{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":759,"total_is_capped":false,"direct_labels_cover":20,"predictions_cover":759,"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":"203cf04dfb66","filters":{"topic":"Psychometric Methodologies and Testing"}},"results":[{"id":"W2125001590","doi":"10.1016/j.jml.2007.12.005","title":"Mixed-effects modeling with crossed random effects for subjects and items","year":2008,"lang":"en","type":"article","venue":"Journal of Memory and Language","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":8560,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"","keywords":"Psychology; Developmental psychology; Cognitive psychology; Social psychology","authors":[{"name":"R. Harald Baayen","is_ca":true},{"name":"Douglas Davidson","is_ca":false},{"name":"Douglas M. Bates","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2151326486886005,"gpt":0.3957748843215883,"spread":0.1806422356329878,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1000281,0.009569657,0.01257648,0.004981029,0.003998905,0.007000261,0.01533097,0.01242222,0.0446192],"category_scores_gemma":[0.1666296,0.007383411,0.02785184,0.005770787,0.003952829,0.006573031,0.005506137,0.01186714,0.01190267],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003660598,"about_ca_system_score_gemma":0.007712726,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02439481,"about_ca_topic_score_gemma":0.01889299,"domain_scores_codex":[0.8658069,0.09041347,0.006290627,0.02490954,0.006812242,0.005767209],"domain_scores_gemma":[0.8329503,0.1144125,0.005157071,0.03364657,0.01195789,0.001875702],"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":[0.1180622,0.009562029,0.1171126,0.006682431,0.09650511,0.003960454,0.01612154,0.2320083,0.008812356,0.1015874,0.03592027,0.2536654],"study_design_scores_gemma":[0.01285894,0.01200004,0.03184861,0.001212617,0.0432143,0.001303463,0.001285902,0.7743653,0.009533629,0.07649315,0.03457789,0.001306108],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1092175,0.001356594,0.8562311,0.0007642049,0.003302377,0.008101828,0.0118382,0.006436487,0.002751798],"genre_scores_gemma":[0.3693222,0.0008172495,0.5106686,0.001651904,0.000854258,0.05535487,0.008775254,0.00357847,0.04897725],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8999719,"threshold_uncertainty_score":0.5290055,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2144487201","doi":"10.1080/10705510903008220","title":"Exploratory Structural Equation Modeling, Integrating CFA and EFA: Application to Students' Evaluations of University Teaching","year":2009,"lang":"en","type":"article","venue":"Structural Equation Modeling A Multidisciplinary Journal","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":1080,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Université de Sherbrooke","funders":"Economic and Social Research Council","keywords":"Structural equation modeling; Confirmatory factor analysis; Goodness of fit; Latent variable; Psychology; Statistics; Measurement invariance; Environmental scanning electron microscope; Exploratory factor analysis; Econometrics; Mathematics","authors":[{"name":"Herbert W. Marsh","is_ca":false},{"name":"Bengt Muthén","is_ca":false},{"name":"Tihomir Asparouhov","is_ca":false},{"name":"Oliver Lüdtke","is_ca":false},{"name":"Alexander Robitzsch","is_ca":false},{"name":"Alexandre J. S. Morin","is_ca":true},{"name":"Ulrich Trautwein","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3055562692897275,"gpt":0.4805804396433995,"spread":0.175024170353672,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03914037,0.001665926,0.001194988,0.004255472,0.001290479,0.002098456,0.001053578,0.0007197165,0.002349801],"category_scores_gemma":[0.09327305,0.0008967317,0.002463948,0.006089066,0.000987209,0.002047525,0.002668435,0.002363415,0.0004246263],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001269716,"about_ca_system_score_gemma":0.003810895,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004991621,"about_ca_topic_score_gemma":0.008192872,"domain_scores_codex":[0.9730912,0.02322436,0.0008199626,0.0008974368,0.001718807,0.0002481682],"domain_scores_gemma":[0.9166477,0.0709399,0.002383546,0.004806654,0.004849082,0.000373092],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0004237629,0.001065445,0.1116333,0.001101263,0.00161919,0.0003817078,0.01423208,0.0429683,0.003519657,0.03624406,0.00553472,0.7812766],"study_design_scores_gemma":[0.0003365748,0.001831736,0.0867308,0.001104878,0.0006486705,0.0006040775,0.007931569,0.7502756,0.004237674,0.1340312,0.01195437,0.0003129267],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1305893,0.0003813372,0.861043,0.0007324921,0.00007245592,0.001529,0.0006180906,0.001459678,0.003574641],"genre_scores_gemma":[0.347307,0.0002941478,0.6488952,0.0001027085,0.00003085823,0.002385013,0.0003963805,0.0001340377,0.0004547477],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.03914037,"threshold_uncertainty_score":0.2069966,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2612259620","doi":"10.22237/jmasm/1177992180","title":"Ordinal Versions of Coefficients Alpha and Theta for Likert Rating Scales","year":2007,"lang":"en","type":"article","venue":"Journal of Modern Applied Statistical Methods","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":1013,"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":"Ordinal data; Mathematics; Statistics; Likert scale; Ordinal Scale; Skewness; Ordinal regression; Reliability (semiconductor); Econometrics","authors":[{"name":"Bruno D. Zumbo","is_ca":true},{"name":"Anne Gadermann","is_ca":true},{"name":"Cornelia Zeisser","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3690266326624584,"gpt":0.5562911121253126,"spread":0.1872644794628542,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03049302,0.001455474,0.001042178,0.005990668,0.0007669404,0.003732337,0.001823975,0.001200576,0.01325978],"category_scores_gemma":[0.1933451,0.0006820192,0.001386844,0.008366137,0.001880305,0.005068689,0.001948344,0.00458364,0.004208262],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001022762,"about_ca_system_score_gemma":0.001632595,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000681075,"about_ca_topic_score_gemma":0.0009016276,"domain_scores_codex":[0.9598738,0.02421307,0.003946974,0.001737678,0.00968252,0.0005459226],"domain_scores_gemma":[0.8165888,0.1317977,0.01641686,0.01122667,0.02275159,0.001218499],"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":[0.0006727772,0.0003045235,0.02193952,0.004074658,0.0006000212,0.0003844988,0.00611002,0.008430227,0.004727452,0.2797953,0.06758585,0.6053751],"study_design_scores_gemma":[0.0003380287,0.001870828,0.06548274,0.003125753,0.0006658945,0.002267405,0.006394194,0.04237999,0.005193891,0.4422587,0.429023,0.0009995356],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0230636,0.005285159,0.930962,0.001335622,0.002800891,0.001481941,0.004233974,0.001690781,0.02914602],"genre_scores_gemma":[0.2064961,0.004207566,0.767921,0.0006798632,0.001248342,0.009875347,0.002863566,0.0006825651,0.006025627],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.03049302,"threshold_uncertainty_score":0.1612644,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2769536523","doi":"10.1002/pfi.21727","title":"Evidence-Based