{"id":"W4399593952","doi":"10.1002/sim.10138","title":"Non‐parametric inference on calibration of predicted risks","year":2024,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Statistical Process Monitoring","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"Canadian Institutes of Health Research","keywords":"Inference; Calibration; Parametric statistics; Computer science; Statistics; Econometrics; Parametric model; Mathematics; Artificial intelligence","routes":{"ca_aff":true,"ca_fund":true,"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":[0.09298708,0.00132846,0.002725279,0.002945612,0.001029304,0.003483158,0.003972234,0.003251076,0.003260972],"category_scores_gemma":[0.3858031,0.001143954,0.003016105,0.002119332,0.007921073,0.007052763,0.006253029,0.005856731,0.0006181214],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001189079,"about_ca_system_score_gemma":0.001406586,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009482535,"about_ca_topic_score_gemma":0.0004873243,"domain_scores_codex":[0.9392059,0.04356159,0.002478638,0.008299084,0.005311753,0.001143048],"domain_scores_gemma":[0.600652,0.3367814,0.01964165,0.0368481,0.004780522,0.001296299],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.001069475,0.0003122411,0.0515007,0.0005286266,0.00156534,0.0005766563,0.001849632,0.2777751,0.003566135,0.503704,0.001979756,0.1555723],"study_design_scores_gemma":[0.00007056189,0.0003759468,0.01336824,0.0001623276,0.0001393741,0.0002351465,0.0002143266,0.564019,0.003368704,0.416634,0.001314314,0.00009809974],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.04777624,0.0001644607,0.9501442,0.0003248208,0.0000380057,0.00009463558,0.0001285579,0.0002587196,0.001070476],"genre_scores_gemma":[0.8707578,0.0002383378,0.1268028,0.000315968,0.0001015657,0.0005015057,0.0004984085,0.0001165171,0.0006671408],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.09298708,"threshold_uncertainty_score":0.4917686,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1888057989843291,"score_gpt":0.5166805173376785,"score_spread":0.3278747183533494,"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."}}