{"id":"W4417409629","doi":"10.13140/rg.2.2.31942.42560","title":"Pupils' self-evaluation biases influence teachers' judgments about their competence","year":2019,"lang":"en","type":"article","venue":"","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"Université du Québec à Montréal","funders":"","keywords":"Competence (human resources); Perception; Affect (linguistics); Cognitive bias","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":["insufficient_payload"],"category_scores_codex":[0.001269824,0.0001230819,0.0001427579,0.00005932048,0.0003012277,0.0001809796,0.0004085139,0.00007470493,0.002543326],"category_scores_gemma":[0.0001376857,0.0001015163,0.00005864729,0.0002925033,0.00009703702,0.000591843,0.00007200224,0.000101761,0.001101858],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001618172,"about_ca_system_score_gemma":0.0002020385,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001018272,"about_ca_topic_score_gemma":0.000498724,"domain_scores_codex":[0.9981142,0.0002281643,0.0001829959,0.0002683093,0.0008803909,0.0003258955],"domain_scores_gemma":[0.9991895,0.0002338533,0.00009413612,0.0002118291,0.0001846119,0.00008607737],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000005337805,0.0001462385,0.9502714,0.00000499598,0.00004516267,4.357612e-7,0.03662391,0.00006212419,0.0007951478,0.006340608,0.001233982,0.004470616],"study_design_scores_gemma":[0.0008778175,0.00007686509,0.9275485,0.00006589633,0.00004544249,2.726171e-7,0.03539413,0.001001624,0.0003326703,0.0007345409,0.03355657,0.0003656525],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7846679,0.00007415088,0.00001106527,0.0003581295,0.0003205128,0.0005776946,0.00000132307,0.0001529451,0.2138363],"genre_scores_gemma":[0.9932405,0.000107793,0.0004992915,0.0003964396,0.000101377,0.00003027991,0.0000065668,0.000007535391,0.005610174],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2085727,"threshold_uncertainty_score":0.9996759,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04345044272216311,"score_gpt":0.3578247868108449,"score_spread":0.3143743440886818,"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."}}