{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":18,"total_is_capped":false,"direct_labels_cover":0,"predictions_cover":18,"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":"83ada4cbf0c3","filters":{"venue":"Studies In Educational Evaluation"}},"results":[{"id":"W2522686050","doi":"10.1016/j.stueduc.2016.08.007","title":"Meta-analysis of faculty's teaching effectiveness: Student evaluation of teaching ratings and student learning are not related","year":2016,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Evaluation of Teaching Practices","field":"Social Sciences","cited_by":584,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Mount Royal University","funders":"","keywords":"Psychology; Set (abstract data type); Meta-analysis; Student achievement; Sample (material); Mathematics education; Artifact (error); Correlation; Academic achievement; Computer science; Mathematics","authors":[{"name":"Bob Uttl","is_ca":true},{"name":"Carmela A. White","is_ca":true},{"name":"Daniela Wong Gonzalez","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4547744883637743,"gpt":0.6028893736354598,"spread":0.1481148852716855,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.09887735,0.004661966,0.01960776,0.005806233,0.001935917,0.006196768,0.004560973,0.004994129,0.003907245],"category_scores_gemma":[0.2658324,0.002864203,0.07199819,0.00536072,0.003117782,0.004019052,0.003437273,0.004587479,0.0005556443],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00371212,"about_ca_system_score_gemma":0.003121431,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009112501,"about_ca_topic_score_gemma":0.01640712,"domain_scores_codex":[0.8366095,0.1067065,0.03048283,0.01319988,0.01124943,0.001751777],"domain_scores_gemma":[0.7019669,0.2393336,0.02390463,0.02377444,0.008627466,0.002393027],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","study_design_scores_codex":[0.003741016,0.0000207989,0.006799675,0.01207503,0.9745392,0.00005652422,0.000132936,0.0001104358,0.0001293513,0.00006449238,0.0002073737,0.002123308],"study_design_scores_gemma":[0.002422841,0.0005270611,0.007319791,0.001700054,0.9867123,0.00008444957,0.00007854356,0.0001860688,0.0001583396,0.0002914832,0.0004880168,0.00003104724],"study_design_candidate":"meta_analysis","study_design_consensus":"meta_analysis","genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.1564968,0.8187205,0.01163206,0.001892864,0.004438356,0.001524181,0.002094174,0.0003873671,0.002813764],"genre_scores_gemma":[0.9614677,0.02909282,0.00475117,0.001426304,0.0006910812,0.0007900942,0.0006814137,0.0002175589,0.0008817526],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9011226,"threshold_uncertainty_score":0.5229197,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2782375796","doi":"10.1016/j.stueduc.2017.12.008","title":"Re-conceptualizing classroom assessment fairness: A systematic meta-ethnography of assessment literature and beyond","year":2018,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":82,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University","funders":"","keywords":"Conceptualization; Accountability; Construct (python library); Educational assessment; Psychology; Ethnography; Standards-based assessment; Process (computing); Pedagogy; Alternative assessment; Mathematics education; Sociology; Computer science; Political science","authors":[{"name":"Amirhossein Rasooli","is_ca":false},{"name":"Hamed Zandi","is_ca":false},{"name":"Christopher DeLuca","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1824586292219303,"gpt":0.5132713437504184,"spread":0.330812714528488,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.277291,0.001611825,0.004838464,0.01287788,0.002855372,0.009569612,0.004006105,0.002162059,0.001645115],"category_scores_gemma":[0.4587441,0.001602621,0.003636483,0.008381753,0.006940768,0.0172155,0.008853901,0.005189211,0.0001038851],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.009211442,"about_ca_system_score_gemma":0.02547554,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01019809,"about_ca_topic_score_gemma":0.01974306,"domain_scores_codex":[0.7444218,0.2075059,0.02722476,0.00859302,0.01034966,0.001904906],"domain_scores_gemma":[0.2614102,0.6871106,0.01856062,0.01928192,0.01277231,0.0008642738],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0007275234,0.0003338827,0.0280406,0.1276552,0.01351476,0.000414186,0.4336383,0.001481132,0.001736386,0.02693461,0.003276421,0.3622469],"study_design_scores_gemma":[0.0005584494,0.0009179476,0.03419867,0.4212689,0.03022719,0.0009804715,0.3814472,0.003945926,0.004883492,0.06634191,0.05475406,0.0004758645],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.3237903,0.5335526,0.1101427,0.01736415,0.001552993,0.006101578,0.00087982,0.0001562077,0.006459626],"genre_scores_gemma":[0.8689556,0.05447753,0.06654858,0.00378761,0.0001951551,0.005161455,0.0003582157,0.0001194692,0.0003963358],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.722709,"threshold_uncertainty_score":0.8912289,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1976592601","doi":"10.1016/j.stueduc.2013.10.006","title":"Towards a culture of inquiry for data use in schools: Breaking down professional learning barriers through intentional