{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":101,"total_is_capped":false,"direct_labels_cover":1,"predictions_cover":101,"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":"4cda1ec5ea24","filters":{"venue":"Educational and Psychological Measurement"}},"results":[{"id":"W2021930444","doi":"10.1177/0013164409355698","title":"The Motivation at Work Scale: Validation Evidence in Two Languages","year":2010,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Motivation and Self-Concept in Sports","field":"Psychology","cited_by":749,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"HEC Montréal; Université du Québec à Montréal; Université de Montréal; Concordia University","funders":"","keywords":"Conceptualization; Scale (ratio); Psychology; Work (physics); Intrinsic motivation; Social psychology; Work motivation; Cognitive psychology; Computer science; Artificial intelligence; Geography; Physics","authors":[{"name":"Marylène Gagné","is_ca":true},{"name":"Jacques Forest","is_ca":true},{"name":"Marie‐Hélène Gilbert","is_ca":true},{"name":"Caroline Aubé","is_ca":true},{"name":"Estelle M. Morin","is_ca":true},{"name":"Angela Malorni","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1709997786686356,"gpt":0.4045055228999631,"spread":0.2335057442313275,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01235528,0.0005180751,0.0005571581,0.002316435,0.001208447,0.001440026,0.0008475003,0.0008103568,0.001609904],"category_scores_gemma":[0.02144663,0.0003471138,0.0008650973,0.00144165,0.00148959,0.001064289,0.002322865,0.001137744,0.0008239723],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007153705,"about_ca_system_score_gemma":0.001624209,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004496078,"about_ca_topic_score_gemma":0.006609085,"domain_scores_codex":[0.9948924,0.002255794,0.0005700627,0.0004449809,0.00154798,0.0002887984],"domain_scores_gemma":[0.9863648,0.006953619,0.001813641,0.0009781569,0.003096293,0.0007934379],"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":[0.001066262,0.002104366,0.8697723,0.0004242138,0.0002921578,0.0002326772,0.01292,0.0003044991,0.003208876,0.002140301,0.002682901,0.1048514],"study_design_scores_gemma":[0.0002346004,0.001179349,0.9813498,0.0002936678,0.00009381588,0.0003448822,0.004276761,0.0004844748,0.001094454,0.0007783664,0.009822729,0.0000469662],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9883897,0.0006711744,0.001576892,0.0003295028,0.0001142546,0.0007625597,0.0008301532,0.00002872603,0.007296975],"genre_scores_gemma":[0.9843428,0.0008496289,0.005889961,0.0005471739,0.00006448535,0.002167891,0.002847465,0.00006962502,0.00322082],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01235528,"threshold_uncertainty_score":0.06534177,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2062055237","doi":"10.1177/00131640021970871","title":"Type I Error Rate Comparisons of Post Hoc Procedures for I j Chi-Square Tables","year":2000,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":444,"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":"Post hoc; Post-hoc analysis; Statistics; Pairwise comparison; Contingency table; Type I and type II errors; Homogeneity (statistics); Mathematics; Chi-square test; Null hypothesis; Monte Carlo method; Econometrics; Medicine; Dentistry","authors":[{"name":"Paul L. MacDonald","is_ca":false},{"name":"R. C. Gardner","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4095890100599183,"gpt":0.4883884670827446,"spread":0.07879945702282637,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.198452,0.003537812,0.005202488,0.007497001,0.002886557,0.004447508,0.006298112,0.004805256,0.02329944],"category_scores_gemma":[0.6248609,0.002307134,0.006165245,0.005827657,0.006466301,0.004888453,0.004396603,0.01080059,0.003782626],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002973573,"about_ca_system_score_gemma":0.003317157,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009747614,"about_ca_topic_score_gemma":0.001322866,"domain_scores_codex":[0.6436089,0.2179832,0.03626953,0.03458102,0.06303941,0.004517992],"domain_scores_gemma":[0.2907483,0.5869867,0.0220949,0.06578185,0.03273215,0.001656107],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.03317827,0.005548229,0.03421296,0.009792011,0.01189895,0.002047775,0.01676309,0.02021973,0.01936716,0.1719225,0.08337733,0.5916721],"study_design_scores_gemma":[0.006713761,0.04278775,0.1057747,0.0059768,0.00848917,0.003053154,0.006532361,0.2107076,0.1336013,0.316906,0.1570494,0.002408186],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.04381328,0.0008301322,0.909312,0.0004903566,0.006088754,0.02264355,0.002541358,0.003681977,0.01059863],"genre_scores_gemma":[0.1622379,0.0003747135,0.7718071,0.0009286735,0.0005666726,0.05494459,0.00106751,0.003327605,0.004745311],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.801548,"threshold_uncertainty_score":0.9884514,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2084244869","doi":"10.1177/0013164403258450","title":"Measurement and Validity Characteristics of the Short Version of the Social and Emotional Loneliness Scale for Adults","year":2004,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Health disparities and outcomes","field":"Social Sciences","cited_by":373,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of New Brunswick","funders":"","keywords":"Loneliness; Psychology; Discriminant validity; Convergent validity; UCLA Loneliness Scale; Construct validity; Concurrent validity; Scale (ratio); Clinical psychology; Test validity; Developmental psychology; Psychometrics; Social psychology; Internal consistency","authors":[{"name":"Enrico DiTommaso","is_ca":true},{"name":"Cyndi Brannen","is_ca":true},{"name":"Lisa A. Best","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2283013155929321,"gpt":0.4013158889867471,"spread":0.1730145733938151,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005822474,0.0002812495,0.0004546905,0.001137127,0.0005490964,0.0005717397,0.000338211,0.0004492574,0.001014729],"category_scores_gemma":[0.02189033,0.0002245251,0.000766283,0.0008826214,0.0003838754,0.0007613309,0.0009653525,0.0005497144,0.0002196868],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003313121,"about_ca_system_score_gemma":0.0005586463,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001185734,"about_ca_topic_score_gemma":0.002992627,"domain_scores_codex":[0.9969499,0.001189,0.0006252472,0.0002379051,0.0008084119,0.0001896821],"domain_scores_gemma":[0.9910183,0.003624795,0.002398518,0.0003492704,0.001989923,0.0006191714],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001616531,0.0001915056,0.9790521,0.00004633026,0.0001070812,0.00004747914,0.001014669,0.0001618585,0.000372907,0.0001662399,0.0004642668,0.0182139],"study_design_scores_gemma":[0.00001590718,0.0003586382,0.9972661,0.00002060154,0.00001784476,0.0001778034,0.0006196273,0.0004506905,0.0001181405,0.0001470634,0.0007982529,0.000009340981],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9955087,0.0003035807,0.0008186268,0.0001466404,0.00003761385,0.0002358226,0.0005531618,0.000009128257,0.002386762],"genre_scores_gemma":[0.9956365,0.0001486069,0.00237014,0.00007255002,0.00002331351,0.0004551828,0.0008975596,0.00000687815,0.0003892479],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.005822474,"threshold_uncertainty_score":0.03079259,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2027051144","doi":"10.1177/00131640021970655","title":"Development and Validation of Scores on a Two-Dimensional Workplace Friendship Scale","year":2000,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Job Satisfaction and Organizational Behavior","field":"Business, Management and Accounting","cited_by":343,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Quest University Canada","funders":"","keywords":"Friendship; Nomological network; Psychology; Scale (ratio); Discriminant validity; Construct validity; Sample (material); Convergent validity; Predictive validity; Construct (python library); Social psychology; Test validity; Likert scale; External validity; Internal consistency; Replicate; Applied psychology; Psychometrics; Developmental psychology; Statistics; Mathematics; Computer science","authors":[{"name":"Ivy K. Nielsen","is_ca":true},{"name":"Steve M. Jex","is_ca":false},{"name":"Gary A. Adams","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.07775349235747162,"gpt":0.2954924167858354,"spread":0.2177389244283638,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008408917,0.0003792092,0.0004030494,0.002636194,0.0005265762,0.0009509084,0.0006612842,0.0005299016,0.001273769],"category_scores_gemma":[0.01974381,0.0003020854,0.0004745615,0.0009175368,0.0006710616,0.0009002446,0.001239544,0.0009486877,0.0005430249],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004229447,"about_ca_system_score_gemma":0.00101546,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001203924,"about_ca_topic_score_gemma":0.002759666,"domain_scores_codex":[0.9958845,0.001423871,0.0006406277,0.000218938,0.001594862,0.0002371761],"domain_scores_gemma":[0.9854075,0.004791366,0.001674762,0.001302891,0.005824737,0.0009987151],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002949539,0.001369815,0.8442301,0.0001151159,0.0001092221,0.000142663,0.00569638,0.000610764,0.006626706,0.001273832,0.00170497,0.1378255],"study_design_scores_gemma":[0.00008959853,0.0012813,0.9845498,0.00004881754,0.00003262536,0.0002140798,0.002823204,0.002926535,0.002408317,0.0006759744,0.004911248,0.00003854684],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9864191,0.00008860528,0.006017594,0.0001289082,0.00004351016,0.001367786,0.0004870731,0.00005129949,0.00539614],"genre_scores_gemma":[0.9680807,0.0001180302,0.02645213,0.00006470244,0.00002189167,0.002356407,0.00118056,0.00002384253,0.001701604],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.008408917,"threshold_uncertainty_score":0.04447114,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2036831393","doi":"10.1177/0013164411409929","title":"Difference Scores From the Point of View of Reliability and Repeated-Measures ANOVA","year":2011,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":167,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia; Carleton University","funders":"","keywords":"Reliability (semiconductor); Repeated measures design; Significant difference; Statistics; Psychology; Analysis of variance; Point (geometry); Mean difference; Mathematics; Econometrics; Confidence interval","authors":[{"name":"David R. Thomas","is_ca":true},{"name":"Bruno D. Zumbo","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7554307011056853,"gpt":0.4653795213841227,"spread":0.2900511797215626,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06165618,0.0007815686,0.001549852,0.003812241,0.0007275421,0.003236886,0.00170689,0.001397329,0.004538093],"category_scores_gemma":[0.2944942,0.0004892759,0.001457901,0.00377799,0.005322215,0.005092952,0.002886595,0.003380897,0.001339384],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009029545,"about_ca_system_score_gemma":0.001228321,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004722405,"about_ca_topic_score_gemma":0.0004831886,"domain_scores_codex":[0.8814825,0.05940815,0.008236702,0.006536648,0.04328181,0.001054229],"domain_scores_gemma":[0.7464682,0.1878972,0.01126345,0.02948123,0.02341162,0.001478129],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.002802676,0.001160572,0.1240618,0.003006235,0.001798047,0.0005434881,0.01266811,0.004557147,0.01125998,0.2832426,0.01802598,0.5368734],"study_design_scores_gemma":[0.0005758,0.008810143,0.2193796,0.001019086,0.001401491,0.002606885,0.01007902,0.03300102,0.02085836,0.6171448,0.08453586,0.0005879815],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.166306,0.002213222,0.7755061,0.003666793,0.001771402,0.001697346,0.001264914,0.00110095,0.04647312],"genre_scores_gemma":[0.6675655,0.0008785118,0.3221648,0.0007993233,0.0007817383,0.003044059,0.0007282104,0.0003357957,0.003702067],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.06165618,"threshold_uncertainty_score":0.3260729,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1763331641","doi":"10.1177/0013164415584576","title":"It Might Not Make a Big DIF","year":2015,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":130,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"York University","funders":"","keywords":"Differential item functioning; Item response theory; Statistics; Polytomous Rasch model; Computer science; Monte Carlo method; Econometrics; Type I and type II errors; Statistical hypothesis testing; Machine learning; Data mining; Artificial intelligence; Psychometrics; Mathematics","authors":[{"name":"R. Philip Chalmers","is_ca":true},{"name":"Alyssa Counsell","is_ca":true},{"name":"David B. Flora","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9118226372485674,"gpt":0.5444501445426981,"spread":0.3673724927058692,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02659532,0.001002635,0.001404838,0.002353047,0.001871793,0.003930259,0.001294903,0.00190273,0.009022163],"category_scores_gemma":[0.1333622,0.0005695508,0.001387743,0.002763754,0.004509213,0.005939245,0.003245976,0.003280876,0.002697587],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001817093,"about_ca_system_score_gemma":0.002064252,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002639211,"about_ca_topic_score_gemma":0.003881221,"domain_scores_codex":[0.9768153,0.0121103,0.002364609,0.003834195,0.003855458,0.001019923],"domain_scores_gemma":[0.9281565,0.04206962,0.005700678,0.01432898,0.008164345,0.001579868],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.000886063,0.0005265265,0.2772764,0.001386972,0.001227916,0.001122885,0.01002279,0.00484191,0.003109658,0.2879284,0.03024185,0.3814286],"study_design_scores_gemma":[0.0002643008,0.0009543953,0.1568173,0.001069598,0.0008158433,0.003130318,0.01112898,0.02436795,0.004302763,0.7105446,0.08626635,0.0003375698],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.507188,0.003613276,0.3201028,0.04222179,0.003926119,0.001132477,0.002983561,0.0009312081,0.1179008],"genre_scores_gemma":[0.9268292,0.0006785416,0.05959286,0.006753273,0.0004100253,0.0005861601,0.0008752696,0.0001615466,0.004113044],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02659532,"threshold_uncertainty_score":0.1406511,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"simulation_or_modeling","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2012610170","doi":"10.1177/0013164403260196","title":"The