{"id":"W2988360744","doi":"10.1097/acm.0000000000003064","title":"A Study of the Validity of the New MCAT Exam","year":2019,"lang":"en","type":"article","venue":"Academic Medicine","topic":"Medical Education and Admissions","field":"Medicine","cited_by":19,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Varimax rotation; Percentile; United States Medical Licensing Examination; Entrance exam; Medicine; Exploratory factor analysis; Regression analysis; Psychology; Statistics; Cronbach's alpha; Clinical psychology; Predictive validity; Medical school; Medical education; Mathematics; Psychometrics","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":[0.0006579073,0.0001044345,0.000373601,0.00004646564,0.00003802578,4.280482e-7,0.0004023962,0.0001234268,0.004441041],"category_scores_gemma":[0.00640394,0.00004213749,0.00006386586,0.0004568165,0.0001848855,0.00001905632,0.0001143376,0.0007486743,0.00001954755],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002355055,"about_ca_system_score_gemma":0.0007469953,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001270471,"about_ca_topic_score_gemma":0.000002932121,"domain_scores_codex":[0.9982444,0.0001506273,0.0005082893,0.0001703182,0.0007824721,0.0001438837],"domain_scores_gemma":[0.9982055,0.0003153987,0.0002680185,0.000718721,0.00007528738,0.000417054],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0000665074,0.0003139875,0.5235817,0.0002245298,0.00009380944,0.000001206889,0.03116442,0.000002293302,0.03055535,0.001100298,0.4104744,0.00242148],"study_design_scores_gemma":[0.009212091,0.001399606,0.7909213,0.002824775,0.0007220602,0.00007006749,0.03330343,0.00005119876,0.01042012,0.001041982,0.1498808,0.0001526168],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9393457,0.0002176844,0.000007941173,0.05353457,0.001179038,0.0008924221,7.643772e-7,0.00001149944,0.004810421],"genre_scores_gemma":[0.9617206,0.00007005167,0.000007472849,0.005595094,0.0004114679,0.000005038872,0.000001296862,0.000009306452,0.03217966],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2673396,"threshold_uncertainty_score":0.996469,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1135842340639894,"score_gpt":0.4053518575735626,"score_spread":0.2917676235095731,"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."}}