{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01268253,0.0006209281,0.0008069329,0.003641541,0.0002896548,0.001788217,0.000830055,0.001281138,0.003969712],"category_scores_gemma":[0.155317,0.0003986244,0.0006053789,0.001538743,0.0008742086,0.001648952,0.001904901,0.001719967,0.0008859332],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005508434,"about_ca_system_score_gemma":0.0002092758,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000440971,"about_ca_topic_score_gemma":0.0006203259,"domain_scores_codex":[0.9888507,0.003946777,0.001653577,0.002176492,0.002812993,0.0005594349],"domain_scores_gemma":[0.7701337,0.1849943,0.02010382,0.01010901,0.01092731,0.003731786],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001086298,0.0008959725,0.9471565,0.0001535461,0.0002252006,0.0001672105,0.002421832,0.0007624115,0.01436656,0.0004664762,0.0004640907,0.0318339],"study_design_scores_gemma":[0.00005982479,0.0007503555,0.9921613,0.00004004696,0.00003659346,0.0003060405,0.0003237686,0.001755076,0.003394905,0.0006418805,0.0005039527,0.00002617158],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.996882,0.0001354535,0.001450087,0.00004779429,0.00001422567,0.00006732566,0.0001247811,0.00002043218,0.001258001],"genre_scores_gemma":[0.9977805,0.00003481614,0.001086972,0.00003651973,0.00002121912,0.00005511937,0.0004919126,0.00003147059,0.0004614695],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01268253,"threshold_uncertainty_score":0.06707239,"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."}}