{"id":"W2795833260","doi":"10.1177/0829573518763484","title":"Common Examiner Scoring Errors on Academic Achievement Measures","year":2018,"lang":"en","type":"article","venue":"Canadian Journal of School Psychology","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":11,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta; University of Victoria","funders":"","keywords":"Psychology; Achievement test; Wechsler Adult Intelligence Scale; Academic achievement; Test (biology); Educational measurement; Measure (data warehouse); Intelligence quotient; Scoring system; Cognition; Clinical psychology; Standardized test; Developmental psychology; Mathematics education; Medicine; Curriculum; Computer science; Psychiatry; Data mining; Pedagogy","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0107063,0.0001985024,0.0005086486,0.002363033,0.0002992098,0.0001170796,0.001854393,0.0002383527,0.001521343],"category_scores_gemma":[0.05718246,0.000143477,0.0001534629,0.001688182,0.0003838918,0.0002412558,0.00003680334,0.001142206,0.0005189711],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001573645,"about_ca_system_score_gemma":0.0004501411,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005765265,"about_ca_topic_score_gemma":0.003639475,"domain_scores_codex":[0.9959539,0.0008056427,0.00133137,0.0004237419,0.0008388755,0.0006464345],"domain_scores_gemma":[0.9923944,0.004077285,0.0008883169,0.0006980397,0.0006200554,0.001321893],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00009265007,0.00001771234,0.7349327,0.000001158779,0.00004103891,0.00009756273,0.0001973927,0.00003947378,0.0009167307,0.0004896768,0.1044548,0.1587191],"study_design_scores_gemma":[0.000638054,0.0008706719,0.8795776,0.00006465941,0.00001303249,0.0002281498,0.0005250504,0.00001164704,0.0001577521,0.0157194,0.1020218,0.0001722151],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9577393,0.0006357699,0.00286902,0.006417863,0.005847323,0.00009537463,0.000005342742,0.00000887971,0.02638111],"genre_scores_gemma":[0.9900287,0.0000320178,0.002767093,0.005306719,0.001344589,0.000001680848,2.784763e-7,0.00001728231,0.0005016819],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1585469,"threshold_uncertainty_score":0.9993914,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6305978026502712,"score_gpt":0.5203701125482403,"score_spread":0.1102276901020309,"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."}}