Survey Design: The Use of a Midpoint on the Likert Scale","year":2017,"lang":"en","type":"article","venue":"Performance Improvement Journal","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":823,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Government of Alberta","funders":"","keywords":"Likert scale; Midpoint; Scale (ratio); Computer science; Survey research; Psychology; Management science; Applied psychology; Mathematics; Engineering; Geography","authors":[{"name":"Seung Youn Chyung","is_ca":false},{"name":"Katherine A. Roberts","is_ca":false},{"name":"Ieva Swanson","is_ca":false},{"name":"Andrea Hankinson","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9048798418730529,"gpt":0.4941430133169829,"spread":0.41073682855607,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3082857,0.001465465,0.00251058,0.00759243,0.001593767,0.004483976,0.002534313,0.00259816,0.007066662],"category_scores_gemma":[0.4539652,0.001029263,0.001920448,0.01281406,0.00301093,0.004440607,0.00489671,0.003763522,0.00277308],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002588435,"about_ca_system_score_gemma":0.006166066,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005166387,"about_ca_topic_score_gemma":0.001100226,"domain_scores_codex":[0.4475228,0.4585544,0.05827814,0.005270687,0.02890119,0.001472843],"domain_scores_gemma":[0.5514313,0.3082899,0.03035886,0.02214663,0.08549897,0.002274364],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.005080674,0.002364351,0.03095015,0.0242531,0.001595389,0.0001797742,0.01660042,0.003668267,0.00350337,0.04925073,0.03999251,0.8225613],"study_design_scores_gemma":[0.01209528,0.03053194,0.1222383,0.03408639,0.002582815,0.0009294313,0.03906372,0.04925349,0.02123807,0.1795289,0.5067422,0.001709432],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.04862363,0.002446677,0.6971897,0.004271337,0.002376658,0.2168714,0.003921825,0.0009047166,0.02339402],"genre_scores_gemma":[0.0722931,0.001208872,0.6004556,0.001685107,0.000231955,0.3219516,0.001150123,0.00009280518,0.0009307237],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.6917143,"threshold_uncertainty_score":0.8530068,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1980472079","doi":"10.1136/bmj.c117","title":"Is a subgroup effect believable? Updating criteria to evaluate the credibility of subgroup analyses","year":2010,"lang":"en","type":"article","venue":"BMJ","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":812,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University","funders":"","keywords":"Subgroup analysis; Credibility; Psychology; Computer science; Information retrieval; Mathematics; Statistics; Political science","authors":[{"name":"Xin Sun","is_ca":true},{"name":"Matthias Briel","is_ca":true},{"name":"Stephen D. Walter","is_ca":true},{"name":"Gordon Guyatt","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5243444902286908,"gpt":0.5884504748589897,"spread":0.06410598463029893,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.7639732,0.003559431,0.009758552,0.01837479,0.006711408,0.01362848,0.01340871,0.01559088,0.003698401],"category_scores_gemma":[0.9515347,0.003588847,0.01241935,0.008588603,0.02177141,0.0201908,0.01524167,0.02763853,0.0008192366],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.007495451,"about_ca_system_score_gemma":0.01157802,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004499687,"about_ca_topic_score_gemma":0.003797256,"domain_scores_codex":[0.1970384,0.5867434,0.1134353,0.01565398,0.08430521,0.002823787],"domain_scores_gemma":[0.01982957,0.8837875,0.0265731,0.03773469,0.03040128,0.001673789],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.01295887,0.0007134472,0.1010499,0.02016523,0.03046883,0.002262407,0.04502948,0.01467919,0.003032966,0.1579844,0.04108193,0.5705733],"study_design_scores_gemma":[0.005087446,0.002983528,0.02595461,0.02437279,0.01538313,0.003384233,0.006683235,0.1162026,0.01021376,0.7343088,0.05357157,0.001854188],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.07414968,0.01638149,0.7961951,0.08433396,0.008638511,0.00721765,0.001308399,0.001192969,0.01058223],"genre_scores_gemma":[0.4443515,0.001680666,0.5304815,0.01302778,0.002807699,0.006045862,0.0005289202,0.0004559538,0.0006200534],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.2360268,"threshold_uncertainty_score":0.2910631,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2126395473","doi":"","title":"Testing for multigroup equivalence of a measuring instrument: a walk through the process.","year":2008,"lang":"en","type":"article","venue":"PubMed","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":737,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Ottawa","funders":"","keywords":"Equivalence (formal languages); Psychology; Covariance; Social psychology; Computer science; Cognitive psychology; Mathematics; Statistics; Pure mathematics","authors":[{"name":"Barbara M. Byrne","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7434047361636618,"gpt":0.4296999713785657,"spread":0.313704764785096,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.438366,0.002804708,0.002883722,0.008451787,0.00627233,0.006777868,0.004586767,0.002652054,0.003813983],"category_scores_gemma":[0.6405976,0.001928754,0.002641802,0.007036416,0.009120357,0.009241832,0.01523074,0.009073304,0.001682597],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003801306,"about_ca_system_score_gemma":0.008521168,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0040482,"about_ca_topic_score_gemma":0.004696504,"domain_scores_codex":[0.5910528,0.3425475,0.01888016,0.009807617,0.03620809,0.001503868],"domain_scores_gemma":[0.3337403,0.5346943,0.01342906,0.06157026,0.05398111,0.002585056],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0009504568,0.00104129,0.04285506,0.003397467,0.001818079,0.001055085,0.06490748,0.002500896,0.008013116,0.1063914,0.00595568,0.7611139],"study_design_scores_gemma":[0.001275505,0.008363539,0.1016677,0.008889744,0.001575679,0.004470618,0.07173017,0.05542101,0.02694753,0.5972708,0.1210459,0.001341836],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0372495,0.002376218,0.941573,0.002034984,0.0004121212,0.01016627,0.0002199804,0.0005386537,0.005429154],"genre_scores_gemma":[0.09839766,0.0004029018,0.8898677,0.0004358026,0.00006597827,0.0100092,0.0002460619,0.0001432298,0.0004314435],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.438366,"threshold_uncertainty_score":0.6925947,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2061477359","doi":"10.1371/journal.pone.0058067","title":"The