Analysis of Repeated Measurements with Mixed-Model Adjusted F Tests","year":2004,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":130,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Manitoba","funders":"","keywords":"Type I and type II errors; Statistics; Repeated measures design; Sample size determination; Mixed model; Mathematics; Covariance; Univariate; Statistic; Homogeneity (statistics); Normality; Analysis of covariance; Heteroscedasticity; Statistical hypothesis testing; Econometrics; Multivariate statistics","authors":[{"name":"Rhonda K. Kowalchuk","is_ca":false},{"name":"H. J. Keselman","is_ca":true},{"name":"James Algina","is_ca":false},{"name":"Russell D. Wolfinger","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7953401039095214,"gpt":0.5668979603801602,"spread":0.2284421435293612,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07126378,0.003573316,0.006168515,0.005836042,0.001638445,0.003557527,0.005671605,0.002628533,0.01905691],"category_scores_gemma":[0.2420478,0.001417872,0.007020762,0.006947887,0.002123102,0.003627313,0.002298709,0.005388203,0.002663373],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001857492,"about_ca_system_score_gemma":0.003905528,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002750842,"about_ca_topic_score_gemma":0.002188408,"domain_scores_codex":[0.9197448,0.04644533,0.005754528,0.01165973,0.01460331,0.001792263],"domain_scores_gemma":[0.8270303,0.131554,0.00803186,0.01911108,0.01366546,0.0006073446],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.009081597,0.002671194,0.02800021,0.007502518,0.009418841,0.001687162,0.006436397,0.02287472,0.008785673,0.08701814,0.04492366,0.7715999],"study_design_scores_gemma":[0.003598529,0.03007411,0.1108951,0.002707224,0.007251528,0.002876336,0.003822005,0.4060759,0.0479278,0.1699814,0.2128988,0.001891261],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02515074,0.0006444898,0.9560199,0.0002916831,0.001390212,0.006146481,0.003025374,0.002942823,0.004388233],"genre_scores_gemma":[0.09191892,0.000411701,0.8748531,0.0002626132,0.0003105478,0.0259806,0.001470807,0.001273559,0.003518119],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.07126378,"threshold_uncertainty_score":0.3768834,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2004549974","doi":"10.1177/0013164403258402","title":"The Effect of Multicollinearity on Multilevel Modeling Parameter Estimates and Standard Errors","year":2003,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":127,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Simon Fraser University","funders":"","keywords":"Multicollinearity; Statistics; Mathematics; Multilevel model; Standard error; Covariance; Variance inflation factor; Variance (accounting); Econometrics; Correlation; Standard deviation; Linear regression","authors":[{"name":"Yann‐Yann Shieh","is_ca":false},{"name":"Rachel T. Fouladi","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3510922864786483,"gpt":0.4841609772101677,"spread":0.1330686907315194,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.265844,0.001877652,0.002347583,0.004364412,0.002174397,0.005124884,0.002341372,0.002287072,0.001741762],"category_scores_gemma":[0.7292315,0.001876165,0.00300123,0.007371744,0.003610288,0.005040203,0.004319954,0.006050773,0.0004840209],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002325292,"about_ca_system_score_gemma":0.002408219,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005476817,"about_ca_topic_score_gemma":0.006574877,"domain_scores_codex":[0.6671812,0.249588,0.02436528,0.01945669,0.03672702,0.002681819],"domain_scores_gemma":[0.1838065,0.7060601,0.02835546,0.06260685,0.01840781,0.0007632436],"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":[0.00132218,0.0004655612,0.4928647,0.0016307,0.0125122,0.0008695704,0.01824788,0.08573009,0.005404716,0.08835412,0.007091534,0.2855067],"study_design_scores_gemma":[0.0004789297,0.001630432,0.4169723,0.002501463,0.004487564,0.002211573,0.005069297,0.3144291,0.02398891,0.2096941,0.01731709,0.001219209],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1676175,0.003018601,0.8206576,0.001950189,0.0005252395,0.0008279909,0.000765846,0.0009323199,0.003704746],"genre_scores_gemma":[0.8504645,0.0006871521,0.1453669,0.0004638892,0.0001303306,0.001056463,0.0005632136,0.0005169563,0.0007505654],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.265844,"threshold_uncertainty_score":0.9053451,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2130969147","doi":"10.1177/0013164412452564","title":"The Relationship Between Root Mean Square Error of Approximation and Model Misspecification in Confirmatory Factor Analysis Models","year":2012,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":123,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Structural equation modeling; Confirmatory factor analysis; Statistics; Mathematics; Population; Cutoff; Factor analysis; Econometrics; Physics","authors":[{"name":"Victoria Savalei","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8678825713483123,"gpt":0.5198963884873063,"spread":0.3479861828610059,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1958023,0.002246962,0.00259669,0.005680326,0.002116004,0.003583558,0.00281203,0.002837265,0.001518196],"category_scores_gemma":[0.608685,0.001662528,0.003064587,0.008241455,0.005318949,0.005609893,0.004452802,0.006098334,0.0005405008],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003170565,"about_ca_system_score_gemma":0.003471608,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003908596,"about_ca_topic_score_gemma":0.005779337,"domain_scores_codex":[0.7399422,0.1757398,0.02536056,0.02031717,0.0369775,0.001662747],"domain_scores_gemma":[0.1662231,0.7492831,0.03236886,0.03559681,0.01582608,0.0007019519],"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":[0.0008403445,0.0005860276,0.4203984,0.003189884,0.008460755,0.002008651,0.01226407,0.1406432,0.003250078,0.08837739,0.007658774,0.3123224],"study_design_scores_gemma":[0.000234396,0.002007475,0.259777,0.003065083,0.002272799,0.003596877,0.003637693,0.4523264,0.009448632,0.2485607,0.01408296,0.0009900036],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1374154,0.005828385,0.8453023,0.002411939,0.0004981753,0.0005119695,0.000483854,0.001213778,0.006334113],"genre_scores_gemma":[0.8385486,0.001001452,0.1566047,0.0007900716,0.0001504039,0.0009084527,0.0006372631,0.0005109898,0.0008479962],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.1958023,"threshold_uncertainty_score":0.9917189,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2064379877","doi":"10.1177/0013164402238082","title":"A Monte Carlo Comparison of Item and Person Statistics Based on Item Response Theory versus Classical Test Theory","year":2002,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":121,"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":"Item response theory; Classical test theory; Monte Carlo method; Statistics; Psychology; Econometrics; Psychometrics; Test theory; Item analysis; Computerized adaptive testing; Test (biology); Differential item functioning; Statistical hypothesis testing; Mathematics","authors":[{"name":"P.B. MacDonald","is_ca":false},{"name":"Sampo V. Paunonen","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7790233134897627,"gpt":0.505780321916664,"spread":0.2732429915730987,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06936503,0.0005999351,0.0009441296,0.003069585,0.0004456388,0.001928622,0.001299463,0.001379331,0.001585485],"category_scores_gemma":[0.357976,0.000467116,0.0007666678,0.001840892,0.002121706,0.003339872,0.001667483,0.001270363,0.0003102547],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001204262,"about_ca_system_score_gemma":0.0009909576,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001463618,"about_ca_topic_score_gemma":0.001560229,"domain_scores_codex":[0.9578832,0.03395642,0.001022514,0.002687836,0.004068042,0.0003818708],"domain_scores_gemma":[0.5083188,0.4516118,0.007853303,0.02348077,0.007784965,0.0009503145],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0067143,0.0007161052,0.2153496,0.0007124517,0.002018879,0.0002911628,0.002946689,0.3399122,0.006033674,0.1758301,0.003214745,0.2462602],"study_design_scores_gemma":[0.0002483882,0.001628936,0.06803903,0.0001685685,0.000224031,0.0007141035,0.0004880556,0.8655397,0.004122929,0.05705947,0.001599572,0.0001672851],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5916283,0.000503034,0.4023365,0.0002718386,0.00006818189,0.0003875229,0.0002574084,0.0005772701,0.00397],"genre_scores_gemma":[0.8921206,0.000105847,0.106539,0.0001065846,0.00001559774,0.0002783439,0.000480659,0.0001026952,0.0002506773],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.06936503,"threshold_uncertainty_score":0.3668417,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2065030780","doi":"10.1177/0013164404273942","title":"Understanding Parameter Invariance in Unidimensional IRT Models","year":2006,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":108,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia; University of Ottawa","funders":"","keywords":"Item response theory; Measurement invariance; Set (abstract data type); Differential item functioning; Psychology; Feature (linguistics); Property (philosophy); Econometrics; Variable (mathematics); Psychometrics; Cognitive psychology; Mathematics; Statistics; Applied mathematics; Computer science; Structural equation modeling; Confirmatory factor analysis; Mathematical analysis; Epistemology","authors":[{"name":"André Rupp","is_ca":true},{"name":"Bruno D. Zumbo","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9212118758900417,"gpt":0.4959208972683364,"spread":0.4252909786217053,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03624367,0.001178445,0.001928838,0.002555957,0.0008773361,0.004707877,0.003231896,0.001562362,0.0028775],"category_scores_gemma":[0.1902227,0.0009335948,0.001839413,0.002790594,0.004701466,0.009011094,0.005262831,0.004177316,0.0006894824],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001635127,"about_ca_system_score_gemma":0.001410566,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002499028,"about_ca_topic_score_gemma":0.0009855054,"domain_scores_codex":[0.9717053,0.01843549,0.001526841,0.003796103,0.003612921,0.0009233668],"domain_scores_gemma":[0.8665839,0.1027165,0.009123728,0.0170979,0.003879113,0.0005990163],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0001376521,0.0001476042,0.02469186,0.0002174236,0.0003398546,0.0002685345,0.00287569,0.09835404,0.001337707,0.7711339,0.001205001,0.0992908],"study_design_scores_gemma":[0.00002117894,0.00006115049,0.005007041,0.00005262039,0.00004577689,0.0001578182,0.0003362887,0.2279376,0.0004639625,0.7648473,0.001008865,0.00006034459],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.04922412,0.0002793258,0.9454246,0.0007559396,0.00003188835,0.00006848929,0.000137931,0.0002871213,0.003790587],"genre_scores_gemma":[0.8198638,0.0004433052,0.1771084,0.0004203608,0.000127845,0.0004631357,0.0004132995,0.0002052843,0.00095458],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.03624367,"threshold_uncertainty_score":0.1916771,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2021178079","doi":"10.1177/0013164403261762","title":"The Nature and Measurement of Emotional Intelligence Abilities: Basic Dimensions and Their Relationships with Other Cognitive Ability and Personality Variables","year":2004,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Emotional Intelligence and Performance","field":"Psychology","cited_by":106,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Psychology; Personality; Alexithymia; Cognition; Emotional intelligence; Developmental psychology; Big Five personality traits; Construct (python library); Congruence (geometry); Social psychology; Cognitive psychology","authors":[{"name":"Kimberly A. Barchard","is_ca":false},{"name":"A. Ralph Hakstian","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1714507792378559,"gpt":0.3521081209423749,"spread":0.180657341704519,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00129615,0.0002952866,0.0001842149,0.0009457853,0.0001654525,0.0005061089,0.0002694406,0.0002086976,0.001254412],"category_scores_gemma":[0.006781095,0.00007871596,0.0001524167,0.0006824437,0.0003970208,0.0004122723,0.0004661084,0.0003748878,0.0002200165],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001478465,"about_ca_system_score_gemma":0.0001846347,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004983226,"about_ca_topic_score_gemma":0.0009671808,"domain_scores_codex":[0.9995497,0.0001464929,0.00007701743,0.00004372542,0.000140134,0.00004301683],"domain_scores_gemma":[0.9969254,0.001341052,0.0008554566,0.0002076,0.0003843718,0.0002861767],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001363542,0.0001802472,0.9422415,0.00006952782,0.00004959101,0.0000859179,0.0004555962,0.0003296997,0.003237933,0.0003976768,0.0001337347,0.05268216],"study_design_scores_gemma":[0.000003032669,0.0001046068,0.9981474,0.000008214619,0.000005918004,0.0001639996,0.00009804771,0.0003700631,0.0006194497,0.0002442928,0.0002323014,0.000002765896],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9964631,0.0001760765,0.0009543368,0.00002860517,0.000003736007,0.00001991353,0.0001302524,0.000006490167,0.00221761],"genre_scores_gemma":[0.9984679,0.00005698545,0.001069983,0.000005788911,0.000004911191,0.00001594469,0.0001416916,0.000001348811,0.0002354133],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.00129615,"threshold_uncertainty_score":0.006854773,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2004501768","doi":"10.1177/0013164402239317","title":"Differential Item Functioning Results May Change Depending On How An Item Is Scored: An Illustration With The Center For Epidemiologic Studies Depression Scale","year":2003,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":97,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Differential item functioning; Psychology; Item response theory; Logistic regression; Statistics; Psychometrics; Ordered logit; Scale (ratio); Clinical psychology; Center for Epidemiologic Studies Depression Scale; Persistence (discontinuity); Depressive symptoms; Mathematics; Cognition; Psychiatry","authors":[{"name":"Michaela N. Gelin","is_ca":true},{"name":"Bruno