{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":12,"total_is_capped":false,"direct_labels_cover":0,"predictions_cover":12,"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":"f042d3dd4442","filters":{"venue":"Applied Measurement in Education"}},"results":[{"id":"W1969608664","doi":"10.1207/s15324818ame1703_4","title":"Comparability of Bilingual Versions of Assessments: Sources of Incomparability of English and French Versions of Canada's National Achievement Tests","year":2004,"lang":"en","type":"article","venue":"Applied Measurement in Education","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":62,"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":"Comparability; Differential item functioning; Psychology; Reading (process); Item response theory; Psychometrics; Developmental psychology; Linguistics; Mathematics","retraction":null,"screen_n_in":null,"score":{"opus":0.2715701847682983,"gpt":0.4269062826454618,"spread":0.1553360978771635,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.008387373,0.0001096597,0.00049297,0.0004315183,0.00004459875,0.000004626017,0.0003663378,0.00005234744,0.00002657251],"category_scores_gemma":[0.02685922,0.00009530023,0.00005264134,0.001388067,0.000232924,0.00006646567,0.0000929767,0.0001000504,6.989185e-8],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000342068,"about_ca_system_score_gemma":0.003462436,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03255771,"about_ca_topic_score_gemma":0.01511523,"domain_scores_codex":[0.9956943,0.0002184989,0.001385108,0.0003144009,0.002249623,0.0001380937],"domain_scores_gemma":[0.9913903,0.003827856,0.001246788,0.0004199934,0.003048027,0.00006706948],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00006717769,0.002282373,0.9710136,0.0001585683,0.00003808071,1.666875e-8,0.002194308,0.004583712,0.01188098,0.002343472,0.0001626945,0.005274998],"study_design_scores_gemma":[0.0007111569,0.0001160835,0.9460227,0.0001031654,0.00001943878,1.014703e-7,0.009303373,0.00003704233,0.02926517,0.01426952,0.00007830784,0.00007388488],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9961377,0.0001505411,0.0004005036,0.000064529,0.0003507019,0.0003676402,0.00003405909,0.000003186382,0.002491179],"genre_scores_gemma":[0.9884601,0.000006330346,0.01148844,0.000008167554,0.00001623357,0.00001060651,0.000005461011,0.000002825405,0.000001867343],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02499086,"threshold_uncertainty_score":0.981338,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2053906918","doi":"10.1207/s15324818ame1703_2","title":"Performance of SIBTEST When the Percentage of DIF Items is Large","year":2004,"lang":"en","type":"article","venue":"Applied Measurement in Education","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 Alberta","funders":"","keywords":"Differential item functioning; Statistics; Psychology; Item response theory; Test (biology); Mathematics; Psychometrics","retraction":null,"screen_n_in":null,"score":{"opus":0.3321836598049474,"gpt":0.4109344435671419,"spread":0.07875078376219452,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008470052,0.00009000178,0.0002022913,0.0002767493,0.00006880549,0.0000233831,0.000544811,0.0000426277,0.0001138011],"category_scores_gemma":[0.004016101,0.00005734235,0.00004555983,0.0009164349,0.00005582198,0.00008708298,0.0000515973,0.000103167,0.00001811303],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000113517,"about_ca_system_score_gemma":0.0002257586,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001225965,"about_ca_topic_score_gemma":0.00002395923,"domain_scores_codex":[0.9974545,0.00008611271,0.0006589605,0.0002464346,0.001372667,0.000181364],"domain_scores_gemma":[0.9977362,0.0009921847,0.0004216503,0.0005337654,0.0002843705,0.00003189881],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0000604955,0.001300807,0.6753676,0.00006746031,0.00001909356,5.382973e-8,0.01170131,0.0003834654,0.04315415,0.005806043,0.002977504,0.259162],"study_design_scores_gemma":[0.0006949356,0.00008189143,0.9369794,0.00009999202,0.00001356206,0.000001068024,0.01124099,0.0001086081,0.02165337,0.02301773,0.005974601,0.0001338866],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9838824,0.0003156606,0.001091426,0.0005940866,0.0005073806,0.0003318789,0.000002686569,0.000007670061,0.01326678],"genre_scores_gemma":[0.9955154,0.00005325194,0.004076695,0.0002081125,0.0000530021,0.00003584725,0.000001453847,0.00000529286,0.0000509333],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2616118,"threshold_uncertainty_score":0.4807938,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2320000683","doi":"10.1080/08957347.2016.1171768","title":"Evaluating