{"id":"W2596054285","doi":"10.22230/ijepl.2016v11n6a727","title":"What They Think About How They Are Evaluated: Perspectives of New York State Physical Educators on Teacher Evaluation Policy","year":2016,"lang":"en","type":"article","venue":"International Journal of Education Policy and Leadership","topic":"Physical Education and Pedagogy","field":"Health Professions","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Test (biology); Physical education; Mathematics education; Psychology; The arts; Sample (material); Stratified sampling; State (computer science); Medical education; Language arts; School district; Pedagogy; Medicine; Political science; Mathematics; Statistics","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001074637,0.0002128711,0.0003163743,0.0007535009,0.0001513765,0.00007299337,0.0004035273,0.0001198417,0.0003540924],"category_scores_gemma":[0.003827587,0.0001423726,0.0001650898,0.0001873637,0.0001352812,0.0007893821,0.00003020687,0.0005090339,0.00006280441],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000860027,"about_ca_system_score_gemma":0.009012492,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000982495,"about_ca_topic_score_gemma":0.0000982787,"domain_scores_codex":[0.9970111,0.0009516968,0.0005354411,0.0002427819,0.0009442415,0.0003147649],"domain_scores_gemma":[0.9955668,0.0009591836,0.001251246,0.0002319388,0.001663455,0.0003273278],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0008457563,0.002056789,0.007275081,0.00008906883,0.0004593004,4.12419e-7,0.6340156,0.00003546763,0.001562093,0.07096235,0.0588956,0.2238024],"study_design_scores_gemma":[0.002999823,0.0004658426,0.1635425,0.003207088,0.0001299998,0.00002100546,0.6518857,0.00005739637,0.0005128725,0.1459512,0.0308339,0.0003926835],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7114675,0.0009794306,0.00002997833,0.2816816,0.003208329,0.0003134282,0.00001637195,0.00001382187,0.002289573],"genre_scores_gemma":[0.9719327,0.0004475414,0.00003070847,0.00315877,0.01182988,0.00002078079,0.000007428858,0.00002632701,0.01254588],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2785228,"threshold_uncertainty_score":0.9966055,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3154698675779187,"score_gpt":0.5328375919706001,"score_spread":0.2173677243926814,"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."}}