{"id":"W4378220707","doi":"10.1177/00131644231165520","title":"The Accuracy of Bayesian Model Fit Indices in Selecting Among Multidimensional Item Response Theory Models","year":2023,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"","keywords":"Curse of dimensionality; Nested set model; Deviance information criterion; Bayesian probability; Item response theory; Statistics; Bayesian information criterion; Computer science; Mathematics; Econometrics; Bayesian inference; Data mining; Psychometrics","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":[0.1734025,0.002581109,0.00311531,0.008536808,0.002141479,0.006639207,0.003311352,0.003517567,0.001935058],"category_scores_gemma":[0.4229036,0.001228505,0.003212894,0.005585087,0.002569459,0.008110695,0.003319731,0.004496467,0.0008401877],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00241061,"about_ca_system_score_gemma":0.003621945,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007304178,"about_ca_topic_score_gemma":0.006899168,"domain_scores_codex":[0.9094523,0.06947468,0.004755134,0.006020337,0.009254299,0.001043213],"domain_scores_gemma":[0.5936813,0.3655339,0.01151342,0.01526703,0.01213119,0.001873108],"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":[0.002345371,0.0007667696,0.1960155,0.001260279,0.00470047,0.0005853856,0.00371073,0.4027022,0.001743331,0.06180993,0.007931056,0.316429],"study_design_scores_gemma":[0.0002848047,0.000617032,0.02237998,0.0007658119,0.000578331,0.0002769298,0.000834453,0.9087815,0.00155182,0.06104468,0.002584863,0.0002998988],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2838096,0.002942679,0.7012675,0.001467168,0.0002012656,0.0007717238,0.0008532475,0.001146448,0.007540292],"genre_scores_gemma":[0.7757697,0.0007244527,0.2200823,0.0004600552,0.0000808296,0.0005922201,0.001452289,0.0003216896,0.0005164143],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.1734025,"threshold_uncertainty_score":0.9170511,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.7180011966789275,"score_gpt":0.5115401562574108,"score_spread":0.2064610404215167,"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."}}