{"id":"W4317794535","doi":"10.5038/1936-4660.16.1.1420","title":"Establishing the Validity and Reliability of the LOCUS Assessments","year":2023,"lang":"en","type":"article","venue":"Numeracy","topic":"Statistics Education and Methodologies","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"Mount Saint Vincent University","funders":"National Science Foundation","keywords":"Psychology; Validity; Reliability (semiconductor); External validity; Test validity; Scale (ratio); Applied psychology; Psychometrics; Social psychology; Clinical psychology","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.001687792,0.00005418447,0.00009802041,0.00001663378,0.0001266173,0.00003612062,0.0001815823,0.00002765992,0.00004830064],"category_scores_gemma":[0.01554664,0.00002799304,0.00002934723,0.0002103125,0.00009007572,0.00006069048,0.0001171582,0.000116781,0.000006123336],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001427731,"about_ca_system_score_gemma":0.00005176259,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004061154,"about_ca_topic_score_gemma":0.000003588074,"domain_scores_codex":[0.9991136,0.0003592958,0.0001615433,0.0001025617,0.0001651903,0.0000977879],"domain_scores_gemma":[0.9922519,0.0071702,0.0001092527,0.0003707273,0.00007976484,0.00001812641],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00003508473,0.0003493449,0.3278666,0.0007369927,0.00005684559,0.000002331601,0.01463643,0.00005508204,0.001353951,0.1841463,0.4213673,0.04939376],"study_design_scores_gemma":[0.0001412779,0.0000146285,0.2285858,0.00002487605,0.00003806497,0.000002017565,0.001790978,0.0003413601,0.001789727,0.7567044,0.01049531,0.0000715238],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9880609,0.000009943254,0.006220188,0.002305688,0.001174533,0.0002179978,0.00003029461,0.00006104011,0.001919398],"genre_scores_gemma":[0.9657927,0.00003584639,0.03190957,0.0001283669,0.00005731624,0.00001803701,0.000002506184,0.00001007095,0.00204555],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5725582,"threshold_uncertainty_score":0.9927458,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2793615220572844,"score_gpt":0.4826761901861915,"score_spread":0.203314668128907,"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."}}