In plain words: They ran 880 language-model raters over 189 interviews to score depression symptoms, checking whether the rater or the patient drove the scores. Model choice explained 30% of score variation versus 10.5% for patient differences, and two equally accurate raters disagreed about 40% of people.
Abstract
Depression has no diagnostic blood test. Language models promise tireless, consistent assessment, but can accurate raters disagree about individuals? We pre-registered 880 language-model raters, crossing 11 open models with prompting and scoring choices, and applied them to 189 interviews against the eight-item Patient Health Questionnaire. Model choice explained 30.0% of summed-symptom score variance, stable participant differences 10.5%. Two randomly drawn raters with area under the receiver operating characteristic curve (AUC) >= 0.70 disagreed on screening decisions for 40% of participants, on average. Average over-rating governed how many were flagged, yet equal-capacity raters chose differently for about one participant in five. A locked analysis of 86 new interviews reproduced the main pre-registered findings. Exploratory recalibration with 40 labelled participants raised accuracy from about 60% to 75% and halved disagreement, leaving one participant in five decided differently. Calibration repaired much of the rater dependence without securing agreement about individuals.
Baihan Lin
arXiv:2610.08501 · cs.CL, cs.AI, q-bio.NC · submitted Oct 6, 2026
abstract · pdf · html