{"id":"W2983818347","doi":"10.3138/ptc-2018-0110","title":"How to Optimize Measurement Protocols: An Example of Assessing Measurement Reliability Using Generalizability Theory","year":2019,"lang":"en","type":"article","venue":"Physiotherapy Canada","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo; McMaster University","funders":"","keywords":"Generalizability theory; Reliability (semiconductor); Computer science; Observational error; Terminology; Reliability engineering; Measurement uncertainty; Standard error; Data mining; Statistics; Power (physics); Mathematics","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.02526036,0.000452353,0.0009118806,0.0001775784,0.0002543094,0.0005061935,0.001515033,0.00009276248,0.0006743096],"category_scores_gemma":[0.001401469,0.0003401339,0.0002397443,0.0009953593,0.0001145528,0.0008211219,0.0001423884,0.0002045448,0.000006246549],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00259437,"about_ca_system_score_gemma":0.002913091,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.3096522,"about_ca_topic_score_gemma":0.130662,"domain_scores_codex":[0.9840025,0.002239758,0.0013324,0.001530988,0.01023045,0.0006638864],"domain_scores_gemma":[0.9919217,0.0003755248,0.0006559339,0.003131849,0.003497272,0.0004177075],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0006956255,0.001336309,0.02988643,0.0003338484,0.0001600933,0.000001869299,0.001031496,0.07897588,0.8466217,0.0007030785,0.002658326,0.03759534],"study_design_scores_gemma":[0.009422516,0.003493333,0.1360393,0.001901445,0.0001534559,0.000005734459,0.01101376,0.0573633,0.5059815,0.08031958,0.1892869,0.005019203],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9360336,0.0001277802,0.04688909,0.00136015,0.0007386068,0.01433082,0.00002325238,0.00004517808,0.0004515043],"genre_scores_gemma":[0.9865654,0.000002273412,0.01148242,0.0008630486,0.0001266951,0.0007971509,0.000001530087,0.00003946224,0.000122069],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3406402,"threshold_uncertainty_score":0.999905,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2942541175076245,"score_gpt":0.3971427089229444,"score_spread":0.1028885914153199,"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."}}