{"id":"W4312069611","doi":"10.31234/osf.io/uwche","title":"Limited evidence of test-retest reliability in infant-directed speech preference in a large pre-registered infant experiment","year":2022,"lang":"en","type":"preprint","venue":"","topic":"Language Development and Disorders","field":"Psychology","cited_by":9,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Manitoba; University of British Columbia","funders":"Leibniz-Gemeinschaft; Social Sciences and Humanities Research Council of Canada; Norges Forskningsråd; National Science Foundation","keywords":"Reliability (semiconductor); Psychology; Preference; Active listening; Test (biology); Context (archaeology); Audiology; Developmental psychology; Speech perception; Sample size determination; Perception; Medicine; Statistics; Communication","routes":{"ca_aff":true,"ca_fund":true,"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":[0.06772371,0.000835495,0.001253892,0.0006724031,0.001501551,0.001332222,0.001540192,0.001152857,0.003091978],"category_scores_gemma":[0.1476054,0.0007842414,0.001548356,0.0005604698,0.002031606,0.001286766,0.001920885,0.001443632,0.001986727],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006491894,"about_ca_system_score_gemma":0.0009611135,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002915093,"about_ca_topic_score_gemma":0.006652614,"domain_scores_codex":[0.9472287,0.02434129,0.004557983,0.01101996,0.01213344,0.0007187381],"domain_scores_gemma":[0.8036338,0.1047581,0.01599038,0.04350542,0.03039895,0.001713322],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.009644457,0.002080295,0.7358293,0.002575543,0.006407958,0.0005867423,0.02805905,0.002334968,0.04957491,0.002303771,0.01055872,0.1500443],"study_design_scores_gemma":[0.0002901843,0.005186194,0.9556158,0.0004366501,0.001130139,0.0008219358,0.001480994,0.003116573,0.01416641,0.001620419,0.01592726,0.0002074174],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9372234,0.002997517,0.04574102,0.0004464625,0.0007960625,0.001806699,0.001480191,0.0005547568,0.008953894],"genre_scores_gemma":[0.9750565,0.0004024438,0.0164466,0.0006728466,0.0001405732,0.003019476,0.001456839,0.0004188772,0.002385817],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9322763,"threshold_uncertainty_score":0.3581615,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06775932560258904,"score_gpt":0.3576466124999326,"score_spread":0.2898872868973436,"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."}}