{"id":"W2769374615","doi":"10.14507/epaa.25.3299","title":"Policy incentives in Canadian large-scale assessment: How policy levers influence teacher decisions about instructional change","year":2017,"lang":"en","type":"article","venue":"Education Policy Analysis Archives","topic":"Educational Assessment and Improvement","field":"Decision Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Incentive; Promotion (chess); Scale (ratio); Curriculum; Work (physics); Test (biology); Psychology; Mathematics education; Pedagogy; Political science; Public relations; Economics","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03831101,0.0004143813,0.0007445152,0.002877898,0.009009651,0.01062086,0.002034879,0.002574662,0.003163447],"category_scores_gemma":[0.1336065,0.0005858947,0.0005435002,0.004679559,0.006045195,0.003254333,0.00405996,0.003284147,0.0002221425],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.1858734,"about_ca_system_score_gemma":0.2181605,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.9641636,"about_ca_topic_score_gemma":0.9737492,"domain_scores_codex":[0.9536079,0.01954415,0.001333067,0.002274462,0.01118727,0.01205318],"domain_scores_gemma":[0.8842438,0.06696609,0.01017161,0.003476964,0.02440489,0.01073662],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001215518,0.0009215016,0.4432571,0.0007552646,0.0004181988,0.0007236606,0.02944569,0.03804927,0.003071921,0.2739883,0.02521549,0.1829382],"study_design_scores_gemma":[0.0002932515,0.0003302341,0.7324972,0.0009186664,0.0003358596,0.0001042881,0.03584824,0.0349705,0.002092566,0.0613171,0.1307453,0.0005469195],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7911488,0.004237601,0.006727955,0.05447545,0.0002054512,0.0007970622,0.001023318,0.0001377662,0.1412466],"genre_scores_gemma":[0.9962714,0.0004285945,0.001036425,0.0006613971,0.00001473998,0.0000523607,0.00006716517,0.00001010837,0.001457851],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1858734,"threshold_uncertainty_score":0.9442715,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07839698053366613,"score_gpt":0.4586197873152658,"score_spread":0.3802228067815996,"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."}}