the Psychometric Characteristics of Generated Multiple-Choice Test Items","year":2016,"lang":"en","type":"article","venue":"Applied Measurement in Education","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":33,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Medical Council of Canada; University of Ottawa; University of Alberta","funders":"","keywords":"Item response theory; Item bank; Item analysis; Licensure; Psychology; Test (biology); Psychometrics; Cognition; Multiple choice; Differential item functioning; Applied psychology; Cognitive psychology; Computer science; Clinical psychology; Statistics; Medicine; Medical education","retraction":null,"screen_n_in":null,"score":{"opus":0.1433803510187834,"gpt":0.4009891954820371,"spread":0.2576088444632537,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.001702891,0.0001067831,0.0002130775,0.0001468083,0.00003640998,0.000009104926,0.00008950831,0.0000664815,0.00005114227],"category_scores_gemma":[0.1396537,0.00006069112,0.00003918018,0.000575805,0.00004764235,0.0000193779,0.00001235349,0.0001046135,0.00003333936],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001921021,"about_ca_system_score_gemma":0.0004330451,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007539517,"about_ca_topic_score_gemma":0.00001383102,"domain_scores_codex":[0.998495,0.00004498626,0.0005079143,0.000222554,0.0005637342,0.0001657986],"domain_scores_gemma":[0.9879235,0.01080629,0.0002755947,0.0004049234,0.0004985972,0.00009109909],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00005615151,0.001016308,0.480281,0.00001842436,0.00001388515,4.796237e-8,0.00008952293,0.000001480237,0.1504353,0.00006678365,0.0005750424,0.367446],"study_design_scores_gemma":[0.001636498,0.0001918504,0.9893171,0.0009796123,0.00007346819,0.000001469535,0.00007776323,0.0000758118,0.006906064,0.0001018148,0.0005525064,0.00008602907],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9957502,0.0002490041,0.0001936668,0.00143513,0.0005333007,0.0006671475,0.000004273586,0.0000203872,0.001146885],"genre_scores_gemma":[0.9982314,0.0001157723,0.0005720544,0.0003872425,0.0003459881,0.0001939985,0.00002002709,0.00001471179,0.0001188512],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5090361,"threshold_uncertainty_score":0.8675934,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2109920068","doi":"10.1080/08957347.2014.905787","title":"Testing Expert-Based Versus Student-Based Cognitive Models for a Grade 3 Diagnostic Mathematics Assessment","year":2014,"lang":"en","type":"article","venue":"Applied Measurement in Education","topic":"Educational and Psychological Assessments","field":"Psychology","cited_by":19,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Medical Council of Canada; University of Alberta","funders":"","keywords":"Cognition; Psychology; Test (biology); Consistency (knowledge bases); Construct (python library); Cognitive test; Cognitive psychology; Statistics; Artificial intelligence; Computer science; Mathematics","retraction":null,"screen_n_in":null,"score":{"opus":0.2185151809927274,"gpt":0.4421785572880547,"spread":0.2236633762953272,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.001295043,0.0002629938,0.0002587121,0.0001695115,0.0001315111,0.00005266006,0.0002411863,0.0001197593,0.0001235269],"category_scores_gemma":[0.0005093385,0.0002513474,0.00006603919,0.0003135667,0.00004841842,0.00005624059,0.00001305043,0.0001806728,0.00004684719],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004200838,"about_ca_system_score_gemma":0.0004521378,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005562872,"about_ca_topic_score_gemma":0.00002090272,"domain_scores_codex":[0.9977715,0.0001417622,0.0005010576,0.0005470612,0.0006419721,0.0003966288],"domain_scores_gemma":[0.9957211,0.003207605,0.0002587611,0.0003149395,0.0003749218,0.0001226806],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.001495234,0.1102479,0.07636062,0.0007610318,0.0006306461,0.000001794093,0.01429826,0.007391716,0.009007501,0.4007683,0.01364861,0.3653884],"study_design_scores_gemma":[0.0332235,0.002544886,0.7977805,0.001513208,0.0005319374,0.000002162036,0.03865654,0.01429783,0.00156852,0.1031844,0.00432783,0.002368664],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5986949,0.0003534977,0.2907467,0.002763523,0.006284835,0.00650803,0.00003065722,0.0002261024,0.09439171],"genre_scores_gemma":[0.9553943,0.000001307105,0.03384065,0.001250344,0.0003034387,0.008989414,0.0001553802,0.00003565005,0.00002956484],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7214199,"threshold_uncertainty_score":0.9999939,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2607007324","doi":"10.1080/08957347.2017.1316276","title":"For