{"id":"W4406051520","doi":"10.1027/1015-5759/a000877","title":"Cracks, Gaps, and Holes in Validation Practice as Evidenced From a Validation Synthesis of the English Version of the Rosenberg Self-Esteem Scale","year":2024,"lang":"en","type":"article","venue":"European Journal of Psychological Assessment","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Psychology; Scale (ratio); Self-esteem; Psychometrics; Test validity; Social psychology; Clinical psychology; Cartography","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01733264,0.0001285091,0.0003263188,0.0002333484,0.00008527304,0.0002105238,0.0009535319,0.00005044482,0.0001103015],"category_scores_gemma":[0.0571999,0.00005958099,0.0001882998,0.001510055,0.0001516458,0.0004614364,0.0002639699,0.000449224,0.000006415753],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004620176,"about_ca_system_score_gemma":0.00005382036,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000009362614,"about_ca_topic_score_gemma":4.203121e-7,"domain_scores_codex":[0.9902671,0.00664806,0.001240814,0.0003381523,0.001368666,0.000137224],"domain_scores_gemma":[0.9739674,0.02365931,0.001373967,0.0004449534,0.0004925448,0.00006186649],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0005465083,0.002306497,0.4898992,0.0001013155,0.0002686663,0.0001984301,0.005430588,0.001771164,0.04752402,0.001329041,0.008242403,0.4423822],"study_design_scores_gemma":[0.0004145455,0.0004495555,0.980904,0.0007235323,0.00009749505,0.00005671591,0.00237842,0.0001654254,0.004588816,0.004338378,0.005783284,0.00009984612],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9861865,0.0006748676,0.001802763,0.002554861,0.001462031,0.0001461966,0.000008672524,0.00001274823,0.007151383],"genre_scores_gemma":[0.9792548,0.0003494282,0.02011041,0.0001005235,0.0001308747,0.000001061329,2.760274e-7,0.000008190003,0.00004437444],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4910048,"threshold_uncertainty_score":0.9507417,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1700920110475426,"score_gpt":0.4586256186510765,"score_spread":0.2885336076035339,"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."}}