{"id":"W2034773786","doi":"10.1007/s10459-014-9529-1","title":"The benefits of testing for learning on later performance","year":2014,"lang":"en","type":"article","venue":"Advances in Health Sciences Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":61,"is_retracted":false,"has_abstract":false,"ca_institutions":"McGill University; Université de Sherbrooke; McMaster University","funders":"","keywords":"Licensure; Context (archaeology); Multiple choice; Test (biology); Educational measurement; Item response theory; Medical education; Medicine; Psychology; Psychometrics; Pedagogy; Curriculum; Internal medicine; Clinical psychology; Significant difference","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":[],"consensus_categories":[],"category_scores_codex":[0.00243985,0.00005405797,0.0001002507,0.0001339878,0.0004253197,0.000010029,0.0001071364,0.00002270906,0.000002896453],"category_scores_gemma":[0.003791964,0.0000367934,0.00001072384,0.000749141,0.0001846153,0.0001687057,0.000007127105,0.0001332068,0.000002689327],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009193271,"about_ca_system_score_gemma":0.0004967374,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000179814,"about_ca_topic_score_gemma":0.000008662492,"domain_scores_codex":[0.9989978,0.00003920528,0.0003635924,0.0001623446,0.0002313493,0.0002057066],"domain_scores_gemma":[0.9988258,0.0005720503,0.0002528499,0.0001180846,0.000207845,0.00002333671],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.00001192137,0.00005097469,0.1167312,0.0002305895,3.500245e-7,1.05723e-9,0.0002615772,0.003121977,0.000009717159,0.007112523,0.0001216463,0.8723475],"study_design_scores_gemma":[0.0007929216,0.006348992,0.6664248,0.003552943,0.000007870895,0.00001101755,0.002992619,0.09849141,0.001086532,0.005806008,0.2142636,0.0002212821],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9839906,0.001215572,0.0003552391,0.007831232,0.001335255,0.0005035802,1.874323e-7,0.00001444966,0.004753891],"genre_scores_gemma":[0.9778172,0.0004092449,0.01971403,0.001520098,0.0002229951,0.0001136104,0.000005284441,0.000004632465,0.0001929024],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8721262,"threshold_uncertainty_score":0.4539609,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02957695616139692,"score_gpt":0.4068746599440601,"score_spread":0.3772977037826631,"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."}}