Which Boys and Which Girls Are Reading Assessment Items Biased Against? Detection of Differential Item Functioning in Heterogeneous Gender Populations","year":2017,"lang":"en","type":"article","venue":"Applied Measurement in Education","topic":"School Choice and Performance","field":"Social Sciences","cited_by":14,"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":"Institute of Education Sciences","keywords":"Differential item functioning; Psychology; Disadvantage; Developmental psychology; Socioeconomic status; Gender bias; Item response theory; Demography; Psychometrics; Social psychology; Population","retraction":null,"screen_n_in":null,"score":{"opus":0.1191182427796675,"gpt":0.3618901915618662,"spread":0.2427719487821987,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001154729,0.0001035847,0.0001530637,0.00018786,0.0006814459,0.0001405372,0.0001199432,0.000103453,0.0000064717],"category_scores_gemma":[0.0003532072,0.000115786,0.00002300619,0.0002033912,0.00002656432,0.0002652766,0.0000158742,0.0001211044,9.379851e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00044154,"about_ca_system_score_gemma":0.0003030027,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001212185,"about_ca_topic_score_gemma":0.04451766,"domain_scores_codex":[0.9987578,0.00006286146,0.0003140557,0.0002479444,0.000402651,0.0002147248],"domain_scores_gemma":[0.9990309,0.00004499794,0.00034063,0.0002171321,0.0003097595,0.00005660948],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00007642937,0.0003425983,0.8873059,0.0001416556,0.00002362272,3.894926e-8,0.004390021,0.0009449901,0.07174831,0.001141685,0.00008596254,0.03379886],"study_design_scores_gemma":[0.0006331923,0.00002042232,0.9895396,0.0001173794,0.00002486791,1.571572e-7,0.003411435,0.001065597,0.003839413,0.0006933319,0.0004971424,0.0001574859],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9932019,0.00006212814,0.001510347,0.0002503477,0.0009722851,0.0007675533,0.000002399528,0.00001631747,0.003216722],"genre_scores_gemma":[0.9987636,0.0000645998,0.0003062924,0.00003016684,0.0003640564,0.0004199852,0.00002083211,0.00001083198,0.0000196769],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1022338,"threshold_uncertainty_score":0.9729174,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3126939008","doi":"10.1080/08957347.2020.1835911","title":"Rethinking Think-Alouds: The Often-Problematic Collection of Response Process Data","year":2021,"lang":"en","type":"article","venue":"Applied Measurement in Education","topic":"Behavioral and Psychological Studies","field":"Psychology","cited_by":14,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Think aloud protocol; Psychology; Unobservable; Process (computing); Test (biology); Cognitive psychology; Data collection; Applied psychology; Computer science; Epistemology; Social science; Sociology","retraction":null,"screen_n_in":null,"score":{"opus":0.3773982575143485,"gpt":0.3923418996521501,"spread":0.01494364213780164,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002978272,0.0001198234,0.0001854807,0.00005499141,0.0001407699,0.00002034048,0.0003806291,0.00009705694,0.0002317394],"category_scores_gemma":[0.0002538009,0.0000846366,0.00002534584,0.0006832059,0.00006446467,0.00005208069,0.00006774547,0.0002306027,0.00003396815],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001092911,"about_ca_system_score_gemma":0.0002416774,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00009469057,"about_ca_topic_score_gemma":0.0001375913,"domain_scores_codex":[0.9982158,0.0003397097,0.0004227982,0.0003947374,0.0004459008,0.0001810672],"domain_scores_gemma":[0.9986216,0.0001302344,0.0002013665,0.0007540135,0.0002664219,0.00002638732],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.005655857,0.01734253,0.03960573,0.0008273451,0.0005530186,0.000007481457,0.339992,0.0000638454,0.1304496,0.03493101,0.178391,0.2521806],"study_design_scores_gemma":[0.001742882,0.0001973437,0.7917764,0.0004510158,0.0002350818,0.00002467244,0.05766438,0.00001303398,0.00487742,0.1349367,0.007600782,0.0004803173],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9648556,0.002447868,0.0001586743,0.006747357,0.001572462,0.001078025,0.000006278628,0.00004942071,0.02308437],"genre_scores_gemma":[0.9982759,0.00003687787,0.000399093,0.0006398546,0.00006975587,0.0002760873,0.00003167002,0.00001029364,0.0002604208],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7521707,"threshold_uncertainty_score":0.345138,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2011276392","doi":"10.1080/08957347.2011.607061","title":"An