{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01088098,0.0003214601,0.000328227,0.001456921,0.0002458315,0.0007128504,0.0004998118,0.0005570082,0.0009483682],"category_scores_gemma":[0.05188836,0.0001880501,0.0005708125,0.0008195036,0.0005780947,0.0007091431,0.0007519159,0.0005862106,0.0002584454],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004392668,"about_ca_system_score_gemma":0.0005934259,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000825451,"about_ca_topic_score_gemma":0.0008886407,"domain_scores_codex":[0.9822155,0.01015377,0.001207244,0.0008972902,0.00502082,0.0005054158],"domain_scores_gemma":[0.958482,0.02824775,0.00681577,0.00159952,0.004024489,0.0008304879],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0007473929,0.001135249,0.9007593,0.0002024869,0.0001671047,0.0001781633,0.006202339,0.0004657916,0.003168957,0.0003595593,0.0003093533,0.08630442],"study_design_scores_gemma":[0.00003394957,0.001962235,0.9875572,0.0000419855,0.00004042357,0.0006145498,0.00235593,0.00286968,0.003020451,0.0003486123,0.001123222,0.00003184954],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9958027,0.0001002423,0.002580822,0.00008309148,0.000015139,0.0001296879,0.00005655155,0.00001844899,0.001213318],"genre_scores_gemma":[0.9965705,0.00004991305,0.002744502,0.00004357685,0.000008396674,0.0001650637,0.00007787647,0.000006081396,0.000334111],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01088098,"threshold_uncertainty_score":0.05754477,"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."}}