Center for Epidemiologic Studies Depression Scale: A Review with a Theoretical and Empirical Examination of Item Content and Factor Structure","year":2013,"lang":"en","type":"review","venue":"PLoS ONE","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":552,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Regina Qu'Appelle Health Region; University of Regina","funders":"","keywords":"Differential item functioning; Anhedonia; Center for Epidemiologic Studies Depression Scale; Clinical psychology; Item response theory; Psychology; PsycINFO; Psychometrics; Confirmatory factor analysis; Scale (ratio); Depressive symptoms; Medicine; Psychiatry; MEDLINE; Structural equation modeling; Anxiety; Statistics","authors":[{"name":"R. Nicholas Carleton","is_ca":true},{"name":"Michel A. Thibodeau","is_ca":true},{"name":"Michelle J. N. Teale","is_ca":true},{"name":"Patrick G. Welch","is_ca":true},{"name":"Murray P. Abrams","is_ca":true},{"name":"Thomas P. Robinson","is_ca":true},{"name":"Gordon J. G. Asmundson","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8519675177471471,"gpt":0.5481175019615244,"spread":0.3038500157856227,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02198706,0.001798403,0.004623587,0.0127085,0.0005714954,0.002366587,0.002867549,0.001524315,0.001647029],"category_scores_gemma":[0.0389832,0.0009001216,0.002895821,0.01682398,0.001435632,0.002251045,0.001543797,0.003104913,0.0009464397],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003072573,"about_ca_system_score_gemma":0.005981646,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007784366,"about_ca_topic_score_gemma":0.01084827,"domain_scores_codex":[0.9866458,0.005584224,0.003094471,0.001000492,0.003460339,0.0002145967],"domain_scores_gemma":[0.974239,0.01445113,0.003177426,0.0007196001,0.007078956,0.0003338756],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0001842733,0.0001079146,0.01119383,0.05173249,0.001336541,0.00008386823,0.0002326862,0.0002608623,0.0001588431,0.001895215,0.03929703,0.8935165],"study_design_scores_gemma":[0.000289838,0.0004132573,0.1061691,0.201495,0.006370586,0.002569171,0.0006919414,0.00121019,0.0004606906,0.007142873,0.6729043,0.0002830246],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0003832171,0.9951521,0.00143926,0.001317424,0.0003555978,0.0001847622,0.0004318882,0.00002833322,0.0007073886],"genre_scores_gemma":[0.003981843,0.9904414,0.003259526,0.0006910844,0.0003077072,0.0004686728,0.0006807862,0.00001159683,0.0001574182],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.02198706,"threshold_uncertainty_score":0.11628,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3095583164","doi":"10.1177/2515245920951747","title":"Your Coefficient Alpha Is Probably Wrong, but Which Coefficient Omega Is Right? A Tutorial on Using R to Obtain Better Reliability Estimates","year":2020,"lang":"en","type":"article","venue":"Advances in Methods and Practices in Psychological Science","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":542,"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":"Omega; Confusion; Reliability (semiconductor); Construct (python library); Alpha (finance); Variety (cybernetics); Quality (philosophy); Confirmatory factor analysis; Cronbach's alpha; Computer science; Correlation coefficient; Statistics; Mathematics; Psychology; Epistemology; Physics; Psychometrics; Structural equation modeling; Philosophy; Programming language; Thermodynamics","authors":[{"name":"David B. Flora","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3780115946846152,"gpt":0.6190029165631815,"spread":0.2409913218785663,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.04655836,0.002976729,0.00215148,0.006803104,0.001727257,0.005904057,0.002986579,0.003524924,0.02069668],"category_scores_gemma":[0.3187602,0.001600518,0.002407527,0.00867289,0.004762382,0.009216439,0.002836953,0.008482761,0.02263991],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00160052,"about_ca_system_score_gemma":0.002400008,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001718185,"about_ca_topic_score_gemma":0.001834657,"domain_scores_codex":[0.9474371,0.03498752,0.005208832,0.003653409,0.008046754,0.0006664316],"domain_scores_gemma":[0.7837873,0.163645,0.01182638,0.01391149,0.02566355,0.001166349],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001289674,0.00006060224,0.002652467,0.003387817,0.0002714937,0.0002773886,0.002945228,0.001438198,0.00233448,0.1291474,0.4379945,0.4193615],"study_design_scores_gemma":[0.00006331888,0.0001643799,0.003616639,0.006868944,0.0002102924,0.0008731795,0.001403902,0.004860607,0.004314288,0.3360133,0.6413276,0.0002835778],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.003815772,0.04787176,0.8112847,0.06743099,0.0168125,0.0004759892,0.002663336,0.009432685,0.04021223],"genre_scores_gemma":[0.05101709,0.04689135,0.8492787,0.0160841,0.007798749,0.002167914,0.002130033,0.00914263,0.01548942],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9534416,"threshold_uncertainty_score":0.2462271,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4214760420","doi":"10.1186/s12913-022-07577-3","title":"Development of a theory-informed questionnaire to assess the acceptability of healthcare interventions","year":2022,"lang":"en","type":"article","venue":"BMC Health Services Research","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":538,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Ottawa Hospital","funders":"","keywords":"Psychological intervention; Medicine; Intervention (counseling); Content validity; Health care; Nursing research; Discriminant validity; Focus group; Health administration; Health informatics; Public health; Nursing; Patient satisfaction; Psychometrics; Clinical psychology","authors":[{"name":"Mandeep Sekhon","is_ca":false},{"name":"Martin Cartwright","is_ca":false},{"name":"Jill Francis","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7507588387515985,"gpt":0.6539609633253661,"spread":0.09679787542623242,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.05114872,0.0006177771,0.0009631609,0.002894074,0.0006626684,0.001606789,0.001620076,0.001267427,0.004871389],"category_scores_gemma":[0.08756323,0.0007049136,0.001770917,0.001661842,0.001078115,0.001897213,0.001983202,0.002426338,0.00105565],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0037843,"about_ca_system_score_gemma":0.007652501,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001357968,"about_ca_topic_score_gemma":0.002015175,"domain_scores_codex":[0.9697289,0.02112361,0.003675754,0.0006771202,0.004277857,0.0005167654],"domain_scores_gemma":[0.9124907,0.05995782,0.004553466,0.002628901,0.01923992,0.001129314],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0007993847,0.002580402,0.08811247,0.01109002,0.0004557414,0.0004614844,0.08071426,0.005355301,0.007626979,0.01097477,0.0226378,0.7691914],"study_design_scores_gemma":[0.00296899,0.01327622,0.4259022,0.02008346,0.001044648,0.001496078,0.08951106,0.04595327,0.0134991,0.03701373,0.3481818,0.001069488],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.4923971,0.002149518,0.282678,0.007711543,0.0006719966,0.1685018,0.005565278,0.0008968897,0.03942779],"genre_scores_gemma":[0.3942757,0.001502448,0.4476348,0.002118714,0.00008342462,0.1494309,0.002190276,0.00007650568,0.002687158],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.05114872,"threshold_uncertainty_score":0.2705036,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2102743086","doi":"10.1016/j.jpsychores.2010.01.006","title":"Introduction