{"id":"W2319075669","doi":"10.1119/perc.2014.pr.027","title":"Measuring the Learning from Two-Stage Collaborative Group Exams","year":2015,"lang":"en","type":"article","venue":"","topic":"Experimental Learning in Engineering","field":"Engineering","cited_by":20,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Test (biology); Diagnostic test; Logistic regression; Collaborative learning; Group tests; Stage (stratigraphy); Mathematics education; Medical physics; Computer science; Medicine; Psychology; Machine learning; Statistical hypothesis testing; Statistics; Mathematics; Pediatrics","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":[0.006711532,0.0008177056,0.001015472,0.001360243,0.0005305285,0.00148652,0.001269605,0.001339196,0.002469521],"category_scores_gemma":[0.05884448,0.0003906701,0.000941821,0.000743717,0.0005650653,0.001171609,0.002411139,0.001633991,0.0005969472],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008100654,"about_ca_system_score_gemma":0.0009020481,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006793986,"about_ca_topic_score_gemma":0.0008880439,"domain_scores_codex":[0.9899943,0.002372755,0.0007491878,0.001450874,0.004807826,0.0006250312],"domain_scores_gemma":[0.9488855,0.02506515,0.01051384,0.005114343,0.005484403,0.004936756],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.01239947,0.04924498,0.4408387,0.001334253,0.001624386,0.0003503232,0.01056726,0.009204525,0.0590931,0.001023762,0.003567459,0.4107518],"study_design_scores_gemma":[0.0005252835,0.04625897,0.9182096,0.0000958052,0.0003692819,0.0002299015,0.001012208,0.005008566,0.0247238,0.0008405638,0.002584423,0.0001416258],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9946331,0.00007239925,0.00232223,0.00004300721,0.00001554167,0.000509378,0.0001638051,0.00008704721,0.002153584],"genre_scores_gemma":[0.9864825,0.0001487724,0.007686583,0.00007615678,0.00002846308,0.00111713,0.0006372708,0.00003303029,0.003790193],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.006711532,"threshold_uncertainty_score":0.03549439,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03139038020399351,"score_gpt":0.2341460360342349,"score_spread":0.2027556558302414,"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."}}