{"id":"W2910137107","doi":"10.7146/dut.v14i26.106217","title":"Intended and unintended test constructs in a Multiple-Mini admission Interview. A validity study.","year":2019,"lang":"en","type":"article","venue":"Dansk Universitetspædagogisk Tidsskrift","topic":"Medical Education and Admissions","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Test (biology); Psychology; Test validity; Unintended consequences; Applied psychology; Clinical psychology; Psychometrics; Political science; Geology","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0004131612,0.0002859044,0.0005467832,0.0004596513,0.0000970121,0.00002591224,0.0002121885,0.0001673982,0.003322274],"category_scores_gemma":[0.003428749,0.0002439122,0.00009762061,0.0005541723,0.0002666909,0.0001718244,0.0002211241,0.0005952077,0.0001361269],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001334045,"about_ca_system_score_gemma":0.000628677,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001677783,"about_ca_topic_score_gemma":0.0000699158,"domain_scores_codex":[0.9980395,0.0002140207,0.0004065662,0.0005975491,0.0003626757,0.0003796429],"domain_scores_gemma":[0.9975163,0.0004321019,0.0001280478,0.0005175406,0.0001613343,0.00124463],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000381875,0.002318734,0.9658872,0.0002053607,0.00008043853,0.0004218211,0.008728913,0.000001005937,0.006628652,0.0006267224,0.002920042,0.01179926],"study_design_scores_gemma":[0.02216507,0.003226415,0.7016608,0.001682546,0.0003050622,0.0003954323,0.1399045,0.002332693,0.001379569,0.0002149381,0.1258918,0.0008412436],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9840398,0.0000916188,0.00003581309,0.003988869,0.0003137128,0.001108395,0.00001930543,0.0001107353,0.01029171],"genre_scores_gemma":[0.9912775,0.00009472749,0.0005084064,0.001161258,0.00004017188,0.000005262876,0.00005797064,0.00002436079,0.006830323],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2642264,"threshold_uncertainty_score":0.9975888,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06462648106310143,"score_gpt":0.3233371688896016,"score_spread":0.2587106878265002,"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."}}