{"id":"W2154633345","doi":"10.1177/0013164403258393","title":"Impact of Test Design, Item Quality, and Item Bank Size on the Psychometric Properties of Computer-Based Credentialing Examinations","year":2004,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":36,"is_retracted":false,"has_abstract":true,"ca_institutions":"Ministry of Education and Child Care","funders":"","keywords":"Credentialing; Computerized adaptive testing; Item bank; Test (biology); Quality (philosophy); Computer science; Consistency (knowledge bases); Test design; Item response theory; Psychology; Psychometrics; Applied psychology; Statistics; Reliability engineering; Artificial intelligence; Medical education; Test method; Medicine; Mathematics; Clinical psychology","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":[0.1527274,0.0009565259,0.001024165,0.001759469,0.0007150308,0.002489357,0.001026183,0.001597321,0.001461806],"category_scores_gemma":[0.4580861,0.0008054582,0.001574596,0.002421489,0.002388373,0.002723113,0.001893225,0.001730985,0.000322833],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001879051,"about_ca_system_score_gemma":0.002214581,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001103556,"about_ca_topic_score_gemma":0.001363959,"domain_scores_codex":[0.8400878,0.1161624,0.0207157,0.005545883,0.01602304,0.001465336],"domain_scores_gemma":[0.1814991,0.7594377,0.02257628,0.01998978,0.01364179,0.002855373],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.01966206,0.002738857,0.7362366,0.0007592489,0.001375723,0.0004749639,0.003241818,0.01281969,0.01646142,0.002482665,0.00084041,0.2029066],"study_design_scores_gemma":[0.002560992,0.02320324,0.915878,0.0005014463,0.001329363,0.001553587,0.0009260727,0.02807233,0.01772671,0.004939726,0.003015033,0.0002934881],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9754636,0.001314844,0.01987076,0.0006619621,0.00008386803,0.0006475701,0.0001481896,0.0001450584,0.001664292],"genre_scores_gemma":[0.9718269,0.0002432617,0.02639953,0.0003247294,0.00005266113,0.0005436009,0.0002271339,0.00008458787,0.0002974592],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8472726,"threshold_uncertainty_score":0.8077092,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.8330425941791564,"score_gpt":0.5172334143779688,"score_spread":0.3158091798011876,"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."}}