Clinical vs' Actuarial Assessments

1
Clinical vs. Actuarial Assessments

• MSc Applied Forensic
• Psychology
• Gareth Norris
• 23rd January, 2006

2
Contradictions in Psychology

• Nature vs. Nurture
• Psychodynamic vs. Cognitive
• Qualitative vs. Quantitative
• (Positivist vs. Classicist Criminology)

3
Types of Assessment

• Assessment can be for one or more purposes
• Treatment
• Recidivism
• Dangerousness/Risk

4
What is risk assessment?

• The Tarasoff case from California in the 1970s
  meant that clinicians became duty bound to inform
  third parties of danger
• Predictions of dangerousness have become
  commonplace and seek to balance individuals
  rights to liberty with those of societies
  protection from harm
• The potential for harm often means that offenders
  may be detained in the CJS for longer than they
  deserve

5
Recent Example

• Damien Hanson convicted of the murder of banker
  John Monckton in 2004
• Granted parole after 7 years of a 12-year
  sentence (three months prior to offence)
• Assessment had shown him to be at a high risk -
  75 or over - of re-offending (91)
• Only monitored by parole board on release
  should have been multi-agency.

6
Clinical Judgment

• Use primarily interviews and case files to
  establish behaviour
• Rely almost exclusively upon the experience and
  judgment of the practitioner
• Can be very subjective
• More than one method

7
Linear Model

• Uses a decision tree to determine risk (from
  Gross et al., 1987)
• Is there a clear and realisable threat?
• Is there serious danger?
• A specific victim?
• Imminent danger?
• Threat to public official, i.e. not
  family/friends?
• Outcome - Clinician should contact police.

8
Hypothetico-deductive model

• Knowledge of previous behaviour guides
  hypothesis building and testing using theory
  (Gross et al., 1987). Combines clinical skill and
  experience to assess importance of
• Previous behaviour in case history e.g. anger
• Observations of others
• Community risk
• Assessment of current psychological state

9
Multi-dimensional Models (e.g. Gottfredson
Gottfredson, 1988)

• In practice, clinical and actuarial methods are
  often used interchangeably. Clinical judgements
  may be quantified and form part of a statistical
  predictor and the latter be only one element in
  clinical decision making.

10
Problems with diagnoses

• Clinicians have been shown bias in many studies
  (e.g. Quinsey and Abtam, 1979 teachers
  psychiatrists also Motandon Harding, 1984)
• Clinicians ability to judge behaviours little
  more than chance
• Actuarial predictions perform better, but still
  overestimate dangerousness
• Also question of whether expert testimony has any
  reliability the hired gun and litigation

11
John Monahan University of Virginia

• When it comes to predicting violence, our
  crystal balls are terribly cloudy (Rosenthal,
  1993, p.1)

12
Fooling the professionals

• Rosenhan (1973) tested the ability of
  psychiatrists to judge normal and insane
  behaviours
• Three psychologists, a psychiatrist, a student, a
  painter, a doctor and housewife all faked
  symptoms
• Seven diagnosed with schizophrenia and one as
  manic depressive
• Only other patients recognised them as normal
  after an average 19 day stay!

13
Actuarial or Statistical Assessment

• Uses a database of predictor variables which
  show a relationship with re-offending variables
• Statistical tests and personality inventories
• Main premise concerns the comparison of the
  current offender with previous, similar offenders
• These predictors can be historical, personality
  and clinically based

14
Use of structured tests

• There are now literally hundreds of predictive
  tests which seek to establish risk of
  dangerousness and re-offending
• Psychopathy Checklist (PCL PCL-R)
• MMPI especially extroversion and impulsivity
• Intelligence tests

15
Reasons for lack of use

• Despite the apparent increased capacity of tests,
  there is a general mistrust by CJS
• Correlations between likelihood of behaviour and
  actual behaviour can be low, or only as much as
  .4
• Best predictor of violence, is usually a past
  history of violence

16
Top Twenty Predictor Variables (Douglas
Webster, 1999)

• Static Predictors
• A history of prior violence
• Younger age at first offence
• History of relationship instability/hostility
• History of employment instability
• Alcohol and drug abuse
• Major mental disorder (e.g. schizophrenia)

17
Predictor Variables (cont.)

• Diagnoses of psychopathy or ASPD
• Early maladjustment at home or school
• Diagnoses of any personality disorder
• History of absconding from custody
• Dynamic Predictors
• Lack of insight into own personality and/or
  misunderstanding of intentions of others
• Tendency to be angry and hostile

18
Predictor Variables (cont.)

• Experiencing psychotic symptoms
• Tendency for impulsiveness
• Resistance to psychiatric treatment
• Risk Management Predictors
• Lack of adequate supervision
• Access to victims, weapons and alcohol/drugs
• Lack of social support and living resources

19
Predictor Variables (cont.)

• Non-compliance with medication
• Excessive stress in family, employment and peers
• Many risk assessment inventories have been based
  on these predictors, e.g. HCR-20 (meaning 20
  Historical, Clinical, Risk) by Webster et al.,
  1997) and the VPS (Violence Prediction Scheme by
  Webster et al. 1994)

