{"id":"W2046660743","doi":"10.1177/1098214008327931","title":"Do Self-Assessments Work to Detect Workshop Success?","year":2009,"lang":"en","type":"article","venue":"American Journal of Evaluation","topic":"Human Resource Development and Performance Evaluation","field":"Psychology","cited_by":10,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Self-assessment; Applied psychology; Psychology; Work (physics); Multilevel model; Self-report study; Computer science; Social psychology; Machine learning; Engineering","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":[0.003664411,0.0001559284,0.0002804245,0.0004410008,0.0001046441,0.00007990584,0.000231025,0.00004819937,0.001890404],"category_scores_gemma":[0.0001235932,0.0001385335,0.0000911048,0.0009472864,0.00002691745,0.0002202855,0.00001323566,0.0002336129,0.0003034644],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002583266,"about_ca_system_score_gemma":0.0001770417,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000002780913,"about_ca_topic_score_gemma":0.000002110707,"domain_scores_codex":[0.9973574,0.0004807336,0.0006121545,0.0001974223,0.001071031,0.0002813087],"domain_scores_gemma":[0.9981852,0.0001280603,0.0007170695,0.0002315806,0.0005798217,0.0001582749],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0004027112,0.00008794501,0.01794258,0.000001115338,0.00009869032,0.000006083123,0.006167089,0.0009088446,0.0001124136,0.00003234247,0.004705535,0.9695346],"study_design_scores_gemma":[0.001278901,0.001911025,0.9890778,0.0001007337,0.0001789889,0.00003951945,0.001594375,0.0002845741,0.00007587009,0.0005906018,0.004612552,0.0002550499],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9900874,0.0001599507,0.002795889,0.0009434809,0.0005975954,0.0003137962,3.842397e-7,0.00002442658,0.00507709],"genre_scores_gemma":[0.9915255,0.00001577766,0.007065607,0.0007594628,0.0004132163,0.00001533213,0.000004903749,0.00001375615,0.0001864354],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9711352,"threshold_uncertainty_score":0.999022,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04228701723909126,"score_gpt":0.418141853921784,"score_spread":0.3758548366826928,"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."}}