{"id":"W2161564801","doi":"10.1109/ccece.2002.1013027","title":"Evaluation of several Efron bootstrap methods to estimate error measures for software metrics","year":2003,"lang":"en","type":"article","venue":"","topic":"Software Reliability and Analysis Research","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary","funders":"","keywords":"Statistics; Kurtosis; Confidence interval; Statistic; Nonparametric statistics; CDF-based nonparametric confidence interval; Skewness; Computer science; Robust confidence intervals; Software quality; Sample size determination; Context (archaeology); Sample (material); Population; Mathematics; Software","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.08866277,0.001856498,0.001648766,0.006290343,0.001032667,0.00148543,0.002711089,0.002360088,0.001621744],"category_scores_gemma":[0.2566027,0.0005922418,0.001493708,0.004413468,0.001474646,0.003621734,0.002619114,0.001788573,0.0003373811],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002035253,"about_ca_system_score_gemma":0.001319298,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003027337,"about_ca_topic_score_gemma":0.003862033,"domain_scores_codex":[0.9451094,0.03587672,0.003448171,0.00182203,0.01305402,0.0006896604],"domain_scores_gemma":[0.649082,0.2997914,0.009980191,0.01303375,0.02713542,0.0009772135],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00252284,0.0006958395,0.03310682,0.001194504,0.0007058395,0.0002938158,0.002751027,0.1535248,0.003385721,0.04928616,0.002833214,0.7496994],"study_design_scores_gemma":[0.000325229,0.001437627,0.03099636,0.0004807827,0.0002903561,0.0005379252,0.000889768,0.9271769,0.00969648,0.02355514,0.004370706,0.0002426824],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1444288,0.002050322,0.8476212,0.0003742288,0.0001382775,0.0004472069,0.0002634146,0.0007140919,0.003962417],"genre_scores_gemma":[0.4075887,0.0006571277,0.5895267,0.00009419907,0.00006213852,0.000637331,0.0003743635,0.000187951,0.0008715476],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.08866277,"threshold_uncertainty_score":0.4688992,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.186371257642096,"score_gpt":0.4842736465247925,"score_spread":0.2979023888826965,"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."}}