{"id":"W2743358023","doi":"10.18260/1-2--28731","title":"PANEL: Gender bias in student evaluations of teaching","year":2018,"lang":"en","type":"article","venue":"","topic":"Evaluation of Teaching Practices","field":"Social Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Set (abstract data type); Gender bias; Promotion (chess); Affect (linguistics); Class (philosophy); Psychology; Gender gap; Institution; Medical education; Social psychology; Computer science; Medicine; Demographic economics; Political science","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01069702,0.00005062237,0.00008956431,0.0001281956,0.000329783,0.00004454289,0.0002420078,0.00004497492,0.001730454],"category_scores_gemma":[0.004363318,0.00004650977,0.00002893408,0.0001685362,0.0001652293,0.0003821235,0.00004620703,0.0001434978,0.0001387551],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009415566,"about_ca_system_score_gemma":0.000208475,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005468353,"about_ca_topic_score_gemma":0.008903145,"domain_scores_codex":[0.9971691,0.001474906,0.0002548231,0.0001404896,0.0008084942,0.0001521803],"domain_scores_gemma":[0.9988115,0.0006809586,0.0001505459,0.0001524838,0.0001617948,0.00004269632],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000008629784,0.0003844128,0.3716256,0.000005786068,0.00003289644,5.473365e-7,0.2482502,0.0001107714,0.0005182339,0.3597895,0.002620885,0.0166526],"study_design_scores_gemma":[0.0007180042,0.0001443016,0.8112476,0.00003879795,0.00005944483,5.631518e-7,0.1035195,0.001036857,0.000341558,0.01337728,0.06925604,0.0002600777],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6185586,0.00002231824,0.0005929165,0.003354995,0.0002011083,0.0001743068,4.311284e-7,0.00004320961,0.377052],"genre_scores_gemma":[0.9923994,0.000008626892,0.004546261,0.0002039354,0.000164212,0.000007989755,8.436325e-7,0.000005061399,0.002663666],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.439622,"threshold_uncertainty_score":0.9991821,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5901902343937442,"score_gpt":0.5874192468082498,"score_spread":0.002770987585494433,"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."}}