{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":14,"total_is_capped":false,"direct_labels_cover":0,"predictions_cover":14,"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":"9f297251ca0a","filters":{"venue":"Educational Evaluation and Policy Analysis"}},"results":[{"id":"W2130850661","doi":"10.3102/01623737027003205","title":"Effects of Kindergarten Retention Policy on Children’s Cognitive Growth in Reading and Mathematics","year":2005,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"Early Childhood Education and Development","field":"Social Sciences","cited_by":277,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Grade retention; Psychology; Propensity score matching; Multilevel model; Limiting; Promotion (chess); Academic achievement; Developmental psychology; Homogeneous; Harm; Reading (process); Mathematics education; Longitudinal study; Cohort; Social psychology; Political science; Mathematics; Statistics","authors":[{"name":"Guanglei Hong","is_ca":true},{"name":"Stephen W. Raudenbush","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.02019279729219308,"gpt":0.3748215988611014,"spread":0.3546288015689083,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009345721,0.0003783884,0.0005315578,0.0006394537,0.000645481,0.001271764,0.001123333,0.001386067,0.005791155],"category_scores_gemma":[0.03450687,0.0002228269,0.001264426,0.0005700415,0.0009828751,0.000855493,0.001450511,0.001743127,0.0003524568],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002691641,"about_ca_system_score_gemma":0.002761843,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0243709,"about_ca_topic_score_gemma":0.03079524,"domain_scores_codex":[0.99164,0.004603877,0.0003837484,0.0006202431,0.0009343594,0.00181784],"domain_scores_gemma":[0.9527479,0.02862362,0.0113795,0.002433895,0.001531814,0.003283339],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.01624548,0.005862192,0.8548398,0.0002605527,0.001860224,0.000529443,0.001007885,0.01634198,0.002112429,0.009576143,0.003534756,0.08782911],"study_design_scores_gemma":[0.0003890938,0.006700874,0.9776378,0.0001322843,0.0008097734,0.0000841529,0.001346112,0.006247287,0.002680264,0.002064398,0.001871486,0.00003655179],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9951586,0.0005697705,0.0003523669,0.001293424,0.000033337,0.00003182413,0.0002197082,0.00003657389,0.002304386],"genre_scores_gemma":[0.9981524,0.0002069735,0.0002129657,0.0001779921,0.00002100305,0.00003492048,0.0001191334,0.000004634525,0.001069859],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0243709,"threshold_uncertainty_score":0.04942548,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3008309614","doi":"10.3102/0162373720906217","title":"Teacher Coaching in a Simulated Environment","year":2020,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"Teacher Education and Leadership Studies","field":"Social Sciences","cited_by":181,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Impact","funders":"National Academy of Education; University of Virginia; Spencer Foundation","keywords":"Coaching; Psychology; Perception; Mathematics education; Medical education; Teacher education; Pedagogy; Applied psychology; Medicine","authors":[{"name":"Julie Cohen","is_ca":false},{"name":"Vivian C. Wong","is_ca":false},{"name":"Anandita Krishnamachari","is_ca":false},{"name":"Rebekah Berlin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.213634107861544,"gpt":0.4796332351777997,"spread":0.2659991273162557,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001439602,0.0003052593,0.0003386776,0.0002902538,0.0006915922,0.0005938126,0.0005248552,0.0006708934,0.002599673],"category_scores_gemma":[0.007190526,0.0002680095,0.0003088529,0.0001920681,0.0005520175,0.0003834477,0.001013811,0.0005670597,0.0004769219],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005211629,"about_ca_system_score_gemma":0.001443174,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001433018,"about_ca_topic_score_gemma":0.002331004,"domain_scores_codex":[0.9965289,0.002491924,0.00009684606,0.0002717856,0.0003507466,0.0002597983],"domain_scores_gemma":[0.9956779,0.002399766,0.0005061206,0.0005936616,0.0002409872,0.000581652],"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":[0.03720787,0.09673452,0.1669813,0.0009961756,0.0004525252,0.001045378,0.01478391,0.1857369,0.06300106,0.004773986,0.003287816,0.4249985],"study_design_scores_gemma":[0.01178811,0.2813003,0.3161496,0.0006579828,0.0005781832,0.001955464,0.01351157,0.2220379,0.09712928,0.005829086,0.04867025,0.0003922612],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9926654,0.00002289468,0.004952177,0.00006715069,0.0000117106,0.0002002962,0.00003019966,0.00005703485,0.001993069],"genre_scores_gemma":[0.9949564,0.00004386033,0.003692398,0.00002742635,0.000005850538,0.0003593938,0.00005223875,0.000006260272,0.0008561254],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.002599673,"threshold_uncertainty_score":0.008696795,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2156855199","doi":"10.3102/01623737022004357","title":"Power