{"id":"W6045406","doi":"10.3138/cjpe.24.004","title":"Interrater Reliability in Content Analysis of Healthcare Service Quality Using Montreal’s Conceptual Framework","year":2009,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal","funders":"Universitetet i Bergen; Thomas Jefferson University","keywords":"Inter-rater reliability; Generalizability theory; Kappa; Reliability (semiconductor); Psychology; Cohen's kappa; Statistics; Content analysis; Service (business); Service quality; Health care; Applied psychology; Mathematics; Business; Sociology; Power (physics); Rating scale; Marketing; Political science","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02729289,0.0001375977,0.0006513182,0.000917774,0.00008740794,0.0001316786,0.000530112,0.000142581,0.0003870192],"category_scores_gemma":[0.008222813,0.000104959,0.0002624835,0.002861633,0.0001302043,0.0003529811,0.00001136831,0.0003265235,0.000002550136],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008731787,"about_ca_system_score_gemma":0.001824081,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.03617253,"about_ca_topic_score_gemma":0.2462581,"domain_scores_codex":[0.9927831,0.002065764,0.002315186,0.0003090913,0.002222935,0.0003039239],"domain_scores_gemma":[0.99168,0.0004216704,0.001259196,0.0005101993,0.005728221,0.0004007746],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001624962,0.0003029451,0.6062438,0.00001773107,0.0001884257,0.000005916031,0.01316161,0.05506147,0.0002522029,0.0006443209,0.0001000176,0.323859],"study_design_scores_gemma":[0.0005272082,0.0004706623,0.9306127,0.0001485563,0.0003121214,0.000001708454,0.008429754,0.03690809,0.00007278523,0.02218506,0.0002054994,0.0001258434],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9903259,0.0004344407,0.001992479,0.006014579,0.0003198284,0.0008104693,0.0000197027,0.00000326706,0.00007933994],"genre_scores_gemma":[0.9958049,0.000003691048,0.003537637,0.0005863696,0.00004461563,0.000007678327,0.000007440222,0.0000031389,0.000004528088],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3243688,"threshold_uncertainty_score":0.9844069,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6590834155962562,"score_gpt":0.5276751899838836,"score_spread":0.1314082256123725,"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."}}