{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0004847371,0.00004522265,0.00006239995,0.00003413166,0.0006196789,0.00002406262,0.0001751117,0.00001448678,0.00001899016],"category_scores_gemma":[0.00008538496,0.00003339635,0.00002168869,0.0001103304,0.00012226,0.0001885226,0.00001568913,0.0000462886,9.18475e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006874365,"about_ca_system_score_gemma":0.0002544023,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001401477,"about_ca_topic_score_gemma":0.00001228786,"domain_scores_codex":[0.9994733,0.00008131644,0.0001628432,0.00005055539,0.0001489012,0.00008310065],"domain_scores_gemma":[0.9993945,0.0001560662,0.0001749941,0.00006840492,0.0001549,0.00005114955],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00002918829,0.00132601,0.8616878,0.000002127447,0.00001313324,1.013575e-7,0.1208418,4.274575e-7,0.01226136,0.0005254963,0.002472043,0.0008405245],"study_design_scores_gemma":[0.000250088,0.0003688071,0.9011548,0.000009190127,0.00004770494,0.00001734554,0.09686705,6.430805e-7,0.0006157577,0.0001355884,0.0004816639,0.0000513832],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9967563,0.0009074518,0.000012243,0.0007078387,0.0002748602,0.0004490142,9.432536e-7,0.000005135684,0.0008862846],"genre_scores_gemma":[0.998654,0.0001076402,0.000507116,0.00003962985,0.000234304,0.0000288449,4.857869e-7,0.000004038719,0.0004239044],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03946698,"threshold_uncertainty_score":0.4766132,"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."}}