{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":4,"total_is_capped":false,"direct_labels_cover":0,"predictions_cover":4,"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","author_layer_release":"2026-06-26"},"query_hash":"3202f460c37a","filters":{"venue":"Large-scale Assessments in Education"}},"results":[{"id":"W2589097440","doi":"10.1186/s40536-017-0042-x","title":"A structural equation modeling approach for examining position effects in large-scale assessments","year":2017,"lang":"en","type":"article","venue":"Large-scale Assessments in Education","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","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 Alberta","funders":"","keywords":"Structural equation modeling; Position (finance); Equivalence (formal languages); Item response theory; Scale (ratio); Multilevel model; Context (archaeology); Computer science; Econometrics; Position paper; Psychometrics; Statistics; Mathematics; Machine learning","authors":[{"name":"Okan Bulut","is_ca":true},{"name":"Qi Quo","is_ca":true},{"name":"Mark J. Gierl","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.344505919788792,"gpt":0.5186875194954802,"spread":0.1741815997066882,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.009607917,0.0003111528,0.0005721632,0.0009971969,0.0007515107,0.0009065805,0.001174132,0.0002569323,0.0000285976],"category_scores_gemma":[0.01053657,0.0002846973,0.0001117053,0.0009741644,0.00003871893,0.001870418,0.0002772745,0.0003906966,0.000005961013],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004606865,"about_ca_system_score_gemma":0.0003419928,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001662073,"about_ca_topic_score_gemma":0.000125741,"domain_scores_codex":[0.9950364,0.0007421898,0.001158389,0.001145311,0.001142215,0.0007755588],"domain_scores_gemma":[0.9939952,0.003336518,0.0009694097,0.00121318,0.0003534399,0.0001322046],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00007775569,0.000780119,0.8626212,0.00008078671,0.00001683883,0.000001196804,0.001313098,0.01017128,0.0004672109,0.0008644145,0.0001294717,0.1234767],"study_design_scores_gemma":[0.001173912,0.00008911853,0.3893444,0.00008713156,0.00001325444,0.000002177104,0.00431423,0.5861042,0.00004314716,0.01859403,0.00003349967,0.0002009212],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5305983,0.00006447565,0.4642662,0.00006488831,0.001620722,0.0007620946,0.00001475858,0.00002584967,0.002582789],"genre_scores_gemma":[0.6921186,0.000008768279,0.3068403,0.0001024708,0.0002001177,0.0003661892,0.0001338015,0.00002349753,0.0002062034],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5759329,"threshold_uncertainty_score":0.9999605,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4384942374","doi":"10.1186/s40536-023-00177-5","title":"Incorporating test-taking engagement into the item selection algorithm in low-stakes computerized adaptive tests","year":2023,"lang":"en","type":"article","venue":"Large-scale Assessments in Education","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","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 Alberta","funders":"","keywords":"Computerized adaptive testing; Test (biology); Selection (genetic algorithm); Psychology; Computer science; Trait; Machine learning; Psychometrics; Developmental psychology","authors":[{"name":"Guher Gorgun","is_ca":true},{"name":"Okan Bulut","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2262113854356192,"gpt":0.4864308640796258,"spread":0.2602194786440065,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01407781,0.000213548,0.0003393969,0.001265616,0.0004059202,0.0003604055,0.0007245248,0.0001067016,0.00004951039],"category_scores_gemma":[0.02325436,0.000160615,0.00006336316,0.00912062,0.00005017805,0.0005403825,0.0002808244,0.0005002024,0.00004263187],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000397177,"about_ca_system_score_gemma":0.000406474,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003796664,"about_ca_topic_score_gemma":0.0004709535,"domain_scores_codex":[0.995218,0.001500358,0.001047131,0.0007287934,0.001030548,0.0004751189],"domain_scores_gemma":[0.9764963,0.02179012,0.0008792846,0.0004760729,0.0002884004,0.00006980212],"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":[0.000006795857,0.000335582,0.467436,0.000007495893,0.000006207496,0.000002244031,0.002794293,0.0009741139,0.0002856888,0.0001810277,0.0004704804,0.5275],"study_design_scores_gemma":[0.0003787735,0.00007571436,0.6898748,0.0001148638,0.000004088748,0.000002052246,0.01880105,0.274985,0.00007480349,0.01492445,0.0006138769,0.0001504761],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8694857,0.00008672995,0.123967,0.0006881715,0.002685391,0.0007779659,0.000006555131,0.0001475382,0.002154988],"genre_scores_gemma":[0.7597377,0.00001795944,0.2393305,0.000209192,0.0002083132,0.0001912189,0.00002429381,0.00001715325,0.0002636486],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5273496,"threshold_uncertainty_score":0.9849732,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3011227050","doi":"10.1186/s40536-020-00080-3","title":"Are