{"id":"W4312094293","doi":"10.1037/spq0000527","title":"Potential scoring and predictive bias in interim and summative writing assessments.","year":2022,"lang":"en","type":"article","venue":"School Psychology","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"Iowa Department of Education","keywords":"Interim; Summative assessment; Rubric; Psychology; PsycINFO; Test (biology); Scale (ratio); Mathematics education; Medical education; Formative assessment; Medicine; MEDLINE; Political science","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":[],"consensus_categories":[],"category_scores_codex":[0.001111172,0.00009235267,0.0001560286,0.0001661655,0.0004808505,0.00007765598,0.0001692099,0.00005145599,0.0005080285],"category_scores_gemma":[0.00007843638,0.0001048495,0.00002174572,0.0002509191,0.0002074557,0.0002346474,0.0002580246,0.0003765204,0.000005977596],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001024868,"about_ca_system_score_gemma":0.00004999572,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002546858,"about_ca_topic_score_gemma":0.0001937153,"domain_scores_codex":[0.9984127,0.0005065579,0.0002036111,0.0003381095,0.0002461249,0.0002928741],"domain_scores_gemma":[0.9996176,0.00009070896,0.0000876706,0.0000885429,0.00002500217,0.00009047173],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001183712,0.00009240854,0.984849,0.000004839682,0.00003467344,0.00003248347,0.006695512,0.000006626067,0.0004090577,0.002090669,0.0008921977,0.004774112],"study_design_scores_gemma":[0.001175193,0.0002690016,0.9225516,0.00001711676,0.0000131837,0.000005635533,0.07048182,0.00009168391,0.000003239372,0.002972649,0.002270423,0.0001484859],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9654967,0.0002252558,0.0002087722,0.001191032,0.0006709421,0.0002679231,0.000008900325,0.00003150347,0.03189896],"genre_scores_gemma":[0.9986435,0.0001653965,0.0003108448,0.0003494874,0.0001580689,0.00006981349,0.000005519419,0.000008474062,0.0002888553],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.06378631,"threshold_uncertainty_score":0.5562556,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0793813733520772,"score_gpt":0.4492791782884478,"score_spread":0.3698978049363706,"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."}}