and Politics in the Adoption of School Reform Models","year":2000,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"Educational Assessment and Improvement","field":"Decision Sciences","cited_by":174,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Education reform; Power (physics); Perspective (graphical); Politics; Process (computing); Public administration; Sustainability; Set (abstract data type); Political science; Public relations; Economic growth; Economics; Higher education","authors":[{"name":"Amanda Datnow","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1507326485546578,"gpt":0.4970251549872353,"spread":0.3462925064325775,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07002693,0.000376814,0.000630178,0.003787206,0.01262676,0.01654815,0.001449938,0.004016387,0.003496798],"category_scores_gemma":[0.09801622,0.0008000367,0.00076829,0.003242868,0.05412749,0.01202507,0.008708885,0.005678686,0.0003318735],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01254067,"about_ca_system_score_gemma":0.00920659,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006338988,"about_ca_topic_score_gemma":0.007360427,"domain_scores_codex":[0.8915767,0.08318547,0.001969084,0.004531371,0.0112259,0.007511373],"domain_scores_gemma":[0.9104323,0.06912699,0.006537407,0.007216725,0.004185674,0.002500847],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"qualitative","study_design_scores_codex":[0.0000576151,0.00006388123,0.005142997,0.00003814086,0.00002080918,0.0001851291,0.04398913,0.001924024,0.0001990691,0.9354656,0.0005697942,0.01234375],"study_design_scores_gemma":[0.0001324753,0.0001826888,0.008328092,0.0002701736,0.00004822079,0.0001960462,0.06864978,0.007591295,0.0009754474,0.8537781,0.05976505,0.00008261079],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5752639,0.001899393,0.04827517,0.03957317,0.0002063845,0.0003346324,0.00003931175,0.00007198798,0.334336],"genre_scores_gemma":[0.9970085,0.0001304555,0.001150416,0.0002979616,0.00002820824,0.00005223693,0.00000416608,0.00001134794,0.001316772],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07002693,"threshold_uncertainty_score":0.3703422,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2124978570","doi":"10.3102/0162373707309073","title":"Early-Grade Retention and Children’s Reading and Math Learning in Elementary Years","year":2007,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"Early Childhood Education and Development","field":"Social Sciences","cited_by":99,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Grade retention; Psychology; Reading (process); Retention rate; Mathematics education; Early childhood; Intervention (counseling); Developmental psychology; Academic achievement; Computer science","authors":[{"name":"Guanglei Hong","is_ca":true},{"name":"Bing Yu","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.02636358941870153,"gpt":0.382660152524002,"spread":0.3562965631053004,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002104271,0.0002158193,0.0004943275,0.0009594502,0.0007418306,0.001133839,0.0007185733,0.0006194994,0.002550444],"category_scores_gemma":[0.008703552,0.0001614537,0.0005241547,0.0006735441,0.0005963788,0.000672209,0.001119254,0.001072668,0.0003570763],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001523535,"about_ca_system_score_gemma":0.001958682,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03732921,"about_ca_topic_score_gemma":0.07455914,"domain_scores_codex":[0.9983645,0.0003716087,0.00008318997,0.0001291803,0.0003694493,0.0006821274],"domain_scores_gemma":[0.9928894,0.001340244,0.00283644,0.0003910871,0.0007457184,0.001797143],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000679583,0.002064277,0.9480659,0.0001129803,0.0001251444,0.0003200187,0.003141521,0.0004599061,0.001079315,0.001184843,0.0007491029,0.04201749],"study_design_scores_gemma":[0.00001412619,0.0005190185,0.9974213,0.00003591808,0.00003463345,0.00003902699,0.0008418362,0.00007873848,0.0003853979,0.0001019784,0.0005221823,0.000005919917],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9983647,0.0002313177,0.00003174589,0.0001623734,0.000004934292,0.00000717285,0.00007070316,0.000004929743,0.001122023],"genre_scores_gemma":[0.9990833,0.0001358509,0.00004473642,0.00002859611,0.00000665035,0.000009122727,0.00008990963,0.0000017146,0.000600039],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03732921,"threshold_uncertainty_score":0.07422382,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2144062847","doi":"10.3102/01623737024003201","title":"Do