{"id":"W2562775190","doi":"10.1037/xge0000079","title":"Generalization versus contextualization in automatic evaluation revisited: A meta-analysis of successful and failed replications.","year":2015,"lang":"en","type":"review","venue":"Journal of Experimental Psychology General","topic":"Social and Intergroup Psychology","field":"Social Sciences","cited_by":25,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Social Sciences and Humanities Research Council of Canada; Universiteit Gent","keywords":"Contextualization; Generalization; Psychology; Meta-analysis; Cognitive psychology; Social psychology; Epistemology; Linguistics","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1192339,0.002594149,0.007831405,0.008394056,0.001096418,0.005533773,0.003560281,0.001956955,0.002781893],"category_scores_gemma":[0.2354645,0.001684615,0.02590285,0.009483352,0.003128746,0.004598348,0.003098932,0.002352548,0.000288161],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003496186,"about_ca_system_score_gemma":0.00360241,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006167033,"about_ca_topic_score_gemma":0.01153598,"domain_scores_codex":[0.9160632,0.04837399,0.01901858,0.008801211,0.007005443,0.0007374443],"domain_scores_gemma":[0.6539298,0.2817842,0.03592052,0.01680487,0.01043224,0.001128294],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","study_design_scores_codex":[0.002282977,0.00009777477,0.03765111,0.1377407,0.7890244,0.0002215944,0.001900682,0.0003612911,0.0004594126,0.0009746155,0.0006190462,0.02866642],"study_design_scores_gemma":[0.001124528,0.0007747404,0.03511557,0.03342362,0.920689,0.0002461669,0.0009527616,0.000463472,0.0008165187,0.001714305,0.004585965,0.00009340493],"study_design_candidate":"meta_analysis","study_design_consensus":"meta_analysis","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.05338919,0.9357058,0.00590206,0.0007210517,0.0004702316,0.001465369,0.0009594362,0.00007932571,0.001307559],"genre_scores_gemma":[0.832925,0.1485694,0.01139755,0.001128029,0.000329354,0.003350163,0.001653301,0.0001517457,0.0004954847],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.880766,"threshold_uncertainty_score":0.6305769,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3501963841645199,"score_gpt":0.5638342725341257,"score_spread":0.2136378883696057,"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."}}