{"id":"W2947059084","doi":"10.1007/s40617-019-00364-3","title":"A Preliminary Evaluation of Interrater Reliability and Concurrent Validity of Open-Ended Indirect Assessment","year":2019,"lang":"en","type":"article","venue":"Behavior Analysis in Practice","topic":"Behavioral and Psychological Studies","field":"Psychology","cited_by":20,"is_retracted":false,"has_abstract":false,"ca_institutions":"Surrey Place Centre; Brock University","funders":"","keywords":"Inter-rater reliability; Concurrent validity; Psychology; Context (archaeology); Clinical psychology; Reliability (semiconductor); External validity; Psychometrics; Developmental psychology; Rating scale; Social psychology; Internal consistency","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.004916362,0.0001583233,0.0006452041,0.0001999505,0.00003042265,0.00002492781,0.0002929593,0.0001347247,0.002284652],"category_scores_gemma":[0.0002515402,0.0001257099,0.0001404729,0.0007612713,0.0001236889,0.0003091006,0.0002756289,0.0002910087,0.0000120724],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008439967,"about_ca_system_score_gemma":0.00003413379,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00104414,"about_ca_topic_score_gemma":0.00009951227,"domain_scores_codex":[0.9960352,0.002055737,0.0007165388,0.0005336084,0.0004882439,0.0001706349],"domain_scores_gemma":[0.9978297,0.000552503,0.0006149299,0.0005272731,0.0004308163,0.00004481031],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0003078024,0.003355299,0.9675049,0.00001285643,0.0003364846,0.000004777021,0.001223221,0.00001925693,0.0005586754,0.000159407,0.0001114406,0.02640587],"study_design_scores_gemma":[0.001212078,0.0006980273,0.9881494,0.00001681114,0.00734787,0.00000391088,0.001886439,0.0001220376,0.0001472984,0.00006472917,0.0002166243,0.0001347435],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9857815,0.0002372447,0.00003987277,0.0001465017,0.0003485167,0.0009768683,0.00003011767,0.00000839556,0.01243099],"genre_scores_gemma":[0.999056,0.00003284708,0.0004129634,0.00003317848,0.000006699448,0.0002821752,0.00002642981,0.000006079697,0.0001436541],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02627113,"threshold_uncertainty_score":0.9986274,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3528654404956125,"score_gpt":0.5112076358369838,"score_spread":0.1583421953413713,"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."}}