D. Zumbo","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8833118557082188,"gpt":0.5341469955629262,"spread":0.3491648601452926,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1101033,0.0008094083,0.0008445587,0.002549205,0.001054431,0.001443788,0.0007531806,0.0007448714,0.001844434],"category_scores_gemma":[0.2425298,0.0004457911,0.001092978,0.003879474,0.002100545,0.001605029,0.001790942,0.001480562,0.0006207458],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000876496,"about_ca_system_score_gemma":0.0005070374,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008041074,"about_ca_topic_score_gemma":0.001961787,"domain_scores_codex":[0.8474614,0.1299964,0.008542595,0.003551157,0.009527577,0.0009207956],"domain_scores_gemma":[0.785395,0.1791364,0.006884169,0.015334,0.01290511,0.0003454437],"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":[0.001323521,0.0003024794,0.5199795,0.0009178384,0.0009734336,0.0007584551,0.0149724,0.001881381,0.006808796,0.02379938,0.008826286,0.4194565],"study_design_scores_gemma":[0.0002638736,0.001222141,0.8773533,0.0009196252,0.0005125468,0.004674551,0.005046878,0.01571639,0.01163876,0.04985918,0.03251719,0.0002755079],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5910213,0.003950861,0.3736634,0.005951896,0.0008108044,0.001328371,0.001269942,0.0004487605,0.02155464],"genre_scores_gemma":[0.8847415,0.0004709029,0.1105906,0.001060207,0.0001390784,0.001433196,0.000396056,0.0001573922,0.00101113],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1101033,"threshold_uncertainty_score":0.5822889,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2033678552","doi":"10.1177/0013164409344548","title":"The Impact of Outliers on Cronbach’s Coefficient Alpha Estimate of Reliability: Ordinal/Rating Scale Item Responses","year":2009,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":96,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Cronbach's alpha; Outlier; Categorical variable; Ordinal data; Statistics; Likert scale; Mathematics; Econometrics; Reliability (semiconductor); Sample (material); Rating scale; Psychology; Psychometrics","authors":[{"name":"Yan Liu","is_ca":true},{"name":"Amery D. Wu","is_ca":true},{"name":"Bruno D. Zumbo","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2803599964946322,"gpt":0.5147372816258765,"spread":0.2343772851312443,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2349381,0.001367934,0.002252339,0.004011393,0.001685933,0.003515607,0.001865773,0.001717782,0.001096761],"category_scores_gemma":[0.673364,0.001537164,0.00231982,0.005022848,0.00538587,0.003676472,0.003403699,0.004681607,0.0006833184],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002172529,"about_ca_system_score_gemma":0.001541218,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003054631,"about_ca_topic_score_gemma":0.002215591,"domain_scores_codex":[0.6485559,0.272423,0.01771374,0.01437103,0.04497788,0.001958389],"domain_scores_gemma":[0.1683448,0.7197886,0.02839601,0.04657044,0.03605353,0.0008465383],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.003311686,0.0004248194,0.4238647,0.002575114,0.006758851,0.001175733,0.02514847,0.0568971,0.007098335,0.06567974,0.01707786,0.3899876],"study_design_scores_gemma":[0.0005454983,0.003308326,0.4327799,0.003803499,0.002876073,0.003983109,0.007874336,0.2882398,0.03342257,0.179863,0.04200542,0.001298519],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2405229,0.004403801,0.7396466,0.002160801,0.0008428625,0.0007223861,0.0005052238,0.0009265595,0.01026884],"genre_scores_gemma":[0.8805726,0.0005458063,0.1155273,0.0006674029,0.0001529904,0.0009533029,0.0003851949,0.0004763059,0.0007190965],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7650619,"threshold_uncertainty_score":0.9434575,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1501970797","doi":"10.1177/0013164415593777","title":"Testing Mediation in Structural Equation Modeling","year":2015,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Attachment and Relationship Dynamics","field":"Psychology","cited_by":93,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Ottawa; Carleton University","funders":"","keywords":"Bootstrapping (finance); Structural equation modeling; Monte Carlo method; Latent variable; Mediation; Econometrics; Contrast (vision); Statistical hypothesis testing; Statistics; Computer science; Latent variable model; Mathematics; Artificial intelligence","authors":[{"name":"Craig Leth‐Steensen","is_ca":true},{"name":"Elena Gallitto","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4340302585753368,"gpt":0.4648977677929242,"spread":0.03086750921758735,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2370327,0.002350702,0.003025418,0.00442196,0.002324741,0.003390098,0.005358595,0.002943888,0.004911608],"category_scores_gemma":[0.5670361,0.001787514,0.004604875,0.00604154,0.005515979,0.006506245,0.004751142,0.00496945,0.0003896622],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001445974,"about_ca_system_score_gemma":0.00488297,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002881621,"about_ca_topic_score_gemma":0.002461278,"domain_scores_codex":[0.6715239,0.3090743,0.003748598,0.006581169,0.007957934,0.00111407],"domain_scores_gemma":[0.2881594,0.6790677,0.007022022,0.02112352,0.004073235,0.0005540933],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.002433249,0.001590104,0.1540186,0.003688148,0.02077321,0.0009969682,0.007425691,0.1390159,0.002862118,0.3238761,0.002147924,0.341172],"study_design_scores_gemma":[0.001060028,0.003078465,0.02350548,0.0008670971,0.00249086,0.0004498237,0.001222461,0.6202002,0.004063192,0.3398099,0.00301269,0.0002397675],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.09694225,0.001064893,0.8975167,0.0008658039,0.0001695774,0.0008669677,0.0001743534,0.0004537797,0.001945607],"genre_scores_gemma":[0.6784458,0.0005557506,0.3181223,0.0002254667,0.00007185145,0.002140904,0.000173665,0.00009444813,0.0001698958],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.2370327,"threshold_uncertainty_score":0.9408746,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2758319537","doi":"10.1177/0013164417719111","title":"Recommendations on the Sample Sizes for Multilevel Latent Class Models","year":2017,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","cited_by":93,"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":"Latent class model; Sample size determination; Statistics; Sample (material); Rule of thumb; Nested set model; Computer science; Model selection; Econometrics; Data mining; Mathematics; Algorithm","authors":[{"name":"Jungkyu Park","is_ca":true},{"name":"Hsiu‐Ting Yu","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4657980968857349,"gpt":0.4110614061339838,"spread":0.05473669075175114,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1888869,0.002541752,0.003536518,0.006723372,0.002762413,0.005053332,0.009194675,0.01082546,0.01321644],"category_scores_gemma":[0.6177962,0.002996607,0.003564536,0.004481142,0.005070776,0.007952618,0.00445952,0.01326516,0.004996398],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003184277,"about_ca_system_score_gemma":0.008553986,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006847762,"about_ca_topic_score_gemma":0.008592522,"domain_scores_codex":[0.7965466,0.1518825,0.01594535,0.005603977,0.02840852,0.001613041],"domain_scores_gemma":[0.2837661,0.6245825,0.01107376,0.02065424,0.05594732,0.003976072],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.002170605,0.001044005,0.009650072,0.005767098,0.0003505098,0.0006716263,0.004993355,0.01822845,0.00432466,0.1582399,0.2538105,0.5407491],"study_design_scores_gemma":[0.006149118,0.002063908,0.02600897,0.04394633,0.001128208,0.001684277,0.006370808,0.1377866,0.02014394,0.4186434,0.3350785,0.0009958874],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01623651,0.01023639,0.824937,0.09214284,0.006509634,0.00919601,0.004324317,0.003798852,0.03261851],"genre_scores_gemma":[0.03606828,0.00208343,0.9355454,0.008439772,0.0008675208,0.0133793,0.0009884235,0.0005949481,0.002032962],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.1888869,"threshold_uncertainty_score":0.9989415,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2036299295","doi":"10.1177/0013164405275668","title":"A Comparison of Four Methods for Detecting Differential Item Functioning in Ordered Response Items","year":2005,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":92,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia; University of Ottawa","funders":"","keywords":"Differential item functioning; Statistics; Psychology; Logistic regression; Item response theory; Type I and type II errors; Linear discriminant analysis; Ordinal regression; Skewness; Ordered logit; Discriminant function analysis; Sample (material); Psychometrics; Mathematics","authors":[{"name":"Elizabeth Kristjansson","is_ca":true},{"name":"Richard Aylesworth","is_ca":true},{"name":"Ian McDowell","is_ca":true},{"name":"Bruno D. Zumbo","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8267139799800711,"gpt":0.6035828191741548,"spread":0.2231311608059163,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1690222,0.002426725,0.002066674,0.0122248,0.001000558,0.003131854,0.002697342,0.00170354,0.002332076],"category_scores_gemma":[0.3703425,0.001081281,0.002805476,0.008472043,0.002625512,0.004866692,0.004246565,0.002573623,0.001313606],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002796349,"about_ca_system_score_gemma":0.00259297,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002046957,"about_ca_topic_score_gemma":0.004030386,"domain_scores_codex":[0.7528875,0.1907703,0.01297969,0.005789268,0.03652649,0.001046679],"domain_scores_gemma":[0.560478,0.3696348,0.02183003,0.01748674,0.0292569,0.001313556],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.003809076,0.0006111019,0.1127061,0.002918369,0.003328169,0.00007035809,0.00304068,0.003614762,0.001788334,0.01610527,0.003874764,0.8481329],"study_design_scores_gemma":[0.00445029,0.01463205,0.6568034,0.005783773,0.004053189,0.002889804,0.007900243,0.1392165,0.01371605,0.1078717,0.04060506,0.002077854],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1566545,0.01165999,0.8044974,0.00205714,0.001011632,0.006342469,0.001127946,0.001722658,0.01492627],"genre_scores_gemma":[0.3093086,0.003924056,0.6738115,0.0004698023,0.0001620112,0.009068754,0.0008828549,0.0002808028,0.002091746],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.1690222,"threshold_uncertainty_score":0.8938854,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1977904348","doi":"10.1177/0013164404264840","title":"Testing the Structure of Psychological Empowerment: Does Gender Make a Difference?","year":2004,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Community Health and Development","field":"Health Professions","cited_by":83,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Western University; Université de Montréal","funders":"","keywords":"Psychology; Confirmatory factor analysis; Competence (human resources); Social psychology; Construct validity; Structural equation modeling; Developmental psychology; Group analysis; Empowerment; Psychometrics; Statistics; Mathematics","authors":[{"name":"Jean‐Sébastien Boudrias","is_ca":true},{"name":"Patrick Gaudreau","is_ca":true},{"name":"Heather K. Spence Laschinger","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3919868797244075,"gpt":0.4792874805514811,"spread":0.08730060082707353,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01368605,0.0003938175,0.0004500757,0.0007823133,0.0007229849,0.0009626962,0.0007855517,0.0006173751,0.007777721],"category_scores_gemma":[0.05409899,0.0002311115,0.0008664705,0.0006442892,0.002812572,0.002193307,0.001477542,0.0008404255,0.0004378335],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006900464,"about_ca_system_score_gemma":0.001575295,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002920931,"about_ca_topic_score_gemma":0.002154206,"domain_scores_codex":[0.9947532,0.003001253,0.0002903087,0.0008057023,0.0007337735,0.0004156906],"domain_scores_gemma":[0.9582815,0.03156959,0.003719569,0.003129241,0.002041653,0.001258586],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001545209,0.000668004,0.9084852,0.000169673,0.00026707,0.0003342053,0.02085996,0.0002100211,0.004344411,0.007506823,0.0007367145,0.05487264],"study_design_scores_gemma":[0.0001550895,0.001994877,0.9674203,0.0001449834,0.0001465266,0.0003196214,0.01790333,0.002032691,0.001781135,0.005901234,0.0021683,0.00003183193],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9936842,0.0001530321,0.00198389,0.0007740734,0.0001136076,0.00008537213,0.0001324053,0.000006263367,0.003067089],"genre_scores_gemma":[0.9984881,0.00005081164,0.000823182,0.000128296,0.00002019044,0.00008298163,0.00009424624,0.000007207558,0.000304956],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01368605,"threshold_uncertainty_score":0.07237965,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3011484504","doi":"10.1177/0013164420911136","title":"Evaluating the Performances of Missing Data Handling Methods in Ability Estimation From Sparse Data","year":2020,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":79,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"","keywords":"Missing data; Imputation (statistics); Statistics; Cart; Data set; Regression; Mean squared error; Mathematics; Computer science; Regression analysis; Data mining; Engineering","authors":[{"name":"Jiaying Xiao","is_ca":true},{"name":"Okan Bulut","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9728668887097844,"gpt":0.6848813766788585,"spread":0.2879855120309259,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.121788,0.001206168,0.001533128,0.002694903,0.000921079,0.001442569,0.00179431,0.002019501,0.001014554],"category_scores_gemma":[0.295025,0.000678451,0.001957717,0.002130731,0.001413927,0.003063027,0.002059276,0.002317348,0.0002548246],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001187276,"about_ca_system_score_gemma":0.002335244,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004171839,"about_ca_topic_score_gemma":0.004536853,"domain_scores_codex":[0.9308642,0.06032222,0.002447632,0.002448373,0.003346636,0.0005710047],"domain_scores_gemma":[0.5272529,0.4387398,0.01153651,0.01281785,0.008629246,0.001023603],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.003897093,0.001058958,0.1590557,0.001802928,0.003817595,0.0002984569,0.003755312,0.4232414,0.002707918,0.01739972,0.001759979,0.3812049],"study_design_scores_gemma":[0.0003627292,0.001463851,0.03057854,0.0004206361,0.000426618,0.0002771145,0.0005108831,0.9443923,0.00409321,0.01585543,0.001400216,0.0002184953],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.3289253,0.001500845,0.6660938,0.0006489591,0.00008296368,0.0007101856,0.0002841722,0.0004362183,0.001317426],"genre_scores_gemma":[0.6100484,0.0004926162,0.3876494,0.0001566788,0.00003487807,0.0008507068,0.0004367455,0.00007363135,0.0002569568],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.121788,"threshold_uncertainty_score":0.6440841,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1981571713","doi":"10.1177/0013164403258407","title":"Balancing