{"id":"W2965155714","doi":"10.1080/00273171.2019.1633617","title":"Evaluating Equivalence Testing Methods for Measurement Invariance","year":2019,"lang":"en","type":"article","venue":"Multivariate Behavioral Research","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":57,"is_retracted":false,"has_abstract":true,"ca_institutions":"York University","funders":"","keywords":"Equivalence (formal languages); Goodness of fit; Statistics; Mathematics; Type I and type II errors; Sample size determination; Chi-square test; Measurement invariance; Test (biology); Applied mathematics; Econometrics; Structural equation modeling; Discrete mathematics; Confirmatory factor analysis","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":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2280217,0.002463226,0.002918602,0.006405384,0.002252233,0.004655917,0.003912705,0.002244075,0.01136835],"category_scores_gemma":[0.6430418,0.0008559084,0.005894138,0.007666233,0.007109332,0.006966186,0.006220888,0.00550283,0.001284851],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00341638,"about_ca_system_score_gemma":0.005256088,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002654135,"about_ca_topic_score_gemma":0.001636559,"domain_scores_codex":[0.6772435,0.2536435,0.01469097,0.01891593,0.03391831,0.001587751],"domain_scores_gemma":[0.3567798,0.557525,0.02004818,0.03570717,0.02848747,0.001452397],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001566957,0.0008330707,0.07181586,0.003089864,0.006572539,0.0003218921,0.004483234,0.01845852,0.001652407,0.2650743,0.01088409,0.6152472],"study_design_scores_gemma":[0.001652166,0.006668821,0.08814544,0.004165506,0.003018064,0.0008082587,0.004363057,0.2241672,0.005990836,0.6167332,0.0436898,0.0005976394],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.03443237,0.001307077,0.9436523,0.001066411,0.0007413863,0.003623297,0.0006196107,0.0006028673,0.01395463],"genre_scores_gemma":[0.3967861,0.0008474952,0.5840261,0.000841618,0.0003900083,0.01477967,0.0009367811,0.0003730967,0.001019204],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7719783,"threshold_uncertainty_score":0.9519867,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9836237612469282,"score_gpt":0.7740472692881768,"score_spread":0.2095764919587514,"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."}}