{"id":"W2002384848","doi":"10.1111/j.1745-3992.2002.tb00103.x","title":"What Do School‐Level Scores From Large‐Scale Assessments Really Measure?","year":2002,"lang":"en","type":"article","venue":"Educational Measurement Issues and Practice","topic":"Cognitive Abilities and Testing","field":"Psychology","cited_by":9,"is_retracted":false,"has_abstract":true,"ca_institutions":"Toronto Public Health","funders":"","keywords":"Reading (process); Scale (ratio); Variance (accounting); Psychology; Cognition; Measure (data warehouse); Subject (documents); Illusion; Mathematics education; Cognitive psychology; Computer science; Linguistics; Data mining","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001877834,0.000220596,0.0002008242,0.00007329907,0.0003204797,0.0004953707,0.0001628792,0.00008800816,0.01894274],"category_scores_gemma":[0.003928123,0.0002174774,0.00005441325,0.0001521079,0.00005930189,0.001283786,0.00005854323,0.0002858386,0.0007662129],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001246504,"about_ca_system_score_gemma":0.0001007373,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001231755,"about_ca_topic_score_gemma":0.0001656324,"domain_scores_codex":[0.9972222,0.0005882959,0.0003513517,0.0005590451,0.0009232163,0.0003558886],"domain_scores_gemma":[0.9964151,0.001779913,0.0002287165,0.0003565527,0.001006664,0.0002130283],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0009737561,0.009947849,0.2923798,0.0001307143,0.00208951,0.0000371961,0.05270398,0.000004340163,0.001516728,0.0250965,0.3709186,0.244201],"study_design_scores_gemma":[0.00144311,0.0002332184,0.4472019,0.0005086889,0.0002868849,0.00006039243,0.05200572,0.00002712387,0.00004230738,0.007411547,0.4902259,0.000553186],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.2435073,0.2163925,0.0009069377,0.1508336,0.01259159,0.001651161,0.0002695092,0.0002038083,0.3736435],"genre_scores_gemma":[0.9768193,0.002192085,0.007094012,0.002845784,0.001894177,0.0001691541,0.0000607303,0.00004162074,0.008883161],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.733312,"threshold_uncertainty_score":0.9848372,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2066488833662569,"score_gpt":0.420967314211943,"score_spread":0.2143184308456861,"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."}}