{"id":"W4387445954","doi":"10.3928/01484834-20230815-05","title":"Scoring Methods in Script Concordance Tests: An Exploratory Psychometric Study","year":2023,"lang":"en","type":"article","venue":"Journal of Nursing Education","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université du Québec","funders":"","keywords":"Concordance; Spearman's rank correlation coefficient; Test (biology); Psychology; Clinical psychology; Medicine; Statistics; Mathematics; Internal medicine","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":[0.002322529,0.00008120167,0.0002886043,0.0009405836,0.00003618376,0.00002831338,0.00008719521,0.00004585964,0.00001185629],"category_scores_gemma":[0.03005037,0.00007067157,0.0000557426,0.001829958,0.00003450773,0.0001968226,0.000005290417,0.0003152282,0.00001127317],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002849103,"about_ca_system_score_gemma":0.0005826494,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001184882,"about_ca_topic_score_gemma":0.000002573835,"domain_scores_codex":[0.9986546,0.0002313381,0.0005453911,0.000143562,0.0002731165,0.0001520118],"domain_scores_gemma":[0.9967013,0.002332374,0.0002866155,0.0001995672,0.0002790756,0.0002010775],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.000126718,0.005983666,0.3145961,0.00001163357,0.00001791552,0.00002087798,0.007531285,0.00003045972,0.001251548,0.0000173267,0.005961327,0.6644511],"study_design_scores_gemma":[0.001234553,0.001437858,0.9594638,0.004570497,0.00008576967,0.0001047704,0.03155895,0.0000953485,0.0001596256,0.001034425,0.0001788565,0.00007551384],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9927437,0.0008174794,0.0004171472,0.001135669,0.004257588,0.0001677444,2.090971e-7,0.00002577252,0.0004347179],"genre_scores_gemma":[0.9853379,0.0001238574,0.01343017,0.0001902922,0.0007567468,0.000007872664,0.000002465778,0.00001691659,0.0001338338],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6643756,"threshold_uncertainty_score":0.9781199,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1869778306894005,"score_gpt":0.549873423595076,"score_spread":0.3628955929056755,"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."}}