{"id":"W2792554342","doi":"10.1080/00313831.2017.1420687","title":"PISA Country Rankings Valid? Results for Canada and Finland","year":2018,"lang":"en","type":"article","venue":"Scandinavian Journal of Educational Research","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Concordia University","funders":"","keywords":"Item response theory; Normalization (sociology); Contrast (vision); Econometrics; Mathematics education; Mathematics; Statistics; Psychology; Psychometrics; Computer science; Sociology; Social science; Artificial intelligence","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02208154,0.00008299651,0.0002399219,0.0007038841,0.0004939617,0.0002682183,0.0007097055,0.00004848757,0.0004726285],"category_scores_gemma":[0.3344989,0.00005669032,0.00004841912,0.001616349,0.0003417683,0.0001800623,0.00007281687,0.0002859748,0.000008414539],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001833444,"about_ca_system_score_gemma":0.00261913,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01761689,"about_ca_topic_score_gemma":0.01776331,"domain_scores_codex":[0.9959332,0.0004726055,0.0007508629,0.0002687254,0.002193736,0.0003808908],"domain_scores_gemma":[0.853502,0.1402593,0.0005619987,0.0003253886,0.004975795,0.0003755508],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"observational","study_design_scores_codex":[0.0006542996,0.00004135155,0.3355217,0.00001071224,0.00002736115,0.000003399172,0.0004053838,0.000003341798,0.0001095592,0.001955919,0.6396221,0.02164484],"study_design_scores_gemma":[0.001335116,0.0006752784,0.7036319,0.0001328496,0.000007689758,0.000206241,0.001316697,0.0001072711,0.0001334601,0.1131853,0.1791448,0.0001233749],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9700267,0.0009327077,0.0005759,0.0130886,0.002034926,0.0001913328,0.0000894661,0.000001083367,0.0130593],"genre_scores_gemma":[0.9801551,0.00004256595,0.009715707,0.0001489176,0.001661493,0.000004872612,0.00000250377,0.000007347913,0.008261486],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4604774,"threshold_uncertainty_score":0.9912346,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5919838220888495,"score_gpt":0.574699651340387,"score_spread":0.01728417074846245,"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."}}