{"id":"W2001581855","doi":"10.1103/physrevstper.10.020120","title":"Comparison of integrated testlet and constructed-response question formats","year":2014,"lang":"en","type":"article","venue":"Physical Review Special Topics - Physics Education Research","topic":"Science Education and Pedagogy","field":"Social Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"Trent University","funders":"Brock University; Trent University","keywords":"Inter-rater reliability; Reliability (semiconductor); Set (abstract data type); Proxy (statistics); Multiple choice; Baseline (sea)","routes":{"ca_aff":true,"ca_fund":true,"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":[0.003157441,0.00009809324,0.0003036936,0.00008894435,0.0004674986,0.00008314777,0.0002952913,0.00004339555,0.0002822311],"category_scores_gemma":[0.004726247,0.00009055174,0.00005739541,0.001020072,0.0008831986,0.0002440801,0.00005042629,0.0003669898,0.00007149734],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001296051,"about_ca_system_score_gemma":0.001693869,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006383997,"about_ca_topic_score_gemma":0.00008791644,"domain_scores_codex":[0.9964215,0.001992141,0.0002932646,0.0002336245,0.000770084,0.0002893863],"domain_scores_gemma":[0.9973342,0.001110719,0.0001453297,0.0002448309,0.0009547954,0.0002101382],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00001247504,0.0003027486,0.003728778,0.0001120823,0.000002621321,1.506526e-8,0.00475713,3.94694e-7,0.0001947376,0.6263855,0.006805187,0.3576983],"study_design_scores_gemma":[0.0001212246,0.000141383,0.01333796,0.0005732998,0.000017644,4.073138e-7,0.006600355,0.0002792916,0.0009023044,0.1226348,0.855204,0.000187296],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8322589,0.002368865,0.0005160827,0.02709088,0.001991271,0.001495637,0.00001432436,0.00006268912,0.1342013],"genre_scores_gemma":[0.9903128,0.00167578,0.0003175008,0.0003762759,0.006440476,0.00005605241,0.00002164306,0.000008209545,0.0007913187],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8483989,"threshold_uncertainty_score":0.5658101,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1877132760863169,"score_gpt":0.5850961455017246,"score_spread":0.3973828694154077,"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."}}