{"id":"W3081786720","doi":"10.5539/ies.v13n9p10","title":"Investigation of Three-Tier Diagnostic and Multiple Choice Tests on Chemistry Concepts with Response Change Behaviour","year":2020,"lang":"en","type":"article","venue":"International Education Studies","topic":"Science Education and Pedagogy","field":"Social Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Test (biology); Mathematics education; Descriptive statistics; Disadvantage; Psychology; Chemistry; Mathematics; Statistics; Computer science","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0004400277,0.00008031627,0.00009973185,0.0000495504,0.0002009266,0.00003452716,0.0001504896,0.00003465006,0.0001107886],"category_scores_gemma":[0.01413891,0.00007191717,0.00001650303,0.0002141779,0.000485334,0.0002196553,0.0000268004,0.00008019871,0.00001086403],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008294677,"about_ca_system_score_gemma":0.0004468128,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007791963,"about_ca_topic_score_gemma":0.0005314872,"domain_scores_codex":[0.9990429,0.0001368361,0.000152335,0.0002063139,0.0003642619,0.00009734624],"domain_scores_gemma":[0.9966009,0.002209088,0.0001493629,0.00006966119,0.0008542538,0.0001166802],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00004496945,0.00004798183,0.9301788,0.00001171384,0.00002354624,2.795838e-7,0.06346703,0.000003455104,0.000487438,0.0004589283,0.001726795,0.003549066],"study_design_scores_gemma":[0.0002325363,0.00006982886,0.9445508,0.00008344527,0.00001409393,8.095403e-7,0.03954926,0.00001874542,0.0009743069,0.0002240031,0.01417459,0.000107543],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9518839,0.0003380557,0.00001391149,0.04615887,0.0007610391,0.0001812145,0.00001065139,0.00002358197,0.0006287927],"genre_scores_gemma":[0.9967555,0.0001116808,0.0001720101,0.001838327,0.0005266536,0.00008521885,0.000009688096,0.000005234809,0.0004956644],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.04487164,"threshold_uncertainty_score":0.9941654,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2687678070582653,"score_gpt":0.4708770704796423,"score_spread":0.202109263421377,"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."}}