the Need for Reliability and Time Efficiency: Short Forms of the Wechsler Adult Intelligence Scale-III","year":2004,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Cognitive Abilities and Testing","field":"Psychology","cited_by":72,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Windsor","funders":"","keywords":"Wechsler Adult Intelligence Scale; Wechsler Preschool and Primary Scale of Intelligence; Psychology; Reliability (semiconductor); Short Forms; Scale (ratio); Wechsler Intelligence Scale for Children; Intelligence quotient; Statistics; Psychometrics; Raw score; Developmental psychology; Clinical psychology; Mathematics; Cognition; Psychiatry; Cartography; Raw data","authors":[{"name":"Sharon L. E. Jeyakumar","is_ca":true},{"name":"Erin M. Warriner","is_ca":true},{"name":"Vaishali V. Raval","is_ca":true},{"name":"Saadia Ahmad","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.08200673823526182,"gpt":0.3386034828471236,"spread":0.2565967446118618,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02619875,0.0006137763,0.0009955235,0.003469749,0.0005825135,0.001305811,0.000777333,0.0005056787,0.003048874],"category_scores_gemma":[0.104757,0.0003574907,0.0008439365,0.003069424,0.0009644304,0.001573907,0.001730004,0.001213692,0.002049203],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005579282,"about_ca_system_score_gemma":0.001264191,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001138794,"about_ca_topic_score_gemma":0.003561597,"domain_scores_codex":[0.9768085,0.01190674,0.004041205,0.0007886215,0.00611239,0.0003424324],"domain_scores_gemma":[0.943731,0.02782471,0.003903775,0.007793777,0.01623251,0.0005142044],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0009915286,0.0003349303,0.1176771,0.0007296276,0.0005297781,0.0002406372,0.003592804,0.001560298,0.004045282,0.01508148,0.08286619,0.7723504],"study_design_scores_gemma":[0.0004216919,0.001694474,0.7553015,0.000844778,0.0005345889,0.002312943,0.002061295,0.008213628,0.009901731,0.03809904,0.1803253,0.0002891016],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.4606813,0.007760885,0.3659449,0.007617565,0.003553404,0.007930641,0.01315319,0.002848829,0.1305093],"genre_scores_gemma":[0.5906684,0.002706852,0.364077,0.001539044,0.0009082091,0.01121201,0.0118621,0.0009722202,0.01605422],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9738013,"threshold_uncertainty_score":0.1385539,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1977836768","doi":"10.1177/0013164402239324","title":"Psychometric Properties Of The French Canadian Version Of The State-Trait Anxiety Inventory For Children","year":2003,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Child and Adolescent Psychosocial and Emotional Development","field":"Psychology","cited_by":70,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"Institut universitaire en santé mentale de Montréal","funders":"","keywords":"State-Trait Anxiety Inventory; Psychology; Anxiety; Psychometrics; Trait anxiety; Trait; Clinical psychology; Developmental psychology; Internal consistency; Psychiatry","authors":[{"name":"Lyse Turgeon","is_ca":true},{"name":"Elise Chartrand","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.07757421657594854,"gpt":0.276566351871343,"spread":0.1989921352953945,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003483572,0.0004766224,0.000413977,0.001818897,0.0008316724,0.0009079008,0.0004594401,0.0002588701,0.001648554],"category_scores_gemma":[0.01086822,0.0001681787,0.0008608153,0.001834474,0.0004141398,0.0003222626,0.0005184744,0.0005183707,0.000243551],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004738129,"about_ca_system_score_gemma":0.005592669,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.4904724,"about_ca_topic_score_gemma":0.5938983,"domain_scores_codex":[0.9981006,0.0003551083,0.0001598874,0.0001521022,0.001003726,0.0002285836],"domain_scores_gemma":[0.9943594,0.001124856,0.0006072246,0.0002210142,0.003337338,0.0003502088],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002503413,0.0001141629,0.8829523,0.0001988961,0.0002068608,0.00009587646,0.001600142,0.000736399,0.000794677,0.0008992028,0.003926785,0.1082243],"study_design_scores_gemma":[0.0000121825,0.0001114419,0.9950787,0.00004048004,0.00005368332,0.00009082413,0.000319608,0.0003783731,0.0001995279,0.00009927675,0.003597699,0.00001829609],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9768642,0.002717087,0.001782937,0.0003840223,0.0001016739,0.0002815708,0.00404109,0.0001018809,0.01372555],"genre_scores_gemma":[0.9870948,0.001636907,0.004636479,0.00009480058,0.0000357681,0.0002769849,0.004067116,0.00002241979,0.002134651],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4904724,"threshold_uncertainty_score":0.975235,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1995196170","doi":"10.1177/0013164403256358","title":"Approximate Confidence Intervals for Effect Sizes","year":2003,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":70,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Manitoba","funders":"","keywords":"Statistics; Confidence interval; Interval (graph theory); Contrast (vision); Coverage probability; Mathematics; Population; Credible interval; Tolerance interval; Econometrics; Demography; Computer science; Combinatorics","authors":[{"name":"James Algina","is_ca":false},{"name":"H. J. Keselman","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7704275307910567,"gpt":0.5976349069448019,"spread":0.1727926238462548,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1591818,0.002345489,0.004802814,0.01024157,0.001259721,0.006258448,0.006613303,0.006251829,0.006979182],"category_scores_gemma":[0.6684543,0.001219833,0.004188057,0.006348432,0.00562123,0.008308349,0.005761041,0.009521582,0.001783609],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003063993,"about_ca_system_score_gemma":0.001960214,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001588459,"about_ca_topic_score_gemma":0.0006210378,"domain_scores_codex":[0.7937118,0.1450487,0.01160324,0.01383749,0.03381471,0.001984033],"domain_scores_gemma":[0.2389018,0.6879552,0.02306334,0.02797077,0.02116275,0.0009460434],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.001935054,0.0002394091,0.008565735,0.006265158,0.002912611,0.0006605287,0.004052635,0.05857016,0.001909634,0.5783422,0.01508408,0.3214628],"study_design_scores_gemma":[0.0009828587,0.00203349,0.01126534,0.008276918,0.001844389,0.002851969,0.001568923,0.1890578,0.007830199,0.6869249,0.0867114,0.0006518218],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.005809091,0.006815638,0.9772785,0.0007973336,0.0006662614,0.0004883201,0.0004032902,0.0007844003,0.006957075],"genre_scores_gemma":[0.286919,0.004937032,0.6955484,0.001735625,0.001549387,0.005922907,0.001138693,0.0006969864,0.001551965],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8408182,"threshold_uncertainty_score":0.8418438,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2158599033","doi":"10.1177/0013164407313368","title":"Factorial Invariance of the Academic Amotivation Inventory (AAI) Across Gender and Grade in a Sample of Canadian High School Students","year":2008,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Motivation and Self-Concept in Sports","field":"Psychology","cited_by":65,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Ottawa; Université du Québec en Outaouais","funders":"","keywords":"Amotivation; Psychology; Measurement invariance; Metric (unit); Sample (material); Internal consistency; Factorial; Construct validity; Confirmatory factor analysis; Developmental psychology; Test validity; Psychometrics; Social psychology; Structural equation modeling; Statistics; Mathematics; Intrinsic motivation","authors":[{"name":"Isabelle Green‐Demers","is_ca":true},{"name":"Lisa Legault","is_ca":true},{"name":"Daniel Pelletier","is_ca":true},{"name":"Luc G. Pelletier","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2520589651374427,"gpt":0.3812125439226223,"spread":0.1291535787851796,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002792601,0.0002856881,0.0003624962,0.001417303,0.001802284,0.001176139,0.0006702704,0.0002420973,0.0009586598],"category_scores_gemma":[0.007057243,0.0002411834,0.0004828005,0.00133291,0.001298583,0.0002583816,0.0007076694,0.0005474114,0.0001272587],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003934702,"about_ca_system_score_gemma":0.005175908,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.6154601,"about_ca_topic_score_gemma":0.695967,"domain_scores_codex":[0.9988783,0.0001454206,0.00007556102,0.0001573654,0.0004986961,0.0002446429],"domain_scores_gemma":[0.9981541,0.0003689644,0.0003310364,0.0002055712,0.000647123,0.0002931643],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00006350619,0.00009226078,0.9856532,0.00001032542,0.00002969515,0.00003231023,0.003027346,0.00005370378,0.0005702708,0.0002236951,0.000183238,0.01006049],"study_design_scores_gemma":[0.000002812128,0.00003815333,0.9988519,0.000002785623,0.000004845682,0.00001880285,0.0007160751,0.00009559465,0.00006161974,0.00003613084,0.0001680605,0.000003198835],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9992296,0.00002169815,0.00008800055,0.00001837399,0.000003247445,0.00002499312,0.00009717373,0.000002301079,0.0005145942],"genre_scores_gemma":[0.9991077,0.00003792578,0.0002620755,0.00001097765,0.000001731642,0.00002596942,0.0003084019,0.000002811113,0.0002424267],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3845399,"threshold_uncertainty_score":0.773609,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1978985895","doi":"10.1177/0013164409344498","title":"Testing the Second-Order Factor Structure and Measurement Equivalence of the Wong and Law Emotional Intelligence Scale Across Gender and Ethnicity","year":2009,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Emotional Intelligence and Performance","field":"Psychology","cited_by":65,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Marriott International (Canada)","funders":"","keywords":"Equivalence (formal languages); Psychology; Ethnic group; Emotional intelligence; Scale (ratio); Social psychology; Context (archaeology); Mathematics; Law; Political science","authors":[{"name":"Daniel S. Whitman","is_ca":false},{"name":"David La Rooy","is_ca":true},{"name":"Chockalingam Viswesvaran","is_ca":false},{"name":"Eyran Kraus","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3337503138767035,"gpt":0.4099216136148018,"spread":0.07617129973809833,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007973208,0.0004068105,0.000351134,0.001040742,0.0004651953,0.0006708449,0.0003545989,0.0002541443,0.00226442],"category_scores_gemma":[0.032617,0.0001690022,0.0005768852,0.0006397177,0.001065657,0.0009207876,0.001088681,0.0005580245,0.0002830949],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003865029,"about_ca_system_score_gemma":0.0009307953,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001835144,"about_ca_topic_score_gemma":0.002091873,"domain_scores_codex":[0.9971015,0.001301347,0.0002856052,0.000348555,0.0007849341,0.0001780251],"domain_scores_gemma":[0.9868141,0.007458281,0.001815868,0.001406595,0.002063241,0.0004418828],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0003924284,0.0002877862,0.9345315,0.00004181516,0.0001768993,0.0001416251,0.004328909,0.0007791134,0.003397892,0.003606178,0.0003511756,0.05196474],"study_design_scores_gemma":[0.00008166151,0.001224604,0.9800262,0.00003729769,0.00005053072,0.0002683688,0.00261682,0.008527734,0.001928833,0.004002926,0.001212373,0.00002271962],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9940212,0.00004785146,0.003982076,0.00006261039,0.0000225361,0.00005777606,0.00004591993,0.00001075953,0.00174909],"genre_scores_gemma":[0.9978786,0.00001871533,0.001706077,0.00001459428,0.000007510943,0.00006521506,0.0001252061,0.00000462648,0.0001795477],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.007973208,"threshold_uncertainty_score":0.04216689,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2143769021","doi":"10.1177/0013164406296976","title":"The Impact of Outliers on Cronbach's Coefficient Alpha Estimate of Reliability: Visual Analogue Scales","year":2007,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":63,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Cronbach's alpha; Outlier; Statistics; Sample (material); Population; Reliability (semiconductor); Statistic; Context (archaeology); Mathematics; Econometrics; Psychometrics; Demography; Geography; Chemistry","authors":[{"name":"Yan Liu","is_ca":true},{"name":"Bruno D. Zumbo","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2458833365065614,"gpt":0.5254813432007601,"spread":0.2795980066941987,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1008727,0.000892806,0.001740259,0.003892526,0.001015753,0.003302963,0.001740069,0.0009557324,0.001886758],"category_scores_gemma":[0.5628919,0.0007039043,0.001363107,0.005540367,0.003526385,0.002901489,0.00280266,0.002470445,0.000626791],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001053608,"about_ca_system_score_gemma":0.001031365,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001263715,"about_ca_topic_score_gemma":0.001463829,"domain_scores_codex":[0.8172595,0.1171513,0.01428018,0.007310149,0.04273529,0.001263574],"domain_scores_gemma":[0.2725553,0.6298522,0.02966625,0.03317141,0.03356296,0.001191899],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.002955871,0.0003310863,0.4529665,0.003083132,0.003239311,0.0008181168,0.02436249,0.01480646,0.008272148,0.03304342,0.01630089,0.4398205],"study_design_scores_gemma":[0.0002348227,0.003235075,0.7670975,0.002037335,0.00161297,0.002683948,0.01256075,0.04821462,0.02148692,0.106354,0.03375062,0.0007314953],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5899045,0.005353726,0.3698321,0.001689881,0.001327753,0.0007005275,0.0009409719,0.0009265426,0.02932412],"genre_scores_gemma":[0.9416931,0.0005289436,0.05515165,0.0002951319,0.0001649416,0.0005582468,0.0003411137,0.0003657097,0.0009012345],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8991273,"threshold_uncertainty_score":0.5334722,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2160937459","doi":"10.1177/0013164406288161","title":"Confidence