{"id":"W4223957752","doi":"10.1177/01466216221084210","title":"A Comparison of Modern and Popular Approaches to Calculating Reliability for Dichotomously Scored Items","year":2022,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Multi-Criteria Decision Making","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University; Université de Montréal","funders":"","keywords":"Cronbach's alpha; Reliability (semiconductor); Monte Carlo method; Statistics; Mathematics; Econometrics; Psychology; Computer science; Psychometrics","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":[0.01210779,0.0002534355,0.000780143,0.0002300769,0.0004282589,0.0001298151,0.0009510498,0.00009398497,0.0001344537],"category_scores_gemma":[0.002734555,0.0001911907,0.000162053,0.0006935742,0.0001100831,0.00005216681,0.0005461869,0.0002716189,0.000009343168],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001381618,"about_ca_system_score_gemma":0.00001821389,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001102629,"about_ca_topic_score_gemma":0.000007912944,"domain_scores_codex":[0.9926737,0.0004134008,0.001562578,0.001372703,0.00355502,0.000422586],"domain_scores_gemma":[0.9972423,0.0007453648,0.0004855465,0.001044187,0.0002493871,0.0002331832],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.003321746,0.003234946,0.0456148,0.00005482446,0.00006921624,0.000004112278,0.00647421,0.03845524,0.123373,0.01170195,0.005301985,0.762394],"study_design_scores_gemma":[0.009644388,0.00388258,0.1886887,0.0000912305,0.0001444445,0.00002820234,0.00765675,0.2416549,0.00550148,0.4883482,0.05216604,0.002193017],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8124874,0.0001932912,0.1834043,0.0004866672,0.0002820351,0.002199067,0.00003920848,0.00005090927,0.0008571647],"genre_scores_gemma":[0.9604758,3.609496e-7,0.03814869,0.0003553931,0.00004084373,0.0009342032,0.000004925385,0.0000190524,0.00002070085],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.760201,"threshold_uncertainty_score":0.7796531,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.7494451842918853,"score_gpt":0.4901284049147518,"score_spread":0.2593167793771335,"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."}}