to health measurement scales","year":2010,"lang":"en","type":"review","venue":"Journal of Psychosomatic Research","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":499,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Baycrest Hospital; University Health Network; University of Toronto","funders":"","keywords":"Scale (ratio); Psychometrics; Psychology; Applied psychology; Psychometric testing; Clinical psychology; Internal consistency; Geography; Cartography","authors":[{"name":"András Keszei","is_ca":false},{"name":"Márta Novák","is_ca":true},{"name":"David L. Streiner","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8911488147396245,"gpt":0.6819205607602203,"spread":0.2092282539794041,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007980867,0.002523982,0.003960003,0.01034592,0.0005966747,0.002451161,0.001693714,0.002078984,0.01308506],"category_scores_gemma":[0.02099795,0.0009656344,0.002397323,0.009379188,0.002522,0.003755082,0.00273382,0.006359246,0.005398457],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002305091,"about_ca_system_score_gemma":0.006085076,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003452684,"about_ca_topic_score_gemma":0.004296904,"domain_scores_codex":[0.9944032,0.002151991,0.001481465,0.0005842639,0.001235356,0.0001437537],"domain_scores_gemma":[0.9861323,0.01044408,0.0007555095,0.0003964989,0.002055342,0.0002163304],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00009296756,0.00006773655,0.0004082455,0.02729462,0.0001773732,0.00006923101,0.0001731608,0.0001626306,0.0002827136,0.01444252,0.08756664,0.8692622],"study_design_scores_gemma":[0.00007430427,0.0001165974,0.003613897,0.02972444,0.0003328781,0.0006442373,0.0001795928,0.0001240316,0.0001973895,0.01389143,0.95104,0.00006122166],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0001469638,0.9825782,0.003217536,0.004340205,0.005285021,0.0001952356,0.0005149394,0.00004441231,0.003677478],"genre_scores_gemma":[0.002406761,0.967313,0.009530888,0.009185914,0.005658773,0.0007899733,0.0005987461,0.00004264028,0.004473289],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.01308506,"threshold_uncertainty_score":0.04377389,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2087062758","doi":"10.1207/s15327752jpa8501_02","title":"Factor Analytic Models: Viewing the Structure of an Assessment Instrument From Three Perspectives","year":2005,"lang":"en","type":"review","venue":"Journal of Personality Assessment","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":492,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Ottawa","funders":"","keywords":"Confirmatory factor analysis; Psychology; Structural equation modeling; CLARITY; Exploratory factor analysis; Psychometrics; Notional amount; Social psychology; Clinical psychology; Statistics","authors":[{"name":"Barbara M. Byrne","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6885502201466005,"gpt":0.588745809497927,"spread":0.09980441064867351,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02916664,0.002261475,0.001586692,0.006826308,0.001295036,0.008855291,0.003490037,0.002322088,0.002022418],"category_scores_gemma":[0.05067703,0.0009476557,0.001258386,0.00643191,0.00600112,0.009528581,0.00278885,0.004944168,0.001521173],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002594434,"about_ca_system_score_gemma":0.005584041,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002390605,"about_ca_topic_score_gemma":0.00230072,"domain_scores_codex":[0.9836673,0.01086051,0.001046259,0.00100612,0.003213317,0.0002065225],"domain_scores_gemma":[0.9703676,0.02244066,0.001659943,0.001753182,0.003440494,0.0003381628],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00002913207,0.00007579415,0.003419351,0.002286046,0.0003285309,0.0002672499,0.007056375,0.004724594,0.001065451,0.6783119,0.008770205,0.2936653],"study_design_scores_gemma":[0.00002958825,0.0001171143,0.001915866,0.002889346,0.0001555396,0.0004758246,0.003704698,0.01965288,0.0008089002,0.8797676,0.09035829,0.0001243235],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"review","genre_scores_codex":[0.002711502,0.02059438,0.9571236,0.007543818,0.0003301142,0.0004252231,0.000199597,0.0005431639,0.01052868],"genre_scores_gemma":[0.05155867,0.03007638,0.913357,0.001324823,0.0004987489,0.001548673,0.0002941777,0.0001330722,0.001208355],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.02916664,"threshold_uncertainty_score":0.1542498,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1972479177","doi":"10.1186/1756-0500-3-267","title":"Structural equation modeling in medical research: a primer","year":2010,"lang":"en","type":"article","venue":"BMC Research Notes","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":481,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Calgary","funders":"","keywords":"Structural equation modeling; Latent variable; Computer science; Medical research; Set (abstract data type); Data science; Linear regression; Measure (data warehouse); Management science; Data mining; Machine learning; Medicine; Pathology","authors":[{"name":"Tanya Beran","is_ca":true},{"name":"Claudio Violato","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9628487103961001,"gpt":0.6876974379363604,"spread":0.2751512724597397,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02760125,0.001203256,0.001624469,0.004843024,0.001027058,0.004174825,0.002315684,0.005008689,0.009231743],"category_scores_gemma":[0.0292089,0.001135726,0.001492184,0.007032516,0.005509799,0.005466542,0.002602441,0.01082756,0.002315046],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002778908,"about_ca_system_score_gemma":0.005782072,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002090515,"about_ca_topic_score_gemma":0.001856874,"domain_scores_codex":[0.9853431,0.01187978,0.0007938854,0.0005507391,0.001260475,0.0001719785],"domain_scores_gemma":[0.9389548,0.05603892,0.001452771,0.001016638,0.002059545,0.0004773712],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00002405304,0.00012396,0.001427828,0.002119042,0.00009850193,0.0003321781,0.002369715,0.005151961,0.0002935961,0.8377829,0.02280849,0.1274677],"study_design_scores_gemma":[0.00003093714,0.00009882329,0.0008851787,0.004777368,0.00004064067,0.0005164428,0.0006846864,0.007981343,0.0001928564,0.7824855,0.2022509,0.0000552546],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.001679964,0.1160948,0.74726,0.1087066,0.002334059,0.0008233075,0.000838165,0.000334242,0.02192884],"genre_scores_gemma":[0.03935033,0.1709885,0.7633828,0.01429937,0.003915932,0.003680868,0.0008566345,0.0001848451,0.003340762],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9723988,"threshold_uncertainty_score":0.1459711,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2134792407","doi":"10.1186/1471-2288-12-159","title":"Should