Interval Coverage for Cohen's Effect Size Statistic","year":2006,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":62,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Manitoba","funders":"","keywords":"Statistics; Confidence interval; Percentile; Statistic; Mathematics; Sample size determination; Coverage probability; Population; Robust confidence intervals; Truncated mean; Econometrics; Medicine","authors":[{"name":"James Algina","is_ca":false},{"name":"H. J. Keselman","is_ca":true},{"name":"Randall D. Penfield","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2409555901132229,"gpt":0.467946665240317,"spread":0.2269910751270941,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2153805,0.00141115,0.003852905,0.01727617,0.001728183,0.004466175,0.005948187,0.005171883,0.007743477],"category_scores_gemma":[0.795851,0.001021055,0.004870539,0.009613401,0.005667309,0.006883324,0.006537952,0.00608347,0.001606253],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003245538,"about_ca_system_score_gemma":0.002606722,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002646982,"about_ca_topic_score_gemma":0.00142325,"domain_scores_codex":[0.6795232,0.1931442,0.02911885,0.02451041,0.07151844,0.002184798],"domain_scores_gemma":[0.1578421,0.7421102,0.02172553,0.03116147,0.04595248,0.001208201],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.004078426,0.0003809217,0.04664113,0.0128397,0.006749926,0.0007237263,0.007473453,0.01872005,0.00216035,0.2945437,0.06424128,0.5414473],"study_design_scores_gemma":[0.002153245,0.002880708,0.08668202,0.01996419,0.006014736,0.00447019,0.004583667,0.1006359,0.01950638,0.5448179,0.2068474,0.001443637],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.03021964,0.0206468,0.8899279,0.005307443,0.003589815,0.00320445,0.003186631,0.002367435,0.04154989],"genre_scores_gemma":[0.5017838,0.002669081,0.4755859,0.003621018,0.001035875,0.01102162,0.00163327,0.001185004,0.001464462],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.7846195,"threshold_uncertainty_score":0.9675755,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2309234137","doi":"10.1177/0013164415596421","title":"Improving the Factor Structure of Psychological Scales","year":2015,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Child and Adolescent Psychosocial and Emotional Development","field":"Psychology","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 British Columbia","funders":"","keywords":"Likert scale; Psychology; Scale (ratio); Confirmatory factor analysis; Exploratory factor analysis; Acquiescence; Guttman scale; Psychometrics; Social psychology; Statistics; Clinical psychology; Structural equation modeling; Developmental psychology; Mathematics","authors":[{"name":"Xijuan Zhang","is_ca":true},{"name":"Victoria Savalei","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1701546745592259,"gpt":0.3575145933061743,"spread":0.1873599187469484,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0535632,0.001996635,0.001312564,0.003138192,0.0009470576,0.002637942,0.001032499,0.000788486,0.007156626],"category_scores_gemma":[0.1877156,0.0007389382,0.002557775,0.003254511,0.0008403275,0.003166,0.002004024,0.00225735,0.002157954],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007449705,"about_ca_system_score_gemma":0.001581419,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001096207,"about_ca_topic_score_gemma":0.001483559,"domain_scores_codex":[0.9750011,0.01643933,0.00270147,0.002469364,0.002844632,0.0005441285],"domain_scores_gemma":[0.8963133,0.06616926,0.002407154,0.01113603,0.02347692,0.0004973845],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0008101315,0.0007638417,0.06505643,0.001308967,0.0007088138,0.0002212357,0.008032576,0.007400376,0.01372488,0.01544623,0.01193886,0.8745876],"study_design_scores_gemma":[0.002013457,0.004677021,0.3428464,0.002876335,0.001788698,0.0008077943,0.01223043,0.3392175,0.03521235,0.1513902,0.1060306,0.0009092003],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1531443,0.0004681765,0.8334032,0.0008256822,0.0005725066,0.00398598,0.001204719,0.001748186,0.00464723],"genre_scores_gemma":[0.3315845,0.0002868309,0.6579429,0.0001694395,0.0001430538,0.005953148,0.002278843,0.0003948694,0.001246442],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.0535632,"threshold_uncertainty_score":0.2832727,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2568498297","doi":"10.1177/0013164416685599","title":"Retrofitting Diagnostic Classification Models to Responses From IRT-Based Assessment Forms","year":2017,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":55,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"","keywords":"Retrofitting; Item response theory; Computer science; Process (computing); Test (biology); Data mining; Engineering; Psychometrics; Mathematics; Statistics","authors":[{"name":"Ren Liu","is_ca":false},{"name":"Anne Corinne Huggins‐Manley","is_ca":false},{"name":"Okan Bulut","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8783454710157835,"gpt":0.5694813197426731,"spread":0.3088641512731104,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1992908,0.002216241,0.001713756,0.005093104,0.001462497,0.006025801,0.003623856,0.001901868,0.001732066],"category_scores_gemma":[0.526748,0.001193552,0.001916026,0.005317295,0.003018213,0.005076195,0.005054854,0.004693659,0.001769192],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004165726,"about_ca_system_score_gemma":0.003505987,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003010024,"about_ca_topic_score_gemma":0.003855574,"domain_scores_codex":[0.7431158,0.2016473,0.01485981,0.00947623,0.02925548,0.001645367],"domain_scores_gemma":[0.5182543,0.3248301,0.02558102,0.07802348,0.05236841,0.0009426157],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001100025,0.001110781,0.1449798,0.001385561,0.0009200572,0.0004110956,0.0153471,0.03496497,0.004592529,0.04599425,0.008160531,0.7410332],"study_design_scores_gemma":[0.0004636929,0.002700757,0.08964564,0.002594061,0.0009024887,0.001656148,0.01344286,0.5743397,0.02716703,0.2440321,0.04231981,0.0007355958],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1258086,0.0004190543,0.8598197,0.001759964,0.0003091298,0.00310381,0.0003874993,0.001645021,0.006747192],"genre_scores_gemma":[0.4852329,0.0002584057,0.5063646,0.0009392323,0.00009100777,0.003972671,0.001176371,0.0003630456,0.001601881],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.1992908,"threshold_uncertainty_score":0.987417,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2081705014","doi":"10.1177/0013164414527449","title":"Seeing Perfectly Fitting Factor Models That Are Causally Misspecified","year":2014,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":53,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"","keywords":"Factor (programming language); Factor analysis; Causal model; Econometrics; Causal structure; Psychology; Statistics; Mathematics; Computer science","authors":[{"name":"Leslie A. Hayduk","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8554756584236994,"gpt":0.4901287910866036,"spread":0.3653468673370959,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.09443253,0.002555092,0.001953439,0.005058376,0.004117135,0.01071129,0.002775316,0.003482291,0.00839173],"category_scores_gemma":[0.4382975,0.002060717,0.002308214,0.004933755,0.01635314,0.02390906,0.007653606,0.009037802,0.001599362],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00474916,"about_ca_system_score_gemma":0.006783951,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006084021,"about_ca_topic_score_gemma":0.00622576,"domain_scores_codex":[0.9075344,0.05963896,0.004521483,0.008298874,0.01846919,0.001537138],"domain_scores_gemma":[0.6456573,0.2354351,0.028209,0.06843469,0.01994619,0.002317831],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0003749962,0.0001955212,0.0405574,0.001036733,0.0008070141,0.001044998,0.02703976,0.01258655,0.001606913,0.7771694,0.01715642,0.1204243],"study_design_scores_gemma":[0.00004095855,0.00009241998,0.005282768,0.0006207416,0.0001327624,0.0005185347,0.005971455,0.01983306,0.001397055,0.9487453,0.01721033,0.0001547479],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.07365671,0.002192219,0.872153,0.02603553,0.001227382,0.0002558283,0.0005775931,0.001326683,0.02257516],"genre_scores_gemma":[0.780479,0.00159673,0.2101876,0.003590531,0.0003634364,0.0003910061,0.0007021316,0.000564537,0.002125056],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9055675,"threshold_uncertainty_score":0.499413,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2941810097","doi":"10.1177/0013164419844305","title":"A Measurement Is a Choice and Stevens’ Scales of Measurement Do Not Help Make It: A Response to Chalmers","year":2019,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":51,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Relevance (law); Confusion; Reliability (semiconductor); Task (project management); Psychology; Test (biology); Epistemology; Measure (data warehouse); Computer science; Philosophy; Psychoanalysis; Data mining; Economics","authors":[{"name":"Bruno D. Zumbo","is_ca":true},{"name":"Edward Kroc","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4019735413709748,"gpt":0.46501596832292,"spread":0.06304242695194523,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1031428,0.001764592,0.003213799,0.004982799,0.007084665,0.01488848,0.005996244,0.03047504,0.002487695],"category_scores_gemma":[0.2958417,0.001480288,0.002091922,0.005532173,0.07282513,0.02587717,0.007806927,0.07595263,0.002315706],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01161717,"about_ca_system_score_gemma":0.00843697,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008271144,"about_ca_topic_score_gemma":0.005951272,"domain_scores_codex":[0.858235,0.09077865,0.009191961,0.01178312,0.02819259,0.001818522],"domain_scores_gemma":[0.5638769,0.3698992,0.008753917,0.01270807,0.04181478,0.002947015],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00008702729,0.00003268319,0.0004102307,0.0002545789,0.00004474529,0.0001848956,0.008900225,0.000280496,0.0001803366,0.4930561,0.4773862,0.01918242],"study_design_scores_gemma":[0.00006843042,0.00005380696,0.0006852086,0.001458877,0.00004010555,0.000578208,0.006588859,0.001737564,0.0006086336,0.6069768,0.3809377,0.0002656783],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.0005104172,0.004641213,0.004253727,0.983027,0.005949965,0.00001142323,0.00002354885,0.00004350796,0.001539173],"genre_scores_gemma":[0.0521115,0.006485448,0.01524166,0.8958993,0.02597964,0.000241471,0.00004430398,0.000235657,0.003761056],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8968572,"threshold_uncertainty_score":0.5454777,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2139940744","doi":"10.1177/0013164404268668","title":"Reliability Generalization of Responses by Care Providers to the Center for Epidemiologic Studies-Depression Scale","year":2004,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Mental Health Treatment and Access","field":"Psychology","cited_by":51,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Simon Fraser University","funders":"","keywords":"Center for Epidemiologic Studies Depression Scale; Reliability (semiconductor); Scale (ratio); Generalization; Psychology; Clinical psychology; Depression (economics); Psychometrics; Item response theory; Gerontology; Psychiatry; Depressive symptoms; Medicine; Cognition; Mathematics","authors":[{"name":"Norm O’Rourke","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2445451191917063,"gpt":0.4761799233039707,"spread":0.2316348041122644,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.08669668,0.0003750985,0.001257024,0.003145633,0.0005195044,0.001034383,0.0008813088,0.0004933975,0.0008299953],"category_scores_gemma":[0.1548018,0.0005264113,0.002255648,0.00273546,0.0008700893,0.0006812155,0.001541067,0.0008895873,0.000236981],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006927575,"about_ca_system_score_gemma":0.0006277083,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002090549,"about_ca_topic_score_gemma":0.003649184,"domain_scores_codex":[0.9422205,0.03797644,0.008360803,0.003224885,0.007757574,0.0004598459],"domain_scores_gemma":[0.8782243,0.07817225,0.01307853,0.01097328,0.01907425,0.0004774173],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0006282303,0.0001103454,0.9388577,0.001046299,0.007064506,0.00005195525,0.003167863,0.0004801608,0.0008395207,0.0004826029,0.002073949,0.04519677],"study_design_scores_gemma":[0.0001369886,0.0005596632,0.9878953,0.0005001519,0.00246601,0.0002190051,0.001477504,0.001237251,0.001219747,0.0009072947,0.00332334,0.00005780763],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"methods","genre_scores_codex":[0.9439468,0.01271504,0.02640793,0.0009366084,0.0003435469,0.00163531,0.004295266,0.0001614118,0.009558076],"genre_scores_gemma":[0.9868742,0.001656137,0.00719308,0.0002625107,0.00007922967,0.0008399384,0.002628219,0.00003274979,0.0004339395],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.08669668,"threshold_uncertainty_score":0.4585014,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1973081171","doi":"10.1177/0013164408322017","title":"A Cross-Cultural Examination of the Psychometric Properties of Responses to the Achievement Goal Questionnaire","year":2008,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Education, Achievement, and Giftedness","field":"Psychology","cited_by":51,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"Simon Fraser University","funders":"","keywords":"Psychology; Confirmatory factor analysis; Structural equation modeling; Measurement invariance; Metric (unit); Psychometrics; Test validity; Academic achievement; Developmental psychology; Social psychology; Statistics; Mathematics","authors":[{"name":"Kou Murayama","is_ca":false},{"name":"Mingming Zhou","is_ca":true},{"name":"John