{"id":"W2091897869","doi":"10.1515/1548-923x.2487","title":"High Time for a Change: Psychometric Analysis of Multiple-Choice Questions in Nursing","year":2012,"lang":"en","type":"article","venue":"International Journal of Nursing Education Scholarship","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":14,"is_retracted":false,"has_abstract":true,"ca_institutions":"Dalhousie University","funders":"","keywords":"Multiple choice; Test (biology); Psychology; Cognition; Educational measurement; MEDLINE; Nursing; Medical education; Medicine; Curriculum; Pedagogy","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":[0.06439494,0.0004425272,0.000622264,0.00209387,0.0008146123,0.001357615,0.001091564,0.001053494,0.001660777],"category_scores_gemma":[0.2220396,0.0003810723,0.001933814,0.001879354,0.001144536,0.003081501,0.001732212,0.00150872,0.0002883004],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001115522,"about_ca_system_score_gemma":0.00126278,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009648381,"about_ca_topic_score_gemma":0.001201499,"domain_scores_codex":[0.9568343,0.02731832,0.004110416,0.001325338,0.009868195,0.0005435192],"domain_scores_gemma":[0.7012893,0.2580992,0.01484899,0.007911776,0.01549496,0.002355828],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003103349,0.0022444,0.7430353,0.000500907,0.0008413538,0.0001015028,0.01204349,0.002253021,0.0019416,0.001211674,0.001055852,0.2316675],"study_design_scores_gemma":[0.0001983348,0.006660456,0.9715837,0.0002906505,0.0001640055,0.00021179,0.004555841,0.009995096,0.002199672,0.001456939,0.002578062,0.0001054849],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9878708,0.0002863753,0.008277191,0.0002522527,0.00009283389,0.0009129812,0.0001300942,0.0000434764,0.002134148],"genre_scores_gemma":[0.9893722,0.00009173104,0.009034175,0.00009857394,0.000029095,0.0009380291,0.0001419712,0.00001981082,0.0002743845],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.935605,"threshold_uncertainty_score":0.340557,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1392415383421562,"score_gpt":0.5151177927329297,"score_spread":0.3758762543907735,"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."}}