{"id":"W4395002215","doi":"10.1111/ijsa.12474","title":"Rating accuracy, leniency, and rater perceptions when using the RPM and BARS","year":2024,"lang":"en","type":"article","venue":"International Journal of Selection and Assessment","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"Western University; Université de Montréal","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Psychology; Perception; Applied psychology; Social psychology","routes":{"ca_aff":true,"ca_fund":true,"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":[0.0348722,0.0004669013,0.0005222766,0.001500766,0.0003273417,0.001449225,0.0006057704,0.0005913805,0.002023272],"category_scores_gemma":[0.1541632,0.0002294404,0.000624514,0.0007285053,0.0008176566,0.001476686,0.0009412339,0.000645521,0.0004749025],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003907644,"about_ca_system_score_gemma":0.0002334299,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000547467,"about_ca_topic_score_gemma":0.0008489063,"domain_scores_codex":[0.9619928,0.02331694,0.002523634,0.001656657,0.01006141,0.0004485243],"domain_scores_gemma":[0.847838,0.1047126,0.02192889,0.008324473,0.01574494,0.001451052],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.004860671,0.0009463872,0.6918819,0.0008326722,0.0009252721,0.0002832946,0.01967336,0.003192586,0.02730419,0.003725402,0.002945773,0.2434284],"study_design_scores_gemma":[0.0002117446,0.004702012,0.9414645,0.0003967669,0.0002585039,0.0007768652,0.008948659,0.02107827,0.01424056,0.002703846,0.004954998,0.0002631985],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9738142,0.0002893919,0.01674522,0.0001379958,0.0001156746,0.0002561866,0.0001490223,0.00009374897,0.008398637],"genre_scores_gemma":[0.9900745,0.00009667915,0.008769181,0.00005200425,0.00005811228,0.0002073966,0.00008591904,0.00001878805,0.0006374548],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0348722,"threshold_uncertainty_score":0.184424,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.368191937024431,"score_gpt":0.5496742645416532,"score_spread":0.1814823275172223,"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."}}