{"id":"W2808863842","doi":"10.1145/3197091.3197116","title":"Gender, confidence, and mark prediction in CS examinations","year":2018,"lang":"en","type":"article","venue":"","topic":"Teaching and Learning Programming","field":"Computer Science","cited_by":32,"is_retracted":false,"has_abstract":true,"ca_institutions":"The Scarborough Hospital; University of Toronto","funders":"","keywords":"Mistake; Work (physics); Computer science; Mathematics education; Artificial intelligence; Psychology; Engineering","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":[0.008974926,0.0002343864,0.0003162918,0.001221188,0.0006461355,0.002244072,0.000905412,0.001357576,0.006159787],"category_scores_gemma":[0.1003485,0.0002447606,0.0004647762,0.000831695,0.0009760611,0.001570326,0.001483631,0.001897754,0.00124838],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007098334,"about_ca_system_score_gemma":0.0007397333,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006246316,"about_ca_topic_score_gemma":0.006346276,"domain_scores_codex":[0.9961902,0.001359722,0.0003368859,0.0005149527,0.001089073,0.0005091002],"domain_scores_gemma":[0.9181633,0.05106075,0.01571107,0.002894766,0.004705844,0.007464239],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00022064,0.0001914826,0.9915478,0.00000579,0.00002024408,0.00005634258,0.0005237439,0.0001848301,0.00005418072,0.0001501687,0.0003754199,0.006669391],"study_design_scores_gemma":[0.000008596822,0.0002443966,0.9954442,0.00003806482,0.00002200109,0.0001537925,0.001240445,0.001585935,0.0001715281,0.0005656325,0.0005071153,0.00001846014],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9942134,0.0004113365,0.0003212379,0.0009689291,0.0000559588,0.000009824657,0.0001522081,0.00001495549,0.003852067],"genre_scores_gemma":[0.9990773,0.00004372027,0.00007604972,0.00005407946,0.0000157593,0.000003698421,0.0000808064,0.000006466896,0.0006420074],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.008974926,"threshold_uncertainty_score":0.04746449,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02749782075312115,"score_gpt":0.2666272239754067,"score_spread":0.2391294032222856,"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."}}