{"id":"W4318715460","doi":"10.2139/ssrn.4294644","title":"Assessing the Practice Sensitivity and Test-Retest Reliability of Behavioral and N400 Measures in a Particular Semantic Categorization Task","year":2022,"lang":"en","type":"article","venue":"SSRN Electronic Journal","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"McGill University; Douglas Mental Health University Institute","funders":"","keywords":"Categorization; N400; Reliability (semiconductor); Task (project management); Psychology; Test (biology); Sensitivity (control systems); Natural language processing; Cognitive psychology; Computer science; Artificial intelligence; Cognition; Event-related potential; Neuroscience; Engineering","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":[0.01287154,0.0007162341,0.0006404067,0.0008027385,0.0007871876,0.0008301084,0.0006955963,0.001381516,0.001107564],"category_scores_gemma":[0.05618419,0.0006709977,0.001131712,0.0004883014,0.0007883928,0.001184578,0.001233314,0.001511008,0.0009668897],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005115697,"about_ca_system_score_gemma":0.0005325081,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001755071,"about_ca_topic_score_gemma":0.003528542,"domain_scores_codex":[0.9922019,0.002051498,0.001063264,0.002042898,0.002395118,0.0002452965],"domain_scores_gemma":[0.9367241,0.03391708,0.005241685,0.01102618,0.01232688,0.0007641276],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0105043,0.00698048,0.6744513,0.0004916597,0.003108996,0.0002779254,0.009244728,0.0044402,0.1436146,0.0008630201,0.001734751,0.1442882],"study_design_scores_gemma":[0.0002646242,0.01140617,0.9456411,0.00004783462,0.0007137871,0.0007121403,0.0004840878,0.005734183,0.03123578,0.000613678,0.003044848,0.0001018913],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.981819,0.0004405997,0.01373794,0.0000598478,0.0001698322,0.0005438154,0.0001987399,0.0001546718,0.002875475],"genre_scores_gemma":[0.9882243,0.0001295993,0.008722109,0.0001056989,0.00006282091,0.0004931145,0.0004582939,0.0001043937,0.00169978],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9871284,"threshold_uncertainty_score":0.06807208,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.150310422322928,"score_gpt":0.4364639250692351,"score_spread":0.286153502746307,"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."}}