Low-Achieving Students Benefit More from Small Classes? Evidence from the Tennessee Class Size Experiment","year":2002,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"School Choice and Performance","field":"Social Sciences","cited_by":96,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Class size; Differential effects; Class (philosophy); Mathematics education; Academic achievement; Randomized experiment; Psychology; Reading (process); Differential (mechanical device); Student achievement; Scale (ratio); Quarter (Canadian coin); Mathematics; Statistics; Computer science; Physics; Geography; Political science","authors":[{"name":"Barbara Nye","is_ca":false},{"name":"Larry V. Hedges","is_ca":false},{"name":"Spyros Konstantopoulos","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1088547008286126,"gpt":0.4471015832571122,"spread":0.3382468824284997,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006134393,0.0003073455,0.0005833994,0.0004417527,0.0007762135,0.000917598,0.0006253648,0.0009631999,0.00662087],"category_scores_gemma":[0.0151731,0.0002063853,0.000402642,0.0004569355,0.001360226,0.0008095966,0.0007983164,0.001030253,0.0004482079],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007276824,"about_ca_system_score_gemma":0.001012592,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01684054,"about_ca_topic_score_gemma":0.02535941,"domain_scores_codex":[0.9968713,0.001996562,0.00009843424,0.0003809338,0.0003708652,0.0002819361],"domain_scores_gemma":[0.9793115,0.009989089,0.005889975,0.001919487,0.0006382299,0.002251737],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.05879787,0.01330478,0.7599488,0.0004089286,0.001643381,0.0004901119,0.003852191,0.001220878,0.02742796,0.01028966,0.008169155,0.1144463],"study_design_scores_gemma":[0.001561205,0.00360437,0.9839946,0.00003815668,0.0002603054,0.00004509285,0.0007062461,0.0006714426,0.002160277,0.003214041,0.003705913,0.00003839218],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9953337,0.0001712823,0.0002445278,0.001093411,0.0000143356,0.00002517229,0.0001629542,0.00001272367,0.002941941],"genre_scores_gemma":[0.9976825,0.0001194795,0.0002149698,0.0004827484,0.00001449717,0.00005247002,0.0001270574,0.000005404474,0.00130092],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01684054,"threshold_uncertainty_score":0.03348505,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1980838979","doi":"10.3102/01623737027001001","title":"Success for All: First-Year Results From the National Randomized Field Trial","year":2005,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"School Choice and Performance","field":"Social Sciences","cited_by":84,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Randomized experiment; Quarter (Canadian coin); Test (biology); Psychology; Randomized controlled trial; Multilevel model; Control (management); Treatment and control groups; Reading (process); Mathematics education; Statistics; Medicine; Mathematics; Computer science; Political science; Geography","authors":[{"name":"Geoffrey D. Borman","is_ca":false},{"name":"Robert E. Slavin","is_ca":false},{"name":"Alan Cheung","is_ca":false},{"name":"Anne Chamberlain","is_ca":false},{"name":"Nancy A. Madden","is_ca":false},{"name":"Bette Chambers","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.09148252164579562,"gpt":0.4777966616051301,"spread":0.3863141399593344,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06038212,0.001616487,0.005239245,0.0007287365,0.001035736,0.002021582,0.001533099,0.004055249,0.007281774],"category_scores_gemma":[0.08964807,0.0006732393,0.003287322,0.0007445031,0.002152862,0.003940534,0.001313666,0.002997088,0.0008735263],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002266148,"about_ca_system_score_gemma":0.005275934,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004120848,"about_ca_topic_score_gemma":0.005516115,"domain_scores_codex":[0.9221931,0.06749941,0.002829418,0.002435351,0.002694634,0.002348178],"domain_scores_gemma":[0.9149394,0.05801324,0.01138565,0.006492062,0.004574348,0.004595287],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","study_design_scores_codex":[0.8965062,0.02420082,0.004051873,0.001713895,0.004039147,0.0001242363,0.0005262081,0.001363262,0.0008047683,0.001538168,0.004533382,0.06059806],"study_design_scores_gemma":[0.5942767,0.3680224,0.01725183,0.0006433667,0.006819093,0.000170709,0.0004615657,0.002574908,0.002227514,0.001857881,0.005549598,0.0001444111],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9411751,0.007828031,0.007619614,0.003494604,0.001840008,0.02427666,0.004145139,0.0004900869,0.00913085],"genre_scores_gemma":[0.9711731,0.002114063,0.006105206,0.001418777,0.0005451731,0.01177762,0.001751316,0.00008865775,0.005026002],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.06038212,"threshold_uncertainty_score":0.3193349,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2153308398","doi":"10.3102/01623737023003251","title":"Class