Experimental Test of Student Verbal Reports and Teacher Evaluations as a Source of Validity Evidence for Test Development","year":2011,"lang":"en","type":"article","venue":"Applied Measurement in Education","topic":"Science Education and Pedagogy","field":"Social Sciences","cited_by":10,"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":"Test (biology); Psychology; CLARITY; Test validity; Set (abstract data type); Association (psychology); Standards for Educational and Psychological Testing; Mathematics education; Content validity; Applied psychology; Social psychology; Psychometrics; Computer science; Higher education; Clinical psychology; Education theory","retraction":null,"screen_n_in":null,"score":{"opus":0.4437717804134738,"gpt":0.4947924416640539,"spread":0.05102066125058008,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004424853,0.00007828305,0.0001137153,0.0001015613,0.0001945582,0.00001913482,0.0001488858,0.00004661857,0.0001539545],"category_scores_gemma":[0.001175666,0.00008236722,0.00001752254,0.00020666,0.00014949,0.0001719296,0.00001461044,0.00004613694,0.000002121533],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003331387,"about_ca_system_score_gemma":0.002664418,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001525512,"about_ca_topic_score_gemma":0.0008193405,"domain_scores_codex":[0.9983953,0.00009194715,0.0003899654,0.0002525097,0.0007218582,0.0001483917],"domain_scores_gemma":[0.9989337,0.000182288,0.0002681196,0.0001871187,0.0003333165,0.00009548009],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00002256877,0.004755626,0.5436107,0.00002435073,0.000005885652,3.222234e-8,0.3546158,0.000002871036,0.07450552,0.001630919,0.000203308,0.02062247],"study_design_scores_gemma":[0.0002192968,0.0001927495,0.6756202,0.00008735392,0.00002496963,7.994364e-7,0.2036429,0.000005275479,0.1176183,0.0008852561,0.001544873,0.0001580306],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9936035,0.000167909,0.0001996531,0.0001120276,0.0003284138,0.0009284915,2.9693e-7,0.00001150483,0.004648169],"genre_scores_gemma":[0.9968539,0.00001215565,0.002461854,0.00004254041,0.00007325351,0.0004754603,0.000003369563,0.000005368463,0.00007210209],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1509729,"threshold_uncertainty_score":0.4726562,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4395110387","doi":"10.1080/08957347.2024.2345594","title":"A Critical Review of Fairness from Multiple Perspectives: Implications for Classroom Assessment Theory","year":2024,"lang":"en","type":"review","venue":"Applied Measurement in Education","topic":"Student Assessment and Feedback","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":"Queen's University","funders":"","keywords":"Psychology; Item response theory; Critical theory; Mathematics education; Political science; Developmental psychology; Psychometrics","retraction":null,"screen_n_in":null,"score":{"opus":0.1496372500266458,"gpt":0.4788594395403927,"spread":0.3292221895137469,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.003264273,0.0003242063,0.001007217,0.0002476863,0.0001904406,0.00008077655,0.0004897895,0.0002165713,0.000105574],"category_scores_gemma":[0.001015158,0.0002849067,0.0003772509,0.0006210738,0.0001528189,0.00007751377,0.00004381668,0.0003250773,0.00003907253],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002387168,"about_ca_system_score_gemma":0.00442754,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000104128,"about_ca_topic_score_gemma":0.0001284276,"domain_scores_codex":[0.9970886,0.0003801141,0.0008040908,0.0006489864,0.0007720101,0.0003062323],"domain_scores_gemma":[0.9974158,0.001194392,0.0003159571,0.0004480115,0.0005344437,0.00009142701],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000004710579,0.0008212969,0.00002729918,0.02249685,0.0001215305,2.710359e-8,0.002366043,4.434713e-8,0.000001487969,0.5168834,0.006791314,0.450486],"study_design_scores_gemma":[0.0001307088,0.00002482247,0.0001499932,0.06344053,0.001504282,6.400764e-8,0.01305445,3.471596e-7,2.99927e-7,0.03059733,0.8907909,0.0003062453],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[6.322704e-7,0.9658984,0.001097889,0.001218979,0.000915042,0.004639262,0.00009748145,0.00004514722,0.02608715],"genre_scores_gemma":[0.001294066,0.988445,0.001183698,0.0001555839,0.0006002989,0.007940052,0.0002379684,0.00004502038,0.00009835713],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.8839996,"threshold_uncertainty_score":0.9999603,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2202994334","doi":"10.1080/08957347.2015.1102913","title":"Analyzing