{"id":"W2418897741","doi":"10.55016/ojs/ajer.v59i4.55750","title":"Fairness in Computerized Testing: Detecting Item Bias using CATSIB with Impact Present","year":2014,"lang":"en","type":"article","venue":"Alberta Journal of Educational Research","topic":"Technology and Data Analysis","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Psychology; Item response theory; Item analysis; Applied psychology; Psychometrics; Statistics; Econometrics; Social psychology; Clinical psychology; Mathematics","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"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":[0.1143379,0.0009588433,0.001087354,0.003355223,0.0007381683,0.00218832,0.001448545,0.0008663314,0.002310907],"category_scores_gemma":[0.3076606,0.0004767987,0.001360466,0.002613223,0.002101909,0.002635993,0.002936952,0.001238047,0.0004267372],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001354902,"about_ca_system_score_gemma":0.001406537,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002115098,"about_ca_topic_score_gemma":0.003346644,"domain_scores_codex":[0.8809421,0.08638761,0.007567857,0.005633567,0.01845993,0.001008935],"domain_scores_gemma":[0.6089255,0.3081982,0.02811392,0.03184826,0.02090486,0.0020094],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.005568933,0.0006859877,0.7507052,0.0002617975,0.0006125584,0.0001786489,0.003087586,0.009550896,0.004928175,0.005750624,0.0009854699,0.217684],"study_design_scores_gemma":[0.0005968247,0.006936027,0.6627411,0.0003306198,0.0005906507,0.001090737,0.001475386,0.2700217,0.02982246,0.02067359,0.00545791,0.0002629416],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"methods","genre_scores_codex":[0.7178646,0.0003615115,0.2704538,0.0002445337,0.0001276264,0.001458269,0.0002116915,0.0007331412,0.008544833],"genre_scores_gemma":[0.9334691,0.00006437205,0.06469294,0.0001646233,0.00003384216,0.0005834061,0.0001986986,0.00007149309,0.0007215556],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.1143379,"threshold_uncertainty_score":0.6046838,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1773974531280598,"score_gpt":0.4178530397295161,"score_spread":0.2404555866014563,"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."}}