{"id":"W2569145478","doi":"10.5539/ijps.v9n1p71","title":"Accuracy of Measurement in the Classical and the Modern Test Theory: An Empirical Study on a Children Intelligence Test","year":2017,"lang":"en","type":"article","venue":"International Journal of Psychological Studies","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"King Saud University","keywords":"Test (biology); Item response theory; Psychology; Standard error; Statistics; Classical test theory; Sample (material); Statistical hypothesis testing; Descriptive statistics; IBM; Computer science; Psychometrics; Mathematics; Developmental psychology","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03741598,0.0004090031,0.0005659687,0.002004955,0.000487925,0.001448962,0.001036148,0.000915612,0.0008797756],"category_scores_gemma":[0.1709928,0.0003648055,0.000878001,0.002316906,0.00192785,0.00241689,0.001421492,0.001119066,0.0002756179],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001017288,"about_ca_system_score_gemma":0.0009575042,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002181126,"about_ca_topic_score_gemma":0.001556037,"domain_scores_codex":[0.9567201,0.01841314,0.005183447,0.00368682,0.01505627,0.0009402051],"domain_scores_gemma":[0.7843191,0.1642626,0.02031246,0.01244635,0.0175557,0.001103801],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001664403,0.0001422716,0.9678702,0.00007777899,0.0001122545,0.0001165725,0.007887639,0.0003377024,0.0005741545,0.0004943663,0.0001372319,0.02208336],"study_design_scores_gemma":[0.00002164876,0.0009939598,0.9823496,0.0001203358,0.0001024181,0.0006302232,0.006884082,0.003833889,0.002823553,0.0006200846,0.001574808,0.00004538386],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9951266,0.0002450525,0.003125996,0.0000665227,0.00001607163,0.00004447209,0.00005054481,0.00001107546,0.00131371],"genre_scores_gemma":[0.9977542,0.00007911739,0.001864589,0.00001927761,0.00000744016,0.00003786249,0.00006755695,0.000009814719,0.0001602244],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03741598,"threshold_uncertainty_score":0.197877,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.7675113121022065,"score_gpt":0.6166518025269109,"score_spread":0.1508595095752956,"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."}}