{"id":"W4416309139","doi":"10.2139/ssrn.5757617","title":"Do Test Scores Misrepresent Test Results? An Item-by-Item Analysis","year":2025,"lang":"en","type":"preprint","venue":"SSRN Electronic Journal","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"Simon Fraser University","funders":"","keywords":"Test (biology); Download; Aggregate (composite); Aggregate data; Ex-ante; Test data","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":[0.1844217,0.00193472,0.002572026,0.007308999,0.001266701,0.004096,0.002856113,0.003275043,0.002042946],"category_scores_gemma":[0.4860315,0.0009996226,0.003850823,0.01120157,0.004858047,0.00605124,0.003648595,0.003037252,0.0009442061],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001477261,"about_ca_system_score_gemma":0.001211967,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001733094,"about_ca_topic_score_gemma":0.002331128,"domain_scores_codex":[0.7409117,0.183822,0.01936533,0.01530316,0.03864813,0.001949701],"domain_scores_gemma":[0.2956253,0.5859047,0.03891054,0.05292127,0.02547785,0.00116033],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0008734698,0.0003176161,0.8950808,0.0003013887,0.006481865,0.0001517287,0.003235036,0.001734723,0.0009323966,0.005355621,0.003031485,0.08250386],"study_design_scores_gemma":[0.0002945535,0.001740801,0.9173828,0.000351456,0.004756383,0.001061259,0.003532,0.02533838,0.003905844,0.03488321,0.006528141,0.000225198],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7579142,0.002763395,0.2195571,0.003086968,0.0006835752,0.00134562,0.001990179,0.0005044688,0.01215429],"genre_scores_gemma":[0.9678563,0.0002826574,0.02835961,0.001027424,0.0002028486,0.0006019765,0.0007934766,0.0001655731,0.0007101945],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8155783,"threshold_uncertainty_score":0.9753268,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2133188991517597,"score_gpt":0.4592982282682351,"score_spread":0.2459793291164755,"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."}}