{"id":"W4390848655","doi":"10.1016/j.fishres.2024.106950","title":"Are state-space stock assessment model confidence intervals accurate? Case studies with SAM and Barents Sea stocks","year":2024,"lang":"en","type":"article","venue":"Fisheries Research","topic":"Marine and fisheries research","field":"Environmental Science","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"Memorial University of Newfoundland","funders":"Canada First Research Excellence Fund; Ocean Frontier Institute; Memorial University of Newfoundland","keywords":"Confidence interval; Frequentist inference; Statistics; Confidence distribution; Stock assessment; Robust confidence intervals; Credible interval; Econometrics; Inference; Coverage probability; Point estimation; Mathematics; Bayesian inference; Bayesian probability; Computer science; Fishery; Fishing","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":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001427298,0.0003088295,0.0003666673,0.0001565915,0.0006142241,0.0008617174,0.0004612896,0.0000943695,0.003519907],"category_scores_gemma":[0.0003342346,0.0002311283,0.00004828111,0.0008234741,0.001710173,0.001046622,0.002161242,0.001081038,0.0000861929],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005067455,"about_ca_system_score_gemma":0.0001484712,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001696662,"about_ca_topic_score_gemma":0.004344889,"domain_scores_codex":[0.995806,0.0003773417,0.0003116978,0.0009216046,0.001602085,0.0009813074],"domain_scores_gemma":[0.9982445,0.0005701988,0.00005658454,0.000579925,0.0001857161,0.000363136],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0006181525,0.000265787,0.7035003,0.001656174,0.0004823406,0.01800894,0.0111176,0.001470498,0.0003770899,0.0004150199,0.2147085,0.04737958],"study_design_scores_gemma":[0.002448024,0.004616207,0.0876049,0.001018357,0.0001045163,0.003417314,0.05723043,0.5340862,0.001013922,0.007596883,0.2980204,0.002842866],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9426275,0.0003386978,0.001437246,0.003461236,0.0001081336,0.001208465,0.0001786142,0.0001491724,0.05049096],"genre_scores_gemma":[0.9644052,0.0009139359,0.0009446098,0.00009394846,0.00004359183,0.0004364924,0.00001368838,0.00006053854,0.03308794],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6158955,"threshold_uncertainty_score":0.997391,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1524408359253686,"score_gpt":0.4243890374158151,"score_spread":0.2719482014904465,"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."}}