{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1196293,0.001126055,0.001203972,0.003725112,0.002531468,0.002201476,0.001735694,0.001089287,0.00282322],"category_scores_gemma":[0.2205352,0.001030009,0.002208625,0.002370788,0.002532796,0.002221927,0.002834632,0.001627775,0.001717991],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001504548,"about_ca_system_score_gemma":0.003114029,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002306176,"about_ca_topic_score_gemma":0.006519067,"domain_scores_codex":[0.8934014,0.06372333,0.01567775,0.006068304,0.01960994,0.001519213],"domain_scores_gemma":[0.5605583,0.2841591,0.01181117,0.02482164,0.1162695,0.002380403],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.008667936,0.005000741,0.628227,0.002358861,0.001338279,0.0004029903,0.04668094,0.00384925,0.04954374,0.007344734,0.002982339,0.2436032],"study_design_scores_gemma":[0.001461291,0.01720284,0.8422792,0.000883967,0.001457774,0.001453006,0.01508669,0.04247836,0.05326233,0.006399251,0.01753493,0.0005002744],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8279483,0.001113009,0.1315844,0.0002794694,0.0004469191,0.01115685,0.0007991669,0.0006318596,0.02603995],"genre_scores_gemma":[0.8363984,0.0003933858,0.1468212,0.0002212937,0.0001862398,0.01072912,0.001015145,0.0002722724,0.003962961],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1196293,"threshold_uncertainty_score":0.6326679,"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."}}