researchers use single indicators, best indicators, or multiple indicators in structural equation models?","year":2012,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":422,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"","keywords":"Structural equation modeling; Latent variable; Path analysis (statistics); Econometrics; Computer science; Statistical model; Performance indicator; Confounding; Statistics; Risk analysis (engineering); Mathematics; Machine learning; Medicine; Economics","authors":[{"name":"Leslie A. Hayduk","is_ca":true},{"name":"Levente Littvay","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9615285550475937,"gpt":0.6606017935692375,"spread":0.3009267614783562,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2574797,0.001776271,0.00417749,0.007120336,0.002419317,0.008777756,0.004008269,0.005025247,0.00348789],"category_scores_gemma":[0.6055329,0.001877756,0.002341317,0.01630974,0.009656526,0.01903723,0.003433183,0.008730528,0.00156782],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004862583,"about_ca_system_score_gemma":0.01228566,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00559758,"about_ca_topic_score_gemma":0.009159629,"domain_scores_codex":[0.7665657,0.1963497,0.01258584,0.008021927,0.01467111,0.001805852],"domain_scores_gemma":[0.4687927,0.4182127,0.03451028,0.0321401,0.04152459,0.004819661],"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":[0.0007761905,0.0003492728,0.0600244,0.01266664,0.001346642,0.0001789541,0.01107639,0.002998991,0.0004210423,0.2636076,0.07650662,0.5700472],"study_design_scores_gemma":[0.0005509321,0.0006009435,0.01850602,0.03139874,0.001121598,0.0003586623,0.01168502,0.0168661,0.00137599,0.8480015,0.06899107,0.0005433378],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"methods","genre_scores_codex":[0.0452807,0.1019777,0.3352064,0.4931153,0.009347826,0.0007615174,0.001920106,0.001049328,0.01134105],"genre_scores_gemma":[0.4612672,0.04068171,0.4484985,0.04012851,0.003254541,0.003544133,0.0009686174,0.0006098332,0.00104677],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.2574797,"threshold_uncertainty_score":0.9156598,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2016005933","doi":"10.1037/0021-9010.87.3.517","title":"Measurement equivalence: A comparison of methods based on confirmatory factor analysis and item response theory.","year":2002,"lang":"en","type":"article","venue":"Journal of Applied Psychology","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":412,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Ottawa","funders":"","keywords":"Equivalence (formal languages); Item response theory; Differential item functioning; Confirmatory factor analysis; Psychology; Raw score; Psychometrics; Factor analysis; Social psychology; Econometrics; Classical test theory; Test (biology); Statistics; Raw data; Mathematics; Structural equation modeling; Developmental psychology","authors":[{"name":"Nambury S. Raju","is_ca":false},{"name":"Larry J. Laffitte","is_ca":false},{"name":"Barbara M. Byrne","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6434044207719166,"gpt":0.5717001743794533,"spread":0.07170424639246331,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2533992,0.002587471,0.003425288,0.01842771,0.001729009,0.005643495,0.003682692,0.002583069,0.003291898],"category_scores_gemma":[0.5798608,0.001575937,0.002975835,0.01345731,0.006206943,0.01079559,0.007398331,0.004401159,0.001247103],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003852178,"about_ca_system_score_gemma":0.007515079,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003407026,"about_ca_topic_score_gemma":0.003420797,"domain_scores_codex":[0.6418476,0.3011094,0.01051021,0.007335272,0.03807989,0.001117584],"domain_scores_gemma":[0.4947452,0.4271049,0.01618025,0.02479298,0.03542021,0.001756447],"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":[0.001059433,0.0006244205,0.02535968,0.006162503,0.004011861,0.0002135359,0.006687079,0.005685774,0.001078497,0.1607889,0.009425433,0.7789028],"study_design_scores_gemma":[0.001656466,0.00336962,0.09621309,0.00987167,0.003331885,0.002101942,0.00818832,0.0967858,0.004197635,0.7102599,0.0629157,0.001108019],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02033929,0.01927831,0.9398441,0.0029397,0.001099964,0.003091025,0.00043843,0.001132084,0.01183705],"genre_scores_gemma":[0.2157222,0.006435843,0.7682955,0.001044471,0.0004554919,0.006100015,0.0005951336,0.000479015,0.0008723782],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.2533992,"threshold_uncertainty_score":0.9206917,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2905628759","doi":"10.20982/tqmp.14.4.p242","title":"A review of effect sizes and their confidence intervals, Part I: The Cohen's d family","year":2018,"lang":"en","type":"review","venue":"The Quantitative Methods for Psychology","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":365,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Ottawa","funders":"","keywords":"Confidence interval; Statistics; Mathematics; Psychology","authors":[{"name":"Jean‐Christophe Goulet‐Pelletier","is_ca":true},{"name":"Denis Cousineau","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7976858719762024,"gpt":0.6853861465175152,"spread":0.1122997254586873,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1005164,0.002866679,0.005953809,0.02473859,0.001478517,0.00499353,0.005594056,0.005105907,0.005532848],"category_scores_gemma":[0.3150442,0.002401508,0.005272816,0.01758167,0.005812485,0.008808348,0.003254226,0.006829354,0.002644349],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.007603148,"about_ca_system_score_gemma":0.008524415,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006290224,"about_ca_topic_score_gemma":0.004071156,"domain_scores_codex":[0.9018679,0.04459881,0.0227931,0.006927137,0.0231721,0.0006410028],"domain_scores_gemma":[0.6298082,0.3105837,0.01559329,0.006070835,0.03694713,0.0009968024],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000217797,0.00006051346,0.001784224,0.05538763,0.001775267,0.0001310477,0.0007720058,0.001640983,0.0005813651,0.03079012,0.0761269,0.8307322],"study_design_scores_gemma":[0.0001931663,0.0004888971,0.01463196,0.08990684,0.002953642,0.001896241,0.0005584265,0.002604318,0.001376379,0.1014414,0.7834633,0.0004854699],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0002533141,0.9605427,0.03067471,0.003329175,0.001888651,0.0001960036,0.0004398041,0.0001420846,0.002533523],"genre_scores_gemma":[0.008396911,0.9105874,0.07079218,0.003473955,0.003008274,0.00203074,0.0005941457,0.0002183119,0.0008980859],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.8994836,"threshold_uncertainty_score":0.5315881,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2029517174","doi":"10.1007/s11336-005-1295-9","title":"Limited