{"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":"codex-gemma-dda1882f352a","candidate_categories":["sts"],"consensus_categories":[],"category_scores_codex":[0.004217168,0.0001026513,0.000131468,0.0001512817,0.0006038258,0.0002140546,0.001076728,0.000124798,0.0005515186],"category_scores_gemma":[0.003424054,0.00006362227,0.00005692539,0.001981858,0.00514395,0.0003396605,0.000122336,0.0003558764,0.00002798544],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003272787,"about_ca_system_score_gemma":0.01698859,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005144376,"about_ca_topic_score_gemma":0.001717096,"domain_scores_codex":[0.9951572,0.0001777482,0.0003227386,0.0003636829,0.003483995,0.0004946636],"domain_scores_gemma":[0.998613,0.0004097871,0.00003574921,0.0001802429,0.0001658179,0.000595387],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.00002091661,0.002007755,0.3309079,0.00003450269,0.0000394527,0.00001828575,0.0497055,4.934384e-7,0.00004409982,0.3957878,0.04725529,0.174178],"study_design_scores_gemma":[0.0002567682,0.0001938011,0.649101,0.001193204,0.00004052146,0.000009089805,0.08070467,0.000991024,0.00005314792,0.008516166,0.258505,0.00043559],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8271158,0.001942451,0.00009146461,0.1376122,0.01045476,0.0005733565,0.000002910582,0.0001061272,0.02210095],"genre_scores_gemma":[0.9979402,0.0003633618,0.0000613451,0.0001751573,0.000526085,0.0001042137,0.000001910147,0.00000600422,0.0008216641],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3872716,"threshold_uncertainty_score":0.9975635,"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."}}