{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001959428,0.0003753989,0.0006416554,0.0002406916,0.000453617,0.00007368631,0.0003733335,0.0003846164,0.002448255],"category_scores_gemma":[0.0006353511,0.0003313408,0.0003563652,0.0003706305,0.0003547266,0.0002671828,0.000139443,0.000708905,0.0001184998],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001644273,"about_ca_system_score_gemma":0.000161059,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001821139,"about_ca_topic_score_gemma":0.00008209648,"domain_scores_codex":[0.9965687,0.000597693,0.0006453771,0.0009393401,0.0006414783,0.0006074285],"domain_scores_gemma":[0.9924117,0.006144322,0.0002594354,0.0006935123,0.0001956898,0.0002953275],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003118789,0.002733576,0.7220964,0.000124454,0.001124804,0.0001229939,0.03424519,0.00003071066,0.00895769,0.1710846,0.007537018,0.04882385],"study_design_scores_gemma":[0.006549423,0.0009065124,0.963694,0.0002310588,0.0002879598,0.0001110489,0.01099519,0.001082489,0.00001174811,0.004196446,0.01129429,0.000639831],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9443826,0.0004588299,0.003410961,0.00144735,0.002842312,0.0004321299,0.0002087709,0.0001006624,0.04671637],"genre_scores_gemma":[0.9957651,0.00006367789,0.001350033,0.000306594,0.0002218231,0.00004139686,0.00004671522,0.00004011764,0.002164487],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2415976,"threshold_uncertainty_score":0.9999139,"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."}}