C. Nesbit","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.205160770575988,"gpt":0.388235285087074,"spread":0.183074514511086,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01084872,0.0003016804,0.0003621715,0.001540828,0.0008746164,0.000646867,0.0003467855,0.0003401025,0.0008524554],"category_scores_gemma":[0.02387096,0.0002379649,0.0006608287,0.001555097,0.0009375964,0.00044922,0.001005455,0.0007438599,0.000199613],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006670722,"about_ca_system_score_gemma":0.0008729558,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008786448,"about_ca_topic_score_gemma":0.01359873,"domain_scores_codex":[0.9964877,0.001483602,0.0004499592,0.0003729876,0.001038771,0.0001669417],"domain_scores_gemma":[0.9737784,0.01179053,0.003597766,0.002700607,0.007170162,0.0009625412],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001865621,0.0002869925,0.9473292,0.00007109255,0.000373139,0.0001282482,0.01531192,0.0002060017,0.003654974,0.0003757741,0.0005183576,0.03155771],"study_design_scores_gemma":[0.0000136054,0.0002383494,0.9926687,0.00002251953,0.00004542047,0.0001844154,0.004259252,0.0003632246,0.0008077688,0.000115851,0.001257172,0.00002357756],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9977609,0.00006449643,0.0005011855,0.0000510754,0.00001898899,0.00004491585,0.0001287152,0.000004809339,0.00142478],"genre_scores_gemma":[0.9980737,0.00007499825,0.000958677,0.00006033144,0.000006890254,0.00006666215,0.0004092933,0.000008249874,0.0003410634],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01084872,"threshold_uncertainty_score":0.05737424,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1966169084","doi":"10.1177/0013164410387382","title":"Invariance of the Measurement Model Underlying the Wechsler Adult Intelligence Scale-IV in the United States and Canada","year":2010,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Cognitive Abilities and Testing","field":"Psychology","cited_by":48,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Calgary; BC Mental Health & Substance Use Services; University of British Columbia","funders":"American Psychological Association; American Educational Research Association","keywords":"Psychology; Wechsler Adult Intelligence Scale; Measurement invariance; Normative; Generality; Construct validity; Intelligence quotient; Latent variable; Wechsler Preschool and Primary Scale of Intelligence; Psychometrics; Cognition; Developmental psychology; Construct (python library); Scale (ratio); Set (abstract data type); Latent variable model; Wechsler Intelligence Scale for Children; Sample (material); Statistics; Confirmatory factor analysis; Structural equation modeling; Mathematics","authors":[{"name":"Stephen C. Bowden","is_ca":false},{"name":"Donald H. Saklofske","is_ca":true},{"name":"Lawrence G. Weiss","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2390830827106653,"gpt":0.3553275337429329,"spread":0.1162444510322676,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01735444,0.000391271,0.0006133231,0.001868889,0.002531677,0.002120054,0.001182841,0.0003104113,0.001457305],"category_scores_gemma":[0.04863909,0.000339772,0.00106156,0.002779175,0.003011089,0.0008633182,0.001425598,0.001061662,0.0002110184],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01884584,"about_ca_system_score_gemma":0.03347472,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.921281,"about_ca_topic_score_gemma":0.8928343,"domain_scores_codex":[0.9857866,0.004543201,0.0007904061,0.001877903,0.005724705,0.001277139],"domain_scores_gemma":[0.9847156,0.004395668,0.001009148,0.001624433,0.007751822,0.0005033769],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002769996,0.0001165694,0.7684911,0.0001542829,0.0005787726,0.0002519818,0.00571142,0.00925592,0.001010599,0.08587023,0.006813721,0.1214684],"study_design_scores_gemma":[0.00005056174,0.00009403777,0.9510754,0.0002153244,0.0001686306,0.0001785599,0.002794189,0.01726325,0.000737808,0.01816459,0.009164607,0.00009299513],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8356461,0.001879073,0.09027081,0.003158946,0.000299779,0.001717631,0.004481708,0.0002851632,0.06226087],"genre_scores_gemma":[0.9835427,0.0004042878,0.01259496,0.0001749809,0.00002123092,0.0003237398,0.001741427,0.00004732166,0.001149336],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07871902,"threshold_uncertainty_score":0.1583652,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2026405046","doi":"10.1177/0013164404268675","title":"Effect Sizes and their Intervals: The Two-Level Repeated Measures Case","year":2005,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":45,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Manitoba","funders":"","keywords":"Statistics; Mathematics; Confidence interval; Estimator; Sample size determination; Variance (accounting); Coverage probability; Population; Econometrics; Demography","authors":[{"name":"James Algina","is_ca":false},{"name":"H. J. Keselman","is_ca":true},{"name":"Randall D. Penfield","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7635696793982186,"gpt":0.5823470726920558,"spread":0.1812226067061629,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3078378,0.001797252,0.005857956,0.008233403,0.001059847,0.005130376,0.006019872,0.006889583,0.005577095],"category_scores_gemma":[0.7641568,0.001339737,0.006169863,0.005609395,0.009275529,0.008243088,0.004456731,0.005810344,0.0005949762],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002338836,"about_ca_system_score_gemma":0.001607798,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007842929,"about_ca_topic_score_gemma":0.0004478932,"domain_scores_codex":[0.6321734,0.2598607,0.02269395,0.03030143,0.05198852,0.002981982],"domain_scores_gemma":[0.1286213,0.7836609,0.03310798,0.04648754,0.007257686,0.000864533],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.01418844,0.0010359,0.1256952,0.008694645,0.02440117,0.005261194,0.01161068,0.04332822,0.005362151,0.2858289,0.006517736,0.4680758],"study_design_scores_gemma":[0.003053326,0.01147009,0.1264007,0.003984662,0.01432763,0.01421122,0.004010124,0.2861941,0.01250022,0.4969081,0.02581421,0.001125635],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1286011,0.01278393,0.8431151,0.001761673,0.0006078964,0.003166807,0.00101799,0.0007673077,0.008178186],"genre_scores_gemma":[0.7579745,0.001632148,0.2308839,0.0005611402,0.0004463163,0.00698944,0.0005935504,0.0001822816,0.0007366608],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.6921622,"threshold_uncertainty_score":0.8535593,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2016558976","doi":"10.1177/0013164402062004013","title":"Stability of the Reliability of LibQual+™ Scores a Reliability Generalization Meta-Analysis Study","year":2002,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Digital Mental Health Interventions","field":"Psychology","cited_by":45,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Generalization; Reliability (semiconductor); Meta-analysis; Psychology; Statistics; Scale (ratio); Mathematics; Medicine; Geography","authors":[{"name":"Bruce Thompson","is_ca":false},{"name":"Colleen Cook","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4337064491574427,"gpt":0.4399387765720935,"spread":0.006232327414650796,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.07466488,0.001155581,0.003517151,0.002156355,0.0005585457,0.001998423,0.001141231,0.0007994667,0.0007195793],"category_scores_gemma":[0.1154929,0.0009972483,0.006367256,0.00195047,0.001332661,0.001056572,0.001269878,0.0009715636,0.0001602433],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001105849,"about_ca_system_score_gemma":0.001376928,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002793301,"about_ca_topic_score_gemma":0.004481302,"domain_scores_codex":[0.9606385,0.02816046,0.003225075,0.003655705,0.00396653,0.000353744],"domain_scores_gemma":[0.9039112,0.06902353,0.008787248,0.01156251,0.006215275,0.0005002064],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","study_design_scores_codex":[0.01260765,0.0004418436,0.353006,0.01540041,0.4665436,0.0003550049,0.002314871,0.004547618,0.004034784,0.0004895355,0.001939077,0.1383197],"study_design_scores_gemma":[0.003077442,0.007377129,0.4514095,0.002778956,0.5135196,0.001166263,0.0007628944,0.003838215,0.005951783,0.001551839,0.008320473,0.0002458745],"study_design_candidate":"meta_analysis","study_design_consensus":"meta_analysis","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8137792,0.16052,0.01862338,0.0007927153,0.0004653942,0.000929633,0.001584247,0.0002664912,0.003039032],"genre_scores_gemma":[0.9877511,0.007093575,0.003496045,0.0002447343,0.0001309329,0.0002661803,0.0007359965,0.00004414699,0.0002372175],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9964828,"threshold_uncertainty_score":0.3948703,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2168109053","doi":"10.1177/0013164413492419","title":"Effectiveness of Combining Statistical Tests and Effect Sizes When Using Logistic Discriminant Function Regression to Detect Differential Item Functioning for Polytomous Items","year":2013,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":43,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Statistics; Differential item functioning; Polytomous Rasch model; Discriminant function analysis; Sample size determination; Mathematics; Logistic regression; Linear discriminant analysis; Psychology; Statistical hypothesis testing; Econometrics; Psychometrics; Item response theory","authors":[{"name":"Juana Gómez‐Benito","is_ca":false},{"name":"Bruno D. Zumbo","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5320350509278347,"gpt":0.4801596014508845,"spread":0.05187544947695016,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2512179,0.004031109,0.006580664,0.00864746,0.001239314,0.004941293,0.003548208,0.00416986,0.001864635],"category_scores_gemma":[0.5401216,0.00186396,0.004293303,0.005805308,0.00399727,0.008270468,0.003881269,0.004516764,0.0009083062],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001213176,"about_ca_system_score_gemma":0.001985318,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004987631,"about_ca_topic_score_gemma":0.0008603135,"domain_scores_codex":[0.634201,0.2915775,0.02262489,0.02150294,0.02832512,0.001768529],"domain_scores_gemma":[0.198022,0.7619671,0.01571555,0.01638482,0.006590351,0.00132017],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.02357868,0.002277807,0.211456,0.002210897,0.01002684,0.001415915,0.001530517,0.07319367,0.02161825,0.006829483,0.002567287,0.6432947],"study_design_scores_gemma":[0.003335606,0.02418924,0.1426817,0.001076868,0.006908021,0.003266482,0.001338742,0.7064618,0.06194783,0.0397512,0.007429078,0.001613383],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3835739,0.002459612,0.6021752,0.001065653,0.0003191205,0.002289174,0.000376511,0.002926589,0.004814204],"genre_scores_gemma":[0.534894,0.000280232,0.462112,0.000388493,0.00015415,0.001363795,0.0002354433,0.0002615452,0.0003103474],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7487822,"threshold_uncertainty_score":0.9233817,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1975494033","doi":"10.1177/0013164413487738","title":"False Positives in Multiple Regression","year":2013,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":42,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Type I and type II errors; Statistics; Regression; False positive paradox; Regression analysis; Econometrics; Observational error; Regression diagnostic; Nominal level; Inflation (cosmology); Computer science; Linear regression; False positives and false negatives; Mathematics; Polynomial regression; Confidence interval","authors":[{"name":"Benjamin R. Shear","is_ca":true},{"name":"Bruno D. Zumbo","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.450798048209222,"gpt":0.4850311437518074,"spread":0.03423309554258541,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3902152,0.002433944,0.00471727,0.006002978,0.002561534,0.007766079,0.005515475,0.008163368,0.003059601],"category_scores_gemma":[0.7631719,0.00173635,0.003706186,0.005525352,0.01343503,0.008979673,0.006331258,0.01024615,0.001264196],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004068804,"about_ca_system_score_gemma":0.003253435,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001535182,"about_ca_topic_score_gemma":0.001260135,"domain_scores_codex":[0.3761152,0.4796037,0.0294088,0.0374252,0.07504416,0.002402927],"domain_scores_gemma":[0.1266476,0.7662559,0.04363348,0.04157297,0.02087446,0.001015699],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.002400151,0.0004415598,0.08810469,0.006320099,0.005298319,0.00610731,0.01504061,0.02103519,0.002027546,0.4490681,0.03849889,0.3656576],"study_design_scores_gemma":[0.0005227712,0.0008230583,0.02734658,0.005658079,0.001559893,0.006082083,0.00260744,0.06385528,0.007832984,0.8337375,0.04934424,0.0006301118],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.03463321,0.01377558,0.9046436,0.01618336,0.006716312,0.001814586,0.000626518,0.00103919,0.02056756],"genre_scores_gemma":[0.5604682,0.004587308,0.3962283,0.02347581,0.003724508,0.004779649,0.0004845632,0.0005397956,0.005711957],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.6097848,"threshold_uncertainty_score":0.7519732,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2037678252","doi":"10.1177/00131640021970998","title":"Testing Repeated Measures Hypotheses When Covariance Matrices are Heterogeneous: Revisiting the Robustness of the Welch-James Test Again","year":2000,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":42,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Manitoba","funders":"","keywords":"Statistics; Bootstrapping (finance); Estimator; Mathematics; Covariance; Robustness (evolution); Type I and type II errors; Sample size determination; Repeated measures design; Covariance matrix; Statistical hypothesis testing; Analysis of covariance; Econometrics","authors":[{"name":"H. J. Keselman","is_ca":true},{"name":"James Algina","is_ca":false},{"name":"Rand R. Wilcox","is_ca":false},{"name":"Rhonda