{"id":"W2158828425","doi":"10.1111/j.1365-2923.2010.03863.x","title":"Script concordance testing: a review of published validity evidence","year":2011,"lang":"en","type":"review","venue":"Medical Education","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":170,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal; Centre de Santé et de Services Sociaux Cavendish; McGill University","funders":"","keywords":"Concordance; Psychology; External validity; Construct validity; Competence (human resources); PsycINFO; Content validity; Validity; Clinical psychology; Internal validity; MEDLINE; Social psychology; Applied psychology; Psychometrics; Medicine; Statistics; 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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1349308,0.001680733,0.005789269,0.02635229,0.001836077,0.007866478,0.00616896,0.004039169,0.006787244],"category_scores_gemma":[0.5323513,0.001782085,0.007394798,0.02176824,0.006385782,0.008149602,0.005514028,0.003231848,0.001232272],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.007060837,"about_ca_system_score_gemma":0.02854516,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006593475,"about_ca_topic_score_gemma":0.01117816,"domain_scores_codex":[0.8602933,0.0616538,0.04173452,0.005083439,0.03012577,0.001109181],"domain_scores_gemma":[0.2682825,0.6388925,0.04002311,0.01102386,0.04035018,0.001428015],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0002928285,0.00009722132,0.005971752,0.5158415,0.003207797,0.0002429531,0.001793089,0.0002917946,0.0001834837,0.003093387,0.007933994,0.4610502],"study_design_scores_gemma":[0.0001103554,0.0002050445,0.006350537,0.9348115,0.007205732,0.0007778299,0.00109811,0.0002852524,0.0004397424,0.002618128,0.04601133,0.00008632829],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.001289583,0.9913737,0.001889891,0.002065679,0.0006183042,0.0004336495,0.0002915983,0.0000390281,0.001998611],"genre_scores_gemma":[0.03396352,0.9510877,0.009559684,0.002573389,0.0006536053,0.0012123,0.000685073,0.00008686939,0.0001777438],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.8650692,"threshold_uncertainty_score":0.713591,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2272363726862678,"score_gpt":0.4794160776581745,"score_spread":0.2521797049719066,"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."}}