{"id":"W4398258877","doi":"10.7910/dvn/ocniwd","title":"Replication Data for: Do Voters Benchmark Economic Performance?","year":2018,"lang":"en","type":"dataset","venue":"Harvard Dataverse","topic":"Media Influence and Politics","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal","funders":"","keywords":"Replication (statistics); Benchmark (surveying); Computer science; Biology; Geography; Virology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"dataset","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"dataset","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"}],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.001809016,0.0002293426,0.0002840175,0.0001240538,0.000553863,0.0002292637,0.003076479,0.000376664,0.01390305],"category_scores_gemma":[0.001101615,0.0002435265,0.00006363392,0.00007569489,0.0005034224,0.0009932166,0.0005262272,0.0002087328,0.04885983],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002625599,"about_ca_system_score_gemma":0.0009845842,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003702987,"about_ca_topic_score_gemma":0.002922898,"domain_scores_codex":[0.997799,0.00009606717,0.0003936461,0.0008593788,0.0003043271,0.0005476009],"domain_scores_gemma":[0.9931858,0.0002232096,0.0003268565,0.005936299,0.00008526844,0.000242539],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00001693707,0.00001358438,0.00004063663,0.00007873233,0.0000312589,0.000001096785,0.0001663891,6.909281e-7,3.513964e-7,0.0003588899,0.9986158,0.0006756465],"study_design_scores_gemma":[0.0001810017,0.00004233935,0.00001917467,0.00006391132,0.0001211068,0.000001163311,0.0001854597,0.00005662093,0.000003526526,0.00009975563,0.9989295,0.0002964519],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"dataset","genre_gemma":"dataset","genre_scores_codex":[0.0002637151,0.00000135694,0.000007477269,0.00004192165,0.002329028,0.0006092325,0.9954637,0.00003634866,0.001247276],"genre_scores_gemma":[0.00001985034,0.001994454,0.0003712993,0.0006055085,0.003402696,0.0000642437,0.9931484,0.00001549806,0.0003779797],"genre_candidate":"dataset","genre_consensus":"dataset","teacher_disagreement_score":0.03495679,"threshold_uncertainty_score":0.9930723,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05475970465636836,"score_gpt":0.3426111790318015,"score_spread":0.2878514743754331,"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."}}