{"id":"W2095178943","doi":"10.3200/jexe.76.3.243-258","title":"Subset Testing: Prevalence and Implications for Study Behaviors","year":2008,"lang":"en","type":"article","venue":"The Journal of Experimental Education","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"Mount Allison University","funders":"","keywords":"Psychology; Set (abstract data type); Clinical psychology; Computer science","routes":{"ca_aff":true,"ca_fund":false,"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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.055103,0.0003965979,0.0005966429,0.00383681,0.001106505,0.001711718,0.001407504,0.00127274,0.001725848],"category_scores_gemma":[0.2326916,0.0006139776,0.0007775064,0.00167876,0.003226799,0.003424362,0.002770925,0.001393162,0.0002242406],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006703786,"about_ca_system_score_gemma":0.0009749765,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001884641,"about_ca_topic_score_gemma":0.00233031,"domain_scores_codex":[0.9372506,0.03261997,0.008453901,0.005985531,0.01433068,0.001359391],"domain_scores_gemma":[0.6079253,0.2920294,0.06633477,0.01770272,0.01227258,0.003735288],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001797952,0.0002149031,0.9812491,0.0001028341,0.00006969346,0.00008261226,0.002918707,0.0000320911,0.000362632,0.0002015386,0.0001128546,0.01447317],"study_design_scores_gemma":[0.00001443413,0.0004292411,0.9925891,0.0002197516,0.00008305231,0.0009121221,0.003611244,0.0005395629,0.0005003671,0.0006344689,0.0004474783,0.00001912022],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9925388,0.001283242,0.002663318,0.0006749217,0.00004150004,0.0001517482,0.00008656266,0.00003478083,0.002525223],"genre_scores_gemma":[0.9984685,0.0002111815,0.0009559287,0.00009889009,0.00002280469,0.00009872217,0.0000508012,0.00001053162,0.00008259117],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.944897,"threshold_uncertainty_score":0.2914159,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1021623673174721,"score_gpt":0.4520509840703397,"score_spread":0.3498886167528676,"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."}}