{"id":"W2977609233","doi":"10.1002/bimj.201800146","title":"Latent variable models for harmonization of test scores: A case study on memory","year":2019,"lang":"en","type":"article","venue":"Biometrical Journal","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University Health Centre; McMaster University; Impact","funders":"FP7 Health","keywords":"Equating; Harmonization; Test (biology); Observational study; Latent variable; Econometrics; Reliability (semiconductor); Statistics; Variable (mathematics); Computer science; Mathematics; Data mining","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06508822,0.0007061721,0.001398652,0.003148862,0.001821561,0.003846698,0.003197236,0.003224643,0.004047616],"category_scores_gemma":[0.1368747,0.0005697536,0.003234199,0.007390135,0.00416127,0.005029399,0.003997764,0.004003447,0.0003692581],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004312935,"about_ca_system_score_gemma":0.002826409,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01890564,"about_ca_topic_score_gemma":0.02251958,"domain_scores_codex":[0.9277104,0.06602583,0.001074357,0.00212913,0.002265132,0.0007950552],"domain_scores_gemma":[0.8211567,0.160632,0.004762222,0.009559055,0.003345491,0.0005445346],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0003996467,0.0005117008,0.04538435,0.0004684933,0.0008569771,0.002200326,0.01687294,0.0549752,0.0004534838,0.6968694,0.004088025,0.1769194],"study_design_scores_gemma":[0.000256029,0.0002915542,0.01601436,0.0003843095,0.000332446,0.0009979368,0.005194293,0.2449356,0.0008083597,0.7139614,0.0166898,0.0001338723],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.09170463,0.00127271,0.8933064,0.005223949,0.00005660379,0.0004594583,0.0003477936,0.0001579302,0.007470501],"genre_scores_gemma":[0.6099134,0.0007142833,0.3855415,0.000408371,0.00005614109,0.001031981,0.0003627116,0.0001010877,0.001870553],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.06508822,"threshold_uncertainty_score":0.3442235,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.161248548609808,"score_gpt":0.3960955340183089,"score_spread":0.2348469854085009,"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."}}