{"id":"W2125797924","doi":"10.1207/s15328023top2801_03","title":"Difficulty and Discriminability of Introductory Psychology Test Items","year":2001,"lang":"en","type":"article","venue":"Teaching of Psychology","topic":"Innovative Teaching Methodologies in Social Sciences","field":"Social Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"Keyano College; University of Calgary","funders":"","keywords":"Psychology; Test (biology); Class (philosophy); Multiple choice; Mathematics education; Social psychology; Significant difference; Applied psychology; Statistics; Artificial intelligence","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":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.009780156,0.0001563389,0.0004679975,0.0001647752,0.0004779221,0.0000118991,0.0007053663,0.0002386761,0.00009959008],"category_scores_gemma":[0.01329541,0.0001384811,0.00007494867,0.0004346247,0.005232554,0.0001665787,0.0001229986,0.0005765582,0.000002828759],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000569945,"about_ca_system_score_gemma":0.00006264848,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001672418,"about_ca_topic_score_gemma":0.0002650859,"domain_scores_codex":[0.9953746,0.002760424,0.0005693501,0.0005234007,0.0003224374,0.0004498457],"domain_scores_gemma":[0.9969367,0.001941904,0.0004445614,0.0004479844,0.0001523744,0.00007647224],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00007104516,0.0005696932,0.719705,0.00002335719,0.0000245083,0.00000335147,0.05393001,8.744221e-7,0.02304643,0.09746268,0.003499695,0.1016634],"study_design_scores_gemma":[0.0004300658,0.0004105249,0.8888888,0.00001972123,0.00002095098,0.00001162888,0.00937307,0.000002579124,0.00007546387,0.057747,0.04279283,0.0002273524],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9035768,0.000159241,0.003497928,0.007149804,0.001198967,0.0002004036,0.000006337659,0.00008317792,0.08412734],"genre_scores_gemma":[0.9872842,0.0001375253,0.01156715,0.0004032809,0.0003446172,0.000007413795,0.000001946994,0.000009605466,0.000244243],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1691838,"threshold_uncertainty_score":0.9974746,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1279364515669357,"score_gpt":0.4971193239900201,"score_spread":0.3691828724230844,"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."}}