{"id":"W3138069980","doi":"10.6339/jds.201207_10(3).0006","title":"A Data Mining Approach to Comparing American and Canadian Grade 10 Students’ PISA Science Test Performance","year":2021,"lang":"en","type":"article","venue":"Journal of Data Science","topic":"Online Learning and Analytics","field":"Computer Science","cited_by":20,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Test (biology); Academic achievement; Mathematics education; Achievement test; Psychology; Political science; Standardized test","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":[0.003953649,0.0009223919,0.0009794482,0.01435851,0.002737439,0.00236425,0.002102226,0.0005432669,0.001937718],"category_scores_gemma":[0.01459839,0.0003054438,0.001150182,0.01797942,0.0007365387,0.000637143,0.001225321,0.001298329,0.0004933893],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006591027,"about_ca_system_score_gemma":0.01396246,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.8147897,"about_ca_topic_score_gemma":0.8174941,"domain_scores_codex":[0.9966583,0.0004503145,0.0003577799,0.0004650523,0.001675486,0.0003930616],"domain_scores_gemma":[0.9895488,0.003159348,0.001225019,0.0004569375,0.00505081,0.0005590643],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0005833365,0.001005358,0.8490982,0.0004160466,0.0006989811,0.0004558622,0.007107997,0.002731172,0.003188734,0.00209324,0.008569917,0.1240512],"study_design_scores_gemma":[0.00003561252,0.0004066226,0.9553539,0.0001359936,0.0001673471,0.0002041498,0.01499338,0.01517406,0.002732206,0.0007564807,0.009959345,0.00008099418],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9323123,0.0003497888,0.01508754,0.0006982412,0.00009246008,0.001799525,0.03505397,0.0005384019,0.01406766],"genre_scores_gemma":[0.9503489,0.0002395953,0.0311072,0.0000993879,0.00001672191,0.001207415,0.01394662,0.00003611526,0.002998038],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1852103,"threshold_uncertainty_score":0.3726022,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1098757292940929,"score_gpt":0.3657505192465182,"score_spread":0.2558747899524254,"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."}}