{"id":"W2924341464","doi":"10.2505/4/jcst19_048_04_64","title":"Two-Stage (Collaborative) Testing in Science Teaching: Does It Improve Grades on Short-Answer Questions and Retention of Material?","year":2019,"lang":"en","type":"article","venue":"Journal of College Science Teaching","topic":"Innovative Teaching Methods","field":"Social Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia; University of Guelph","funders":"","keywords":"Mathematics education; Science education; Teaching method; Psychology; Pedagogy","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":["metaresearch","sts"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.05575238,0.0002184158,0.0004875069,0.001675621,0.002067947,0.0004435427,0.001065093,0.00007892623,0.00002433355],"category_scores_gemma":[0.0154107,0.0001635918,0.00006285538,0.002805766,0.002486634,0.003340617,0.0001891101,0.001211992,0.000001228183],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007719107,"about_ca_system_score_gemma":0.002136012,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006789943,"about_ca_topic_score_gemma":0.0002372541,"domain_scores_codex":[0.994278,0.001688939,0.0009976822,0.0004990762,0.001887521,0.0006488109],"domain_scores_gemma":[0.996756,0.0007952309,0.0009746071,0.0002905655,0.0009826703,0.0002009039],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"qualitative","study_design_scores_codex":[0.00004096285,0.0001204629,0.02685155,0.00002046428,0.000006214923,0.00001004259,0.01662559,0.0001350141,0.7692889,0.1825743,0.00001191935,0.004314564],"study_design_scores_gemma":[0.00498043,0.005361386,0.3521699,0.006567902,0.0001573084,0.0001284092,0.3559779,0.008486263,0.2165029,0.04115422,0.005671443,0.002841935],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9691067,0.00001567292,0.0005294846,0.0009805629,0.001676171,0.0004568069,0.00001777972,0.00002683485,0.02719002],"genre_scores_gemma":[0.9439825,0.000005886747,0.0552286,0.00007059051,0.000203627,0.000003392034,1.884762e-7,0.00001357158,0.0004916419],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.552786,"threshold_uncertainty_score":0.9992312,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04291436430538549,"score_gpt":0.3999190650102065,"score_spread":0.357004700704821,"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."}}