interruption","year":2013,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Educational Assessment and Improvement","field":"Decision Sciences","cited_by":78,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Status quo; Context (archaeology); Interrupt; Psychology; Cognition; Professional development; Professional learning community; Social psychology; Pedagogy; Computer science; Political science","authors":[{"name":"Steven T. Katz","is_ca":true},{"name":"Lisa Ain Dack","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6261356046479121,"gpt":0.6152237617204723,"spread":0.01091184292743974,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2388426,0.0007802983,0.001123295,0.003550278,0.01183689,0.03436575,0.005737445,0.006161383,0.001881421],"category_scores_gemma":[0.3569797,0.001761553,0.001157629,0.002012283,0.04201611,0.02226395,0.03110527,0.01694323,0.0007313841],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.008506613,"about_ca_system_score_gemma":0.03320034,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004682065,"about_ca_topic_score_gemma":0.003121349,"domain_scores_codex":[0.6036176,0.3305199,0.01731797,0.01039383,0.03296066,0.005190037],"domain_scores_gemma":[0.3914138,0.4070042,0.03313931,0.08809414,0.06083727,0.01951125],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0001188873,0.0006642709,0.02527545,0.0006117945,0.00008418776,0.0001774399,0.8474998,0.0003147302,0.002600784,0.06031346,0.001993015,0.06034618],"study_design_scores_gemma":[0.0001711409,0.0009560458,0.01980256,0.003613509,0.0002285177,0.001235234,0.7452911,0.004737841,0.008387514,0.1174035,0.09783654,0.0003366186],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.4556724,0.001400309,0.3881127,0.1015556,0.0005802963,0.001400293,0.00008948642,0.001178932,0.05000998],"genre_scores_gemma":[0.9237673,0.0002340558,0.0697734,0.003569662,0.00005111349,0.0007684944,0.00002862226,0.0002050749,0.001602261],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2388426,"threshold_uncertainty_score":0.9386426,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3128610517","doi":"10.1016/j.stueduc.2021.100977","title":"The long-term washback effects of the National Matriculation English Test on college English learning in China: Tertiary student perspectives","year":2021,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":31,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Matriculation; Active listening; Competence (human resources); Psychology; Test (biology); College English; Curriculum; Perception; Mathematics education; China; Medical education; Pedagogy; Medicine; Political science","authors":[{"name":"Hao Zhang","is_ca":true},{"name":"Monique Bournot-Trites","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.03124067406168235,"gpt":0.419786253443572,"spread":0.3885455793818897,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009496037,0.0002442111,0.00071884,0.001113078,0.001486496,0.001189498,0.001014821,0.0009780076,0.00194772],"category_scores_gemma":[0.03168723,0.0001302554,0.0007373074,0.0009764917,0.001075919,0.001061709,0.001447081,0.001189751,0.000204725],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003728223,"about_ca_system_score_gemma":0.005394956,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.04799857,"about_ca_topic_score_gemma":0.05884648,"domain_scores_codex":[0.9964724,0.001296135,0.0002696676,0.0002377039,0.0009606995,0.0007635829],"domain_scores_gemma":[0.977668,0.007243372,0.003860228,0.0009597057,0.004833297,0.00543532],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"qualitative","study_design_scores_codex":[0.0022453,0.003221685,0.9122018,0.00009058748,0.0001366357,0.000434548,0.005854032,0.0002236993,0.0008791875,0.0003170896,0.0009238222,0.07347168],"study_design_scores_gemma":[0.0000483264,0.002295156,0.9930386,0.00004275153,0.0001009703,0.00004911122,0.003224903,0.0002531842,0.0004475462,0.00008496662,0.0003958574,0.00001865558],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9982646,0.0003549914,0.00001883549,0.0004816446,0.00001436872,0.00001383727,0.00003073175,0.000002287056,0.0008186174],"genre_scores_gemma":[0.9995789,0.00008093091,0.000016335,0.00006223969,0.00001257283,0.000007621938,0.00003008168,8.14016e-7,0.000210384],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.04799857,"threshold_uncertainty_score":0.09543836,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2072591409","doi":"10.1016/s0191-491x(00)00011-0","title":"The Brazilian National Evaluation System of Basic Education: Context, process, and impact","year":2000,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"School Choice and Performance","field":"Social Sciences","cited_by":23,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Université de Montréal","funders":"","keywords":"Context (archaeology); Process (computing); Process management; Mathematics education; Computer science; Psychology; Business; Geography","authors":[{"name":"Manuel Crespo","is_ca":true},{"name":"JoséFrancisco Soares","is_ca":false},{"name":"Alberto de Mello e