{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":24,"total_is_capped":false,"direct_labels_cover":0,"predictions_cover":24,"direct_label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline (scores rank; they never assert a category)","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12"},"query_hash":"5003ca732a82","filters":{"venue":"International Journal of Testing"}},"results":[{"id":"W2126572511","doi":"10.1207/s15327574ijt0403_1","title":"Validating the Beck Depression Inventory-II for Hong Kong Community Adolescents","year":2004,"lang":"en","type":"article","venue":"International Journal of Testing","topic":"Child and Adolescent Psychosocial and Emotional Development","field":"Psychology","cited_by":160,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Ottawa","funders":"","keywords":"Psychology; Confirmatory factor analysis; Beck Depression Inventory; Structural equation modeling; Exploratory factor analysis; Clinical psychology; Internal consistency; Depression (economics); Psychometrics; Developmental psychology; Psychiatry; Anxiety; Statistics; Mathematics","retraction":null,"screen_n_in":null,"score":{"opus":0.08080420797592054,"gpt":0.3520861026986411,"spread":0.2712818947227206,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0007777925,0.0001064887,0.0001219009,0.0000805428,0.0005307435,0.00006273699,0.0005875397,0.00004630611,0.00003781456],"category_scores_gemma":[0.0005263126,0.00007491035,0.0001376689,0.00008796711,0.00004279002,0.0001336248,0.0001178013,0.0004288875,0.000008790073],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000130052,"about_ca_system_score_gemma":0.00007318931,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001225394,"about_ca_topic_score_gemma":0.00001248923,"domain_scores_codex":[0.9988092,0.00009078533,0.0004655648,0.0000920369,0.0003753382,0.0001671088],"domain_scores_gemma":[0.9987196,0.0001430873,0.0004888257,0.00009205696,0.0004923472,0.00006408704],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0005736196,0.002532501,0.7958207,0.00009302758,0.0006980319,0.00007072114,0.01467152,0.00217007,0.00623008,0.004805382,0.007904283,0.1644301],"study_design_scores_gemma":[0.004016054,0.0002193157,0.9679355,0.003896252,0.00004422721,0.0003778235,0.001876871,0.00006621151,0.0006035228,0.01732546,0.003403598,0.0002351582],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9880078,0.000143137,0.002937176,0.002367306,0.003918729,0.0001256231,0.000007744297,0.00001658313,0.002475924],"genre_scores_gemma":[0.9956763,0.000004265466,0.001883793,0.0008302392,0.001412565,0.000003206587,0.000006996836,0.00001209953,0.0001705399],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1721148,"threshold_uncertainty_score":0.4082104,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2142413505","doi":"10.1080/15305058.2011.635830","title":"The Role of Item Models in Automatic Item Generation","year":2012,"lang":"en","type":"article","venue":"International Journal of Testing","topic":"Intelligent Tutoring Systems and Adaptive Learning","field":"Computer Science","cited_by":73,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"","keywords":"Computer science; Item bank; Item response theory; Field (mathematics); Process (computing); Task (project management); Item analysis; Artificial intelligence; Machine learning; Psychometrics; Psychology","retraction":null,"screen_n_in":null,"score":{"opus":0.05584327543240598,"gpt":0.2802359057096199,"spread":0.224392630277214,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001123365,0.0000549732,0.00009084748,0.0001210565,0.00004866486,0.0000937959,0.000565279,0.00001937475,0.000001522759],"category_scores_gemma":[0.000806732,0.0000386334,0.00004654743,0.0001100318,0.000009873956,0.0007630216,0.00007615348,0.0001355219,0.000002585245],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008477976,"about_ca_system_score_gemma":0.00004694521,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003488693,"about_ca_topic_score_gemma":0.000001858335,"domain_scores_codex":[0.9988478,0.00007309751,0.0004880591,0.00005200132,0.0004131384,0.0001258494],"domain_scores_gemma":[0.9982786,0.000619226,0.0005416789,0.00007565037,0.000453362,0.00003143993],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000005637917,0.0001025387,0.118131,0.000009463947,0.00008560302,0.00001707304,0.004145116,0.06390452,0.03249722,0.3308772,0.00004594932,0.4501787],"study_design_scores_gemma":[0.000101048,0.00003617009,0.006687047,0.0002224879,0.000002250874,0.0001477198,0.0001720109,0.9844022,0.004600285,0.002245387,0.001328538,0.00005486505],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7609475,0.001763976,0.2330461,0.0002231673,0.001914477,0.00005622113,3.418423e-7,0.00001633788,0.002031922],"genre_scores_gemma":[0.9800276,0.000006450313,0.01930923,0.00001579494,0.0005468738,0.000001011035,1.110438e-7,0.000003748071,0.00008919115],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9204977,"threshold_uncertainty_score":0.1575424,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1966255248","doi":"10.1080/15305050701438058","title":"The Beck Depression Inventory-II: Testing for Measurement Equivalence and Factor Mean Differences Across Hong Kong and American Adolescents","year":2007,"lang":"en","type":"article","venue":"International Journal of Testing","topic":"Child and Adolescent Psychosocial and Emotional Development","field":"Psychology","cited_by":68,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Ottawa","funders":"","keywords":"Beck Depression Inventory; Psychology; Confirmatory factor analysis; Equivalence (formal languages); Measurement invariance; Clinical psychology; Structural equation modeling; Statistics; Psychiatry; Mathematics","retraction":null,"screen_n_in":null,"score":{"opus":0.1069246491615596,"gpt":0.3636558479182934,"spread":0.2567311987567337,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0009962987,0.0001259,0.000145132,0.00005544481,0.0004761248,0.0001147237,0.0002579681,0.00002989873,0.000003297399],"category_scores_gemma":[0.0008506386,0.00008573256,0.000050087,0.000084094,0.0001415539,0.0001061204,0.0001121039,0.0001819774,6.175622e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000773565,"about_ca_system_score_gemma":0.00003466824,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000574244,"about_ca_topic_score_gemma":0.00005979474,"domain_scores_codex":[0.9985005,0.00003503224,0.0004612431,0.0001738001,0.0005652653,0.0002642126],"domain_scores_gemma":[0.9982393,0.0003688261,0.0005525593,0.00004988929,0.0006700638,0.000119381],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000109385,0.00006882136,0.7241206,0.00001261568,0.00006214406,0.000007684909,0.00130008,9.998074e-7,0.001760072,0.00005615812,0.00006232781,0.2724392],"study_design_scores_gemma":[0.0006437299,0.0001556933,0.9957511,0.001149258,0.00001015396,0.00008753024,0.001145775,0.00005745627,0.00009501604,0.0005020068,0.0002961125,0.0001061326],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9957341,0.0006210701,0.001341504,0.0005205977,0.001453618,0.000106153,0.000007439304,0.00001095232,0.0002045827],"genre_scores_gemma":[0.9977431,0.00002702644,0.001276429,0.0001537503,0.0007369042,0.000001707961,5.677012e-7,0.00000913053,0.00005137243],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2723331,"threshold_uncertainty_score":0.3662015,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2076222948","doi":"10.1080/15305058.2010.509554","title":"Using