K. Kowa","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.350129595397415,"gpt":0.4037648988950279,"spread":0.05363530349761286,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2089843,0.00155614,0.003132551,0.003951907,0.002216222,0.004192012,0.004479904,0.002602583,0.003888748],"category_scores_gemma":[0.6522658,0.00103556,0.003373113,0.00387179,0.01074369,0.007337268,0.004064492,0.005482898,0.0005417804],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001435237,"about_ca_system_score_gemma":0.003367355,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002990412,"about_ca_topic_score_gemma":0.001495474,"domain_scores_codex":[0.8005704,0.1401763,0.009676896,0.02329201,0.02426554,0.002018895],"domain_scores_gemma":[0.1736051,0.753233,0.01609406,0.04393583,0.01179147,0.001340568],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00681205,0.001006067,0.1935474,0.002190396,0.01322395,0.003774162,0.02073506,0.03481223,0.01319567,0.224079,0.005391664,0.4812324],"study_design_scores_gemma":[0.001060476,0.008105994,0.2260604,0.0008354219,0.003150727,0.002095971,0.005293248,0.2826138,0.02327819,0.4279141,0.01873584,0.0008559331],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1722465,0.0009351235,0.8184868,0.001336311,0.0008260189,0.0005486699,0.0002423145,0.0005504141,0.004827939],"genre_scores_gemma":[0.8181394,0.0002382751,0.1783733,0.0005051754,0.0004649899,0.0007610613,0.0001964887,0.0002867738,0.001034459],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.2089843,"threshold_uncertainty_score":0.9754632,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2066906964","doi":"10.1177/0013164409344520","title":"Athletes’ Perceptions of Coaching Competency Scale II-High School Teams","year":2009,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Sport Psychology and Performance","field":"Psychology","cited_by":40,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Psychology; Coaching; Athletes; Confirmatory factor analysis; Applied psychology; Construct validity; Scale (ratio); Construct (python library); Exploratory factor analysis; Structural equation modeling; Clinical psychology; Psychometrics; Statistics; Physical therapy; Mathematics; Medicine","authors":[{"name":"Nicholas D. Myers","is_ca":false},{"name":"Melissa A. Chase","is_ca":false},{"name":"Mark R. Beauchamp","is_ca":true},{"name":"Ben Jackson","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.06080880170417884,"gpt":0.3496966861001845,"spread":0.2888878843960057,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002792011,0.0002903626,0.0004133336,0.001266055,0.0007082658,0.0009402641,0.0003416002,0.0004807165,0.002049123],"category_scores_gemma":[0.004918126,0.0002180563,0.0004366648,0.0005134539,0.000506934,0.00048293,0.001023834,0.0009032162,0.0003171445],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003912984,"about_ca_system_score_gemma":0.0006694653,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003249413,"about_ca_topic_score_gemma":0.007531168,"domain_scores_codex":[0.9983596,0.0004431729,0.0002165803,0.0001310005,0.0005866638,0.00026299],"domain_scores_gemma":[0.9953464,0.0007175156,0.001771759,0.0001854449,0.0009630831,0.001015825],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001498092,0.00066522,0.9807085,0.00005569101,0.0001027161,0.00008435518,0.003361312,0.0002412093,0.002065177,0.0001786668,0.0008009726,0.0115864],"study_design_scores_gemma":[0.00001832469,0.0003158365,0.9960184,0.00002059807,0.00001335639,0.00007373343,0.002401187,0.0002756041,0.0002504856,0.00007590636,0.0005287919,0.000007897911],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9971956,0.00002854276,0.0001897992,0.00003914843,0.000007892757,0.0000976186,0.00007944158,0.000003741086,0.00235828],"genre_scores_gemma":[0.9979931,0.00005957083,0.0007074401,0.00004324927,0.00001086112,0.0001764093,0.000249638,0.000003711237,0.000756104],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.003249413,"threshold_uncertainty_score":0.01476574,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2019602618","doi":"10.1177/0013164404267277","title":"Validity Studies Psychometric Properties of Scores on the French and Korean Versions of the Hexaco Personality Inventory","year":2004,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Personality Traits and Psychology","field":"Psychology","cited_by":39,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Université du Québec à Montréal; Brock University; University of Calgary; Concordia University","funders":"Agence Nationale de la Recherche","keywords":"Psychology; Personality; Internal consistency; Psychometrics; Personality Assessment Inventory; California Psychological Inventory; Personality test; Big Five personality traits; Scale (ratio); Test validity; Social psychology; Clinical psychology","authors":[{"name":"Kathleen Boies","is_ca":true},{"name":"Tae-Yong Yoo","is_ca":false},{"name":"Annik Ebacher","is_ca":true},{"name":"Kibeom Lee","is_ca":true},{"name":"Michael C. Ashton","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4726173115767044,"gpt":0.3908928201723001,"spread":0.08172449140440435,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005514913,0.0004522269,0.000332344,0.001527986,0.0006603553,0.00064984,0.0003289717,0.0002902629,0.00233787],"category_scores_gemma":[0.01849396,0.0001617668,0.0007341148,0.001241715,0.0007650724,0.0006009899,0.0008485633,0.0005169191,0.0004354028],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006409987,"about_ca_system_score_gemma":0.000823434,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004604862,"about_ca_topic_score_gemma":0.007101,"domain_scores_codex":[0.9973532,0.0007795158,0.0004103574,0.0003812853,0.0008797082,0.0001960644],"domain_scores_gemma":[0.9893065,0.003604712,0.00221294,0.000776089,0.003486939,0.0006128203],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002688726,0.0002229138,0.9551677,0.0001177362,0.0003075471,0.0001016776,0.001351481,0.0002494905,0.001346413,0.000669534,0.0007041495,0.03949258],"study_design_scores_gemma":[0.00002616262,0.0002126275,0.9958175,0.00003936226,0.00005665447,0.0001800256,0.0008523804,0.0003617322,0.0005431607,0.0002790442,0.001617249,0.00001425528],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9899151,0.0004432228,0.001345246,0.0001485704,0.00005861596,0.0002187022,0.0006800599,0.00001874928,0.007171618],"genre_scores_gemma":[0.9942926,0.000284741,0.002203256,0.00009489939,0.00003193413,0.0003144673,0.001411434,0.0000153506,0.001351365],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.005514913,"threshold_uncertainty_score":0.02916604,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2912880345","doi":"10.1177/0013164419829855","title":"Item Response Tree Models to Investigate Acquiescence and Extreme Response Styles in Likert-Type Rating Scales","year":2019,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":39,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Acquiescence; Psychology; Likert scale; Rating scale; Item response theory; Social psychology; Explanatory model; Scale (ratio); Psychometrics; Statistics; Developmental psychology; Mathematics","authors":[{"name":"Minjeong Park","is_ca":true},{"name":"Amery D. Wu","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7658315465227954,"gpt":0.4754891438004715,"spread":0.2903424027223239,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06369835,0.002504179,0.001673731,0.003937178,0.0008310747,0.003113451,0.003140342,0.001951387,0.006341824],"category_scores_gemma":[0.1603064,0.001201342,0.004128161,0.007071619,0.00173175,0.004632549,0.002464921,0.004146539,0.00242376],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002111912,"about_ca_system_score_gemma":0.001527028,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002596328,"about_ca_topic_score_gemma":0.003159026,"domain_scores_codex":[0.9230632,0.06641823,0.002183153,0.003842017,0.003832211,0.0006612543],"domain_scores_gemma":[0.790507,0.1781452,0.01193205,0.01040522,0.008137131,0.0008733628],"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":[0.001007857,0.001627261,0.258229,0.001487194,0.003558364,0.0005062705,0.01476259,0.2186975,0.002265572,0.23736,0.01544191,0.2450565],"study_design_scores_gemma":[0.0001478544,0.0006051795,0.02725408,0.0003049467,0.0002820228,0.0003017643,0.00160474,0.8135861,0.000684037,0.1469285,0.008102952,0.0001979088],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.04385599,0.0002409396,0.9494154,0.00044379,0.00007822646,0.001292647,0.001173436,0.0007140195,0.002785609],"genre_scores_gemma":[0.4924958,0.0004435661,0.4938554,0.0004402772,0.00009646725,0.007609114,0.002529388,0.0002634081,0.002266539],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.06369835,"threshold_uncertainty_score":0.3368731,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2014421030","doi":"10.1177/00131640021970970","title":"Structural Equation Models and the Regression Bias for Measuring Correlates of Change","year":2000,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":38,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Lakehead University; University of Manitoba","funders":"","keywords":"Structural equation modeling; Skewness; Analysis of covariance; Regression analysis; Regression; Statistics; Psychology; Linear regression; Econometrics; Regression toward the mean; Covariance; Mathematics","authors":[{"name":"Robert A. Cribbie","is_ca":true},{"name":"John Jamieson","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9410559504515326,"gpt":0.523874003364629,"spread":0.4171819470869036,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1004338,0.001931366,0.002333045,0.004468876,0.001239184,0.00342319,0.002126246,0.002248523,0.004834599],"category_scores_gemma":[0.3616139,0.001337728,0.002057039,0.008421897,0.004220687,0.005493256,0.002759126,0.005865569,0.001266781],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002631168,"about_ca_system_score_gemma":0.003884929,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005290819,"about_ca_topic_score_gemma":0.006231091,"domain_scores_codex":[0.8753643,0.103792,0.003659196,0.004413667,0.01208899,0.0006818061],"domain_scores_gemma":[0.6912038,0.2677454,0.01620073,0.01619937,0.00808679,0.0005639958],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0003323886,0.000217271,0.06514169,0.001349979,0.002199035,0.0002089863,0.002116113,0.04955772,0.0008098585,0.6477726,0.009570892,0.2207235],"study_design_scores_gemma":[0.0001913747,0.0005900718,0.03242708,0.0008211828,0.0006084162,0.0005128214,0.0009114787,0.1766082,0.001696205,0.7650746,0.02034733,0.0002113236],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02071483,0.002267118,0.9650319,0.003183194,0.0004155697,0.0005871979,0.0007240956,0.0006868838,0.006389298],"genre_scores_gemma":[0.3769737,0.003943465,0.6074966,0.001596755,0.0005959453,0.004823543,0.001396728,0.0003770715,0.002796114],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.1004338,"threshold_uncertainty_score":0.5311509,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1981221874","doi":"10.1177/0013164403261051","title":"A Note on How to Quantify and Report Whether Irt Parameter Invariance Holds: When Pearson Correlations are Not Enough","year":2004,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":37,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Item response theory; Measurement invariance; Pearson product-moment correlation coefficient; Mathematics; Econometrics; Statistics; Scale (ratio); Measure (data warehouse); Correlation; Psychology; Psychometrics; Confirmatory factor analysis; Computer science; Structural equation modeling; Data mining","authors":[{"name":"André Rupp","is_ca":true},{"name":"Bruno D. Zumbo","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7338274295775041,"gpt":0.4936132193884701,"spread":0.240214210189034,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1506162,0.002078257,0.003199474,0.004878646,0.003489299,0.008190199,0.004610933,0.00501013,0.004168706],"category_scores_gemma":[0.619444,0.001119033,0.001745022,0.007038257,0.01318301,0.01757051,0.006083035,0.01261694,0.003001609],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001741396,"about_ca_system_score_gemma":0.002869083,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004051124,"about_ca_topic_score_gemma":0.003172886,"domain_scores_codex":[0.8392582,0.1059054,0.01657138,0.01173045,0.02479434,0.001740307],"domain_scores_gemma":[0.4757834,0.3784335,0.02895502,0.07097142,0.04410402,0.001752648],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.000563,0.0003209857,0.08642982,0.001723165,0.001145714,0.00212811,0.01508588,0.003987019,0.007481706,0.4071038,0.1081945,0.3658364],"study_design_scores_gemma":[0.00008690739,0.0007593319,0.04623691,0.001648657,0.0002893867,0.004174621,0.00951839,0.02332229,0.009988233,0.8226894,0.08062448,0.0006612969],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.03634246,0.005625928,0.8888081,0.04407723,0.005641289,0.0004114594,0.0009260071,0.001871537,0.016296],"genre_scores_gemma":[0.4353876,0.002320107,0.5333365,0.01665519,0.003369789,0.00156164,0.001196935,0.001766185,0.004405999],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8493838,"threshold_uncertainty_score":0.7965441,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2140588557","doi":"10.1177/0013164414548894","title":"A Cautionary Note on the Use of the Vale and Maurelli Method to Generate Multivariate, Nonnormal Data for Simulation Purposes","year":2014,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":37,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Kurtosis; Skewness; Statistics; Univariate; Mathematics; Monte Carlo method; Population; Variable (mathematics); Multivariate statistics; Sample (material); Econometrics; Algorithm","authors":[{"name":"Oscar L. Olvera Astivia","is_ca":true},{"name":"Bruno D. Zumbo","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6134090036945616,"gpt":0.4939607930004505,"spread":0.1194482106941111,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1491941,0.00230462,0.003113925,0.003500537,0.004273951,0.007271471,0.009923424,0.006520729,0.005656888],"category_scores_gemma":[0.5352716,0.001504408,0.003632827,0.004173778,0.01008162,0.006067095,0.00423345,0.02753326,0.005631064],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003158704,"about_ca_system_score_gemma":0.00454936,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01127342,"about_ca_topic_score_gemma":0.02307963,"domain_scores_codex":[0.825312,0.1374769,0.01176348,0.005738563,0.01877249,0.000936607],"domain_scores_gemma":[0.4301991,0.4683334,0.009434226,0.02976644,0.06047335,0.001793531],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001486446,0.0002861936,0.01656172,0.002196977,0.0009433393,0.002238629,0.0117669,0.009386298,0.003030066,0.1334917,0.6966916,0.1219202],"study_design_scores_gemma":[0.0007343158,0.0007249418,0.01609884,0.006065868,0.0005281931,0.003119849,0.004953667,0.07260272,0.01192517,0.2396467,0.6422994,0.001300311],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02174474,0.00885955,0.4870799,0.3831297,0.07107913,0.001499467,0.002179073,0.006398845,0.01802966],"genre_scores_gemma":[0.1306675,0.003528149,0.646181,0.1690458,0.01745993,0.003478504,0.000462292,0.003129644,0.02604728],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8508059,"threshold_uncertainty_score":0.7890231,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2154633345","doi":"10.1177/0013164403258393","title":"Impact