Information Goodness-of-fit Testing in Multidimensional Contingency Tables","year":2006,"lang":"en","type":"article","venue":"Psychometrika","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":327,"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":"Goodness of fit; Statistics; Contingency table; Mathematics; Statistical hypothesis testing; Estimator; Null hypothesis; Econometrics; Applied mathematics","authors":[{"name":"Alberto Maydeu‐Olivares","is_ca":false},{"name":"Harry Joe","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4253619074110371,"gpt":0.4361651235139779,"spread":0.01080321610294077,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.05319666,0.00119974,0.002312744,0.005170514,0.0009277111,0.002990239,0.003451155,0.001640082,0.006048574],"category_scores_gemma":[0.3967118,0.0006130905,0.001911549,0.006218298,0.00343165,0.005299823,0.002899144,0.002503595,0.0007703181],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008189302,"about_ca_system_score_gemma":0.001300806,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006622046,"about_ca_topic_score_gemma":0.0004686974,"domain_scores_codex":[0.9216641,0.06159827,0.004218635,0.00421771,0.007502471,0.0007988947],"domain_scores_gemma":[0.435635,0.5181814,0.01320561,0.02376022,0.007912561,0.001305254],"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":[0.002018103,0.0006301978,0.1509763,0.001981519,0.002244707,0.002035506,0.004062804,0.120296,0.00279005,0.283954,0.008855048,0.4201558],"study_design_scores_gemma":[0.0002917637,0.001341857,0.0325557,0.0005660803,0.0002363109,0.001239024,0.00103858,0.4660544,0.003032344,0.4878576,0.005557036,0.0002293352],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.06821945,0.0005887458,0.926567,0.0004325706,0.00008459529,0.0002700322,0.0009560951,0.0005774916,0.002304024],"genre_scores_gemma":[0.6577025,0.0002963842,0.3382555,0.0002580415,0.0001224017,0.00147932,0.001413796,0.0001400218,0.0003319301],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.05319666,"threshold_uncertainty_score":0.2813342,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2037416776","doi":"10.3758/brm.42.3.847","title":"A general framework and an R package for the detection of dichotomous differential item functioning","year":2010,"lang":"en","type":"article","venue":"Behavior Research Methods","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":288,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Université du Québec à Montréal","funders":"","keywords":"Differential item functioning; R package; Set (abstract data type); Differential (mechanical device); Psychometrics; Item response theory; Computer science; Psychology; Cognitive psychology; Artificial intelligence; Developmental psychology; Programming language","authors":[{"name":"David Magis","is_ca":false},{"name":"Sébastien Béland","is_ca":true},{"name":"Francis Tuerlinckx","is_ca":false},{"name":"Paul De Boeck","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7240286491433732,"gpt":0.6707538589519908,"spread":0.05327479019138237,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04601972,0.004963258,0.002884641,0.006722646,0.001259428,0.003610482,0.004279596,0.00191059,0.02089152],"category_scores_gemma":[0.1625387,0.003058325,0.006447131,0.007073926,0.002103312,0.002439942,0.004302854,0.003810372,0.01378511],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00131742,"about_ca_system_score_gemma":0.005073532,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003787803,"about_ca_topic_score_gemma":0.003687981,"domain_scores_codex":[0.9615539,0.02315685,0.004898923,0.00413188,0.00538222,0.0008761411],"domain_scores_gemma":[0.8918447,0.08152804,0.004886547,0.01393364,0.00731067,0.0004964182],"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":[0.001025437,0.0009109039,0.05400797,0.005086186,0.003818564,0.0008588733,0.00282194,0.02294393,0.01180458,0.1604346,0.1167145,0.6195725],"study_design_scores_gemma":[0.001783015,0.002039363,0.1098167,0.002566334,0.002812503,0.004047345,0.0009959253,0.2510715,0.02852013,0.3240361,0.2710752,0.00123575],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.004188087,0.0001231201,0.9784245,0.0003109301,0.00005518332,0.002109685,0.004802698,0.008407597,0.00157827],"genre_scores_gemma":[0.01686679,0.00006859654,0.9706581,0.0001187925,0.00002911277,0.007190875,0.002439614,0.001648836,0.0009792296],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.04601972,"threshold_uncertainty_score":0.2433785,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2089944619","doi":"10.1080/00224540009600503","title":"Confirmatory Factor Analyses of the Full and Short Versions of the Marlowe-Crowne Social Desirability Scale","year":2000,"lang":"en","type":"article","venue":"The Journal of Social Psychology","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":277,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Lethbridge","funders":"","keywords":"Scale (ratio); Confirmatory factor analysis; Short Forms; Psychology; Social desirability; Full scale; Social psychology; Statistics; Econometrics; Structural equation modeling; Computer science; Mathematics; Clinical psychology","authors":[{"name":"Robert Loo","is_ca":true},{"name":"Karran Thorpe","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5809835667547556,"gpt":0.540996276488838,"spread":0.03998729026591763,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01527882,0.0006590757,0.0008158903,0.002215764,0.001175204,0.0009573395,0.0006676081,0.0002608546,0.002427155],"category_scores_gemma":[0.04985528,0.000354337,0.001308024,0.001866504,0.0007932485,0.0007114493,0.0009502989,0.001644546,0.0004565887],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001437917,"about_ca_system_score_gemma":0.005915619,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03449119,"about_ca_topic_score_gemma":0.06026383,"domain_scores_codex":[0.994687,0.001221556,0.0005108515,0.000383686,0.002976629,0.0002203486],"domain_scores_gemma":[0.9666191,0.01041841,0.001490752,0.001609045,0.01910886,0.000753901],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001160702,0.001323106,0.6165396,0.0007855216,0.0006794665,0.0003370671,0.02072537,0.001963409,0.00797815,0.007036158,0.01197939,0.3294921],"study_design_scores_gemma":[0.0002477407,0.0009648497,0.9650822,0.0002556848,0.0002663927,0.0001535818,0.006698729,0.009038779,0.002920359,0.004107811,0.0101399,0.0001240874],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9507841,0.0003362917,0.03306015,0.0006800299,0.0002146314,0.004251547,0.002805284,0.0002020044,0.007665757],"genre_scores_gemma":[0.9379354,0.0003716853,0.05116094,0.0001275385,0.00004714517,0.004481863,0.003092611,0.00005631356,0.00272654],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03449119,"threshold_uncertainty_score":0.0808031,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2089004370","doi":"10.1080/00273171.2014.911075","title":"Assessing