Integrated Decentering: Validating an Automated Measure of Decentering Expression Across Seven Independent Samples
Research Square August 14, 2026 preprint DOI: 10.21203/rs.3.rs-10410851/v1 (opens in new tab)
Study at a glance
AI-extracted from the abstract| Characteristics | Secondary analysis of multiple samples, including a post-hoc secondary analysis of a registered randomized trial |
|---|---|
| Population | Seven independent samples: self-report, community, essay, diary, and online-argument samples, and two contemplative-practice cohorts |
| Duration | 12-month stability assessment; 24-month practice-exposure analysis |
| Measures | EQ-D, MPoDT, Big Five |
| Registration | NCT02628548 |
| Key findings | The authors report that one fixed scoring model, Integrated Decentering, transferred unchanged across seven samples and showed convergent, criterion, discriminant, incremental, reliability, and practice-linked validity, including correlations with established decentering scales and prediction of empathic concern, wise-reasoning search-for-resolution, and value-behavior coherence. They propose it as a scalable measure of decentering expression in reflective text. |
Abstract
Abstract Objectives. Decentering---relating to one's thoughts and emotions as passing mental events rather than being fused with them---is difficult to measure at scale. We test whether Integrated Decentering (ID), an automated measure of the decentering expressed in short reflective text, shows construct validity as one fixed scoring model. Methods. A purpose-trained language-model ensemble scored texts from seven independent samples (self-report, community, essay, diary, and online-argument samples, and two contemplative-practice cohorts). One cohort was a post-hoc secondary analysis of a registered randomized trial; ID was not a prespecified trial outcome. Aggregate-level analyses tested convergent, criterion, discriminant, incremental, reliability, and practice-linked validity, controlling for text length. Results. ID converged with established decentering scales (EQ-D r=.36, 95% CI [.20,.50], n=124; MPoDT r=.41[.09,.65], n=35); no prompted general-purpose model, frontier included, reproduced the EQ-D convergence. ID predicted empathic concern (r=.40[.36,.44]), wise-reasoning search-for-resolution (β=.18[.05,.30]), and value--behavior coherence (r=.33[.24,.41]; ΔR²=.027 beyond the Big Five). Predicted nulls held (adjacent wise-reasoning subscales, rumination, negative affect), and length-matched impersonal argument showed no positive discrimination (AUC =.45[.40,.49]). Aggregates were stable over 12 months (r=.69[.52,.81]). ID distinguished practitioners from matched non-practitioner controls (meditation g=1.83[1.08,2.89]; yoga g=1.38[0.68,2.29]); in a post-hoc non-randomized exposure analysis, reported sitting practice tracked 24-month gains (g=.80[0.13,1.56]). ID overlapped with an expert developmental score, as theory predicts, yet predicted decentering-relevant self-report beyond it. Conclusions. One fixed scoring model transferred unchanged across all seven samples: ID offers a scalable measure of decentering expression in reflective text for contemplative science and other fields concerned with metacognitive distance. Trial registration. ClinicalTrials.gov NCT02628548; registered 11 December 2015.