L9 Psychological AssesSMent Overview: Screening & Prediction?

參考文獻(xiàn)：Pomerantz, A. (2013). Clinical psychology: science, practice, and culture (3rd ed.). Thousand Oaks, CA: SAGE Publications.

9.1 Goals & Strategies of AssesSMent?

Common Goals of AssesSMent

1. Screening: Identification of those experiencing clinically?significant difficulties

2. Prediction/Prognosis: Estimation of the likelihood of a?future event (e.g., clinically significant problems; violence)

3. Diagnosis/Case Formulation: Specification of DSM?diagnoses and conceptualization of client experience and distress

4. Treatment Monitoring and Evaluation: Examination of?how well prevention or intervention is working

?

Common AssesSMent Strategies

- Self-Report?Questionnaires

- Interviews

- Observational?Methods?(lab, clinic; school, home, etc.)

- Performance-Based?Tests?(commonly timely tests)

?

Fundamental Questions

? Are assesSMent approaches:

- Standardized?

? Minimize influence of testing situation and assessor

? Comparable materials/items across assessors, detailed?instructions about administration and scoring

- Reliable?

? (Statistically) Internal consistency, test-retest, inter-rater

- Valid?

? Content?validity, convergent?validity, discriminant?validity, concurrent?validity, predictive?validity,?incremental?validity

- Useful?

? Practical utility of approach, relative to others?(eg cheaper, faster)

?

9.2 Example of a Screening Tool

Alcohol Use Disorders Identification Test?(AUDIT)

? Developed by World Health?Organization (WHO) to screen?alcohol problems in the past year?(Babor et al., 2001)

? Best-available brief screener for?alcohol use disorder (AUD) and?high-risk/problem drinking?- not specifically to diagnose AUD

? Used around the world in variety?of contexts?- designed in health care settings

?

AUDIT - typically self-reported questions

??Basic Information

Items cover 3 domains

- Alcohol consumption (Q1-3)

- Drinking dependence (Q4-6)

- Alcohol-related problems (Q7-10)

scoring

Qs 1-8: scored 0 - 4

Qs 9-10: scored 0, 2, or 4

Total score: 0 - 40

? Fundamental Questions?for the AUDIT

AUDIT is standardized?- exact instruction & question, clear scoring

AUDIT is reliable (i.e., consistent and trustworthy)

- Internal consistency

? Consistency of responses across items on measure

? Cronbach’s α across studies = .82?(Cronbach’s α?is used for internal consistency: .6-.7 acceptable, .7-.8 good, .8-1.0 excellent)

- Test-retest reliability

? Consistency of responses over time (r > .7 good)

? Median r across studies .80?- consistent

- Inter-rater reliability?(not statistically applied here, normally no interviews)

? Consistency of judgments across raters?(kappa > .75 excellent)

AUDIT is valid (i.e., it measures what it purports to measure)

- Content validity?(no statistic test)

? Extent to which item content reflects concept of interest

- Convergent validity

? Moderate-to-strong association with measures intended to?assess same or similar concepts

- Discriminant validity

? Weaker association with measures intended to assess different?concepts

- Concurrent validity

? Association of measure with another presumably related?measure at same point in time (i.e., cross-sectional)

- Predictive validity

? Association of measure with another presumably related?measure at future point in time

- Incremental validity

? Extent to which measure predicts more than what previous?measures could predict

AUDIT is useful

- Low cost & brief?(just 10Qs)

- More accurate than similar measures

? Is AUDIT accurate?

- What cutoff value to use?

- How do we select it?

- Sensitivity?(correct positive identified), Specificity?(negative identified)

?

9.3 Issues in Prediction

Using AssesSMents to Predict Behavior

? Clinical vs. Actuarial approaches to predicting behavior

Clinical approaches:

? Overwhelmingly used in practice

? Prone to errors and decision-making biases?- own judgement

“Providing service that assumes that clinicians?‘can do better’ simply based on?self-confidence or plausibility in the?absence of evidence that they actually?can do so is simply unethical.”?- Robyn M. Dawes, 2005

?

Evidence-based prediction (Actuarial Prediction):

? Evidence overwhelmingly supports actuarial approaches

? Need to provide information on (in)accuracy of?predictions, so that they are not weighted too heavily in?decision-making processes

? Recognize bias and boundaries of competence

? Don’t overstate powers of assesSMent and prediction

Limitations in predicting behavior:

- No approaches are 100% accurate

- Important not to overstate results

- You can only use the available data to make the prediction

?

Example: Predicting Violent/Aggressive?Behavior

? Legal systems?- Prediction of violence risk as part of decision making

? Mental health systems?- Prediction of violence risk?is relevant to:

? Commitment to?psychiatric settings due to?concerns about risk to self?or others

? Tort liability of clinicians?for patients’ violence

? Approaches range from clinical to actuarial extremes

- Clinical: Completely unstructured

? Clinician selects measure(s) and?combines risk factors to produces?risk estimate based on their personal clinical?experience and judgment

- Actuarial (statistical): Completely structured

? Risk assesSMent completely?determined by prediction equation?and established cutoff scores

?

? Structured components of violence risk?assesSMent/prediction?in Actuarial (statistical)?approach:

- Identify risk factors from empirical literature?based on data

- Measure risk factors using best approaches?based on data

- Combine risk factors using empirically derived prediction?formula?based on data

- Produce final risk estimate based on empirically derived cutoffs?based on data

This method was used to create the original VRAG.

?

Violence Risk Appraisal Guide (VRAG)

? 12-item scale used to make completely actuarial?predictions of violence following release?from an institution for individuals with with mental disorders who engaged in violent crimes in the past (Quinsey et al., 1988)

? Developed on psychiatric sample of 618 males in Canadian?maximum-security hospitals:

- Items with strongest predictive power identified

- Statistical methods used to determine optimal?combination of items to obtain risk score

? VRAG scores moderately predict incarcerated men’s violent?recidiviSM for follow-up periods of up to 10 years

? Results similar for ~ 8 other actuarial methods

- All tap overlapping dimensions:

? Psychopathy?(anti-social)

? Criminal attitudes

? Criminal history

? Substance-abuse-related problems

? Clinical methods are consistently less reliable