{"id":"W4233550164","doi":"10.31234/osf.io/vr8ut","title":"How Many Tiers Do We Need? Type I Errors and Power in Multiple Baseline Designs","year":2020,"lang":"en","type":"preprint","venue":"","topic":"Behavioral and Psychological Studies","field":"Psychology","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal; Institut universitaire en santé mentale de Montréal; Institut Universitaire en Santé Mentale de Québec","funders":"","keywords":"Baseline (sea); Type I and type II errors; Multiple baseline design; Power (physics); Computer science; Replicate; Statistical power; Statistics; Empirical research; Reliability engineering; Mathematics; Psychology; Engineering","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":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0002475323,0.0004033163,0.0005668658,0.0001082685,0.00005955137,0.00007149879,0.000285875,0.0005873549,0.002847857],"category_scores_gemma":[0.00008293356,0.0002970374,0.0001112307,0.0002336517,0.0001675348,0.00003022124,0.0004855075,0.0009729027,0.0002251808],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003094233,"about_ca_system_score_gemma":0.00001216164,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003863241,"about_ca_topic_score_gemma":0.0001060233,"domain_scores_codex":[0.9979962,0.0002243637,0.0003130504,0.0009100284,0.0001516608,0.0004047634],"domain_scores_gemma":[0.9991375,0.0001394837,0.0001084081,0.0003839235,0.0000524552,0.0001782222],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002096068,0.001516556,0.4804679,0.0001840546,0.0005241425,0.000991614,0.01158577,0.00001776921,0.0009940366,0.005746441,0.3870304,0.1088453],"study_design_scores_gemma":[0.004217803,0.001449404,0.5774666,0.0002662752,0.0002284919,0.00003361163,0.01662271,0.0001454852,0.00006274815,0.01431062,0.3827368,0.002459446],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8159276,0.02071748,0.003760984,0.1072925,0.007286787,0.002151109,0.0001971398,0.0005693155,0.04209705],"genre_scores_gemma":[0.9865425,0.0007682327,0.001836444,0.001310177,0.0001250335,0.0000864902,0.00002636055,0.00003456043,0.009270265],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1706148,"threshold_uncertainty_score":0.9999482,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.318544135642937,"score_gpt":0.3610885468397447,"score_spread":0.04254441119680769,"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."}}