Approximate Fit in Categorical Data Analysis","year":2014,"lang":"en","type":"article","venue":"Multivariate Behavioral Research","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":276,"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":"Mathematics; Statistics; Categorical variable; Goodness of fit; Bivariate analysis; Statistic; Ordinal data; Mean squared error; Multivariate statistics; Item response theory; Cutoff; Econometrics; Psychometrics","authors":[{"name":"Alberto Maydeu‐Olivares","is_ca":false},{"name":"Harry Joe","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.959382260741685,"gpt":0.7182746444274037,"spread":0.2411076163142812,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1853167,0.002273474,0.004362448,0.009788658,0.001749372,0.004915506,0.004477897,0.003097555,0.003612205],"category_scores_gemma":[0.6085298,0.001131181,0.004549775,0.01367571,0.004702247,0.00615772,0.005174622,0.004368249,0.001073871],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00222651,"about_ca_system_score_gemma":0.003653375,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002933175,"about_ca_topic_score_gemma":0.002677007,"domain_scores_codex":[0.7544858,0.1847604,0.01659153,0.01338498,0.0294333,0.001343885],"domain_scores_gemma":[0.353338,0.5617746,0.02364955,0.0389792,0.0210546,0.00120413],"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":[0.001500892,0.0005645445,0.2501815,0.0044841,0.00853987,0.001158656,0.007656331,0.1620012,0.002325465,0.1025904,0.01139359,0.4476036],"study_design_scores_gemma":[0.0002012239,0.001550396,0.05908572,0.001561979,0.0007643679,0.001430296,0.003336672,0.6160533,0.002384238,0.2992544,0.01394997,0.0004275152],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0219028,0.0008573756,0.9736075,0.0003492708,0.000137096,0.0005638013,0.0005568796,0.0005503497,0.001474927],"genre_scores_gemma":[0.411172,0.0005857712,0.5809577,0.0006048955,0.0001277327,0.003917444,0.001846521,0.00035836,0.0004296537],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.1853167,"threshold_uncertainty_score":0.9800599,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2128090963","doi":"10.1080/00223980009598225","title":"Common Method Variance and Specification Errors: A Practical Approach to Detection","year":2000,"lang":"en","type":"article","venue":"The Journal of Psychology","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":274,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Calgary","funders":"","keywords":"LISREL; Variance (accounting); Bivariate analysis; Psychology; Social desirability; Econometrics; Construct (python library); Common-method variance; Structural equation modeling; Specification; Point (geometry); Statistics; Correlation; Variable (mathematics); Social psychology; Computer science; Mathematics","authors":[{"name":"Theresa J. B. Kline","is_ca":true},{"name":"Lorne M. Sulsky","is_ca":true},{"name":"Sandra D. Rever-Moriyama","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5461551079132649,"gpt":0.5639502931705979,"spread":0.01779518525733292,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2441803,0.002865961,0.003051249,0.01017861,0.002172271,0.005007853,0.004321051,0.005763153,0.003607426],"category_scores_gemma":[0.6457635,0.0017453,0.003170929,0.01011136,0.007510279,0.006705314,0.006164868,0.009305167,0.0008072755],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00266981,"about_ca_system_score_gemma":0.005171466,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002856154,"about_ca_topic_score_gemma":0.002761868,"domain_scores_codex":[0.6259427,0.2581148,0.02424691,0.02569533,0.06410705,0.00189326],"domain_scores_gemma":[0.2208946,0.6783932,0.0315924,0.04627142,0.02210226,0.0007460663],"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":[0.0008695133,0.0008027409,0.05773655,0.003274925,0.002584041,0.002406515,0.01727094,0.01361441,0.004905047,0.3488749,0.01118671,0.5364737],"study_design_scores_gemma":[0.0007348269,0.001839678,0.06343182,0.005283691,0.001620426,0.005815066,0.006696334,0.2728275,0.01471595,0.5664057,0.05977783,0.0008511132],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.006823203,0.0003914818,0.9878875,0.0005749654,0.0002281833,0.001151927,0.00008365194,0.0004947499,0.002364393],"genre_scores_gemma":[0.2065728,0.0003266392,0.7858532,0.0008425422,0.0002283372,0.004187259,0.0001692622,0.0003199861,0.001500009],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.7558197,"threshold_uncertainty_score":0.9320602,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2089262464","doi":"10.1007/s11205-010-9682-8","title":"Assessing the Unidimensionality of Psychological Scales: Using Multiple Criteria from Factor Analysis","year":2010,"lang":"en","type":"article","venue":"Social Indicators Research","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":252,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Statistics; Item response theory; Factor analysis; Sample (material); Sample size determination; Scale (ratio); Psychology; Mathematics; Curse of dimensionality; Econometrics; Quality of Life Research; Psychometrics","authors":[{"name":"Suzanne L. Slocum-Gori","is_ca":true},{"name":"Bruno D. Zumbo","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8500569143669792,"gpt":0.7003382737954648,"spread":0.1497186405715144,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06284671,0.001714141,0.001897347,0.009105425,0.00208121,0.00492905,0.001194357,0.001035176,0.001175116],"category_scores_gemma":[0.2514673,0.0006562249,0.002431591,0.008014803,0.00272927,0.005482554,0.004147585,0.002034614,0.0002119721],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001473738,"about_ca_system_score_gemma":0.00421912,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002309817,"about_ca_topic_score_gemma":0.002884712,"domain_scores_codex":[0.9562067,0.02463895,0.005649495,0.002372931,0.01046469,0.0006672377],"domain_scores_gemma":[0.7504626,0.1891509,0.01417646,0.01797097,0.02637921,0.001859778],"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":[0.000704113,0.000876567,0.4047709,0.001537121,0.002496508,0.0001411524,0.01404183,0.008494752,0.0061991,0.04343103,0.002413532,0.5148934],"study_design_scores_gemma":[0.0003741152,0.002607579,0.6018821,0.001863047,0.001524756,0.0004004352,0.014584,0.1308163,0.008337888,0.2279599,0.00906724,0.0005825732],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.487497,0.001338443,0.4949277,0.001064311,0.0002337,0.002171875,0.0006792421,0.0004315627,0.01165618],"genre_scores_gemma":[0.7963914,0.0003442871,0.2005405,0.00006112791,0.00005144605,0.001824984,0.0005276382,0.00005920061,0.0001994248],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.06284671,"threshold_uncertainty_score":0.3323691,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2005967696","doi":"10.1080/00273171.2014.933697","title":"Adjusting