Souza","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.08028795285571905,"gpt":0.492372740274014,"spread":0.4120847874182949,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1265112,0.0005402304,0.001626202,0.01043102,0.004089256,0.006270562,0.001820797,0.001667788,0.003009375],"category_scores_gemma":[0.1532077,0.000543949,0.0008708751,0.01123512,0.00397084,0.002853333,0.004508832,0.001445833,0.0004114779],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.03396073,"about_ca_system_score_gemma":0.1187134,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.1547793,"about_ca_topic_score_gemma":0.2098379,"domain_scores_codex":[0.905028,0.05855149,0.008763349,0.00307174,0.02074761,0.003837816],"domain_scores_gemma":[0.7653102,0.102266,0.01293609,0.01372906,0.09477784,0.01098087],"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":[0.0007408187,0.001069117,0.2833598,0.004991592,0.0002800091,0.0002052717,0.02007442,0.002464334,0.002156213,0.118719,0.0359582,0.5299811],"study_design_scores_gemma":[0.0001452611,0.0007197357,0.797285,0.004600459,0.0004271383,0.0002541663,0.01273455,0.004137269,0.002217122,0.01433021,0.1629544,0.0001946703],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5907184,0.02964238,0.02914421,0.05950306,0.00141907,0.004488849,0.008014493,0.0007250669,0.2763444],"genre_scores_gemma":[0.9611255,0.002641059,0.02483892,0.001594448,0.0001720624,0.001362812,0.001773429,0.000127485,0.006364251],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1547793,"threshold_uncertainty_score":0.6690632,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2054415360","doi":"10.1016/j.stueduc.2009.12.005","title":"Evaluation for learning: A cross-case analysis of evaluator strategies","year":2009,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":19,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta; Queen's University","funders":"","keywords":"Process (computing); Citizen journalism; Participatory evaluation; Reflection (computer programming); Program evaluation; Knowledge management; Computer science; Process management; Psychology; Sociology; Political science; Engineering","authors":[{"name":"Christopher DeLuca","is_ca":true},{"name":"Cheryl Poth","is_ca":true},{"name":"Michelle Searle","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5328970254040822,"gpt":0.6892803130688502,"spread":0.1563832876647679,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07237728,0.0006741517,0.0007788126,0.008694193,0.001867255,0.00444684,0.001839485,0.001677882,0.004266267],"category_scores_gemma":[0.2287419,0.0003169738,0.001048429,0.003317113,0.001496888,0.006643938,0.003491953,0.0009560552,0.0004695542],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.005066568,"about_ca_system_score_gemma":0.004261845,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002860046,"about_ca_topic_score_gemma":0.003915768,"domain_scores_codex":[0.9202376,0.06306323,0.004351311,0.001626267,0.008702668,0.002018993],"domain_scores_gemma":[0.5661487,0.3819128,0.009329598,0.008887236,0.0318068,0.001914934],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"qualitative","study_design_scores_codex":[0.004574259,0.003982893,0.3122345,0.003047408,0.001196262,0.001898617,0.1124779,0.004622136,0.004304803,0.02537657,0.002528973,0.5237557],"study_design_scores_gemma":[0.00109065,0.01333143,0.5261942,0.00706409,0.004737012,0.00645377,0.2626217,0.07088315,0.02890502,0.04186464,0.03619789,0.0006564897],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9527079,0.002427216,0.02962194,0.0004079887,0.00002575481,0.001524993,0.0001421793,0.00006327751,0.01307867],"genre_scores_gemma":[0.9855512,0.0004081047,0.01245828,0.00006874457,0.00000509118,0.000465273,0.0001156436,0.00003277055,0.0008948744],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07237728,"threshold_uncertainty_score":0.3827722,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4385393971","doi":"10.1016/j.stueduc.2023.101290","title":"Evaluating student engagement and experiential learning in global classrooms: A qualitative case study","year":2023,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Education and Critical Thinking Development","field":"Social Sciences","cited_by":18,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Western University; University of Toronto","funders":"","keywords":"Experiential learning; Student engagement; Psychology; Interactivity; Pedagogy; Mathematics education; Experiential education; Educational technology; Qualitative research; Sociology; Computer science; Multimedia; Social science","authors":[{"name":"Selina Mae Quibrantar","is_ca":true},{"name":"Obidimma Ezezika","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4136159316534047,"gpt":0.6442659677685991,"spread":0.2306500361151944,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02507923,0.0006983465,0.001052707,0.001972383,0.01237698,0.007092142,0.002886346,0.002683048,0.002833475],"category_scores_gemma":[0.03728277,0.000541829,0.0005684569,0.00182434,0.009622476,0.003330367,0.009121965,0.003437418,0.0003739161],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.007475771,"about_ca_system_score_gemma":0.007755067,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005758805,"about_ca_topic_score_gemma":0.01306193,"domain_scores_codex":[0.9747541,0.01920684,0.0005539595,0.00100183,0.002166396,0.002316885],"domain_scores_gemma":[0.9460093,0.04148366,0.002291509,0.001350854,0.004026296,0.004838276],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