{"id":"W1882962089","doi":"10.1111/j.1467-8489.2011.00552.x","title":"Predicting versus testing: a conditional cross-forecasting accuracy measure for hypothetical bias*","year":2011,"lang":"en","type":"article","venue":"Australian Journal of Agricultural and Resource Economics","topic":"Economic and Environmental Valuation","field":"Economics, Econometrics and Finance","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"Genome Canada; Agriculture Food and Rural Development; University of Alberta","funders":"Genome Alberta; Genome Canada; Alberta Crop Industry Development Fund","keywords":"Measure (data warehouse); Econometrics; Context (archaeology); Preference; Statistics; Divergence (linguistics); Computer science; Mathematics; Data mining","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":[],"consensus_categories":[],"category_scores_codex":[0.0007208404,0.0001783221,0.0003538384,0.00007850413,0.0001975307,0.0001027801,0.0001755691,0.0001376596,0.0001304706],"category_scores_gemma":[0.0004454372,0.0001496545,0.0001806753,0.00005101804,0.0001413753,0.0004846519,0.00003806378,0.0002135351,0.00002374283],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001156648,"about_ca_system_score_gemma":0.00001589604,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000193111,"about_ca_topic_score_gemma":0.00000440524,"domain_scores_codex":[0.9984851,0.00001490412,0.0009580564,0.0002438555,0.00002970303,0.0002683775],"domain_scores_gemma":[0.9983359,0.000271325,0.001079042,0.00008195761,0.00005594188,0.0001758314],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0004295356,0.0001582683,0.9711764,0.00005709323,0.0004138166,0.00000438401,0.001733921,0.004376586,0.0001101173,0.0174015,0.001164675,0.002973696],"study_design_scores_gemma":[0.002535544,0.000579617,0.9859065,0.00004164203,0.00004494081,0.0002846486,0.001003206,0.0004601726,0.0003487445,0.005777987,0.002693883,0.0003230723],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9960908,0.0001415511,0.00008307106,0.0002924559,0.0003243656,0.0001545564,0.0001114309,0.000009587922,0.002792178],"genre_scores_gemma":[0.995661,0.00002418763,0.003360966,0.00004327661,0.000456087,0.00000661233,0.00002124723,0.00001557751,0.0004110216],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01473013,"threshold_uncertainty_score":0.6102733,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3925589470092689,"score_gpt":0.250860995632072,"score_spread":0.1416979513771969,"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."}}