the Attribute Hierarchy Method to Make Diagnostic Inferences about Examinees’ Knowledge and Skills in Mathematics: An Operational Implementation of Cognitive Diagnostic Assessment","year":2010,"lang":"en","type":"article","venue":"International Journal of Testing","topic":"Cognitive and developmental aspects of mathematical skills","field":"Mathematics","cited_by":55,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Alberta Advanced Education; University of Alberta","funders":"","keywords":"Cognition; Hierarchy; Set (abstract data type); Item response theory; Test (biology); Computer science; Computerized adaptive testing; Psychology; Psychometrics; Artificial intelligence; Mathematics education; Developmental psychology","retraction":null,"screen_n_in":null,"score":{"opus":0.0804734892655436,"gpt":0.4593924747952735,"spread":0.3789189855297299,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.001903717,0.0001674974,0.0003165385,0.0002801119,0.00007887088,0.0001151919,0.0003161343,0.00004513674,0.0001267867],"category_scores_gemma":[0.02214393,0.0001201835,0.00005085806,0.0002162159,0.00007663352,0.0002987106,0.0001903634,0.0003140198,0.000001887382],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000755612,"about_ca_system_score_gemma":0.0002415251,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005645281,"about_ca_topic_score_gemma":0.000309258,"domain_scores_codex":[0.997999,0.0001523667,0.0009584316,0.0001571644,0.0005597034,0.000173327],"domain_scores_gemma":[0.9677132,0.03024636,0.0005387815,0.0000755884,0.001318921,0.0001071119],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00004551155,0.003681996,0.5980329,0.0002745439,0.0006712045,0.0002164448,0.02538566,0.0001568973,0.03009779,0.05280675,0.00008702031,0.2885433],"study_design_scores_gemma":[0.002092589,0.0005342165,0.7831923,0.003275137,0.0001728875,0.0006734778,0.006940448,0.003876935,0.01222241,0.1866129,0.0000178667,0.0003887695],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9216115,0.00002985819,0.07713275,0.0002368769,0.0001895804,0.0003508194,0.00004144212,0.000006546134,0.0004006037],"genre_scores_gemma":[0.7185038,0.000006388862,0.2812521,0.0000595543,0.0001378511,0.0000148469,0.000005870912,0.00001124806,0.000008382012],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2881545,"threshold_uncertainty_score":0.986093,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2135248568","doi":"10.1080/15305058.2011.602810","title":"A Generalized Logistic Regression Procedure to Detect Differential Item Functioning Among Multiple Groups","year":2011,"lang":"en","type":"article","venue":"International Journal of Testing","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":46,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Université du Québec à Montréal","funders":"","keywords":"Differential item functioning; Logistic regression; Completeness (order theory); Statistics; Item response theory; Mathematics; Set (abstract data type); Extension (predicate logic); Differential (mechanical device); Flexibility (engineering); Regression; Psychology; Psychometrics; Computer science","retraction":null,"screen_n_in":null,"score":{"opus":0.548040852468272,"gpt":0.4478112341211973,"spread":0.1002296183470748,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.00281299,0.0001768094,0.0003556771,0.001032285,0.0001635482,0.0002770731,0.001328837,0.00007788413,0.0003150204],"category_scores_gemma":[0.3223573,0.0001152148,0.0001832548,0.001027547,0.00005735019,0.0004053769,0.0002914335,0.0003314948,0.0000211593],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009156694,"about_ca_system_score_gemma":0.0000695466,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007911449,"about_ca_topic_score_gemma":0.00001430183,"domain_scores_codex":[0.9961699,0.0002917955,0.00125882,0.0003349288,0.001682195,0.000262326],"domain_scores_gemma":[0.9804429,0.01513539,0.001730888,0.0002358195,0.00222913,0.0002258761],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0003006831,0.00005402902,0.8231879,0.000003978355,0.000070541,0.0001826228,0.0005064168,0.0007168977,0.01062423,0.00006699181,0.0008934885,0.1633922],"study_design_scores_gemma":[0.0009922941,0.0003728979,0.9672252,0.0003933333,0.00002685629,0.0007037269,0.0003964616,0.01520523,0.001952559,0.01235152,0.0001670339,0.000212904],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8289563,0.00006676224,0.1665859,0.0001094609,0.003028139,0.00008221934,0.000002788851,0.00003278925,0.001135651],"genre_scores_gemma":[0.8323534,0.000002317362,0.166683,0.00007861635,0.0007249534,0.000003767961,5.432254e-7,0.00001242281,0.000140975],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3195443,"threshold_uncertainty_score":0.6833507,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2086483769","doi":"10.1080/15305058.2011.552748","title":"Teachers' Perceptions of Large-Scale Assessment Programs Within Low-Stakes Accountability Frameworks","year":2011,"lang":"en","type":"article","venue":"International Journal of Testing","topic":"Educational Assessment and Improvement","field":"Decision Sciences","cited_by":41,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"Centre for Advancing Health Outcomes; University of Alberta; Queen's University","funders":"","keywords":"Accountability; Seriousness; Scale (ratio); Perception; Psychology; Standardized test; Educational assessment; Medical education; Pedagogy; Mathematics education; Political science; Geography; Medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.1565396694431699,"gpt":0.4547954859236611,"spread":0.2982558164804912,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.004412601,0.0001292225,0.0002556754,0.0002778123,0.00008688297,0.0001823994,0.001169253,0.00009539515,0.001052153],"category_scores_gemma":[0.002661093,0.00009419346,0.0001873204,0.0003304415,0.0001094315,0.0007775953,0.0001453055,0.0005656308,0.00001453018],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000184828,"about_ca_system_score_gemma":0.0004309205,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007326438,"about_ca_topic_score_gemma":0.00004993418,"domain_scores_codex":[0.995809,0.0001343233,0.001459755,0.0002205616,0.002199648,0.000176694],"domain_scores_gemma":[0.9935382,0.0008107817,0.001604583,0.0002346497,0.003700829,0.0001109372],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00002431274,0.001204979,0.9524676,0.000003747883,0.00007983964,0.000003655844,0.007977939,0.00009605121,0.001891529,0.000607079,0.0004257336,0.03521755],"study_design_scores_gemma":[0.0005354072,0.0004076211,0.9347942,0.0002473034,0.00003339819,0.00004529777,0.03869751,0.002424833,0.0005604048,0.02169471,0.0004015026,0.0001578059],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9733058,0.00001944042,0.0205525,0.000654172,0.00157114,0.0001191629,0.00001458485,0.00001073099,0.003752508],"genre_scores_gemma":[0.8541455,0.000001585646,0.1452105,0.0001401936,0.0003119141,0.000006120892,0.000004076792,0.000007512785,0.0001725333],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1246581,"threshold_uncertainty_score":0.999861,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2057990449","doi":"10.1080/15305058.2012.738266","title":"Analysis