large surveys of adult literacy skills as comparable over time as we think?","year":2020,"lang":"en","type":"article","venue":"Large-scale Assessments in Education","topic":"Education Systems and Policy","field":"Social Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true},"ca_institutions":"Institut National de la Recherche Scientifique","funders":"Canadian Institutes of Health Research; Social Sciences and Humanities Research Council of Canada; Fonds de Recherche du Québec-Société et Culture","keywords":"Ceteris paribus; Literacy; Cohort; Adult literacy; Cohort effect; Survey data collection; Psychology; Demographic economics; Demography; Medicine; Pedagogy; Sociology; Economics; Statistics","authors":[{"name":"Samuel Vézina","is_ca":true},{"name":"Alain Bélanger","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.02048207184716733,"gpt":0.4094220301978794,"spread":0.388939958350712,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001309349,0.0001651225,0.000339218,0.0001323952,0.0002458285,0.0001379319,0.0003594802,0.0001555897,0.002385242],"category_scores_gemma":[0.0004287607,0.0001748054,0.00008273478,0.0008497009,0.00005409803,0.0006948002,0.00005194241,0.0002122532,0.0006744289],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002096038,"about_ca_system_score_gemma":0.001365283,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00464516,"about_ca_topic_score_gemma":0.0008774623,"domain_scores_codex":[0.997077,0.001050437,0.0005379693,0.0003542646,0.0005463241,0.0004339334],"domain_scores_gemma":[0.9982982,0.0002094209,0.000556651,0.0002987313,0.0003843767,0.0002526045],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001119141,0.002644965,0.747181,0.0001189508,0.00003434353,9.334114e-7,0.1806115,0.000008426026,0.0001383273,0.02239483,0.0448529,0.002002661],"study_design_scores_gemma":[0.0009574575,0.00007850242,0.5208185,0.0004150042,0.00002615075,9.764053e-7,0.08744726,0.0002669477,0.0002607111,0.003954035,0.3853517,0.0004227372],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9500549,0.0001894498,0.0001344871,0.006366874,0.001249998,0.0005611796,0.00008240377,0.00005418037,0.04130647],"genre_scores_gemma":[0.9828734,0.0001558496,0.0006428299,0.003175296,0.0007096202,0.00005894277,0.0001125606,0.00002232453,0.01224919],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3404988,"threshold_uncertainty_score":0.9985267,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4392361000","doi":"10.1186/s40536-024-00194-y","title":"An engagement-aware predictive model to evaluate problem-solving performance from the study of adult skills' (PIAAC 2012) process data","year":2024,"lang":"en","type":"article","venue":"Large-scale Assessments in Education","topic":"Online Learning and Analytics","field":"Computer Science","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":"","keywords":"Task (project management); Scale (ratio); Process (computing); Learning analytics; Computer science; Test (biology); Analytics; Psychology; Artificial intelligence; Data science","authors":[{"name":"Jinnie Shin","is_ca":false},{"name":"Bowen Wang","is_ca":false},{"name":"Wallace Nascimento Pinto","is_ca":false},{"name":"Mark J. Gierl","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.02761815327562231,"gpt":0.3969492216172661,"spread":0.3693310683416438,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001158872,0.0001650356,0.0001661493,0.0001374282,0.000165465,0.0002460766,0.00178163,0.00004685417,0.000008685558],"category_scores_gemma":[0.00005151491,0.0001277997,0.00001906192,0.0007377678,0.00001517844,0.001898885,0.0004461366,0.0003927723,0.000009360037],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009698761,"about_ca_system_score_gemma":0.0007924633,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001255018,"about_ca_topic_score_gemma":0.0002109537,"domain_scores_codex":[0.9977825,0.0002679602,0.0003803789,0.0007231162,0.0005759831,0.0002700163],"domain_scores_gemma":[0.9981644,0.0001067963,0.0001229663,0.00127415,0.0002600917,0.00007155884],"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":[0.00002845836,0.0119956,0.3208501,0.0003288537,0.0002119502,0.000002241656,0.1481641,0.3835073,0.0001029253,0.0006849325,0.002966619,0.1311569],"study_design_scores_gemma":[0.0002283154,0.0002441025,0.04947307,0.0003608333,0.00004781389,3.133714e-7,0.01018419,0.9381561,0.00002029974,0.001068809,0.00008142804,0.0001347107],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8527056,0.00006749991,0.14421,0.001468631,0.0005153617,0.0007110896,0.00006337158,0.00009212965,0.0001663468],"genre_scores_gemma":[0.9721726,0.00003178727,0.02697886,0.0001701772,0.0001480263,0.0001152834,0.0001679701,0.00001820146,0.0001970595],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5546488,"threshold_uncertainty_score":0.5211522,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}