Size and Eighth-Grade Math Achievement in the United States and Abroad","year":2001,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"School Choice and Performance","field":"Social Sciences","cited_by":82,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Class size; Mathematics education; Curriculum; Class (philosophy); Student achievement; Academic achievement; Math education; Psychology; Pedagogy; Computer science","authors":[{"name":"Suet‐ling Pong","is_ca":false},{"name":"Aaron M. Pallas","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.05050335955880637,"gpt":0.4362833388378362,"spread":0.3857799792790298,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0004878906,0.0001769117,0.0002131737,0.001143883,0.000420248,0.0006404073,0.0003273917,0.0002575765,0.00364019],"category_scores_gemma":[0.002139427,0.0001126361,0.0002050632,0.0008530233,0.0003128191,0.0003420152,0.00047332,0.0005027963,0.0004666049],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004502083,"about_ca_system_score_gemma":0.0002744575,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03861517,"about_ca_topic_score_gemma":0.06389808,"domain_scores_codex":[0.9997556,0.00006471726,0.00002371436,0.00004332743,0.00004736465,0.00006535462],"domain_scores_gemma":[0.996577,0.000646204,0.001497861,0.0001538127,0.0002833181,0.0008417583],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00006447223,0.00006191058,0.9973151,0.00000507286,0.00002438942,0.00003853611,0.0003144921,0.00007530268,0.00009721174,0.0001225676,0.0003267236,0.0015543],"study_design_scores_gemma":[0.000002571535,0.00002966427,0.9993312,0.000002870917,0.000004869224,0.00001500973,0.0003038769,0.00007199717,0.00003582496,0.00001597331,0.0001842546,0.000001892211],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9984764,0.00002858332,0.00001924862,0.00004131618,0.000002971717,0.000002511747,0.0004799304,0.000003693938,0.0009453077],"genre_scores_gemma":[0.9986425,0.00003019355,0.00002777489,0.0000110572,0.000003811074,0.000006104907,0.0007238382,0.000002094012,0.0005525967],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03861517,"threshold_uncertainty_score":0.0767808,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2146700645","doi":"10.3102/0162373708328259","title":"Reading Instruction Time and Homogeneous Grouping in Kindergarten: An Application of Marginal Mean Weighting Through Stratification","year":2008,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"School Choice and Performance","field":"Social Sciences","cited_by":58,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Reading (process); Psychology; Homogeneous; Weighting; Mathematics education; Multilevel model; Reading comprehension; Longitudinal study; Set (abstract data type); Computer science; Statistics; Linguistics; Mathematics","authors":[{"name":"Guanglei Hong","is_ca":true},{"name":"Yihua Hong","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.04578867278231784,"gpt":0.3879297195275477,"spread":0.3421410467452298,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04612523,0.0007398627,0.001835551,0.002198159,0.001346407,0.001486956,0.001864211,0.0009084971,0.002909369],"category_scores_gemma":[0.1378046,0.0007986656,0.002755248,0.003153832,0.00355931,0.002657478,0.003985157,0.002489317,0.0001554381],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002478907,"about_ca_system_score_gemma":0.00289924,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01505088,"about_ca_topic_score_gemma":0.01058203,"domain_scores_codex":[0.9641622,0.02817225,0.001050726,0.002995072,0.002609374,0.001010342],"domain_scores_gemma":[0.8547169,0.1214027,0.006912624,0.01293106,0.003128009,0.0009086186],"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":[0.001949008,0.0008710944,0.2277345,0.0005443429,0.003132662,0.0004536364,0.008948515,0.03739807,0.002196159,0.3396659,0.00250743,0.3745986],"study_design_scores_gemma":[0.0003415827,0.002451586,0.1459572,0.000313424,0.002118122,0.0003172055,0.002780394,0.3998668,0.004870671,0.4287201,0.01187447,0.00038845],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2670504,0.0007825299,0.7264994,0.0006007562,0.0001329114,0.0005389336,0.0002548785,0.0002891307,0.00385103],"genre_scores_gemma":[0.7623647,0.0002938814,0.2350521,0.0001236727,0.0000713477,0.0009054888,0.000151753,0.00007927344,0.0009578184],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.04612523,"threshold_uncertainty_score":0.2439364,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1987197661","doi":"10.3102/0162373711424206","title":"Differential