of Sources of Latent Class Differential Item Functioning in International Assessments","year":2013,"lang":"en","type":"article","venue":"International Journal of Testing","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":39,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Comparability; Differential item functioning; Psychology; Multinomial logistic regression; Latent variable; Item response theory; Latent class model; Logistic regression; Matching (statistics); Statistics; Construct validity; Linear discriminant analysis; Construct (python library); Psychometrics; Developmental psychology; Mathematics; Computer science","retraction":null,"screen_n_in":null,"score":{"opus":0.3977077293797573,"gpt":0.4811063293577611,"spread":0.0833985999780038,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.003147237,0.0001107549,0.0005039129,0.003290723,0.00002817976,0.0001674813,0.001285239,0.00005507528,0.0008488807],"category_scores_gemma":[0.05684236,0.00008179507,0.0002889075,0.00214768,0.00006205502,0.0005049806,0.0002293673,0.0002292393,0.000003586134],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008945358,"about_ca_system_score_gemma":0.00006620961,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002839933,"about_ca_topic_score_gemma":0.00001099877,"domain_scores_codex":[0.9952543,0.0002371757,0.00200509,0.000201664,0.002154312,0.0001473965],"domain_scores_gemma":[0.9767593,0.01673236,0.003120872,0.0001629733,0.003159693,0.00006479948],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00003040519,0.00009680779,0.881137,0.000001798368,0.0005914421,0.000008961999,0.0001128459,0.01426439,0.004447816,0.00005694803,0.00005738668,0.09919418],"study_design_scores_gemma":[0.0004315422,0.00008840016,0.8869734,0.00009615309,0.00007016002,0.00002536013,0.0004247887,0.1097161,0.0003744051,0.001699888,0.00003732842,0.00006237831],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9718423,0.00005057207,0.02405866,0.0002545796,0.001819696,0.0000448857,0.00000776966,0.000005119118,0.001916415],"genre_scores_gemma":[0.9781033,0.000007310938,0.02156891,0.00002908267,0.0001760105,0.000001534763,0.000002117392,0.000005326779,0.0001064133],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0991318,"threshold_uncertainty_score":0.9511023,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2063611796","doi":"10.1080/15305058.2011.617475","title":"Methodologies for Investigating Item- and Test-Level Measurement Equivalence in International Large-Scale Assessments","year":2012,"lang":"en","type":"article","venue":"International Journal of Testing","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":32,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of British Columbia","funders":"","keywords":"Differential item functioning; Comparability; Psychology; Item response theory; Equivalence (formal languages); Statistics; Nonparametric statistics; Test (biology); Psychometrics; Consistency (knowledge bases); Item analysis; Logistic regression; Mathematics","retraction":null,"screen_n_in":null,"score":{"opus":0.84777911019985,"gpt":0.5833143230805201,"spread":0.2644647871193299,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03423003,0.0001699518,0.0003777544,0.0009512466,0.00009910401,0.0002991174,0.001320366,0.00007479876,0.0000382111],"category_scores_gemma":[0.6310325,0.0001306996,0.0001127948,0.0006482329,0.00007769925,0.001089666,0.0004233571,0.000315677,0.000002437312],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002263224,"about_ca_system_score_gemma":0.0001260458,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002093083,"about_ca_topic_score_gemma":0.000007369788,"domain_scores_codex":[0.9949619,0.0004122212,0.001518275,0.0002838765,0.002419526,0.0004041574],"domain_scores_gemma":[0.8772222,0.1157515,0.002426522,0.00023604,0.004127027,0.0002367326],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00002370454,0.00008440615,0.8363786,0.000006062034,0.00004110663,0.000005538834,0.0003843894,0.0001581708,0.006958658,0.0001928567,0.0003240832,0.1554424],"study_design_scores_gemma":[0.001385228,0.0002099137,0.9500422,0.0004051996,0.00001986312,0.0003014881,0.002868365,0.01228085,0.001264212,0.02952302,0.001481325,0.0002183349],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6880037,0.0007646934,0.3023771,0.001690685,0.005024903,0.0001876176,0.00004462444,0.00002465008,0.001882109],"genre_scores_gemma":[0.5284334,0.00001264457,0.4708877,0.0001639493,0.000451001,0.000005564054,6.236936e-7,0.000007271503,0.00003780019],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5968025,"threshold_uncertainty_score":0.9944634,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2113879845","doi":"10.1080/15305058.2012.690140","title":"Investigating Sources of Differential Item Functioning in International Large-Scale Assessments Using a Confirmatory Approach","year":2013,"lang":"en","type":"article","venue":"International Journal of Testing","topic":"Educational and Psychological Assessments","field":"Psychology","cited_by":28,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Differential item functioning; Psychology; Equivalence (formal languages); Measurement invariance; Scale (ratio); Item response theory; Pairwise comparison; Cognition; Psychometrics; Differential (mechanical device); Confirmatory factor analysis; Clinical psychology; Developmental psychology; Social psychology; Statistics; Structural equation