{"id":"W4405015817","doi":"10.1007/s40670-024-02197-4","title":"The Predictive Power of Short Answer Questions in Undergraduate Medical Education Progress Difficulty","year":2024,"lang":"en","type":"article","venue":"Medical Science Educator","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"McMaster University","funders":"","keywords":"Formative assessment; Logistic regression; Odds; Medical education; Predictive power; Referral; Psychology; Medicine; Mathematics education; Family medicine; 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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00909804,0.0006951324,0.0006624553,0.0032474,0.000482349,0.001574407,0.001121991,0.001979489,0.003877788],"category_scores_gemma":[0.141902,0.0003318065,0.0009771432,0.001043299,0.0007472034,0.002167719,0.001856317,0.001960572,0.001008268],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004478522,"about_ca_system_score_gemma":0.0007335362,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002449123,"about_ca_topic_score_gemma":0.002109282,"domain_scores_codex":[0.9940845,0.002708897,0.0007617139,0.000540464,0.001606616,0.0002977983],"domain_scores_gemma":[0.6780688,0.2760558,0.02122243,0.005465254,0.01023913,0.008948624],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000306711,0.0002870368,0.9939255,0.00001581466,0.00005564919,0.00002831639,0.0002651142,0.0002015521,0.0001442127,0.00003093845,0.0001353029,0.004603854],"study_design_scores_gemma":[0.00002526304,0.0006097544,0.9952991,0.00002853179,0.00005106579,0.00008398908,0.0004330163,0.002800959,0.0002729215,0.0001627436,0.0002192572,0.00001330603],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9978274,0.0001380258,0.000314901,0.0001546128,0.00003957244,0.00002459582,0.0001731717,0.00002071865,0.001306994],"genre_scores_gemma":[0.9990293,0.00005806311,0.0001849754,0.00003123306,0.00002611734,0.00002446909,0.0001961425,0.000009310549,0.0004404923],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.00909804,"threshold_uncertainty_score":0.04811561,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01476621263406066,"score_gpt":0.4006816139708742,"score_spread":0.3859154013368135,"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."}}