{"id":"W2139397945","doi":"10.1111/j.1745-3992.2007.00090.x","title":"Defining and Evaluating Models of Cognition Used in Educational Measurement to Make Inferences About Examinees' Thinking Processes","year":2007,"lang":"en","type":"article","venue":"Educational Measurement Issues and Practice","topic":"Cognitive Abilities and Testing","field":"Psychology","cited_by":166,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Cognition; Identification (biology); Psychology; Cognitive psychology; Computer science; Management science","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"],"consensus_categories":[],"category_scores_codex":[0.008417928,0.0001711788,0.0001968663,0.0002165649,0.000181962,0.00008648606,0.00008599655,0.00005773254,0.0002431956],"category_scores_gemma":[0.01608524,0.0001759387,0.00001784228,0.0003305279,0.0000618197,0.0003835731,0.00004264727,0.0001597846,0.000007765088],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001107899,"about_ca_system_score_gemma":0.0005294985,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001140399,"about_ca_topic_score_gemma":0.0009434323,"domain_scores_codex":[0.9973978,0.000285515,0.0005391173,0.0003930338,0.001116627,0.0002679361],"domain_scores_gemma":[0.9929942,0.004125259,0.0003084998,0.0001166544,0.002353761,0.000101554],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001501431,0.004118659,0.344027,0.001589671,0.0005811568,0.000004665968,0.2071286,0.0003532573,0.005634312,0.224201,0.0006518011,0.2102084],"study_design_scores_gemma":[0.00121861,0.0005338464,0.8854444,0.001852085,0.0002173331,0.00005943449,0.05349979,0.00008513312,0.0004813083,0.05435985,0.001744185,0.000504002],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.920895,0.02116424,0.001697123,0.01245273,0.0005849478,0.0009382836,0.00001468927,0.00002388924,0.04222906],"genre_scores_gemma":[0.988376,0.00006913002,0.01067155,0.0004137131,0.000226499,0.0001315132,0.00001806014,0.00001444622,0.00007908809],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5414174,"threshold_uncertainty_score":0.9922027,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2732332005606202,"score_gpt":0.4452723569892246,"score_spread":0.1720391564286045,"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."}}