modeling","retraction":null,"screen_n_in":null,"score":{"opus":0.1081283061996754,"gpt":0.4028085569126335,"spread":0.2946802507129581,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0005426865,0.0001592879,0.0002679794,0.0004102665,0.00005731134,0.000101194,0.0006016135,0.00009568985,0.001483085],"category_scores_gemma":[0.0004099965,0.0001402225,0.0001130669,0.0002157831,0.00007873706,0.0004341546,0.0001097158,0.0004281606,0.00001552508],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001329327,"about_ca_system_score_gemma":0.00009565746,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002448435,"about_ca_topic_score_gemma":0.00000240246,"domain_scores_codex":[0.9976323,0.0001644189,0.001050061,0.000218086,0.0006995623,0.0002355973],"domain_scores_gemma":[0.997197,0.0005211098,0.001192564,0.0001012531,0.0008863432,0.0001016838],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00003079099,0.0008229513,0.9822505,0.00001242672,0.0002678721,0.00001275772,0.00165101,0.0008305908,0.008220622,0.0006127974,0.0002408252,0.005046816],"study_design_scores_gemma":[0.001775918,0.0001239491,0.9684172,0.0003553685,0.0000270134,0.0004090928,0.006223802,0.02041273,0.000142482,0.00174366,0.000178378,0.0001904236],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9762044,0.000061361,0.007825714,0.0002183147,0.003229065,0.0001115335,0.00001109994,0.00001223175,0.01232627],"genre_scores_gemma":[0.9676636,0.000001545708,0.03116799,0.0002112895,0.0007186123,0.00001325388,0.00001355278,0.00001537609,0.0001947451],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02334227,"threshold_uncertainty_score":0.9994297,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1999798470","doi":"10.1080/15305050701193520","title":"Using Exploratory and Confirmatory Methods to Identify the Cognitive Dimensions In a Large-Scale Science Assessment","year":2007,"lang":"en","type":"article","venue":"International Journal of Testing","topic":"Cognitive Science and Mapping","field":"Computer Science","cited_by":20,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"","keywords":"Confirmatory factor analysis; Psychology; Scale (ratio); Item response theory; Cognition; Psychometrics; Measurement invariance; Test validity; Applied psychology; Structural equation modeling; Clinical psychology; Statistics; Mathematics","retraction":null,"screen_n_in":null,"score":{"opus":0.1334967275173401,"gpt":0.4887037303380774,"spread":0.3552070028207374,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008387464,0.00007233859,0.0001029588,0.0005455896,0.0002158537,0.0002755904,0.000772516,0.00001473752,0.000002186276],"category_scores_gemma":[0.002015736,0.00005490714,0.000028502,0.0008107795,0.0001516604,0.001154592,0.0005834643,0.0002190364,0.000001333484],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00013703,"about_ca_system_score_gemma":0.0003838694,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001264584,"about_ca_topic_score_gemma":0.00001440903,"domain_scores_codex":[0.9984471,0.00009525609,0.0003742111,0.000185402,0.0006613179,0.0002367264],"domain_scores_gemma":[0.997028,0.001128576,0.0002652369,0.00007922138,0.001377199,0.0001217543],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"observational","study_design_scores_codex":[0.00002220161,0.0001766539,0.1045594,0.000006326739,0.00004373527,0.0004706784,0.01462255,0.0004984385,0.5760795,0.005246677,0.00002387833,0.29825],"study_design_scores_gemma":[0.001122643,0.0001648171,0.8114436,0.001263033,0.00002024404,0.001343753,0.0148466,0.141217,0.0231788,0.00476096,0.0003101043,0.0003283801],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4310994,0.00004182446,0.5675882,0.0002816699,0.0005237249,0.00004819182,5.374561e-7,0.000005026856,0.0004114559],"genre_scores_gemma":[0.7189093,0.000001687164,0.2803968,0.0006191405,0.00006829686,8.110095e-7,3.360316e-8,0.000001988071,0.000001928103],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7068843,"threshold_uncertainty_score":0.2906944,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2089954256","doi":"10.1080/15305058.2010.501536","title":"Evaluating the Bookmark Standard Setting Method: The Impact of Random Item Ordering","year":2011,"lang":"en","type":"article","venue":"International Journal of Testing","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":19,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Dalhousie University","funders":"","keywords":"Licensure; Assertion; Test (biology); Computer science; Psychology; Process (computing); Mathematics education; Statistics; Medical education; Mathematics; Medicine; Programming language","retraction":null,"screen_n_in":null,"score":{"opus":0.7272170666147368,"gpt":0.606127092720526,"spread":0.1210899738942108,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.0506505,0.000156414,0.0004118926,0.0004451472,0.0002394274,0.0002392276,0.002447673,0.0000425364,0.0002829545],"category_scores_gemma":[0.4542646,0.00006730375,0.0004149899,0.001075511,0.0001106196,0.0003514488,0.0003382165,0.0004685088,0.000003091925],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009901107,"about_ca_system_score_gemma":0.0002704283,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002117521,"about_ca_topic_score_gemma":0.000001902563,"domain_scores_codex":[0.9937204,0.001542182,0.001806524,0.0002083034,0.002471321,0.0002513139],"domain_scores_gemma":[0.7766666,0.2104888,0.005999786,0.0005824589,0.006142882,0.0001194781],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0005028751,0.00002647702,0.2080978,0.000002548917,0.0002816883,0.00002421933,0.002053581,0.01270621,0.005844319,0.0001080362,0.000483008,0.7698693],"study_design_scores_gemma":[0.004446464,0.00233301,0.5941211,0.0008088346,0.000159693,0.002877492,0.006850333,0.2586156,0.004439106,0.1245557,0.0003651883,0.0004274721],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8057243,0.0004516366,0.1824968,0.0004695417,0.001360765,0.000138867,0.000007998587,0.00001459468,0.009335478],"genre_scores_gemma":[0.6995415,0.000005330368,0.3000032,0.00005446354,0.0003431174,0.000001524026,1.078831e-7,0.000009296701,0.00004138342],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7694418,"threshold_uncertainty_score":0.9775551,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2058922083","doi":"10.1080/15305058.2014.891223","title":"Uncovering Substantive Patterns in Student Responses in International Large-Scale Assessments—Comparing a Latent Class to a Manifest DIF Approach","year":2014,"lang":"en","type":"article","venue":"International Journal of Testing","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":19,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Generalizability theory; Differential item functioning; Psychology; Item response theory; Latent class model; Homogeneous; Contrast (vision); Homogeneity (statistics); Reading comprehension; Cognitive psychology; Scale (ratio); Comprehension; Social psychology; Psychometrics; Developmental psychology; Reading (process); Statistics; Mathematics","retraction":null,"screen_n_in":null,"score":{"opus":0.3759875606329405,"gpt":0.4999896769691466,"spread":0.1240021163362061,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01271289,0.0001915751,0.0004777049,0.002268738,0.00005397756,0.0004898458,0.002060759,0.0000604468,0.00005155797],"category_scores_gemma":[0.04956838,0.0001566582,0.0001239468,0.001241185,0.00002201252,0.0005337393,0.0006443435,0.0005094996,0.000009028298],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004680891,"about_ca_system_score_gemma":0.00007811494,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001236245,"about_ca_topic_score_gemma":0.0001802323,"domain_scores_codex":[0.9943575,0.0006221823,0.001743144,0.0004423351,0.002468567,0.0003663188],"domain_scores_gemma":[0.9848703,0.01264912,0.001107773,0.0002317816,0.001005467,0.0001355358],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001725475,0.0002701917,0.9451922,0.000002883766,0.00003857865,0.0001312257,0.0009504888,0.03054462,0.0005526301,0.0001358239,0.00002790995,0.02198095],"study_design_scores_gemma":[0.001320004,0.0001434564,0.9347823,0.0003514331,0.000004739696,0.0001692207,0.00272909,0.05844578,0.00006666108,0.001492769,0.0003381016,0.0001564736],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9067892,0.00002264433,0.08631536,0.0004242969,0.001872557,0.0001116952,0.000008516719,0.00001328271,0.004442409],"genre_scores_gemma":[0.9277487,0.000005808057,0.07160754,0.0001918723,0.0003485806,0.000007057259,0.000001502481,0.00001353982,0.00007538901],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03685549,"threshold_uncertainty_score":0.9584375,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2025305573","doi":"10.1080/15305050903106859","title":"Uncovering the Psychometric Properties of Scales Measuring Individualist and Collectivist Orientations","year":2009,"lang":"en","type":"article","venue":"International Journal of Testing","topic":"Cultural Differences and Values","field":"Psychology","cited_by":17,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Calgary","funders":"","keywords":"Collectivism; Psychology; Individualism; Social psychology; Scale (ratio); Psychometrics; Developmental psychology; Geography; Political science","retraction":null,"screen_n_in":null,"score":{"opus":0.2054654203050422,"gpt":0.3769399820431952,"spread":0.171474561738153,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0002629826,0.00005967692,0.0001030162,0.0001660706,0.00008042764,0.0000777671,0.0002156388,0.00001721087,0.00001695396],"category_scores_gemma":[0.0006776754,0.00003492109,0.00004176194,0.0002610757,0.00005223527,0.000126709,0.00002155952,0.0001111195,8.568722e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002737851,"about_ca_system_score_gemma":0.00001799479,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008688558,"about_ca_topic_score_gemma":0.0000117342,"domain_scores_codex":[0.9992223,0.00004114281,0.000305593,0.00006481088,0.0002939605,0.00007217364],"domain_scores_gemma":[0.9989899,0.0001637306,0.0003409382,0.00004300724,0.000437137,0.00002523028],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001895304,0.0003039831,0.729362,0.00001829772,0.0004813231,0.00002042676,0.01062488,0.0005496389,0.03777392,0.002928345,0.0007467756,0.2170009],"study_design_scores_gemma":[0.000385481,0.0001788391,0.995011,0.0002073247,0.00002111629,0.0002774242,0.001283747,0.00006398615,0.0008490552,0.001593008,0.00007365156,0.00005542275],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9956687,0.0007143592,0.0001306691,0.0006449393,0.0004931437,0.00004137046,0.000003248957,0.000005250938,0.002298333],"genre_scores_gemma":[0.998967,0.00001278402,0.0006379812,0.00006633194,0.000197183,9.992967e-7,3.109851e-7,0.00000291386,0.0001144967],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.265649,"threshold_uncertainty_score":0.1424041,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2051773947","doi":"10.1080/15305058.2014.957382","title":"Reading Proficiency and Comparability of Mathematics and Science Scores for Students From English and Non-English Backgrounds: An International Perspective","year":2014,"lang":"en","type":"article","venue":"International Journal of Testing","topic":"School Choice and Performance","field":"Social Sciences","cited_by":17,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Ottawa; University of Victoria; University of British Columbia","funders":"","keywords":"Comparability; Reading (process); Mathematics education; Perspective (graphical); Language proficiency; Meaning (existential); English language; Variance (accounting); Psychology; Point (geometry); Mathematics; Linguistics; Accounting","retraction":null,"screen_n_in":null,"score":{"opus":0.0523961827382444,"gpt":0.4029235874767592,"spread":0.3505274047385147,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.00227978,0.00006963908,0.0001473234,0.0001430993,0.0002167022,0.0003346089,0.0005180565,0.0000282132,0.000005393581],"category_scores_gemma":[0.008623614,0.00006247367,0.00002064012,0.0001069254,0.0004693618,0.001371623,0.0001040863,0.0001137679,9.720587e-8],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001028432,"about_ca_system_score_gemma":0.0001175073,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004833382,"about_ca_topic_score_gemma":0.00008957984,"domain_scores_codex":[0.998654,0.0000296071,0.0003183351,0.000161259,0.0007204232,0.0001163942],"domain_scores_gemma":[0.9955992,0.0008543215,0.0004047806,0.00005743732,0.002986062,0.00009820909],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00002844573,0.00009179992,0.9536287,0.00001083164,0.00002721517,5.760249e-7,0.03902229,0.00001913168,0.00123285,0.003576005,0.00000777588,0.002354434],"study_design_scores_gemma":[0.001201883,0.0003300722,0.9217174,0.0004913532,0.00003486655,0.000009949336,0.05073569,0.01238879,0.0006059324,0.01217692,0.000143581,0.0001635466],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9964788,0.00003570714,0.0007286012,0.0001270986,0.0006701163,0.00009354543,0.00001033043,0.00000693092,0.001848883],"genre_scores_gemma":[0.9793755,0.00003026192,0.0195353,0.00002621357,0.00101947,0.000001737314,6.930403e-7,0.000003838842,0.000007000144],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03191124,"threshold_uncertainty_score":0.9997272,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1983043858","doi":"10.1080/15305058.2001.9669474","title":"Illustrating the Use of Nonparametric Regression to Assess Differential Item and Bundle Functioning Among Multiple Groups","year":2001,"lang":"en","type":"article","venue":"International Journal of Testing","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":15,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"","keywords":"Nonparametric statistics; Differential item functioning; Bundle; Nonparametric regression; Differential (mechanical device); Statistics; Regression analysis; Regression; Smoothing; Psychology; Mathematics; Econometrics; Item response