{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001795804,0.00008187249,0.000221794,0.002466033,0.0001003615,0.0001064909,0.0004191379,0.00008084884,0.0001156685],"category_scores_gemma":[0.002701036,0.00008590621,0.0001861459,0.001747647,0.00009480628,0.001284554,0.000005776787,0.0001812012,0.00000559897],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008025049,"about_ca_system_score_gemma":0.0002574199,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001410657,"about_ca_topic_score_gemma":0.0000336259,"domain_scores_codex":[0.9983528,0.000202547,0.0004889344,0.00009681682,0.0006536152,0.0002053028],"domain_scores_gemma":[0.9971325,0.0009193112,0.0005958323,0.00008167038,0.001145434,0.0001252816],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00008034272,0.002401672,0.8898499,0.000002425127,0.0002820557,9.528326e-8,0.01744288,0.00002140928,0.0004361851,0.005755888,0.0008238234,0.08290329],"study_design_scores_gemma":[0.000450583,0.00003088594,0.9929826,0.0004047514,0.0003146744,0.000002272869,0.002743735,0.00003046297,0.00003914802,0.001457734,0.001456823,0.00008634897],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9858512,0.0008875292,0.0006141335,0.005356002,0.00550088,0.0002126795,0.00001442112,0.000008987773,0.001554118],"genre_scores_gemma":[0.9943713,0.00006326128,0.003625644,0.0001056328,0.001628791,0.00001298078,0.00002359285,0.000007035657,0.000161747],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1031327,"threshold_uncertainty_score":0.3503153,"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."}}