{"id":"W4390315320","doi":"10.1145/3623762.3633498","title":"Multi-Institutional Multi-National Studies of Parsons Problems","year":2023,"lang":"en","type":"article","venue":"","topic":"Teaching and Learning Programming","field":"Computer Science","cited_by":11,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"DePaul University","keywords":"Computer science; Scratch; Coding (social sciences); Code (set theory); Institution; Pair programming; Programming language; Mathematics education; Psychology; Software; Sociology; Software development","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":[0.0007024278,0.00007851816,0.0001098649,0.0001300919,0.0002064469,0.000035415,0.0003131975,0.00002648633,0.000003282904],"category_scores_gemma":[0.0003735842,0.00005938261,0.00005321636,0.000422965,0.00008515313,0.0001832954,0.0001833936,0.000112601,0.00009812297],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002826051,"about_ca_system_score_gemma":0.00008448,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003389717,"about_ca_topic_score_gemma":0.00002206387,"domain_scores_codex":[0.9990995,0.00005142805,0.0001634982,0.0002106595,0.0003054412,0.0001694523],"domain_scores_gemma":[0.9994878,0.0001436505,0.00005717562,0.0001291836,0.0001480185,0.00003421408],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000004826671,0.0009397065,0.03162238,0.0002852454,0.0004664462,0.00003148501,0.03117478,0.1578748,0.003797121,0.3800155,0.009529131,0.3842585],"study_design_scores_gemma":[0.00130114,0.0000958312,0.06341097,0.0001321838,0.000008573039,0.00001885197,0.0007209826,0.826169,0.0003576129,0.000641106,0.1067793,0.0003644949],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01111506,0.0002157158,0.9855402,0.001465182,0.0003894334,0.0001284181,0.00000242491,0.0009074901,0.0002360668],"genre_scores_gemma":[0.5432904,0.00001107002,0.4544848,0.00004533255,0.00002239941,0.00001959242,0.000003915223,0.000003338518,0.002119123],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.6682941,"threshold_uncertainty_score":0.2421553,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1932726159606293,"score_gpt":0.3767914645232791,"score_spread":0.1835188485626498,"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."}}