{"id":"W2041542884","doi":"10.1007/s10459-007-9068-0","title":"Undesired variance due to examiner stringency/leniency effect in communication skill scores assessed in OSCEs","year":2007,"lang":"en","type":"article","venue":"Advances in Health Sciences Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":109,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Calgary","funders":"","keywords":"Variance (accounting); Rasch model; Objective structured clinical examination; Raw score; Psychology; Medical education; Communication skills; Applied psychology; Medicine; Statistics; Developmental psychology; Mathematics; Raw data","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":[],"consensus_categories":[],"category_scores_codex":[0.007596321,0.0001684841,0.0003420397,0.001463137,0.0001678703,0.00002965833,0.0003905031,0.00009572913,0.0000281406],"category_scores_gemma":[0.003123491,0.0001586168,0.00001906915,0.00565175,0.0002421261,0.000833507,0.00004046159,0.0003883488,0.0000119699],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001137316,"about_ca_system_score_gemma":0.001606176,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001957504,"about_ca_topic_score_gemma":0.004673593,"domain_scores_codex":[0.9970395,0.0002609564,0.001061957,0.000517692,0.0005499715,0.0005699188],"domain_scores_gemma":[0.9983483,0.0006313337,0.0003092086,0.0004482039,0.0001537672,0.0001091777],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00006794206,0.0008444515,0.6213294,0.0004796113,0.00000107758,0.000002542171,0.003489248,0.0007940701,0.0001496554,0.005796407,0.0001854197,0.3668602],"study_design_scores_gemma":[0.0005804597,0.0004635019,0.9884116,0.001807426,0.000002448363,0.00001131712,0.002886673,0.0006844192,0.000344213,0.003337198,0.001302959,0.0001678055],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9813979,0.004285762,0.002348737,0.006337939,0.001366633,0.001496256,6.406236e-7,0.00003281467,0.002733287],"genre_scores_gemma":[0.9633992,0.000960944,0.03261641,0.002545832,0.000122062,0.0002415017,0.0000299814,0.00001132321,0.00007279894],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3670821,"threshold_uncertainty_score":0.6468207,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0203232632687586,"score_gpt":0.4305531748638134,"score_spread":0.4102299115950548,"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."}}