{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.008970078,0.0001673623,0.000334348,0.000400802,0.0001984456,0.0001037613,0.0004178979,0.00007152487,0.0001873679],"category_scores_gemma":[0.1097481,0.00008266083,0.000111165,0.002107907,0.0002775501,0.00006438365,0.00004419868,0.0001426403,0.000003959107],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005952946,"about_ca_system_score_gemma":0.0001185307,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005090276,"about_ca_topic_score_gemma":0.000001421721,"domain_scores_codex":[0.996316,0.0007359767,0.0008326067,0.0004527099,0.001474914,0.0001877484],"domain_scores_gemma":[0.9602904,0.03767937,0.0005363902,0.0003807231,0.001000899,0.000112242],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0004194061,0.008656241,0.7963242,0.00008407261,0.0001895068,7.959198e-7,0.0009826015,0.002358467,0.04658718,0.008919579,0.003256941,0.132221],"study_design_scores_gemma":[0.0005210896,0.001110561,0.9796048,0.00009821734,0.00001004679,0.0000048983,0.0001108185,0.0001033368,0.0007538317,0.01755325,0.00002258937,0.0001064941],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.975822,0.0008044172,0.01803228,0.004136453,0.0003508073,0.000358375,0.00001169164,0.00001201578,0.0004719102],"genre_scores_gemma":[0.9853181,0.00002704813,0.01426061,0.0002388922,0.0001006227,0.00003023563,7.114274e-7,0.000004948118,0.00001884103],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1832807,"threshold_uncertainty_score":0.8977509,"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."}}