{"id":"W4322152776","doi":"10.2196/43792","title":"Use of Multiple-Select Multiple-Choice Items in a Dental Undergraduate Curriculum: Retrospective Study Involving the Application of Different Scoring Methods","year":2023,"lang":"en","type":"article","venue":"JMIR Medical Education","topic":"Medical Education and Admissions","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Raw score; Wilcoxon signed-rank test; Checklist; Statistics; Multiple choice; Logistic regression; Raw data; Test (biology); Medicine; Psychology; Mathematics; Significant difference; Mann–Whitney U test","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.000966607,0.0001749378,0.0004234339,0.0003570607,0.00008706833,0.00001344348,0.0002043764,0.0001338874,0.000134979],"category_scores_gemma":[0.04381798,0.0001169042,0.0001029151,0.001571289,0.0001544556,0.0001018676,0.00008886556,0.0005465577,0.000007966289],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002341263,"about_ca_system_score_gemma":0.001298604,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001553374,"about_ca_topic_score_gemma":0.0006660449,"domain_scores_codex":[0.9972362,0.0004334834,0.0007618786,0.0003733607,0.000939284,0.0002558196],"domain_scores_gemma":[0.9965746,0.001811469,0.0003192364,0.0004498822,0.0002431682,0.000601579],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00003621223,0.004007096,0.9587425,0.0001369422,0.00004280419,0.000001005998,0.003460734,0.000003808143,0.004658255,0.0002879531,0.001025743,0.02759692],"study_design_scores_gemma":[0.00119203,0.0001992839,0.9586965,0.0005306431,0.00005680662,0.000006037215,0.008468425,0.02924315,0.0006448546,0.0001769212,0.000676975,0.0001083175],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9923577,0.00008594383,0.0005808329,0.00437867,0.0005404942,0.001952987,0.000002076163,0.00006215757,0.00003912093],"genre_scores_gemma":[0.9977154,0.00006789894,0.0005459848,0.0003623892,0.0002181715,0.0007698435,0.00007603996,0.00002329346,0.0002210327],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.04285137,"threshold_uncertainty_score":0.9642364,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05281666249751181,"score_gpt":0.4341871759580677,"score_spread":0.3813705134605559,"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."}}