{"id":"W4401091997","doi":"10.1111/jep.14114","title":"A many‐facet Rasch measurement model approach to investigating objective structured clinical examination item parameter drift","year":2024,"lang":"en","type":"article","venue":"Journal of Evaluation in Clinical Practice","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Touchstone Exploration (Canada); McMaster University","funders":"","keywords":"Rasch model; Polytomous Rasch model; Facet (psychology); Item response theory; Psychology; Medicine; Psychometrics; Clinical psychology; Social psychology; Developmental psychology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4042676,0.0002214559,0.0008614102,0.001128229,0.0001110465,0.0006565567,0.0007095846,0.0003344839,0.00005930067],"category_scores_gemma":[0.9502617,0.0001560237,0.000420174,0.003181925,0.0001214211,0.001693285,0.0001782091,0.002018962,0.00003573753],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003750153,"about_ca_system_score_gemma":0.001096506,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000009744358,"about_ca_topic_score_gemma":0.000005116469,"domain_scores_codex":[0.9611619,0.02091761,0.007825841,0.001064854,0.008614428,0.000415435],"domain_scores_gemma":[0.4710077,0.5080073,0.006600173,0.001225496,0.01234222,0.000817093],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0003026358,0.0005054242,0.01594948,0.0000220635,0.0001923553,0.0000179011,0.001202199,0.07086243,0.0001095274,0.0008927352,0.007495984,0.9024472],"study_design_scores_gemma":[0.0009582781,0.000559052,0.2854768,0.0001494188,0.0001865683,0.00007340161,0.001794337,0.6637247,0.00001270383,0.04569306,0.001214616,0.0001570341],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5331036,0.0008885557,0.4494665,0.003544661,0.004186637,0.0008095718,0.000003363655,0.00002820312,0.007968972],"genre_scores_gemma":[0.6693767,0.00003704918,0.3290925,0.0009160578,0.000509746,0.0000146821,7.202773e-7,0.00001391868,0.00003859408],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9022902,"threshold_uncertainty_score":0.8771499,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.816087781485377,"score_gpt":0.6373573978864153,"score_spread":0.1787303835989617,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}