{"id":"W2958621914","doi":"10.25035/pad.2019.01.003","title":"Criterion-related Validity of Forced-Choice Personality Measures: A Cautionary Note Regarding Thurstonian IRT versus Classical Test Theory Scoring","year":2019,"lang":"en","type":"article","venue":"Personnel Assessment and Decisions","topic":"Personality Traits and Psychology","field":"Psychology","cited_by":30,"is_retracted":false,"has_abstract":true,"ca_institutions":"Wilfrid Laurier University","funders":"","keywords":"Two-alternative forced choice; Psychology; Item response theory; Personality; Criterion validity; Test (biology); Social psychology; Statistics; Personality test; Econometrics; Psychometrics; Mathematics; Test validity; Cognitive psychology; Construct validity; Clinical psychology","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":[0.231363,0.0005468425,0.001644019,0.002830188,0.001815893,0.002762827,0.002542648,0.0008132685,0.0008020137],"category_scores_gemma":[0.3656109,0.0004943261,0.002420607,0.002464114,0.003668758,0.002255802,0.001838834,0.002833863,0.0002132982],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001442418,"about_ca_system_score_gemma":0.002673399,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006592582,"about_ca_topic_score_gemma":0.01575899,"domain_scores_codex":[0.8735476,0.07876231,0.01610927,0.005986197,0.02465862,0.0009359139],"domain_scores_gemma":[0.588299,0.310214,0.01923743,0.03066944,0.05023496,0.001345],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001123148,0.000278285,0.7342243,0.003563098,0.005673042,0.0004075928,0.01194042,0.001516499,0.003808204,0.01265519,0.01302958,0.2117805],"study_design_scores_gemma":[0.000225834,0.001961049,0.90142,0.00493422,0.002020953,0.001378464,0.005397597,0.01067551,0.0100695,0.0293257,0.03225508,0.0003361084],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7379029,0.03306609,0.1768284,0.01987926,0.005076255,0.001405983,0.001436715,0.0005096074,0.02389478],"genre_scores_gemma":[0.9284143,0.001433986,0.06400712,0.002761422,0.0005997377,0.0008237222,0.0004814708,0.0001826421,0.001295556],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.768637,"threshold_uncertainty_score":0.9478663,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.117747935211502,"score_gpt":0.4145756678622894,"score_spread":0.2968277326507874,"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."}}