{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.06513457,0.0009145313,0.001028806,0.004209639,0.001150349,0.001030984,0.001093978,0.0005659078,0.001038432],"category_scores_gemma":[0.2113008,0.0004866595,0.0009797973,0.002793465,0.001360313,0.000967574,0.002571217,0.0009386806,0.0003780819],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001270288,"about_ca_system_score_gemma":0.001276267,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003260867,"about_ca_topic_score_gemma":0.007178666,"domain_scores_codex":[0.8921097,0.05156363,0.02172642,0.01020503,0.02309302,0.001302291],"domain_scores_gemma":[0.6830058,0.1832212,0.04249237,0.03703179,0.05267739,0.001571383],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001189633,0.0004439034,0.7598459,0.001167099,0.001208518,0.0009357103,0.02608488,0.002360068,0.009341042,0.001527016,0.002305665,0.1935906],"study_design_scores_gemma":[0.00007840669,0.001838412,0.9398411,0.00113202,0.0007468093,0.003094395,0.007741792,0.005708171,0.0210252,0.002478888,0.01613479,0.0001800246],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9503537,0.002535189,0.03914999,0.0002232233,0.0002496061,0.00098725,0.0006788701,0.0003728532,0.00544942],"genre_scores_gemma":[0.9718331,0.0007964732,0.023389,0.000186232,0.00006468285,0.0005879256,0.0006134908,0.0002150438,0.002314161],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9348654,"threshold_uncertainty_score":0.3444687,"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."}}