theory; Psychometrics","retraction":null,"screen_n_in":null,"score":{"opus":0.6770030026306908,"gpt":0.4656857700680094,"spread":0.2113172325626814,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.003321215,0.000127704,0.000295468,0.0009484454,0.0001953412,0.0005606475,0.000788819,0.00005405924,0.0000513324],"category_scores_gemma":[0.3430624,0.0000725881,0.0001092288,0.001785711,0.00008462518,0.000648876,0.0002960926,0.0003205295,0.000001638141],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004678383,"about_ca_system_score_gemma":0.00003778501,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001200014,"about_ca_topic_score_gemma":0.00001471059,"domain_scores_codex":[0.9964563,0.0003425464,0.001184936,0.0002262805,0.00160513,0.00018485],"domain_scores_gemma":[0.9168736,0.078822,0.001945195,0.0002096289,0.002024204,0.00012532],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0000552982,0.00004058563,0.8330714,0.000001549507,0.00003079314,0.00002224936,0.0001307198,0.004335167,0.003604678,0.00002617814,0.0001915206,0.1584899],"study_design_scores_gemma":[0.0004288973,0.00022445,0.9207742,0.0002945865,0.00001687074,0.0004830958,0.0007331368,0.07546449,0.0003314088,0.0007968309,0.0003516593,0.0001004029],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8711365,0.0000608335,0.1267581,0.0002991157,0.001429904,0.00006725309,0.000002397225,0.000009801714,0.0002361536],"genre_scores_gemma":[0.9205379,0.00000749543,0.07886679,0.00005268185,0.0004281008,0.000001279133,3.690768e-7,0.00000779503,0.00009753148],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3397412,"threshold_uncertainty_score":0.6624713,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2137207467","doi":"10.1080/15305058.2014.977444","title":"Explore the Usefulness of Person-Fit Analysis on Large-Scale Assessment","year":2014,"lang":"en","type":"article","venue":"International Journal of Testing","topic":"Mental Health Research Topics","field":"Psychology","cited_by":11,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"Centre for Health Evaluation and Outcome Sciences; University of Alberta","funders":"","keywords":"Item response theory; Scale (ratio); Psychology; Econometrics; Statistics; Psychometrics; Developmental psychology; Mathematics","retraction":null,"screen_n_in":null,"score":{"opus":0.2900361297675365,"gpt":0.4846693249959824,"spread":0.1946331952284459,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001924997,0.00007655103,0.0001942117,0.0003006866,0.00007215852,0.00003059482,0.0005971738,0.00003518239,0.0006824856],"category_scores_gemma":[0.000610082,0.00005136509,0.0001456119,0.0002744756,0.00004367925,0.00007429058,0.00005530899,0.0003693189,0.00001693861],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001241924,"about_ca_system_score_gemma":0.00005717158,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007594217,"about_ca_topic_score_gemma":0.00001998269,"domain_scores_codex":[0.9981012,0.0002750247,0.0004448436,0.0001115006,0.0008809078,0.0001864767],"domain_scores_gemma":[0.9974274,0.00115142,0.0005676359,0.0001753901,0.0005962552,0.00008187839],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002953483,0.001140337,0.8452838,0.00004539468,0.002434213,0.0001212331,0.008144482,0.00376857,0.0006342431,0.01166414,0.003279607,0.1231886],"study_design_scores_gemma":[0.001562949,0.0008910276,0.9613442,0.0002407135,0.0001755205,0.0001260045,0.007283838,0.02326134,0.0004489701,0.0007441327,0.003786406,0.0001348319],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9718126,0.00003514564,0.008321088,0.003216991,0.001005391,0.0000713212,0.00001127412,0.000006269771,0.01551989],"genre_scores_gemma":[0.9958754,0.000001970633,0.002902396,0.000348192,0.0005130898,0.000005673132,0.00000231693,0.000008338189,0.0003426828],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1230538,"threshold_uncertainty_score":0.7472739,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4400119116","doi":"10.1080/15305058.2024.2364174","title":"Can the dark core of personality be measured briefly, multidimensionally, and invariantly? The D25 measure","year":2024,"lang":"en","type":"article","venue":"International Journal of Testing","topic":"Personality Traits and Psychology","field":"Psychology","cited_by":10,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Western University","funders":"","keywords":"Measure (data warehouse); Psychology; Personality; Core (optical fiber); Cognitive psychology; Social psychology; Computer science; Data mining","retraction":null,"screen_n_in":null,"score":{"opus":0.1511349496781008,"gpt":0.3789092397915193,"spread":0.2277742901134185,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002447736,0.0001445446,0.0002121692,0.0001014337,0.0001160081,0.00008040852,0.0005916146,0.000083259,0.0002566168],"category_scores_gemma":[0.0009697941,0.00008112637,0.0001521204,0.0001747505,0.0003223156,0.00008584712,0.00007038043,0.0005807371,0.000003874422],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004533425,"about_ca_system_score_gemma":0.0001885504,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000750357,"about_ca_topic_score_gemma":0.0001718324,"domain_scores_codex":[0.9980631,0.0002574369,0.0005600344,0.0002017154,0.0007417352,0.0001759997],"domain_scores_gemma":[0.9968582,0.001743439,0.000362451,0.0001663351,0.0007975019,0.00007205523],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.002107204,0.00109985,0.2375917,0.0001847221,0.01087515,0.002598187,0.08123536,0.0002559482,0.08023351,0.1754324,0.146575,0.2618109],"study_design_scores_gemma":[0.002238005,0.0004871088,0.9300282,0.001070665,0.0003713148,0.01369967,0.003524882,0.001448921,0.0001485286,0.01651143,0.03008523,0.0003860137],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9028348,0.004201026,0.0003068491,0.07876394,0.003629707,0.0001510182,0.0001671626,0.00002300383,0.009922464],"genre_scores_gemma":[0.9959673,0.00001294204,0.0006731888,0.00240746,0.0006793607,0.00000318186,0.00000362925,0.00001572558,0.000237268],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6924365,"threshold_uncertainty_score":0.3308237,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2956918498","doi":"10.1080/15305058.2019.1632316","title":"Migration Background in PISA’s Measure of Social Belonging: Using a Diffractive Lens to Interpret Multi-Method DIF Studies","year":2019,"lang":"en","type":"article","venue":"International Journal of Testing","topic":"Intergenerational and Educational Inequality Studies","field":"Social Sciences","cited_by":9,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Ordered logit; Differential item functioning; Logistic regression; Immigration; Measure (data warehouse); Psychology; Through-the-lens metering; Econometrics; Social psychology; Measurement invariance; Differential (mechanical device); Item response theory; Lens (geology); Psychometrics; Statistics; Structural equation modeling; Computer science; Mathematics; Geography; Developmental psychology; Confirmatory factor analysis; Data