{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.00920612,0.0003970848,0.0006545453,0.002353533,0.0006243896,0.0008068883,0.0006372205,0.0005050609,0.0008793213],"category_scores_gemma":[0.0315588,0.0005050356,0.0005312883,0.002272306,0.0009474069,0.0009674786,0.001431354,0.0008026777,0.0003156355],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006990473,"about_ca_system_score_gemma":0.0009589056,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009901645,"about_ca_topic_score_gemma":0.001534323,"domain_scores_codex":[0.9898648,0.003176728,0.00159743,0.0008113047,0.003917151,0.0006326251],"domain_scores_gemma":[0.9714977,0.009519977,0.01181638,0.001624238,0.004033988,0.001507703],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00004778177,0.0001195131,0.9940493,0.00003251263,0.00001710941,0.0001466212,0.0007850579,0.0000287488,0.0002619051,0.00001321795,0.00004500555,0.004453233],"study_design_scores_gemma":[0.000006054325,0.0007397021,0.9943781,0.00004294774,0.00003451245,0.001423657,0.00195983,0.0002274856,0.0007120701,0.00002580543,0.0004326074,0.00001725088],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9994097,0.000147563,0.0002172734,0.00001374118,0.000002287828,0.00002886094,0.00002715665,0.000002503806,0.000150899],"genre_scores_gemma":[0.9991252,0.0001564765,0.0005016885,0.00002090508,0.000003533052,0.00002956801,0.00007014581,0.000004443484,0.00008803501],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9907939,"threshold_uncertainty_score":0.04868716,"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."}}