{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0002597162,0.0001942993,0.0001530525,0.00004400413,0.00008813961,0.00009608898,0.0002226298,0.0000505887,0.0001223722],"category_scores_gemma":[0.00008916163,0.0001540733,0.00003528269,0.0002639753,0.0000329341,0.0002026844,0.00007228487,0.0004144506,0.000221883],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001747156,"about_ca_system_score_gemma":0.00001226328,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001657005,"about_ca_topic_score_gemma":0.0000392289,"domain_scores_codex":[0.9990404,0.00006287397,0.0001728243,0.0001759414,0.0002697855,0.0002781931],"domain_scores_gemma":[0.9994645,0.0001271307,0.00002313118,0.0002270243,0.0000357657,0.0001225143],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000004651136,0.000004353138,0.001474771,0.000004590986,0.00005017807,0.000007193009,0.002794297,0.9659069,0.02793187,0.0007176016,0.0002570553,0.0008465684],"study_design_scores_gemma":[0.001176074,0.00007907511,0.001166082,0.00005502918,0.00001886692,0.000004455538,0.01330461,0.799119,0.06722123,0.00007035756,0.1171569,0.0006283394],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.805564,0.003328767,0.05579707,0.00004947515,0.00117547,0.0002994063,0.0000049039,0.002493656,0.1312872],"genre_scores_gemma":[0.9951156,0.0000156965,0.003550091,0.00002174324,0.0002300378,0.00002748226,0.00001226495,0.00007233809,0.000954754],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1895515,"threshold_uncertainty_score":0.6282929,"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."}}