{"id":"W1987169263","doi":"10.1002/tl.38","title":"Using Student Outcomes to Evaluate Teaching: A Cautious Exploration","year":2001,"lang":"en","type":"article","venue":"New Directions for Teaching and Learning","topic":"Teacher Education and Leadership Studies","field":"Social Sciences","cited_by":26,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Mathematics education; Teaching method; Psychology; Higher education; Pedagogy; Medical education; Medicine; 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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1697749,0.001714766,0.002888303,0.006483205,0.00326051,0.01765226,0.004142638,0.002627723,0.002402907],"category_scores_gemma":[0.22032,0.000507135,0.001340827,0.005639242,0.009099313,0.01582998,0.004585291,0.006367688,0.0006227113],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002688255,"about_ca_system_score_gemma":0.00756686,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0101265,"about_ca_topic_score_gemma":0.01566419,"domain_scores_codex":[0.8169331,0.1377758,0.01680882,0.001844557,0.02561606,0.001021681],"domain_scores_gemma":[0.6612958,0.2702704,0.008171197,0.01612091,0.04163824,0.002503383],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"qualitative","study_design_scores_codex":[0.0009205214,0.0008854403,0.2513402,0.00564937,0.0009800015,0.00126537,0.02812356,0.00250978,0.003801811,0.1585708,0.04083901,0.5051142],"study_design_scores_gemma":[0.0002692695,0.005775777,0.1592749,0.02177477,0.001380591,0.005418141,0.123765,0.02549244,0.02029114,0.3255894,0.3096774,0.001291184],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2875482,0.0632102,0.3297561,0.2091079,0.007454142,0.002560885,0.003146605,0.002170233,0.0950458],"genre_scores_gemma":[0.8605105,0.006104842,0.1148598,0.008666951,0.001249965,0.001963288,0.000313294,0.000534787,0.005796616],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.1697749,"threshold_uncertainty_score":0.8978663,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2412671041432465,"score_gpt":0.4862596402215878,"score_spread":0.2449925360783413,"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."}}