Methodological advances in statistical prediction

Area of Science:

• Psychological assessment
• Clinical psychology
• Predictive analytics

Background:

• The seminal 1989 work by Dawes, Faust, and Meehl advocated for routine use of statistical prediction rules in mental health.
• Subsequent research has largely supported the efficacy of statistical prediction over clinical judgment.
• However, the field has evolved, necessitating an update on methodological advancements.

Purpose of the Study:

• To review methodological advances in statistical prediction within psychological assessment.
• To examine the evolving role and applications of statistical prediction rules.
• To highlight key developments and ongoing issues in the field.

Main Methods:

• Literature review focusing on methodological advances in statistical prediction.
• Analysis of statistical prediction's utility in criterion-referenced vs. norm-referenced assessment.
• Examination of prediction of violence and criminal recidivism, comparing clinical judgment, structured professional judgment, and statistical prediction.
• Discussion of issues in building prediction rules, including predictor weighting, machine learning, and theory.

Main Results:

• Statistical prediction rules are highly valuable for criterion-referenced assessment (e.g., violence, recidivism) but less so for some norm-referenced tasks (e.g., personality description).
• Both structured professional judgment and statistical prediction show validity, particularly in predicting violence and recidivism.
• Overriding statistical predictions by professionals consistently resulted in decreased predictive validity.

Conclusions:

• Statistical prediction has advanced significantly, becoming a crucial area in psychological assessment.
• The integration of structured professional judgment alongside statistical prediction is supported.
• Future directions involve refining rule-building, incorporating new statistical analyses like machine learning, and clarifying the role of theory.