{"id":"W2020794294","doi":"10.1186/1472-6920-14-97","title":"Accuracy of portrayal by standardized patients: Results from four OSCE stations conducted for high stakes examinations","year":2014,"lang":"en","type":"article","venue":"BMC Medical Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":25,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary","funders":"","keywords":"Cronbach's alpha; Objective structured clinical examination; Trainer; Rating scale; Reliability (semiconductor); Scale (ratio); Psychology; Medical education; Educational measurement; Applied psychology; Medicine; Medical physics; Psychometrics; Clinical psychology; Computer science; Curriculum; Pedagogy","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":[],"consensus_categories":[],"category_scores_codex":[0.006262153,0.0004172336,0.0003093368,0.001108368,0.0003153834,0.0006456868,0.0004202696,0.0005700728,0.001156554],"category_scores_gemma":[0.04236931,0.0002327004,0.000503131,0.0006976626,0.0006585305,0.000610891,0.001836433,0.0004290839,0.0004643157],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004263036,"about_ca_system_score_gemma":0.0003721639,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001188744,"about_ca_topic_score_gemma":0.001734409,"domain_scores_codex":[0.9937558,0.002595437,0.0009668701,0.000697276,0.001644035,0.0003404787],"domain_scores_gemma":[0.9562545,0.01845501,0.01510208,0.002968124,0.00576844,0.001451839],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00076239,0.0002252746,0.9770979,0.0001161497,0.0001171116,0.0001495861,0.003534351,0.0004980141,0.002157774,0.00001769441,0.0002089666,0.01511483],"study_design_scores_gemma":[0.00002082738,0.001538121,0.9932968,0.00002604955,0.00004849563,0.0002474465,0.001646846,0.0007751487,0.002005533,0.00002216915,0.0003511803,0.00002136177],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.999228,0.00005174817,0.0002834421,0.000009144327,0.000004646511,0.00002634422,0.00009451029,0.000005805757,0.0002963755],"genre_scores_gemma":[0.9991269,0.00004265986,0.000438903,0.000009357821,0.000005438105,0.00002182996,0.0002028354,0.000004001377,0.0001480087],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.006262153,"threshold_uncertainty_score":0.03311783,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02735951081956597,"score_gpt":0.3465033525339843,"score_spread":0.3191438417144183,"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."}}