Effects of Literacy Instruction Time and Homogeneous Ability Grouping in Kindergarten Classrooms","year":2011,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"School Choice and Performance","field":"Social Sciences","cited_by":58,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Psychology; Homogeneous; Literacy; Mathematics education; Developmental psychology; Early literacy; Cognition; Cognitive development; Differential effects; Longitudinal study; Early childhood; Pedagogy; Statistics","authors":[{"name":"Guanglei Hong","is_ca":false},{"name":"Carl Corter","is_ca":true},{"name":"Yihua Hong","is_ca":true},{"name":"Janette Pelletier","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.02560448217574205,"gpt":0.3647464052251619,"spread":0.3391419230494199,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0008270686,0.0006309271,0.0007294081,0.001317135,0.0006563616,0.001700113,0.0008760785,0.0005137148,0.003865096],"category_scores_gemma":[0.006161953,0.0003751459,0.0008209047,0.0008580588,0.001048859,0.0006386708,0.002454622,0.0008717261,0.0003123051],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001274449,"about_ca_system_score_gemma":0.001208715,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01097906,"about_ca_topic_score_gemma":0.02214907,"domain_scores_codex":[0.9977163,0.0002880447,0.0001848007,0.0005651789,0.0005471538,0.000698542],"domain_scores_gemma":[0.992806,0.002683521,0.002171734,0.0004520325,0.0003509432,0.001535839],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003268637,0.002294525,0.9285989,0.0001818692,0.0003805646,0.001488752,0.007230959,0.0004400298,0.02520528,0.0006507494,0.0002773561,0.02998235],"study_design_scores_gemma":[0.000009185836,0.0003378476,0.9973305,0.0000168058,0.00004352142,0.0000275027,0.001359685,0.00005781123,0.0006159807,0.00005673671,0.0001375245,0.000006961694],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9991196,0.0001132239,0.00003965783,0.00001690832,0.000003960237,0.000006520229,0.00002233288,0.000008048183,0.0006698561],"genre_scores_gemma":[0.9991558,0.00006269543,0.0001064933,0.0000154093,0.000003708063,0.00001395049,0.00006130029,0.000006379924,0.0005743616],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01097906,"threshold_uncertainty_score":0.02183032,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2791023047","doi":"10.3102/0162373718760218","title":"School Improvement Grants in Ohio: Effects on Student Achievement and School Administration","year":2018,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"School Choice and Performance","field":"Social Sciences","cited_by":47,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Regression discontinuity design; Quarter (Canadian coin); Psychology; Demographic economics; Turnover; Regression analysis; Mathematics education; Economics; Statistics; Geography; Mathematics; Management","authors":[{"name":"Deven Carlson","is_ca":false},{"name":"Stéphane Lavertu","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.03287081499345695,"gpt":0.4489629340119663,"spread":0.4160921190185093,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006017489,0.0002498442,0.000493577,0.001673142,0.001273347,0.002254682,0.0009258673,0.0007634811,0.003674741],"category_scores_gemma":[0.01613489,0.0002844238,0.0008921276,0.002213042,0.001113796,0.001017556,0.002318818,0.001768213,0.0002252335],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002861711,"about_ca_system_score_gemma":0.003331955,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.08012046,"about_ca_topic_score_gemma":0.08286521,"domain_scores_codex":[0.9934704,0.003414367,0.0002296428,0.0004924152,0.0009855529,0.001407603],"domain_scores_gemma":[0.9776793,0.008843729,0.006787289,0.001176286,0.001731057,0.003782374],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0008891025,0.002311578,0.9697396,0.0000349646,0.0002455282,0.0000959259,0.00075373,0.001884785,0.0002787486,0.001312679,0.001955874,0.02049751],"study_design_scores_gemma":[0.00005129337,0.0005226852,0.9950691,0.00001580897,0.0001018508,0.00001818585,0.001394422,0.001352695,0.0002373838,0.0002233625,0.001000769,0.00001238033],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9983695,0.00007002821,0.0000956693,0.0002743227,0.00001432877,0.00001220993,0.0002247422,0.00001162533,0.00092757],"genre_scores_gemma":[0.9990338,0.00004301993,0.00005925517,0.00003721648,0.000009165719,0.00002249051,0.000195473,0.0000026243,0.0005968575],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08012046,"threshold_uncertainty_score":0.1593082,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1977534634","doi":"10.3102/0162373711398125","title":"Who