{"id":"W2073141549","doi":"10.1177/1098214007312630","title":"Using Self-Assessments to Detect Workshop Success","year":2008,"lang":"en","type":"article","venue":"American Journal of Evaluation","topic":"Communication in Education and Healthcare","field":"Psychology","cited_by":49,"is_retracted":false,"has_abstract":true,"ca_institutions":"Alberta Health Services; University of British Columbia; University of Calgary; University of Saskatchewan","funders":"","keywords":"Psychology; Reliability (semiconductor); Applied psychology; Program evaluation; Scale (ratio); Gold standard (test); Medical education; Medicine; Statistics","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":[0.001268777,0.00008134867,0.0001911866,0.0002856134,0.0001696018,0.00001562298,0.0002286344,0.00002892061,0.001005049],"category_scores_gemma":[0.0001414133,0.00008006016,0.00006282853,0.0006232085,0.00005125233,0.0001225316,0.00002311462,0.0002065507,0.00005731148],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003494075,"about_ca_system_score_gemma":0.0005502186,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001360689,"about_ca_topic_score_gemma":0.00001999902,"domain_scores_codex":[0.9980091,0.0007319096,0.0004989687,0.0001142442,0.000488513,0.0001572469],"domain_scores_gemma":[0.9976409,0.0002010849,0.0006837687,0.0003474801,0.0009653667,0.0001614155],"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":[0.0003702405,0.0005274757,0.1271911,0.000009405673,0.0002855601,0.0000128943,0.04010272,0.002057989,0.0007823758,0.0003624463,0.009171954,0.8191258],"study_design_scores_gemma":[0.001405333,0.00105787,0.9662002,0.00009833491,0.0001793525,0.0008461799,0.01669606,0.001908667,0.00008593634,0.0004171952,0.01080734,0.0002975349],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9857823,0.000252695,0.008492531,0.002211454,0.0008913256,0.0002034763,8.087782e-7,0.00001480777,0.002150582],"genre_scores_gemma":[0.9626825,0.00005365141,0.03612127,0.0008218822,0.0002184056,0.00001450243,0.00000178561,0.00001329922,0.00007263539],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.839009,"threshold_uncertainty_score":0.9999081,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2077316628935172,"score_gpt":0.5490251839928058,"score_spread":0.3412935210992886,"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."}}