0001077174,0.0009362479,0.007332212,0.0002237151,0.00001124715,0.001090312,0.968154,0.0001225387,0.001359357,0.002516825,0.0004919263,0.01765381],"study_design_scores_gemma":[0.00001696311,0.0003096625,0.003252577,0.0001829095,0.00001202409,0.0003014271,0.9890219,0.0001818435,0.001200587,0.0006974479,0.004801237,0.00002139909],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9854866,0.0002484995,0.0035865,0.001190118,0.00004506347,0.0005132441,0.00005446713,0.00001971334,0.008855843],"genre_scores_gemma":[0.9950734,0.0001780207,0.001527369,0.0002412448,0.000009918466,0.0003177896,0.0000212076,0.00001555196,0.002615346],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02507923,"threshold_uncertainty_score":0.1326333,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4210694249","doi":"10.1016/j.stueduc.2022.101126","title":"Item wording effects in self-report measures and reading achievement: Does removing careless respondents help?","year":2022,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":17,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"","keywords":"Reading (process); Psychology; Mathematics education; Academic achievement; Achievement test; Standardized test; Linguistics","authors":[{"name":"Hatice Cigdem Bulut","is_ca":false},{"name":"Okan Bulut","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4604167025598831,"gpt":0.5511538330204064,"spread":0.09073713046052334,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.241551,0.001278105,0.002030071,0.002149851,0.00216518,0.002463259,0.003321736,0.003544767,0.003129437],"category_scores_gemma":[0.4629652,0.001459272,0.004516035,0.002532452,0.00386732,0.006360759,0.002083638,0.003431642,0.0006470223],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001394966,"about_ca_system_score_gemma":0.00271786,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004975556,"about_ca_topic_score_gemma":0.01014684,"domain_scores_codex":[0.8102607,0.1362987,0.02551362,0.008747214,0.01752269,0.001657162],"domain_scores_gemma":[0.342504,0.5609976,0.02859274,0.04600734,0.01991794,0.001980381],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002858169,0.001851829,0.695691,0.001948952,0.002746594,0.0002647895,0.0142723,0.0005270544,0.003095735,0.003544669,0.005233288,0.2679655],"study_design_scores_gemma":[0.0006522964,0.002739404,0.9588671,0.002549889,0.003280166,0.0006680186,0.004505038,0.002246629,0.009854341,0.00549429,0.008910364,0.0002324809],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9271633,0.007903477,0.04878245,0.006741693,0.001259785,0.001231604,0.0006427927,0.000289084,0.005985887],"genre_scores_gemma":[0.9501703,0.00105609,0.0415628,0.002758728,0.0002712315,0.001252824,0.0007810126,0.0001726229,0.001974265],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.758449,"threshold_uncertainty_score":0.9353026,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2089211529","doi":"10.1016/j.stueduc.2007.04.002","title":"ANTIBULLYING PROGRAMS: A SURVEY OF EVALUATION ACTIVITIES IN PUBLIC SCHOOLS","year":2007,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Bullying, Victimization, and Aggression","field":"Psychology","cited_by":15,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Ottawa","funders":"Social Sciences and Humanities Research Council of Canada; Ministère de l’Éducation, Gouvernement de l’Ontario","keywords":"Rigour; Psychological intervention; Program evaluation; Medical education; Poison control; Human factors and ergonomics; Suicide prevention; Medicine; Injury prevention; Psychology; Applied psychology; Nursing; Environmental health; Political science","authors":[{"name":"Jennifer Smith","is_ca":true},{"name":"Wendy Ryan","is_ca":true},{"name":"J. Bradley Cousins","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2152842294067903,"gpt":0.4833329445328567,"spread":0.2680487151260665,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005622582,0.0003284734,0.0006021408,0.004612078,0.001808095,0.001438823,0.0006764196,0.001012361,0.001177773],"category_scores_gemma":[0.01403124,0.0005533772,0.0005045646,0.003386246,0.0008746723,0.001302717,0.001905443,0.001316322,0.0002572778],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002527879,"about_ca_system_score_gemma":0.004774762,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02046633,"about_ca_topic_score_gemma":0.04129516,"domain_scores_codex":[0.9941583,0.001935923,0.0008351754,0.0003742195,0.001533243,0.001163179],"domain_scores_gemma":[0.974779,0.004892254,0.00999663,0.0008181747,0.004931081,0.004582759],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000101533,0.0007968787,0.9742491,0.0001175367,0.00003504243,0.0001068218,0.005450182,0.00002693888,0.0004317019,0.00006609768,0.0006232969,0.01799482],"study_design_scores_gemma":[0.000005067406,0.0002066707,0.9883675,0.00006094783,0.00001268101,0.00008939294,0.01025176,0.00004649486,0.0001289991,0.00001293494,0.0008102614,0.000007337504],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9980549,0.0005858287,0.00007687812,0.0001873938,0.000006030611,0.00007765596,0.0001595333,0.000009534005,0.0008422612],"genre_scores_gemma":[0.9978709,0.0008214549,0.000223471,0.0001862139,0.00000820788,0.00009018416,0.0002350674,0.000009472727,0.0005548833],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02046633,"threshold_uncertainty_score":0.04069442,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4402569962","doi":"10.1016/j.stueduc.2024.101403","title":"Students’ perceptions of online peer feedback in process-oriented L2 writing: A qualitative inquiry","year":2024,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":12,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Western University","funders":"","keywords":"Peer feedback; Perception; Mathematics education; Peer evaluation; Process (computing); Qualitative research; Psychology; Computer science; Pedagogy; Higher education; Sociology; Political science","authors":[{"name":"Pelin İrgin","is_ca":true},{"name":"Zeynep Bilki","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.230095691557151,"gpt":0.6111867833582557,"spread":0.3810910918011047,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01500484,0.0004200139,0.0006484585,0.001518649,0.004136097,0.004323782,0.001127469,0.001840895,0.001538015],"category_scores_gemma":[0.02477091,0.0004552161,0.0005583501,0.001189892,0.005079517,0.003100005,0.003544969,0.002169243,0.0002831641],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002236825,"about_ca_system_score_gemma":0.002921471,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00196305,"about_ca_topic_score_gemma":0.002117445,"domain_scores_codex":[0.9872819,0.009311726,0.000518848,0.0005317061,0.001257487,0.001098509],"domain_scores_gemma":[0.9780919,0.01661763,0.001330087,0.0005274905,0.002064944,0.001367994],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00004729984,0.0002121761,0.009103642,0.0001900457,0.000008766821,0.0006879005,0.9774415,0.0001154393,0.001936682,0.0006125229,0.0001763676,0.009467658],"study_design_scores_gemma":[0.00001271387,0.0003214554,0.007881106,0.0001855184,0.00001200874,0.000410901,0.982278,0.0004324356,0.001694788,0.0005163736,0.006219363,0.00003531458],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9955549,0.0001274042,0.001825137,0.0005350473,0.00001548412,0.0001393815,0.00003334557,0.00001349456,0.001755819],"genre_scores_gemma":[0.997503,0.0001593455,0.001045294,0.000158785,0.000006445446,0.0001971091,0.00001842255,0.00000848891,0.0009032363],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01500484,"threshold_uncertainty_score":0.07935417,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3176398386","doi":"10.1016/j.stueduc.2021.101058","title":"The development and psychometric properties of an educational development impact questionnaire","year":2021,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Evaluation of Teaching Practices","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":"McGill University; Alberta Advanced Education; University of Alberta","funders":"","keywords":"Exploratory factor analysis; Context (archaeology); Psychology; Scholarship; Psychometrics; Medical education; Applied psychology; Mathematics education; Knowledge management; Computer science; Medicine; Developmental psychology","authors":[{"name":"Janice Miller‐Young","is_ca":true},{"name":"Luis Fernando Marín Ardila","is_ca":true},{"name":"Cheryl Poth","is_ca":true},{"name":"Luis Francisco Vargas‐Madriz","is_ca":true},{"name":"Jiaying Xiao","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3343641478523127,"gpt":0.5538289167297261,"spread":0.2194647688774134,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.043182,0.0004870663,0.0007027555,0.003720639,0.0006963392,0.00181903,0.001030078,0.0007111988,0.001748833],"category_scores_gemma":[0.09779668,0.000414615,0.001376046,0.002872629,0.0009118128,0.001995589,0.002261348,0.001341637,0.0005299097],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00130317,"about_ca_system_score_gemma":0.002738321,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001302272,"about_ca_topic_score_gemma":0.001458017,"domain_scores_codex":[0.9733478,0.0132157,0.004200137,0.0007843532,0.007679515,0.0007725439],"domain_scores_gemma":[0.889928,0.07327176,0.007679929,0.004035634,0.02265822,0.002426478],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0004672554,0.001638099,0.5691129,0.0005933691,0.0002688963,0.0001312543,0.004390249,0.002439956,0.002200204,0.002831686,0.003194288,0.4127319],"study_design_scores_gemma":[0.0002098237,0.004352064,0.9527038,0.0005000644,0.0001956929,0.000364641,0.00416931,0.01119994,0.003600409,0.00286047,0.01968721,0.0001566963],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"methods","genre_scores_codex":[0.9268235,0.0006167063,0.04962939,0.001180707,0.0002187207,0.005961272,0.001581674,0.0003866286,0.01360135],"genre_scores_gemma":[0.922534,0.0004610147,0.06828327,0.0002658688,0.00005911929,0.006047794,0.001301751,0.00006663323,0.0009805465],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.043182,"threshold_uncertainty_score":0.228371,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4404020856","doi":"10.1016/j.stueduc.2024.101412","title":"Predicting the Mathematics Literacy of Resilient Students from High‐performing Economies: A Machine Learning Approach","year":2024,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Online Learning and Analytics","field":"Computer Science","cited_by":7,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University","funders":"","keywords":"Mathematics