Benefits From Kindergarten? Evidence From the Introduction of State Subsidization","year":2011,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"Early Childhood Education and Development","field":"Social 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 Toronto","funders":"","keywords":"Socioeconomic status; Subsidy; Immigration; Demographic economics; Revenue; Psychology; State (computer science); Mathematics education; Economic growth; Economics; Demography; Political science; Sociology; Population; Computer science","authors":[{"name":"Elizabeth Dhuey","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.08998701385613952,"gpt":0.3771080661541513,"spread":0.2871210522980118,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01195005,0.0002637073,0.0007717939,0.001847557,0.0009231494,0.002726009,0.001741315,0.001750593,0.01500785],"category_scores_gemma":[0.06080215,0.000436748,0.001110341,0.003230509,0.002347528,0.002058086,0.003267763,0.003337132,0.0005537376],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002725863,"about_ca_system_score_gemma":0.005168403,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0599541,"about_ca_topic_score_gemma":0.07820722,"domain_scores_codex":[0.9895067,0.005100316,0.0007626755,0.001070744,0.001880794,0.001678715],"domain_scores_gemma":[0.8920079,0.07274365,0.02412323,0.002707226,0.005074448,0.00334365],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003089815,0.001036672,0.7960799,0.003539101,0.002010856,0.000480375,0.002990382,0.0005402138,0.0001553118,0.01835962,0.02141076,0.150307],"study_design_scores_gemma":[0.0004727803,0.0004113553,0.9477085,0.006331111,0.002196513,0.0002516793,0.006007126,0.0002781191,0.0002910802,0.002724762,0.0332888,0.00003805041],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7253287,0.1126096,0.0008172876,0.0708279,0.0007911638,0.0002290465,0.005320992,0.00003685566,0.08403845],"genre_scores_gemma":[0.9589731,0.03031562,0.0002750813,0.007101901,0.0003768954,0.00007643498,0.001060079,0.00001419255,0.00180679],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0599541,"threshold_uncertainty_score":0.1192102,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2804370236","doi":"10.3102/0162373718782648","title":"Course Choice and Achievement Effects of a System-Wide Vocational Education and Training Voucher Scheme for Young People","year":2018,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"Education Systems and Policy","field":"Social Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"Queen's University; National Centre for Vocational Education Research; Queen's University Belfast","keywords":"Voucher; Vocational education; Equity (law); School choice; Academic achievement; Training (meteorology); Quality (philosophy); Public economics; Psychology; Mathematics education; Economic growth; Medical education; Pedagogy; Political science; Business; Economics; Accounting","authors":[{"name":"Duncan McVicar","is_ca":false},{"name":"Cain Polidano","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.04401642896527869,"gpt":0.4417211995699809,"spread":0.3977047706047022,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008601577,0.0003587284,0.0007699824,0.0006985059,0.0009866933,0.00182392,0.001256541,0.001465479,0.01567046],"category_scores_gemma":[0.02214515,0.0002667089,0.001075005,0.0008884163,0.001945475,0.001311729,0.002213142,0.001995343,0.0006787724],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006102841,"about_ca_system_score_gemma":0.003623005,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.032858,"about_ca_topic_score_gemma":0.0217399,"domain_scores_codex":[0.9936197,0.003733397,0.0002112833,0.0004561133,0.0005906073,0.001388778],"domain_scores_gemma":[0.9822416,0.009460242,0.00305995,0.001005472,0.0009950943,0.003237666],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.1408151,0.0953505,0.3578343,0.0007683123,0.001721452,0.0007289766,0.00358071,0.05555213,0.006905379,0.09512216,0.008189783,0.2334312],"study_design_scores_gemma":[0.0131968,0.06752162,0.8232177,0.0002218784,0.001948599,0.00009540669,0.004457068,0.05169461,0.007045531,0.0186522,0.01178528,0.0001632938],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.991552,0.0001080248,0.0003149084,0.0005960726,0.00005543403,0.0001582838,0.0001673704,0.00001567604,0.007032339],"genre_scores_gemma":[0.9956762,0.00003745811,0.0001678463,0.0001055163,0.00001675569,0.00005746878,0.00005317661,0.000003161045,0.003882528],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.032858,"threshold_uncertainty_score":0.06533355,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4390877660","doi":"10.3102/01623737231218735","title":"Beyond