{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007440677,0.0004843239,0.000733892,0.004203052,0.005221681,0.003240265,0.001577208,0.0009832194,0.00477814],"category_scores_gemma":[0.02386291,0.0006127111,0.0005924112,0.004174821,0.002726849,0.004933032,0.006053234,0.003029252,0.001005435],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003554841,"about_ca_system_score_gemma":0.003455945,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01914317,"about_ca_topic_score_gemma":0.05023123,"domain_scores_codex":[0.9916245,0.003340978,0.0007868817,0.0009230156,0.001884339,0.001440326],"domain_scores_gemma":[0.9702986,0.007441803,0.008657719,0.002992272,0.005942549,0.004667059],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0004402864,0.006257276,0.8539749,0.0002581307,0.0002642141,0.0006479073,0.07876014,0.0004720913,0.0006790651,0.00571756,0.004995591,0.04753283],"study_design_scores_gemma":[0.00004489058,0.0009413913,0.8349605,0.0003048974,0.00009583454,0.00107187,0.1445998,0.0005688844,0.001326389,0.001541356,0.01443061,0.000113484],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9939848,0.0002009553,0.0002220628,0.0002934269,0.00001679997,0.000034113,0.0001040536,0.000008001441,0.005135829],"genre_scores_gemma":[0.9963111,0.0004108989,0.0004856935,0.0002459958,0.00001735523,0.0001353758,0.0002045741,0.00002510799,0.002163858],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01914317,"threshold_uncertainty_score":0.03935051,"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."}}