{"id":"W4415962483","doi":"10.1128/jmbe.00205-25","title":"Question format is the best predictor of item discrimination: a multivariable analysis","year":2025,"lang":"en","type":"article","venue":"Journal of Microbiology and Biology Education","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Trent University","funders":"","keywords":"Recall; Logistic regression; Item bank; Odds; Multivariable calculus; Correlation; Exploratory analysis; Relation (database); Odds ratio","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":[],"consensus_categories":[],"category_scores_codex":[0.00774546,0.001060717,0.00116965,0.001425185,0.0006790128,0.001878687,0.001512291,0.001277903,0.006149106],"category_scores_gemma":[0.02140334,0.0006425747,0.003872423,0.001631443,0.0004730535,0.001351031,0.001392112,0.002734605,0.0009546184],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004253565,"about_ca_system_score_gemma":0.001303495,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004692957,"about_ca_topic_score_gemma":0.002092576,"domain_scores_codex":[0.9954364,0.002198344,0.0002751664,0.0009478407,0.0005476359,0.0005945651],"domain_scores_gemma":[0.9803739,0.01264889,0.002318993,0.001841179,0.001458948,0.001357982],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0007019911,0.0001468372,0.9940698,0.00002177409,0.0007044644,0.00009129423,0.00008946477,0.0003574847,0.00028355,0.00006227211,0.0005454384,0.002925623],"study_design_scores_gemma":[0.00009159737,0.001153141,0.9711311,0.00004676828,0.00143812,0.0004480898,0.0004988764,0.02314713,0.0005479848,0.0004900846,0.0009620405,0.00004519159],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9962187,0.0003207012,0.001914775,0.0002680282,0.00005592699,0.00003446377,0.0007083308,0.00005306319,0.0004260424],"genre_scores_gemma":[0.9981969,0.00006565353,0.000543157,0.00004615665,0.00004466464,0.00004330669,0.0005168284,0.00003004158,0.0005133123],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.00774546,"threshold_uncertainty_score":0.0409624,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01958637381077954,"score_gpt":0.3725553989119098,"score_spread":0.3529690251011302,"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."}}