{"id":"W4225015069","doi":"10.3758/s13428-022-01861-0","title":"Accuracy of paper-and-pencil systematic observation versus computer-aided systems","year":2022,"lang":"en","type":"article","venue":"Behavior Research Methods","topic":"Behavioral and Psychological Studies","field":"Psychology","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"Brock University","funders":"Universidad Autónoma de Madrid; University of Auckland","keywords":"Pencil (optics); Computer science; Computer-aided; Computer graphics (images); Artificial intelligence; Programming language; 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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0309293,0.0005536611,0.0006661761,0.001000612,0.0003663597,0.001986519,0.0007481673,0.0006531302,0.002260383],"category_scores_gemma":[0.1900405,0.0003869658,0.0005977551,0.00104364,0.001065422,0.00144939,0.001507779,0.0007906913,0.000628586],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006173066,"about_ca_system_score_gemma":0.0009998317,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001639742,"about_ca_topic_score_gemma":0.002917235,"domain_scores_codex":[0.9558015,0.02712214,0.005054812,0.003008941,0.008362402,0.0006501624],"domain_scores_gemma":[0.7367765,0.2045117,0.02413863,0.01621634,0.01721247,0.001144393],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.01295782,0.0008555279,0.283117,0.003476422,0.0009368457,0.0001264174,0.01016352,0.003155702,0.03137652,0.001400555,0.001983984,0.6504497],"study_design_scores_gemma":[0.0005972855,0.01304709,0.8828463,0.001577876,0.000887825,0.0007953519,0.004952413,0.01982813,0.06256787,0.003235768,0.009315594,0.0003484188],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8848554,0.003131823,0.09882665,0.0006164683,0.0003224665,0.0007876658,0.0005609496,0.0006464303,0.01025215],"genre_scores_gemma":[0.9647924,0.0007500268,0.03246757,0.0001512613,0.00006250446,0.0002686053,0.0001613892,0.00008861075,0.001257658],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9690707,"threshold_uncertainty_score":0.1635717,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.8223134969967072,"score_gpt":0.6146735580251322,"score_spread":0.207639938971575,"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."}}