Fairness Among Linguistic Minority Populations Using a Latent Class Differential Item Functioning Approach","year":2015,"lang":"en","type":"article","venue":"Applied Measurement in Education","topic":"Social and Intergroup Psychology","field":"Social Sciences","cited_by":8,"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; Item response theory; Race (biology); Latent class model; Ethnic group; Social psychology; Measurement invariance; Test (biology); Statistics; Developmental psychology; Psychometrics; Confirmatory factor analysis; Structural equation modeling; Mathematics; Sociology","retraction":null,"screen_n_in":null,"score":{"opus":0.1834551797688675,"gpt":0.3591230444973037,"spread":0.1756678647284362,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001061836,0.0001232896,0.0001672743,0.0001818073,0.0003593452,0.00009027188,0.0001435226,0.0001351068,0.00002400477],"category_scores_gemma":[0.0003733235,0.000136002,0.00004853819,0.0004946789,0.0001054089,0.0001165322,0.00002159558,0.0001768158,0.000007301945],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001208218,"about_ca_system_score_gemma":0.0004431281,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008018148,"about_ca_topic_score_gemma":0.003528761,"domain_scores_codex":[0.9983867,0.0001791236,0.0003174195,0.0003025177,0.0005159551,0.0002982145],"domain_scores_gemma":[0.999191,0.00002027935,0.0001610097,0.0001429562,0.000342771,0.0001419465],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00005836376,0.00129745,0.849906,0.00003270233,0.00004694228,2.456595e-7,0.03933227,0.0005436162,0.001920375,0.0957669,0.0007328232,0.01036234],"study_design_scores_gemma":[0.001117942,0.00003953254,0.9223461,0.00009014964,0.0001861201,8.99218e-7,0.04887863,0.002814637,0.00008280448,0.02080049,0.003006028,0.0006366808],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9380074,0.00008841634,0.0115424,0.000149551,0.003559286,0.0004800042,5.555068e-7,0.00006358101,0.04610877],"genre_scores_gemma":[0.9977247,0.000002950681,0.0008631087,0.00003892822,0.001118546,0.0001154045,0.00002592043,0.00001055481,0.00009984752],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07496641,"threshold_uncertainty_score":0.9985875,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4321452091","doi":"10.1080/08957347.2023.2172017","title":"Dissecting Knowledge, Guessing, and Blunder in Multiple Choice Assessments","year":2023,"lang":"en","type":"article","venue":"Applied Measurement in Education","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University; University of Toronto","funders":"Division of Molecular and Cellular Biosciences; National Institute of General Medical Sciences; National Heart, Lung, and Blood Institute","keywords":"Test (biology); Probabilistic logic; Mistake; Psychology; Proxy (statistics); Robustness (evolution); Bayesian probability; Social psychology; Computer science; Artificial intelligence; Machine learning; Political science","retraction":null,"screen_n_in":null,"score":{"opus":0.8152165292421817,"gpt":0.575029629601757,"spread":0.2401868996404247,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.06437974,0.0001732105,0.0008654971,0.0007176268,0.0001015368,0.0003961639,0.0004215868,0.00005801153,0.0002395698],"category_scores_gemma":[0.01557508,0.0001123527,0.0001176422,0.002330443,0.00001892641,0.0001648671,0.00006974534,0.0001303644,0.0006395519],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002026802,"about_ca_system_score_gemma":0.0001754238,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007092561,"about_ca_topic_score_gemma":0.001603428,"domain_scores_codex":[0.9923301,0.001100123,0.003099388,0.0007346888,0.002497937,0.0002378056],"domain_scores_gemma":[0.9957916,0.001565289,0.001019203,0.0009746587,0.0005698685,0.0000793539],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000003525796,0.0003805126,0.8331961,0.00009419455,0.00003188405,1.745126e-7,0.002389649,0.0001240489,0.002812528,0.00128522,0.01001464,0.1496675],"study_design_scores_gemma":[0.0003106279,0.000006513758,0.9585586,0.0001120652,0.00003023288,3.893961e-7,0.007457862,0.00381378,0.0001364606,0.005836555,0.02357472,0.0001621543],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9586925,0.0006102983,0.0006572054,0.0002429759,0.0008178882,0.000970385,5.814333e-7,0.00001129963,0.03799687],"genre_scores_gemma":[0.9969668,0.00001628959,0.0007082941,0.00006144077,0.00009557604,0.000291867,0.000009463004,0.00001011042,0.001840189],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1495054,"threshold_uncertainty_score":0.9927171,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4416211129","doi":"10.1080/08957347.2025.2563889","title":"Establishing