{"id":"W1974633292","doi":"10.1007/s10459-013-9480-6","title":"Validating MMI scores: are we measuring multiple attributes?","year":2014,"lang":"en","type":"article","venue":"Advances in Health Sciences Education","topic":"Medical Education and Admissions","field":"Medicine","cited_by":48,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Calgary; University of Guelph","funders":"","keywords":"Extraversion and introversion; Rubric; Construct (python library); Construct validity; Correlation; Personality; Psychology; Emotionality; Cognition; Clinical psychology; Psychometrics; Developmental psychology; Big Five personality traits; Social psychology; Computer science; Mathematics","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":[],"category_scores_codex":[0.001670014,0.0001053553,0.0002222185,0.0002269719,0.0003639227,0.00002722635,0.0001604011,0.0000398654,0.0002202934],"category_scores_gemma":[0.009874172,0.0000858949,0.00003032382,0.0008098495,0.0001699719,0.0003551902,0.00002157817,0.0002018077,0.00002392195],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001938809,"about_ca_system_score_gemma":0.002754796,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001464513,"about_ca_topic_score_gemma":0.000134074,"domain_scores_codex":[0.9982015,0.0001336261,0.0004764569,0.0003708442,0.0004488982,0.0003686723],"domain_scores_gemma":[0.9981637,0.0004572815,0.0003038477,0.000217496,0.0001060101,0.0007516876],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00001224278,0.0004780024,0.6845729,0.0008841811,0.000001488214,2.599515e-7,0.001267808,0.000446156,0.0001026012,0.002874049,0.002594271,0.3067661],"study_design_scores_gemma":[0.001246331,0.0005143795,0.224486,0.008874803,0.00001049017,0.00003739955,0.01154186,0.01405311,0.0009029632,0.006090767,0.7318379,0.0004039506],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5672354,0.03880254,0.01026987,0.3527808,0.01091039,0.002465733,0.000006267839,0.0003332711,0.01719574],"genre_scores_gemma":[0.9757401,0.002549422,0.01383257,0.00694795,0.0004405117,0.00007126056,0.00001396155,0.000008054096,0.0003961117],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7292436,"threshold_uncertainty_score":0.9984661,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09639065746901428,"score_gpt":0.4350509654003858,"score_spread":0.3386603079313716,"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."}}