{"id":"W2284581455","doi":"10.1080/14926156.2015.1119332","title":"Canadian and United States Students' Performances on the OECD's PISA 2012 Problem-Solving Assessment","year":2015,"lang":"en","type":"article","venue":"Canadian Journal of Science Mathematics and Technology Education","topic":"Education and Critical Thinking Development","field":"Social Sciences","cited_by":8,"is_retracted":false,"has_abstract":false,"ca_institutions":"","funders":"","keywords":"Mathematics education; Set (abstract data type); Reading (process); Scale (ratio); Psychology; Computer science; Political science; Geography","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"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":[0.002551752,0.0005058424,0.0006183328,0.00260102,0.003340718,0.003890893,0.001044267,0.0008003627,0.006022485],"category_scores_gemma":[0.01027043,0.0002403308,0.0007214562,0.005482493,0.0009949356,0.0009972778,0.002063582,0.0009764871,0.001517389],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01997651,"about_ca_system_score_gemma":0.03163076,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.9719391,"about_ca_topic_score_gemma":0.9903148,"domain_scores_codex":[0.997269,0.0001787644,0.0001216472,0.0002142993,0.001585248,0.0006311733],"domain_scores_gemma":[0.989728,0.0008372288,0.0006164669,0.0002453036,0.006751041,0.00182193],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001039531,0.0007325563,0.8373952,0.0001411696,0.0001449802,0.0001612186,0.01634728,0.002321318,0.0006970756,0.004335595,0.05267923,0.08400483],"study_design_scores_gemma":[0.00003188665,0.0001006567,0.974375,0.00006061882,0.0000374511,0.00002139009,0.008160857,0.0007775312,0.0005308981,0.0002248972,0.01563313,0.00004558674],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9566572,0.0003054723,0.00007732651,0.0006688527,0.00007216778,0.00003894563,0.003145361,0.00004744846,0.03898723],"genre_scores_gemma":[0.9806081,0.0003223564,0.0002061232,0.0001163164,0.000008068841,0.00003433487,0.003677943,0.00002040571,0.01500636],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02806091,"threshold_uncertainty_score":0.1449404,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.025483940492606,"score_gpt":0.3353345004961147,"score_spread":0.3098505600035087,"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."}}