{"id":"W4307097012","doi":"10.31234/osf.io/2hswm","title":"The Accuracy of Bayesian Model Fit Indices in Selecting Among Multidimensional Item Response Theory Models","year":2022,"lang":"en","type":"preprint","venue":"","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"","keywords":"Curse of dimensionality; Nested set model; Deviance information criterion; Bayesian probability; Statistics; Bayesian information criterion; Computer science; Mathematics; Econometrics; Bayesian inference; Data mining","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":[0.1731564,0.002590975,0.003094642,0.008282553,0.002164305,0.00666179,0.003314032,0.003413168,0.001937141],"category_scores_gemma":[0.4204609,0.001236164,0.003160531,0.005471537,0.002563859,0.008132798,0.00325123,0.004567339,0.0008504624],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002419467,"about_ca_system_score_gemma":0.003661775,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007286483,"about_ca_topic_score_gemma":0.006937787,"domain_scores_codex":[0.9100174,0.06919102,0.004594719,0.006044928,0.009127465,0.001024475],"domain_scores_gemma":[0.5988175,0.3603705,0.01144335,0.01514574,0.01234993,0.001872922],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.002292377,0.0007603722,0.1945918,0.001277379,0.004707108,0.0005719215,0.003771329,0.4017202,0.001730295,0.06376432,0.007958774,0.3168542],"study_design_scores_gemma":[0.0002840748,0.0006312406,0.02193536,0.0007780072,0.0005718604,0.0002688645,0.0008331212,0.9054415,0.001541737,0.06474907,0.002667085,0.0002981954],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.2754765,0.002924177,0.7095746,0.001483295,0.0002008319,0.000784902,0.0008369558,0.001146779,0.007571858],"genre_scores_gemma":[0.7765105,0.0007237027,0.2193264,0.0004669754,0.00008119401,0.000608294,0.001435605,0.0003266388,0.0005207341],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.1731564,"threshold_uncertainty_score":0.9157493,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4788933920804174,"score_gpt":0.4867495426273737,"score_spread":0.007856150546956375,"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."}}