mining","retraction":null,"screen_n_in":null,"score":{"opus":0.2921427318247282,"gpt":0.5026481723540723,"spread":0.2105054405293441,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001648451,0.000111534,0.0002876292,0.0003290163,0.0001413585,0.00005209349,0.0002918726,0.00005139354,0.00003771357],"category_scores_gemma":[0.004159933,0.0001028848,0.0001116198,0.0002721444,0.00006919645,0.0005205135,0.00008011067,0.0001835637,0.000006171361],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005721112,"about_ca_system_score_gemma":0.0003186176,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001853468,"about_ca_topic_score_gemma":0.001520793,"domain_scores_codex":[0.9979792,0.0002658467,0.0006826817,0.0001419489,0.0007629485,0.000167396],"domain_scores_gemma":[0.9954327,0.001125864,0.0004728673,0.00004110959,0.002887372,0.00004013907],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0004980895,0.0009763211,0.4612355,0.00007251294,0.001487698,0.00002738847,0.3915483,0.01556566,0.08830308,0.02949211,0.001754172,0.009039187],"study_design_scores_gemma":[0.003522657,0.0007221526,0.515481,0.003771774,0.0001967182,0.0001102674,0.4328565,0.02342307,0.005648215,0.008164963,0.004906434,0.001196257],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9861232,0.0001518396,0.004579548,0.006068224,0.001470764,0.0001325095,0.00001094518,0.00000599798,0.001456985],"genre_scores_gemma":[0.9701296,0.00001351382,0.02856488,0.0002317239,0.0007969276,0.000003642938,0.000001734056,0.000008169538,0.0002497978],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08265487,"threshold_uncertainty_score":0.4980129,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3211365142","doi":"10.1207/s15327574ijt0601_7","title":"Book Review","year":2006,"lang":"en","type":"article","venue":"International Journal of Testing","topic":"Behavioral and Psychological Studies","field":"Psychology","cited_by":4,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Psychology","retraction":null,"screen_n_in":null,"score":{"opus":0.2545338026231972,"gpt":0.4185244620667977,"spread":0.1639906594436004,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0002256084,0.00006295963,0.0001329521,0.0000403358,0.00002584121,0.00001416509,0.0002596807,0.00002492314,0.003864827],"category_scores_gemma":[0.0001392192,0.00004384802,0.0000857492,0.00006912563,0.00002988449,0.00006766483,0.0000273407,0.0001475838,0.0001330664],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002229783,"about_ca_system_score_gemma":0.000006426439,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004409533,"about_ca_topic_score_gemma":6.042842e-7,"domain_scores_codex":[0.9991637,0.00003391544,0.0004027553,0.00007159026,0.0002321388,0.00009584388],"domain_scores_gemma":[0.9990789,0.0001337445,0.0003334977,0.00004807322,0.0003800781,0.00002569513],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00001707337,0.000137243,0.02107257,0.00000638765,0.000043452,0.0003209721,0.00001571545,0.00000262402,0.0002298144,0.0003354106,0.9232525,0.05456625],"study_design_scores_gemma":[0.0003571885,0.0001418469,0.1628622,0.0006603207,0.00003378712,0.0009736798,0.00001069937,9.937405e-7,0.00002072783,0.0008049547,0.8340501,0.00008349004],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.03187361,0.7455891,0.0004663343,0.01967066,0.005381399,0.0001119685,0.00000845641,0.00004380757,0.1968547],"genre_scores_gemma":[0.7591633,0.008472298,0.01190691,0.1463037,0.01259487,0.00003044992,0.00001301888,0.0000616227,0.06145386],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7371168,"threshold_uncertainty_score":0.9970458,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4309375705","doi":"10.1080/15305058.2022.2070755","title":"Generating reading comprehension items using automated processes","year":2022,"lang":"en","type":"article","venue":"International Journal of Testing","topic":"Topic Modeling","field":"Computer Science","cited_by":4,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"","keywords":"Reading comprehension; Blank; Comprehension; Computer science; Natural language processing; Test (biology); Reading (process); Artificial intelligence; Salient; Process (computing); Psychology; Linguistics","retraction":null,"screen_n_in":null,"score":{"opus":0.08765853027184589,"gpt":0.3270988022823678,"spread":0.2394402720105219,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0005836127,0.00008139356,0.0001218105,0.0002385512,0.00025172,0.0002066044,0.001026543,0.00001432867,0.00001000299],"category_scores_gemma":[0.0008040593,0.00008133123,0.00004064087,0.0003241671,0.000009164601,0.0005792496,0.0005216599,0.0002393366,8.539359e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002125641,"about_ca_system_score_gemma":0.0002556347,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003755361,"about_ca_topic_score_gemma":5.010996e-7,"domain_scores_codex":[0.9983812,0.00007503158,0.0005051853,0.0001486182,0.0007517376,0.0001382308],"domain_scores_gemma":[0.9980558,0.0002993034,0.000641939,0.0001052889,0.0008513174,0.00004632199],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00000436297,0.000032477,0.004015476,0.000009547412,0.00003194614,0.0002113772,0.0005286653,0.8961634,0.08403037,0.0002903775,0.00008881987,0.0145932],"study_design_scores_gemma":[0.0002036134,0.00004610876,0.0001191007,0.0001097485,0.000004221023,0.002254932,0.00008168027,0.995644,0.0009228606,0.0003159241,0.0002148079,0.00008295094],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5951998,0.0001362838,0.4025631,0.0002968126,0.001492475,0.00003197676,0.000001009492,0.0001196556,0.0001588438],"genre_scores_gemma":[0.665902,8.532843e-7,0.3336343,0.0001611511,0.0002849828,7.930742e-7,4.941603e-7,0.000005806623,0.000009656284],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.09948067,"threshold_uncertainty_score":0.3316591,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4381282730","doi":"10.1080/15305058.2023.2214648","title":"The analysis of TIMSS 2015 data with confirmatory mixture item response theory: A multidimensional approach","year":2023,"lang":"en","type":"article","venue":"International Journal of Testing","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Confirmatory factor analysis; Item response theory; Class (philosophy); Psychology; Mathematics education; Structural equation modeling; Polytomous Rasch model; Sample (material); Latent class model; Cognition; Psychometrics; Mathematics; Statistics; Developmental psychology; Artificial intelligence; Computer science","retraction":null,"screen_n_in":null,"score":{"opus":0.5081435997123916,"gpt":0.5085350331339626,"spread":0.0003914334215709969,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03752734,0.0001204211,0.0003883866,0.001569794,0.0001594976,0.0001718099,0.002859401,0.00004826943,0.00003722357],"category_scores_gemma":[0.3058342,0.00006046657,0.0001436402,0.004647022,0.0001836548,0.0003067629,0.0005449347,0.0002884439,0.000007477617],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003468797,"about_ca_system_score_gemma":0.0002300039,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001192893,"about_ca_topic_score_gemma":0.000002445597,"domain_scores_codex":[0.9944501,0.001279714,0.001120532,0.0002862797,0.002669386,0.000194025],"domain_scores_gemma":[0.8217819,0.1724452,0.002073327,0.0006825376,0.002922592,0.00009438432],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.007351219,0.0002952498,0.5579388,0.00001053637,0.007972754,0.0005736614,0.001747583,0.07577638,0.009435109,0.002778317,0.0448066,0.2913138],"study_design_scores_gemma":[0.0009233812,0.0001829948,0.6400357,0.0001182503,0.0003478175,0.0004757733,0.003734874,0.342062,0.0001253738,0.004025065,0.007783344,0.0001854002],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9578756,0.0005268704,0.03737101,0.001953447,0.0009350386,0.00008875431,0.0001146134,0.00003172991,0.001102905],"genre_scores_gemma":[0.9123429,0.0000135796,0.0869462,0.00009529054,0.0001748027,0.000001326177,0.00001253802,0.00001067046,0.0004026938],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2911284,"threshold_uncertainty_score":0.9910681,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4391876290","doi":"10.1080/15305058.2024.2318424","title":"A