{"id":"W2568498297","doi":"10.1177/0013164416685599","title":"Retrofitting Diagnostic Classification Models to Responses From IRT-Based Assessment Forms","year":2017,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":55,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Retrofitting; Item response theory; Computer science; Process (computing); Test (biology); Data mining; Engineering; Psychometrics; Mathematics; Statistics","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":[0.008993718,0.0001813124,0.0002833432,0.0002246848,0.000899116,0.0006676288,0.0009590128,0.0001014814,0.0004272539],"category_scores_gemma":[0.126592,0.000117094,0.00008495786,0.0003042845,0.0001244911,0.0002398118,0.0001161724,0.0001997486,0.00006889915],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001152383,"about_ca_system_score_gemma":0.0001124176,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006475648,"about_ca_topic_score_gemma":0.00001110335,"domain_scores_codex":[0.9957938,0.0004965072,0.0006600329,0.0008731709,0.001882763,0.0002936843],"domain_scores_gemma":[0.9773806,0.02017188,0.0004968618,0.001054152,0.0006110707,0.0002853962],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002351562,0.0008246595,0.6559843,0.000005435209,0.00003165763,0.000002794897,0.0002034041,0.0001819361,0.002983168,0.02102429,0.01181897,0.3067042],"study_design_scores_gemma":[0.0002782426,0.0001662484,0.7916595,0.00003754191,0.000007388584,0.000001231027,0.000177543,0.000883112,0.00003165029,0.2050371,0.001583425,0.0001369729],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8676457,0.0002523679,0.07320807,0.03922619,0.001484805,0.0003982214,0.00003173963,0.00003583509,0.01771703],"genre_scores_gemma":[0.9380841,0.00001545155,0.06003863,0.001214104,0.0003000869,0.0001395614,0.000006445977,0.000006980553,0.0001946045],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3065673,"threshold_uncertainty_score":0.8807651,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.8783454710157835,"score_gpt":0.5694813197426731,"score_spread":0.3088641512731104,"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."}}