{"id":"W2916706452","doi":"","title":"1st year Psychology Student Calibration on Exams: Insights into confidence calibration, question complexity effects, and (in)consistencies with the Dunning-Kruger effect","year":2018,"lang":"en","type":"article","venue":"2018 Conference of the Canadian Society for the Study of Education","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"Dalhousie University; Mount Saint Vincent University","funders":"","keywords":"Calibration; Psychology; Social psychology; Computer science; Mathematics; Statistics","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.03467361,0.0004880934,0.0008815155,0.003415206,0.0006592806,0.002916976,0.001646449,0.001224147,0.008060073],"category_scores_gemma":[0.3101151,0.0007151558,0.001079079,0.002317765,0.001840882,0.004073876,0.002996397,0.002777069,0.0005620465],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001098155,"about_ca_system_score_gemma":0.001019429,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003620724,"about_ca_topic_score_gemma":0.004389764,"domain_scores_codex":[0.9870646,0.007005162,0.0007775618,0.001482564,0.003274629,0.0003954332],"domain_scores_gemma":[0.4236478,0.5104694,0.02451797,0.02802637,0.0102864,0.003052088],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00110119,0.0009194023,0.9025466,0.0001127188,0.0003790506,0.00008212059,0.006460259,0.002139928,0.001617421,0.006665707,0.0007922215,0.07718345],"study_design_scores_gemma":[0.000025611,0.0002107338,0.9919212,0.00002504558,0.00008054776,0.00006067929,0.0006908587,0.002965708,0.0006977615,0.002847486,0.0004423904,0.0000319915],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9846759,0.0004445922,0.004959633,0.000283686,0.00001809193,0.00006090624,0.0001362149,0.00007147338,0.009349514],"genre_scores_gemma":[0.9979813,0.00004994516,0.001234069,0.00003775049,0.000009452024,0.00002731937,0.0001287988,0.00003527508,0.0004962045],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9653264,"threshold_uncertainty_score":0.1833738,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0409719884851442,"score_gpt":0.3553656221381101,"score_spread":0.3143936336529659,"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."}}