education; Literacy; Computer science; Sociology; Pedagogy; Mathematics","authors":[{"name":"Yimei Zhang","is_ca":true},{"name":"Maria Cutumisu","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.05050611502294976,"gpt":0.4089772244966975,"spread":0.3584711094737477,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001716505,0.00060163,0.0004840161,0.002004758,0.0006751562,0.001684555,0.0004344487,0.0005539844,0.001830195],"category_scores_gemma":[0.007184986,0.0002145465,0.001019866,0.001362976,0.0004032831,0.0008441033,0.001570135,0.001612232,0.0004324598],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005474114,"about_ca_system_score_gemma":0.001102345,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00618603,"about_ca_topic_score_gemma":0.007529396,"domain_scores_codex":[0.9993947,0.0002694241,0.0000524969,0.0001140955,0.00008330146,0.00008597443],"domain_scores_gemma":[0.9981738,0.0009318782,0.0003044483,0.0001261095,0.0002701121,0.0001935289],"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":[0.0002568655,0.0007361661,0.8965698,0.0001278199,0.0002619267,0.0001653679,0.00140176,0.00959467,0.000616683,0.00120061,0.001630786,0.08743751],"study_design_scores_gemma":[0.0000628135,0.000754127,0.829231,0.0003136494,0.0002971445,0.0001791615,0.006596399,0.1503724,0.002179219,0.005666405,0.00425266,0.00009508375],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9931301,0.0002329938,0.004519769,0.0004662746,0.00001941266,0.00006172779,0.000402285,0.00004683695,0.001120602],"genre_scores_gemma":[0.9951234,0.0001432572,0.003628975,0.00005135736,0.000009235517,0.00009483479,0.0005479526,0.000004628558,0.0003963776],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.00618603,"threshold_uncertainty_score":0.01230007,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4387762598","doi":"10.1016/j.stueduc.2023.101309","title":"Predicting reading comprehension performance based on student characteristics and item properties","year":2023,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Reading and Literacy Development","field":"Psychology","cited_by":3,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Centre for Advancing Health Outcomes; University of Alberta","funders":"","keywords":"Reading comprehension; Reading (process); Affect (linguistics); Test (biology); Psychology; Multilingualism; Comprehension; Mathematics education; Cognition; Computer science; Cognitive psychology; Linguistics; Pedagogy; Communication","authors":[{"name":"Mehrdad Yousefpoori-Naeim","is_ca":true},{"name":"Okan Bulut","is_ca":true},{"name":"Bin Tan","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1548524743564766,"gpt":0.4384727952419538,"spread":0.2836203208854772,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003282058,0.0006775444,0.0004951226,0.001959161,0.0002577846,0.001556087,0.0003933065,0.001149673,0.002734881],"category_scores_gemma":[0.0320767,0.0001997074,0.0007783767,0.001395342,0.0002971913,0.001486496,0.0006504689,0.0007873503,0.001232336],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002509395,"about_ca_system_score_gemma":0.0004387941,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001284164,"about_ca_topic_score_gemma":0.001565966,"domain_scores_codex":[0.9986997,0.0005976933,0.0001832515,0.0001661987,0.0002485117,0.0001046912],"domain_scores_gemma":[0.9447168,0.04313578,0.005129605,0.0014745,0.003442844,0.002100479],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002299957,0.0002609878,0.9900137,0.00001118103,0.00006664135,0.00003619784,0.0001564916,0.0003936042,0.001214308,0.00002177847,0.0001089278,0.007486061],"study_design_scores_gemma":[0.00002567444,0.001078198,0.9871367,0.00001422156,0.00009657328,0.0001261515,0.000430145,0.008376122,0.00241516,0.0001781157,0.0001107398,0.00001222746],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9991968,0.00002590749,0.0003834307,0.00001614774,0.000002067613,0.000007721726,0.00006841623,0.00001720176,0.0002824282],"genre_scores_gemma":[0.9990234,0.00002134611,0.0004248909,0.00000775893,0.000003783143,0.00001086512,0.0002729582,0.00001022808,0.0002246675],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.003282058,"threshold_uncertainty_score":0.01735735,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1982497397","doi":"10.1016/s0191-491x(02)00014-7","title":"Matching the grade 8 TIMSS item pool to the Ontario curriculum","year":2002,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":2,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Toronto; Lakehead University","funders":"","keywords":"Mathematics education; Curriculum; Matching (statistics); Secondary education; Psychology; Pedagogy; Mathematics; Statistics","authors":[{"name":"Alexandra Lawson","is_ca":true},{"name":"Catherine Bordignon","is_ca":true},{"name":"Philip