Cognitive Item Models for Fair and Theory-Grounded Automatic Item Generation: A Large-Scale Assessment Study with Image-Based Math Items","year":2025,"lang":"en","type":"article","venue":"Applied Measurement in Education","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 Alberta","funders":"Fonds National de la Recherche Luxembourg","keywords":"Item response theory; Cognition; Item analysis; Differential item functioning; Test (biology); Equating; Item bank; Standardized test","retraction":null,"screen_n_in":null,"score":{"opus":0.2627134757698749,"gpt":0.4411525673993796,"spread":0.1784390916295048,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01886035,0.0002276633,0.0003931785,0.0007776427,0.0003368564,0.000779983,0.0003158481,0.00006958295,0.00001297015],"category_scores_gemma":[0.009214113,0.0001720643,0.00004061659,0.001873682,0.00004696043,0.000436351,0.00005854543,0.0001633395,0.000001403128],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003303513,"about_ca_system_score_gemma":0.0007973295,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003823319,"about_ca_topic_score_gemma":0.0001851927,"domain_scores_codex":[0.9963232,0.0005662334,0.0007674625,0.0007831723,0.001241731,0.0003181808],"domain_scores_gemma":[0.9875879,0.01056697,0.0003995275,0.0004351857,0.0009371789,0.00007319247],"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":[0.000544945,0.006991009,0.1052943,0.0002907903,0.000233188,0.000001314129,0.01132314,0.002451187,0.003906493,0.06665377,0.002266254,0.8000436],"study_design_scores_gemma":[0.006734626,0.0005339031,0.1844523,0.0004455379,0.0002422017,0.000001999072,0.1769689,0.4873459,0.0007758313,0.1416061,0.0002737096,0.0006190226],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3994448,0.0001269324,0.5934358,0.0002438402,0.0003385436,0.002177852,0.000005382751,0.00004937891,0.004177491],"genre_scores_gemma":[0.9039317,0.000002710794,0.093821,0.0003348278,0.0000735252,0.001724798,0.00001946402,0.00001446291,0.00007752952],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7994246,"threshold_uncertainty_score":0.9991317,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4365813887","doi":"10.1080/08957347.2023.2201703","title":"Multi-Group Generalizations of SIBTEST and Crossing-SIBTEST","year":2023,"lang":"en","type":"article","venue":"Applied Measurement in Education","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":false},"ca_institutions":"York University","funders":"","keywords":"Type I and type II errors; Differential item functioning; Statistics; Mathematics; Monte Carlo method; Logistic regression; Set (abstract data type); Population; Group (periodic table); Statistical power; Applied mathematics; Item response theory; Computer science; Psychometrics; Demography","retraction":null,"screen_n_in":null,"score":{"opus":0.5905972101310687,"gpt":0.4788309874553456,"spread":0.1117662226757231,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.007057887,0.0001025635,0.0002019157,0.0007882515,0.0001604936,0.0001532554,0.0002666744,0.00005727429,0.00002914561],"category_scores_gemma":[0.01823118,0.00008741942,0.00002710293,0.003610788,0.0001004703,0.00009875149,0.00006686222,0.00007633861,0.00002926074],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006746985,"about_ca_system_score_gemma":0.0001975211,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001091461,"about_ca_topic_score_gemma":0.0001017184,"domain_scores_codex":[0.9977792,0.000128563,0.0006096356,0.000381448,0.0009000583,0.0002010751],"domain_scores_gemma":[0.9970587,0.001891373,0.0002825612,0.0003727963,0.0003295388,0.00006495764],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001186769,0.000496547,0.4914772,0.00003178439,0.00001011715,2.278856e-7,0.001793361,0.0007562803,0.1121447,0.005127875,0.004949647,0.3832004],"study_design_scores_gemma":[0.0003832292,0.00002104808,0.9767169,0.00003482462,0.000007164126,9.450978e-7,0.00252461,0.001435452,0.001216855,0.01303161,0.004500809,0.0001265428],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.975493,0.0005826306,0.01754891,0.0004522743,0.001038431,0.0004651044,0.000004358495,0.00007983523,0.00433544],"genre_scores_gemma":[0.9632537,0.00004917711,0.03631192,0.00009164553,0.00006309922,0.0000756052,0.000009985295,0.000009390904,0.0001355049],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4852397,"threshold_uncertainty_score":0.9900387,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}