20
Predictive efficiency

• Risk assessment can be judged by a range of
  possible outcomes
• A true positive is when the predictor variable
  accurately forecasts the outcome
• Similarly, a true negative occurs when the
  absence of a predictor corresponds to the absence
  of recidivism
• False positives negatives are erroneous
  judgments with opposing consequences

21

• The accuracy equation (ad/N) shows the
  correctness of the predictor a desirable
  predictor is one that maximises correct hits (a
  d) and minimises error or misses (b c)

Criterion -
False Positive b
True Positive a
Predictor -
c False Negative
d True Negative
22
Consequences of Error

• A high false negative rate means that many
  dangerous offenders will be released into society
• A high false positive rate means the opposite
  many people will be detained despite being at a
  low or no risk of re-offending

23
Which predictor is best?

• Many predictors are employed in risk
  assessments. The efficiency of this variable at
  determining future events is its ability to do
  this above chance. This is defined by its
  correlation with the predictor variable and its
  base rate. A base rate is the occurrence of this
  criterion in any given population.

24
Different populations have different base rates,
hence any predictor variable will need to improve
this beyond chance
Criterion -
Base rate 50 Predictor present in 40 of 50
recidivists and absent in 40 of 50 non-recidivists
40 (80 of 50) a
10 (20 of 50) b
Predictor -
c 10 (20 of 50)
50
d 40 (80 of 50) 50
N100
25
Base rate 90
Criterion -
72 (80 of 90) a
2 (20 of 10) b
Predictor -
c 18 (20 of 90)
90
d 8 (80 of 10) 10
N100
26
Base rate 10
Criterion -
8 (80 of 10) a
18 (20 of 90) b
Predictor -
c 2 (20 of 10) 10
d 72 (80 of 90) 90
N100
27
Common errors made in predicting dangerousness
(Hall, 1987)

• Failure to recognize dangerousness as salient
  feature of assessment
• Inadequate forensic database(s)
• Distortions by the clinician
• Predicting dangerousness in the absence of
  previous dangerousness
• Accepting illusionary correlation

28
Problems Cont.

• Predicting from clinical diagnosis
• Not accounting for triggering stimuli
• Not considering opportunity
• Not assessing inhibitory variables e.g. age
• Ignoring base rates
• Vague conclusions
• Failure to consider disconfirmatory evidence

29
Improving Risk Assessment

• Reliability can be improved by averaging over
  judges
• Reducing cognitive bias by training not relying
  on intuition
• Use of decision aids objective reporting of
  clinical judgments statistically
• Appreciating the limitations
• Risk assessment is context specific (see over)

30
Clinical or Statistical?
Assessing risk of dangerousness or re-offending?
Risk/danger to whom staff, public, self
Context of risk Community, prison etc.
Time scale Short to long
Drug dependency/ Therapy aversion?
Have effects of treatment been accounted for?
What level of risk acceptable to Institution?
Potential benefits of risk?
Aspects of assessing risk (Meehle, 1954)
31
Similar debates in other forensic areas

• The controversy over clinical judgment and
• statistical reasoning are not only confined
• to assessment of offenders. Two other
• areas include
• Jury Selection
• Offender Profiling

32
Jury Selection

• Sometimes called jury profiling, it is actually
  a
• process of de-selection, whereby the
• prosecution and defence can preempt the
• inclusion of certain individuals from being on a
• jury. The idea is to only use those which will
• likely be most sympathetic to your case.

33
Methods

• Intuition looks for general character
  dispositions or implicit personality theories.
  Often basic stereotypes of human behaviour
  derived by consultant.
• Scientific examines specific theories known to
  influence jury behaviour, e.g. authoritarianism
  or locus of control. Tested using mock juries and
  focus groups.

34
Most successful?

• As with predictions of offender risk, it is
  impossible to account for all the variables which
  become central features of a trial. Hence, the
  controversial process of jury selection is also
  confounded by the fact that it is not possible to
  realistically evaluate the outcomes. One method
  or practitioner could be more successful
  depending upon the merits of the case or even
  luck.

35
Offender Profiling

• Offender profiling (or psychological/criminal
  profiling etc.) is a technique the police
  sometimes use during the investigation of
  (usually serious) crime. Profilers analyse
  behaviours from crime scenes in order to make
  inferences about the person who is likely to have
  committed the offence. The style and use of these
  is so varied, again it is difficult to appraise
  how well an individual is at performing this task.

36
Methods

• Clinical/Diagnostic Evaluations
• - based upon practitioners experience and
  judgment from past cases
• - use personality/crime theories
• - part science, part art
• Investigative Psychology
• - developed using offender databases
• - utilises statistical averages
• - follows rigorous scientific method
• NB. Also FBI approach mixture of statistics and
  investigative skill.

37
Most successful?

• Neither method has shown itself to possess the
  most utility and/or validity both have had some
  moments of glory and also spectacular failures.
  Police often skeptical about the place of
  profiling within an investigation. Many studies
  attempt to assess utility, but again the presence
  of many confounding variables makes this
  difficult.

38
Talking Point Sex Offender Risk

• With the recent debate regarding sex offenders
  working with children, what are your feelings
  regarding the assessment of these individuals?
• As it is generally accepted that there will
  always be a chance of re-offending, is there an
  acceptable level of risk that can be taken?