meta-analysis of the relationship between Wonderlic test scores and school success","year":2024,"lang":"en","type":"article","venue":"International Journal of Testing","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Calgary; Wilfrid Laurier University","funders":"","keywords":"Psychology; Test (biology); Clinical psychology; Meta-analysis; Medicine; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.8254326103415369,"gpt":0.5486442116530132,"spread":0.2767883986885237,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.008293644,0.00009919257,0.0005317974,0.001711094,0.00008821171,0.0004819253,0.001228756,0.0000434242,0.0002053544],"category_scores_gemma":[0.3987939,0.00005089274,0.0006360447,0.004328137,0.0001002503,0.000403762,0.0002230583,0.0003444205,0.000004927601],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003661703,"about_ca_system_score_gemma":0.000121672,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004962062,"about_ca_topic_score_gemma":0.00000721717,"domain_scores_codex":[0.9963987,0.0003315995,0.001251556,0.0002060519,0.00170403,0.0001080083],"domain_scores_gemma":[0.6980101,0.2977378,0.001810444,0.0003590336,0.001953068,0.0001295745],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000002372324,0.000008815602,0.9838892,0.000003232847,0.008704969,0.000010292,0.00006588923,0.000689731,0.0001072305,0.0003426697,0.0002173723,0.005958269],"study_design_scores_gemma":[0.00006650066,0.0000323046,0.9566522,0.00004321203,0.01108514,0.00006421448,0.00007879938,0.002732071,0.0000489491,0.02905101,0.00009391495,0.00005165529],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9805211,0.00359482,0.01040544,0.003314737,0.0006866471,0.0000543615,0.00003829553,0.00001440307,0.00137014],"genre_scores_gemma":[0.9845148,0.000003811748,0.01490318,0.00006268499,0.0002051774,9.5627e-7,3.843719e-7,0.000005459469,0.0003035723],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3905003,"threshold_uncertainty_score":0.6062703,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2266762729","doi":"10.1080/15305058.2015.1057826","title":"Developing a Validity Argument Through Abductive Reasoning with an Empirical Demonstration of the Latent Class Analysis","year":2015,"lang":"en","type":"article","venue":"International Journal of Testing","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of British Columbia","funders":"","keywords":"Abductive reasoning; Test (biology); Argument (complex analysis); Psychology; Class (philosophy); Empirical research; Computer science; Latent class model; Cognitive psychology; Empirical evidence; Artificial intelligence; Social psychology; Mathematics education; Natural language processing; Machine learning; Epistemology","retraction":null,"screen_n_in":null,"score":{"opus":0.7377369571260765,"gpt":0.5358460428535601,"spread":0.2018909142725164,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.007215058,0.000105304,0.0003173408,0.0004148937,0.00008865364,0.0001709959,0.00102877,0.00004207063,0.0000109213],"category_scores_gemma":[0.06794868,0.00005595946,0.000147818,0.002425856,0.00009148374,0.0005930354,0.0001450725,0.0002305286,9.078845e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002079772,"about_ca_system_score_gemma":0.0004092807,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007664078,"about_ca_topic_score_gemma":0.00002285039,"domain_scores_codex":[0.9957125,0.0006047211,0.0009970593,0.0002107803,0.002336496,0.0001384586],"domain_scores_gemma":[0.9865242,0.0068156,0.002148016,0.0002162373,0.004211663,0.00008427414],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00007231005,0.00004794783,0.9649629,7.471823e-7,0.0002529074,0.00001551817,0.0006857917,0.02209044,0.0001310376,0.0006477346,0.00006195086,0.0110307],"study_design_scores_gemma":[0.0006638151,0.0004855698,0.9168498,0.0001838666,0.0002119164,0.0004663201,0.002960981,0.03074628,0.001515046,0.04555478,0.0002032493,0.0001583927],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7036539,0.00003053196,0.2938888,0.001117067,0.0004894002,0.00003831409,0.000002083735,0.000005969108,0.0007738993],"genre_scores_gemma":[0.6698086,0.000001300306,0.3299562,0.00007730206,0.0001409055,6.081061e-7,4.063648e-7,0.000003173852,0.00001154036],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.06073362,"threshold_uncertainty_score":0.9399024,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2119846802","doi":"10.1080/15305058.2014.976865","title":"Impact of Inclusion of Varying Percentages of Repeaters on Equating","year":2015,"lang":"en","type":"article","venue":"International Journal of Testing","topic":"School Choice and Performance","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Centre for Advancing Health Outcomes; University of Alberta","funders":"","keywords":"Equating; Statistics; Representativeness heuristic; Population; Sample (material); Psychology; Mathematics; Demography; Econometrics; Chemistry","retraction":null,"screen_n_in":null,"score":{"opus":0.1350884152148236,"gpt":0.4379068574010745,"spread":0.3028184421862509,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001456594,0.00004946928,0.0001422066,0.0001704061,0.00006479395,0.00001212893,0.0003508331,0.00002845683,0.0000307446],"category_scores_gemma":[0.004038692,0.00003954358,0.00009488443,0.0001322415,0.00005921013,0.0003046598,0.0001088632,0.000114627,7.680667e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001557491,"about_ca_system_score_gemma":0.0002913076,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001865762,"about_ca_topic_score_gemma":0.00001762389,"domain_scores_codex":[0.9984651,0.00006917466,0.0004736999,0.00004831524,0.0008588582,0.00008481293],"domain_scores_gemma":[0.9972957,0.0004513183,0.0009989538,0.0000475302,0.001145635,0.00006087114],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001724807,0.00009232168,0.9150168,0.00001323292,0.00008222034,0.00001266722,0.01848077,0.009972522,0.03498427,0.00008623977,0.0001318605,0.02095463],"study_design_scores_gemma":[0.006498565,0.006218181,0.872264,0.01080186,0.0001620353,0.0002152815,0.01968079,0.01550744,0.06030405,0.007288465,0.0003861577,0.000673145],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9928899,0.00005411323,0.00007607526,0.0001857524,0.0003190213,0.00002321455,0.000003427452,0.000002991123,0.006445519],"genre_scores_gemma":[0.9971341,0.00001557391,0.002412941,0.00001886547,0.0003940438,1.271094e-7,5.581431e-7,0.000003756156,0.00002002193],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.04275276,"threshold_uncertainty_score":0.4834983,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}