of Test Design, Item Quality, and Item Bank Size on the Psychometric Properties of Computer-Based Credentialing Examinations","year":2004,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":36,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Ministry of Education and Child Care","funders":"","keywords":"Credentialing; Computerized adaptive testing; Item bank; Test (biology); Quality (philosophy); Computer science; Consistency (knowledge bases); Test design; Item response theory; Psychology; Psychometrics; Applied psychology; Statistics; Reliability engineering; Artificial intelligence; Medical education; Test method; Medicine; Mathematics; Clinical psychology","authors":[{"name":"Dehui Xing","is_ca":true},{"name":"Ronald K. Hambleton","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8330425941791564,"gpt":0.5172334143779688,"spread":0.3158091798011876,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1527274,0.0009565259,0.001024165,0.001759469,0.0007150308,0.002489357,0.001026183,0.001597321,0.001461806],"category_scores_gemma":[0.4580861,0.0008054582,0.001574596,0.002421489,0.002388373,0.002723113,0.001893225,0.001730985,0.000322833],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001879051,"about_ca_system_score_gemma":0.002214581,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001103556,"about_ca_topic_score_gemma":0.001363959,"domain_scores_codex":[0.8400878,0.1161624,0.0207157,0.005545883,0.01602304,0.001465336],"domain_scores_gemma":[0.1814991,0.7594377,0.02257628,0.01998978,0.01364179,0.002855373],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.01966206,0.002738857,0.7362366,0.0007592489,0.001375723,0.0004749639,0.003241818,0.01281969,0.01646142,0.002482665,0.00084041,0.2029066],"study_design_scores_gemma":[0.002560992,0.02320324,0.915878,0.0005014463,0.001329363,0.001553587,0.0009260727,0.02807233,0.01772671,0.004939726,0.003015033,0.0002934881],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9754636,0.001314844,0.01987076,0.0006619621,0.00008386803,0.0006475701,0.0001481896,0.0001450584,0.001664292],"genre_scores_gemma":[0.9718269,0.0002432617,0.02639953,0.0003247294,0.00005266113,0.0005436009,0.0002271339,0.00008458787,0.0002974592],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8472726,"threshold_uncertainty_score":0.8077092,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2083403230","doi":"10.1177/0013164406292030","title":"Confidence Intervals for an Effect Size Measure in Multiple Linear Regression","year":2007,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":34,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Manitoba","funders":"","keywords":"Statistics; Confidence interval; Coverage probability; Mathematics; CDF-based nonparametric confidence interval; Percentile; Linear regression; Robust confidence intervals; Sample size determination; Regression analysis; Tolerance interval; Measure (data warehouse); Confidence distribution; Regression; Computer science","authors":[{"name":"James Algina","is_ca":false},{"name":"H. J. Keselman","is_ca":true},{"name":"Randall D. Penfield","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4771766274732445,"gpt":0.5316271763859817,"spread":0.05445054891273715,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1854605,0.001900691,0.003906098,0.008597727,0.001285836,0.005265768,0.00603358,0.005526269,0.005894057],"category_scores_gemma":[0.6981112,0.001098303,0.004342582,0.006609645,0.004612898,0.005823536,0.00464481,0.007146753,0.001123901],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002173609,"about_ca_system_score_gemma":0.001300452,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002088189,"about_ca_topic_score_gemma":0.0008680857,"domain_scores_codex":[0.8202316,0.1224042,0.009804853,0.01498025,0.03113352,0.001445582],"domain_scores_gemma":[0.1475472,0.7981626,0.0175567,0.02172023,0.01419647,0.0008167226],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.005988445,0.0005676103,0.05738403,0.009282594,0.009814113,0.001625606,0.005300634,0.09336489,0.004027056,0.3540374,0.01964965,0.438958],"study_design_scores_gemma":[0.001180267,0.003584222,0.0845717,0.008856179,0.004981779,0.003662605,0.002150237,0.3553486,0.01336814,0.4671459,0.05414916,0.001001123],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.04675525,0.0148658,0.9203819,0.001441017,0.0009720125,0.0009755917,0.001728675,0.001761071,0.01111876],"genre_scores_gemma":[0.6331522,0.002930839,0.353181,0.0009742263,0.0008474754,0.004548928,0.002284708,0.0007157833,0.001364869],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8145395,"threshold_uncertainty_score":0.9808207,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2154413108","doi":"10.1177/0013164406299132","title":"Correction for Attenuation With Biased Reliability Estimates and Correlated Errors in Populations and Samples","year":2007,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":34,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Carleton University","funders":"","keywords":"Statistics; Reliability (semiconductor); Monte Carlo method; Correction for attenuation; Attenuation; Sample size determination; Observational error; Mathematics; Population; Sampling error; Sample (material); Sampling (signal processing); Random variable; Econometrics; Computer science; Physics; Demography; Optics","authors":[{"name":"Donald W. Zimmerman","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3426256692564635,"gpt":0.449982568421525,"spread":0.1073568991650615,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1796756,0.001982975,0.00242138,0.003913106,0.001409593,0.002722097,0.003444837,0.002861029,0.002790107],"category_scores_gemma":[0.728368,0.001598114,0.002294332,0.004668631,0.006598153,0.004778251,0.005397541,0.004548201,0.0005412932],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002468494,"about_ca_system_score_gemma":0.004260959,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004496467,"about_ca_topic_score_gemma":0.004130697,"domain_scores_codex":[0.8185282,0.1486198,0.006939017,0.009681499,0.0147216,0.001509792],"domain_scores_gemma":[0.2282462,0.6851268,0.02486764,0.04816501,0.01298158,0.0006127477],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.001047031,0.0002048066,0.0856735,0.001273759,0.002158774,0.001829605,0.007250446,0.1665146,0.002332433,0.4726632,0.004436365,0.2546155],"study_design_scores_gemma":[0.0001897547,0.0005734818,0.03255387,0.001053528,0.0008245007,0.001858893,0.0006719728,0.4408232,0.004562397,0.5090791,0.007545725,0.0002635826],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01719697,0.0004183285,0.9802776,0.0004524113,0.0001077734,0.0002469445,0.00006198874,0.0002934617,0.000944518],"genre_scores_gemma":[0.4775363,0.0006049871,0.5177509,0.000426511,0.0001908099,0.0014355,0.0001732543,0.0003125251,0.001569213],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.1796756,"threshold_uncertainty_score":0.9502265,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2406198090","doi":"10.1177/0013164416651116","title":"Multidimensional Extension of Multiple Indicators Multiple Causes Models to Detect DIF","year":2016,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Advanced Statistical Modeling Techniques","field":"Computer Science","cited_by":33,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"","keywords":"Differential item functioning; Item response theory; Type I and type II errors; Statistics; Statistical power; Sample size determination; Context (archaeology); Correlation; Mathematics; Econometrics; Psychometrics","authors":[{"name":"Soo In Lee","is_ca":false},{"name":"Okan Bulut","is_ca":true},{"name":"Youngsuk Suh","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1697549661461686,"gpt":0.3494013428645477,"spread":0.1796463767183791,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.08211504,0.002328187,0.002368686,0.005586877,0.001685117,0.002482831,0.002811942,0.001769171,0.005597267],"category_scores_gemma":[0.1934433,0.001190295,0.005828461,0.004226541,0.002964966,0.004270985,0.005312273,0.002884104,0.0006434841],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001825316,"about_ca_system_score_gemma":0.002654686,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003051388,"about_ca_topic_score_gemma":0.002737946,"domain_scores_codex":[0.9064398,0.07745803,0.002694277,0.006508716,0.00581317,0.001086015],"domain_scores_gemma":[0.8164668,0.1418611,0.01387746,0.01991356,0.006934769,0.0009462706],"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":[0.002303083,0.001110382,0.3860543,0.001118733,0.005962396,0.001090459,0.00910757,0.1237672,0.001646117,0.1780763,0.00421683,0.2855465],"study_design_scores_gemma":[0.0004358294,0.001474097,0.06490911,0.0002966225,0.001040354,0.000901202,0.001075885,0.6904556,0.001397057,0.233487,0.0043044,0.0002228146],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1164754,0.0002397476,0.8780479,0.000457709,0.0001000294,0.001400478,0.0004172962,0.0005200906,0.002341294],"genre_scores_gemma":[0.7277907,0.0001900864,0.266775,0.0002056655,0.00006615571,0.00330363,0.0006409167,0.00009170046,0.0009362183],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.08211504,"threshold_uncertainty_score":0.434271,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2166456938","doi":"10.1177/0013164406292089","title":"A Test of the Age-Based Measurement Invariance and Temporal Stability of Antonovsky's Sense of Coherence Scale","year":2007,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Health, psychology, and well-being","field":"Health Professions","cited_by":33,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of British Columbia","funders":"","keywords":"Measurement invariance; Structural equation modeling; Psychology; Scale (ratio); Stability (learning theory); Population; Test (biology); Statistics; Developmental psychology; Mathematics; Demography; Confirmatory factor analysis; Geography; Computer science","authors":[{"name":"Chris G. Richardson","is_ca":true},{"name":"Pamela A. Ratner","is_ca":true},{"name":"Bruno D. Zumbo","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2646435922938717,"gpt":0.4371949212201265,"spread":0.1725513289262547,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004154149,0.0002252967,0.0002521723,0.001303478,0.0006471851,0.000510381,0.0004183827,0.0003186548,0.001531556],"category_scores_gemma":[0.02536483,0.0001782004,0.0005675007,0.0009951029,0.0007182953,0.000920279,0.0009302549,0.0006224179,0.0002596301],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004563216,"about_ca_system_score_gemma":0.001032594,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00712412,"about_ca_topic_score_gemma":0.009033336,"domain_scores_codex":[0.9981024,0.0004109424,0.0003569307,0.0002393366,0.0007089585,0.0001815525],"domain_scores_gemma":[0.9815297,0.005769949,0.006509701,0.001768569,0.00374199,0.0006800669],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002075995,0.0001738548,0.9733056,0.00003213765,0.0001229485,0.00005347452,0.001304798,0.0003113514,0.001523538,0.001123936,0.0005201775,0.02132048],"study_design_scores_gemma":[0.00001526519,0.0002797901,0.9967306,0.00000903456,0.00001341519,0.00008389886,0.000523873,0.0008684878,0.0003145657,0.0004979428,0.0006529749,0.0000101452],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.991791,0.0001101181,0.003158648,0.0001202108,0.00003007748,0.0001409037,0.0005169128,0.00001798317,0.004114236],"genre_scores_gemma":[0.9968055,0.00004643168,0.001877079,0.00003302587,0.00001613912,0.0001907135,0.0007035364,0.000008554026,0.0003188995],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.00712412,"threshold_uncertainty_score":0.0219695,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4388140417","doi":"10.1177/00131644231202949","title":"Comparing RMSEA-Based Indices for Assessing Measurement Invariance in Confirmatory Factor Models","year":2023,"lang":"en","type":"article","venue":"Educational and Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":33,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"York University","funders":"","keywords":"Structural equation modeling; Confirmatory factor analysis; Mathematics; Goodness of fit; Statistics","authors":[{"name":"Nataly Beribisky","is_ca":true},{"name":"Gregory R. Hancock","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9282427079429465,"gpt":0.5543654165441689,"spread":0.3738772913987776,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1121801,0.002888104,0.002913631,0.01038013,0.002016999,0.004185782,0.003308165,0.00204172,0.005334449],"category_scores_gemma":[0.3865162,0.001339988,0.006656585,0.01550242,0.002121592,0.006655567,0.003335142,0.006055458,0.001978029],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002640394,"about_ca_system_score_gemma":0.006553366,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007166058,"about_ca_topic_score_gemma":0.01304815,"domain_scores_codex":[0.9152467,0.04605097,0.0108441,0.006209433,0.02057266,0.001076043],"domain_scores_gemma":[0.6570148,0.2573234,0.02017212,0.02641605,0.03758247,0.001491],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0007612634,0.001032193,0.2251646,0.005049767,0.01054815,0.0005456497,0.008905067,0.03428402,0.004553065,0.05133266,0.03417374,0.6236499],"study_design_scores_gemma":[0.001121492,0.003924928,0.3733792,0.009590582,0.007076245,0.001544136,0.01345148,0.2747931,0.01724139,0.1952801,0.1008276,0.00176985],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.125526,0.005802119,0.8225994,0.002056546,0.001331836,0.003449945,0.005687872,0.003085379,0.03046088],"genre_scores_gemma":[0.4353667,0.002405558,0.5449249,0.0007539476,0.0001663326,0.007094022,0.006125889,0.00119554,0.001967156],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.1121801,"threshold_uncertainty_score":0.5932723,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}