A social inference model of idealization and devaluation.

Area of Science:

• Cognitive Psychology
• Computational Psychiatry
• Psychiatric Disorders

Background:

• Polarized beliefs, termed dichotomous thinking or "splitting," are common and characteristic of several psychiatric disorders.
• Existing theories are descriptive; a quantitative model is needed to explain the emergence and updating of these rigid beliefs.

Purpose of the Study:

• To introduce a Bayesian model of splitting that quantifies the tendency to categorize individuals or objects as strictly "Good" or "Bad."
• To predict how dichotomous beliefs are formed and updated based on new information, accounting for context-dependency and temporal stability.
• To measure individual differences in splitting susceptibility and its manifestation in clinical populations.

Main Methods:

• Developed a Bayesian computational model to parameterize rigid categorization of objects as "Good" or "Bad."
• Modeled the attribution of counter-evidence to external factors to consolidate idealized or devalued beliefs.
• Fitted the model to empirical data from healthy participants and individuals with borderline personality disorder.

Main Results:

• The model demonstrates how splitting can be context-dependent yet stable over time, with belief shifts triggered by sufficient counter-evidence.
• Healthy participants showed less changeable character impressions when holding a "Good" belief, while those with borderline personality disorder exhibited higher, symmetric splitting.
• The model successfully quantifies individual susceptibility to relational instability.

Conclusions:

• The proposed Bayesian framework offers a quantitative account of dichotomous thinking and its update mechanisms.
• Splitting dynamics differ significantly between healthy individuals and those with borderline personality disorder, highlighting clinical relevance.
• The generative model has potential applications in understanding and modeling relational and affective dynamics in psychotherapy.