Prescriptive Reforms: An Examination of North Carolina’s Flexible School Restart Program","year":2024,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"School Choice and Performance","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Autonomy; Flexibility (engineering); Principal (computer security); Quarter (Canadian coin); School choice; Mathematics education; South carolina; Psychology; Pedagogy; Public administration; Sociology; Political science; Economics; Management; Computer science; Law; History","authors":[{"name":"Lam Pham","is_ca":false},{"name":"Gage F. Matthews","is_ca":false},{"name":"Timothy A. Drake","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.05610033606833933,"gpt":0.4434867261868464,"spread":0.3873863901185071,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01134418,0.0002816302,0.0006487787,0.002091363,0.004068503,0.003838946,0.003378499,0.001369524,0.003902109],"category_scores_gemma":[0.03281372,0.0002782962,0.0007339935,0.003472056,0.00217818,0.002115438,0.003779605,0.003235268,0.0002791827],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.02534153,"about_ca_system_score_gemma":0.03476017,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.5157947,"about_ca_topic_score_gemma":0.693387,"domain_scores_codex":[0.986878,0.00492812,0.0002933908,0.000839769,0.004072329,0.002988422],"domain_scores_gemma":[0.9437372,0.02711708,0.009185608,0.002801137,0.01002806,0.007130953],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001241481,0.003881023,0.6943843,0.0005581112,0.0005290398,0.001532593,0.03591042,0.01645435,0.002465585,0.03997318,0.0378795,0.1651903],"study_design_scores_gemma":[0.00007322142,0.0007475404,0.9359057,0.0002895643,0.000133063,0.00005679467,0.02828898,0.005048931,0.0006293174,0.001741196,0.02700811,0.00007746404],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.967962,0.0005274668,0.0005720648,0.008815016,0.00004216389,0.0002072737,0.0004295606,0.00004726318,0.02139741],"genre_scores_gemma":[0.9920284,0.000276927,0.0005347296,0.002211719,0.00003991707,0.00019986,0.0004477242,0.00002028726,0.00424028],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5157947,"threshold_uncertainty_score":0.9741138,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4407384694","doi":"10.3102/01623737241311537","title":"Measuring Grading Standards at High Schools: A Methodological Contribution, an Example, and Some Policy Implications","year":2025,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"Higher Education Learning Practices","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true},"ca_institutions":"Wilfrid Laurier University","funders":"Canadian Institutes of Health Research; University of Alberta","keywords":"Grading (engineering); Academic standards; Mathematics education; Policy analysis; Higher education; Econometrics; Psychology; Political science; Economics; Economic growth; Public administration; Engineering","authors":[],"retraction":null,"screen_n_in":null,"score":{"opus":0.2785140261361238,"gpt":0.5528497145082416,"spread":0.2743356883721178,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1796068,0.001322268,0.0009950617,0.005620823,0.006129848,0.005798927,0.005487874,0.003682861,0.001369215],"category_scores_gemma":[0.3328149,0.0008652242,0.001020097,0.01695175,0.009230321,0.003834601,0.004735205,0.003123393,0.0003538526],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01415679,"about_ca_system_score_gemma":0.02172728,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.1647845,"about_ca_topic_score_gemma":0.210035,"domain_scores_codex":[0.6873401,0.2623847,0.01288765,0.008039678,0.02712839,0.002219548],"domain_scores_gemma":[0.5959744,0.2640416,0.0376239,0.03943378,0.06116933,0.001757047],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0003167406,0.001717795,0.5035846,0.002421131,0.000688808,0.0008925075,0.02744992,0.007343292,0.002009349,0.2383397,0.02108605,0.1941501],"study_design_scores_gemma":[0.0006088793,0.001319707,0.432938,0.006716062,0.001288073,0.001764183,0.06831056,0.03182545,0.01047529,0.298472,0.1453036,0.0009781318],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.2619008,0.01260074,0.5255494,0.1308374,0.002842539,0.008584947,0.002845983,0.0002364733,0.05460172],"genre_scores_gemma":[0.6648045,0.00274859,0.310485,0.008142317,0.0008160557,0.007279728,0.0007280841,0.00009947094,0.00489619],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.1796068,"threshold_uncertainty_score":0.9498628,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}