{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02568955,0.0001532238,0.000489846,0.0002127324,0.0002683671,0.0003400527,0.003592705,0.0000495767,0.0000163134],"category_scores_gemma":[0.4017467,0.00005541871,0.0001445064,0.0002401752,0.0009860847,0.0002088786,0.0003702681,0.0004988121,0.000002622527],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003557484,"about_ca_system_score_gemma":0.00001659916,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000008475662,"about_ca_topic_score_gemma":0.00001523429,"domain_scores_codex":[0.9943213,0.001540244,0.001095063,0.0003266511,0.00255578,0.0001609632],"domain_scores_gemma":[0.9295058,0.06742401,0.001399664,0.0005551923,0.001064667,0.00005072367],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0006256668,0.002093138,0.745751,4.505978e-7,0.0001500733,0.00004837186,0.004145852,0.0000360458,0.00002367802,0.001081306,0.0007165103,0.2453279],"study_design_scores_gemma":[0.00100665,0.00135612,0.7974323,0.00003454604,0.00001708287,0.00009479608,0.003768692,0.00009923508,0.000007916899,0.19608,0.0000482694,0.00005440991],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9738327,0.001227976,0.003383399,0.01954987,0.0006298966,0.0002943981,0.000006649846,0.000004710376,0.001070411],"genre_scores_gemma":[0.9982919,0.0002781371,0.0005521408,0.0005956863,0.000258189,0.000007833627,6.482843e-8,0.000003395103,0.000012619],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3760572,"threshold_uncertainty_score":0.8903538,"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."}}