate how
  difficult it is likely to be to assess the
  behaviour in question.
• Spousal homicide is so rare as to make it all but
  impossible to predict.
• It is an irony within the field that where it is
  needed most, risk assessment technology is at its
  weakest.
• See Conroy and Murrie 2007, Maden 2007, Beaumont
  1999.

7
Research findings the limitations of RA
technology

• 2. Predicting onset is not the same as predicting
  recidivism
• Many approaches fail to differentiate between
  family violence onset and recidivism.
• The risk factors that predict onset are not
  necessarily those that predict recidivism.
• For example, exposure to domestic violence in
  childhood is a powerful predictor of onset but a
  weak predictor of recidivism.
• By contrast, mental health problems are weak
  predictors of onset but reliable predictors of
  recidivism.

8
Research findings the limitations of RA
technology

• 3. Predicting severe domestic violence may not be
  the same as predicting lethality.
• Some commentators argue that lethal domestic
  violence is a crime of cumulation.
• Others argue that there are discrete factors that
  can predict lethality.
• On the other hand, research by the Dobash team
  found that a significant minority of homicide
  perpetrators exhibited surprisingly few risk
  indicators for either (suggesting that early
  attachment difficulties, untapped by most
  assessments, may be more significant than
  dominant assessment tools recognise).