Nagy","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1656116291663467,"gpt":0.4685884905270746,"spread":0.3029768613607279,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006091932,0.0004552029,0.0008759584,0.00277918,0.00163197,0.0013194,0.001001705,0.0005862346,0.01532464],"category_scores_gemma":[0.02805678,0.0003527284,0.001192516,0.003118737,0.0006877443,0.0005131838,0.001616825,0.0005514121,0.005923534],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.007896907,"about_ca_system_score_gemma":0.0245979,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.4808241,"about_ca_topic_score_gemma":0.7813034,"domain_scores_codex":[0.9945759,0.001079833,0.0008775836,0.0004309253,0.002261057,0.0007746096],"domain_scores_gemma":[0.9823869,0.001877518,0.001344974,0.001608822,0.01178872,0.0009930335],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002099798,0.001468109,0.6317484,0.0006047713,0.0002453816,0.0002201629,0.006314293,0.001601158,0.01074683,0.002586069,0.0780332,0.2643319],"study_design_scores_gemma":[0.0001586651,0.0003837085,0.9559973,0.0001328983,0.00007836178,0.00004161184,0.001492595,0.000821356,0.001849925,0.0003503727,0.03866275,0.00003048803],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8653688,0.0002822176,0.008903453,0.0009755732,0.0002629223,0.01019339,0.01556348,0.0004717539,0.09797835],"genre_scores_gemma":[0.8571397,0.0004773014,0.02360236,0.0005475727,0.00007735432,0.01971358,0.02412329,0.0002740551,0.07404473],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9921031,"threshold_uncertainty_score":0.9560509,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W7104543262","doi":"10.1016/j.stueduc.2025.101530","title":"Stages of teaching expertise from routine to adaptive: A model for advancing teaching effectiveness","year":2025,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Teacher Education and Leadership Studies","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Calgary","funders":"","keywords":"Conceptualization; Perspective (graphical); Focus group; Conceptual framework; Teaching method; Qualitative research; Multimethodology; Resource (disambiguation); Qualitative property; Professional development","authors":[],"retraction":null,"screen_n_in":null,"score":{"opus":0.2695009569417753,"gpt":0.5581773688283811,"spread":0.2886764118866058,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01227469,0.0007132697,0.0004109962,0.00462292,0.002267929,0.004616901,0.001666443,0.001111227,0.001865921],"category_scores_gemma":[0.02364601,0.0004959232,0.0008391036,0.002050952,0.005866442,0.006094378,0.003243729,0.001609688,0.0003002354],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01138942,"about_ca_system_score_gemma":0.01294864,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03144831,"about_ca_topic_score_gemma":0.03314358,"domain_scores_codex":[0.9910252,0.005222564,0.0005022247,0.0006254356,0.001723231,0.0009013834],"domain_scores_gemma":[0.9841068,0.008702374,0.001785976,0.0007774504,0.003329714,0.001297678],"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":[0.0006284483,0.001579658,0.3673045,0.001735128,0.0001297733,0.000584371,0.1430151,0.00994901,0.003880269,0.1923387,0.002434741,0.2764203],"study_design_scores_gemma":[0.0003871083,0.006813926,0.5078673,0.00214938,0.000530772,0.001391515,0.1518037,0.08813436,0.01146299,0.1863373,0.04279386,0.0003278104],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7997546,0.001531158,0.1096424,0.005877857,0.00007686522,0.002590698,0.0003553623,0.0003895953,0.07978158],"genre_scores_gemma":[0.9621238,0.000345616,0.0358473,0.00009965812,0.000007088312,0.0004047425,0.00008091175,0.00001454832,0.001076389],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03144831,"threshold_uncertainty_score":0.08263642,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1838440795","doi":"10.1016/j.stueduc.2015.09.002","title":"Evaluating intake variables for a teacher education programme: Improving student success and process efficiency","year":2015,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Teacher Education and Leadership Studies","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of the Fraser Valley","funders":"","keywords":"Practicum; Psychology; Medical education; Focus group; Process (computing); Program evaluation; Mathematics education; Pedagogy; Medicine; Computer science; Political science; Sociology","authors":[{"name":"Sheryl MacMath","is_ca":true},{"name":"Barbara Salingré","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.433369157431125,"gpt":0.5880200743970577,"spread":0.1546509169659327,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02281243,0.001145608,0.002119662,0.001846494,0.001348534,0.003148105,0.001369392,0.001407605,0.003083957],"category_scores_gemma":[0.09686782,0.0005061709,0.001902958,0.00255609,0.0007339064,0.002380538,0.002926305,0.00168838,0.0006925259],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001561319,"about_ca_system_score_gemma":0.003212402,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001708067,"about_ca_topic_score_gemma":0.002150745,"domain_scores_codex":[0.9767542,0.01413923,0.002246544,0.0008687625,0.005423336,0.0005679429],"domain_scores_gemma":[0.9228987,0.05401522,0.009599442,0.00361466,0.007284225,0.002587887],"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":[0.03879545,0.03866427,0.3964611,0.001907739,0.001649521,0.0001098199,0.008295959,0.005792647,0.01078062,0.001397465,0.002153815,0.4939916],"study_design_scores_gemma":[0.003843792,0.05520778,0.8612326,0.0007754723,0.002836989,0.00009639277,0.00439524,0.02104604,0.04417512,0.002454042,0.003648822,0.0002877532],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9915596,0.0001496151,0.005459988,0.0001233987,0.00002598629,0.0009166984,0.0002535834,0.0001432841,0.001367856],"genre_scores_gemma":[0.9741794,0.0002272987,0.02035488,0.00006555116,0.00003062124,0.002257356,0.0007614035,0.0001073947,0.002016178],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02281243,"threshold_uncertainty_score":0.1206451,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4413515290","doi":"10.1016/j.stueduc.2025.101512","title":"Stage-based