{"id":"W4386066764","doi":"10.5811/westjem.58153","title":"Gender and Inconsistent Evaluations: A Mixed-methods Analysis of Feedback for Emergency Medicine Residents","year":2023,"lang":"en","type":"article","venue":"Western Journal of Emergency Medicine","topic":"Diversity and Career in Medicine","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"National Institutes of Health; National Center for Advancing Translational Sciences; University of Southern California","keywords":"Medicine; Demography; Family medicine; Internal medicine","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":[0.04587318,0.0006853215,0.00115622,0.002677187,0.001416734,0.001809448,0.00131039,0.0009599687,0.001424931],"category_scores_gemma":[0.09926189,0.0006622514,0.002445047,0.002123115,0.001062745,0.001487013,0.002063925,0.0008053822,0.0002553546],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00178537,"about_ca_system_score_gemma":0.001771456,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004716835,"about_ca_topic_score_gemma":0.007282418,"domain_scores_codex":[0.9555829,0.03187424,0.003101932,0.003357714,0.004782862,0.001300502],"domain_scores_gemma":[0.9110843,0.05770126,0.0175992,0.005362335,0.007286758,0.0009661625],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"qualitative","study_design_scores_codex":[0.001464235,0.0004497903,0.9561013,0.000570572,0.001193761,0.0002216344,0.0209118,0.0001830106,0.0007430399,0.0002423862,0.000674027,0.01724444],"study_design_scores_gemma":[0.0001229279,0.001942221,0.9620961,0.0004041991,0.0006098729,0.0003382556,0.02847965,0.003119362,0.0008335578,0.0004657728,0.001511874,0.00007630907],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9943785,0.0004773751,0.003382944,0.0001036685,0.00003420654,0.0006841485,0.0004929308,0.00001716287,0.0004289717],"genre_scores_gemma":[0.9931555,0.0001282301,0.004079296,0.000141771,0.00002597371,0.001782818,0.0003905933,0.0000218414,0.0002739187],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9541268,"threshold_uncertainty_score":0.2426035,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2103788623131993,"score_gpt":0.5010196562997677,"score_spread":0.2906407939865684,"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."}}