professional development needs of canadian high school teachers: Insights from the dynamic model of educational effectiveness","year":2025,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Teacher Education and Leadership Studies","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"Université du Québec à Montréal; Université du Québec en Outaouais","funders":"","keywords":"Professional development; Mathematics education; Faculty development; Psychology; Pedagogy","authors":[{"name":"Jérôme St‐Amand","is_ca":true},{"name":"Panayiotis Antoniou","is_ca":false},{"name":"Martine DeGrandpré","is_ca":true},{"name":"Victoria Michaelidou","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1620902212337623,"gpt":0.4620496655656992,"spread":0.2999594443319369,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004909663,0.0002287691,0.000418053,0.002140489,0.0041985,0.00364263,0.00196166,0.0009306303,0.002001462],"category_scores_gemma":[0.02448374,0.0003194471,0.0003491899,0.002386384,0.002772209,0.001921069,0.001623884,0.0008871336,0.0001019913],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.05231545,"about_ca_system_score_gemma":0.06785889,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.9350998,"about_ca_topic_score_gemma":0.9602492,"domain_scores_codex":[0.9959766,0.0008595632,0.0001350829,0.0002126864,0.001833353,0.0009826677],"domain_scores_gemma":[0.9871679,0.005230997,0.001170396,0.0003156328,0.004273195,0.001841861],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0004208517,0.00065901,0.5525212,0.0002727778,0.00005832797,0.0004875509,0.2631759,0.003677647,0.00156829,0.05687887,0.003700448,0.1165792],"study_design_scores_gemma":[0.00002907124,0.0001740708,0.7946155,0.0001327831,0.00004443908,0.0001407906,0.1847015,0.005352943,0.0004098531,0.007864252,0.006474135,0.00006072121],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9818283,0.0002419831,0.000746459,0.001984693,0.000007116016,0.00009812739,0.0001423511,0.000006836133,0.01494415],"genre_scores_gemma":[0.9986567,0.0001085439,0.0003239695,0.0000292869,0.000001070678,0.00002418508,0.00004183764,0.000002225907,0.0008120167],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9476845,"threshold_uncertainty_score":0.379577,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W7117577854","doi":"10.1016/j.stueduc.2025.101557","title":"Assessing school readiness domains in a large cohort of refugee children: Validation and links with family factors","year":2025,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Early Childhood Education and Development","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Child, Adolescent and Family Mental Health","funders":"Universiti Sains Malaysia; British Academy","keywords":"Rasch model; Refugee; Polytomous Rasch model; Construct validity; Psychometrics; Sample (material); Cohort; Item analysis; Item response theory","authors":[{"name":"Alexandra Cheah","is_ca":false},{"name":"Kimberley Kong","is_ca":false},{"name":"Jean Anne Heng","is_ca":true},{"name":"Katharina Ereky-Stevens","is_ca":false},{"name":"Iram Siraj","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.05304186638219593,"gpt":0.4417620280352073,"spread":0.3887201616530114,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002833219,0.0005138634,0.0004243592,0.001033045,0.001035725,0.0008173566,0.0005019925,0.0004427208,0.001039988],"category_scores_gemma":[0.004597633,0.0003415979,0.0006586754,0.0008064827,0.0005627976,0.0006579407,0.001286349,0.000932421,0.0003205558],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005653026,"about_ca_system_score_gemma":0.001199352,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01297008,"about_ca_topic_score_gemma":0.02428985,"domain_scores_codex":[0.9992427,0.0002885619,0.00009428501,0.0001053554,0.0001328569,0.000136269],"domain_scores_gemma":[0.9987212,0.0003348761,0.0003446805,0.000187021,0.0002628122,0.0001494498],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00003798613,0.00007734609,0.993481,0.00002321605,0.00002105771,0.000158789,0.0021244,0.00003251069,0.0003448664,0.00002048964,0.00007688176,0.003601466],"study_design_scores_gemma":[0.000004095092,0.0001199754,0.9945005,0.00003123133,0.000009838514,0.000288784,0.004483268,0.00007295973,0.0001423562,0.00001651579,0.0003246059,0.000005873965],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9995659,0.00005257219,0.00008055155,0.00001806258,0.000002190236,0.00003140717,0.0001180352,0.000001249328,0.0001299805],"genre_scores_gemma":[0.9984986,0.0001921701,0.0005738037,0.00002448447,0.000001772033,0.0001328577,0.0003988141,0.000003493464,0.0001739767],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01297008,"threshold_uncertainty_score":0.0257892,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}