Incremental Fit Indices for Nonnormality","year":2014,"lang":"en","type":"article","venue":"Multivariate Behavioral Research","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":223,"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":"Statistics; Mathematics; Index (typography); Structural equation modeling; Population; Goodness of fit; Econometrics; Value (mathematics); Sample (material); Computer science","authors":[{"name":"Patricia É. Brosseau-Liard","is_ca":true},{"name":"Victoria Savalei","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9278926090737645,"gpt":0.6852634534042872,"spread":0.2426291556694773,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04316852,0.001566154,0.00114202,0.004633424,0.001213809,0.002194243,0.002775757,0.001134119,0.004116185],"category_scores_gemma":[0.3161058,0.0006254057,0.00172642,0.005068489,0.001520983,0.003415432,0.002648064,0.003057046,0.001027576],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001444643,"about_ca_system_score_gemma":0.003116029,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004267641,"about_ca_topic_score_gemma":0.00536179,"domain_scores_codex":[0.9777879,0.01171357,0.001685803,0.002522154,0.005684163,0.0006064865],"domain_scores_gemma":[0.8086097,0.1333339,0.01034589,0.02092125,0.0256705,0.001118872],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0004798074,0.0005691434,0.2177963,0.0005020864,0.0008188697,0.0005161248,0.003474875,0.03407717,0.005394509,0.05477135,0.01638276,0.665217],"study_design_scores_gemma":[0.0003429895,0.001788676,0.2174446,0.0006119922,0.0009187235,0.001730022,0.002228731,0.5344338,0.02892286,0.1697866,0.04106936,0.0007217167],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1603017,0.0004567674,0.8275301,0.0007078297,0.0003492974,0.0008014873,0.0005832554,0.002466782,0.006802689],"genre_scores_gemma":[0.4684195,0.0001684594,0.5269958,0.0002638111,0.00005912183,0.001201782,0.0008328964,0.0008002404,0.001258379],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.04316852,"threshold_uncertainty_score":0.2282997,"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":"simulation_or_modeling","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"agree"},{"id":"W1991374387","doi":"10.1080/00273171.2012.715252","title":"An Investigation of the Sample Performance of Two Nonnormality Corrections for RMSEA","year":2012,"lang":"en","type":"article","venue":"Multivariate Behavioral Research","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":221,"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; Statistics; Mathematics; Goodness of fit; Sample (material); Sample size determination; Population; Econometrics; Chemistry","authors":[{"name":"Patricia É. Brosseau-Liard","is_ca":true},{"name":"Victoria Savalei","is_ca":true},{"name":"Libo Li","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.873633224993975,"gpt":0.6517804000643616,"spread":0.2218528249296134,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.09786294,0.001688239,0.001413734,0.002417909,0.001352223,0.001811532,0.002832839,0.002313857,0.002268395],"category_scores_gemma":[0.47649,0.0005993673,0.002089496,0.003684755,0.002165822,0.004547614,0.002640578,0.004068864,0.0006493673],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001556032,"about_ca_system_score_gemma":0.002540605,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00435038,"about_ca_topic_score_gemma":0.005149835,"domain_scores_codex":[0.9299946,0.05018006,0.003375462,0.006143354,0.009534615,0.0007718529],"domain_scores_gemma":[0.3961395,0.5184228,0.01358113,0.04320867,0.02760549,0.001042464],"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":[0.003253576,0.0009339309,0.1314121,0.001353998,0.003470416,0.001040125,0.006687087,0.0753549,0.01188696,0.07043649,0.01554027,0.6786301],"study_design_scores_gemma":[0.0008872827,0.003111039,0.1646909,0.001028733,0.001913422,0.002589957,0.003081357,0.6937407,0.03022802,0.07004828,0.02801123,0.0006691295],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2886908,0.003953224,0.6913357,0.003676161,0.001431756,0.0009420037,0.0008372507,0.002715178,0.0064179],"genre_scores_gemma":[0.6560919,0.0005464632,0.3382654,0.0006654463,0.0001658892,0.0009682995,0.001021964,0.000930325,0.00134427],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.902137,"threshold_uncertainty_score":0.5175549,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2159759566","doi":"10.1177/1094428114529165","title":"Reflections on Partial Least Squares Path Modeling","year":2014,"lang":"en","type":"article","venue":"Organizational Research Methods","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":217,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Public Safety Canada","funders":"","keywords":"Structural equation modeling; Partial least squares regression; Econometrics; Path analysis (statistics); Path (computing); Null hypothesis; Sample (material); Computer science; Key (lock); Management science; Data mining; Mathematics; Machine learning; Economics","authors":[{"name":"Cameron N. McIntosh","is_ca":true},{"name":"Jeffrey R. Edwards","is_ca":false},{"name":"John Antonakis","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8883201917735504,"gpt":0.7223769447242043,"spread":0.165943247049346,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1035089,0.001671571,0.00152891,0.00290899,0.002099999,0.006931147,0.003948605,0.004557236,0.007882873],"category_scores_gemma":[0.2283186,0.001157272,0.001979869,0.005499428,0.02183926,0.0139216,0.004845647,0.01672653,0.001909669],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004884267,"about_ca_system_score_gemma":0.00634241,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007017645,"about_ca_topic_score_gemma":0.004247389,"domain_scores_codex":[0.9021078,0.08213529,0.001957067,0.003709983,0.009367031,0.000722856],"domain_scores_gemma":[0.6101912,0.3597538,0.004564693,0.01008735,0.0142757,0.001127286],"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":[0.00001965541,0.00002839426,0.0007391574,0.0002197441,0.00003680778,0.00003540043,0.001293962,0.004480122,0.00003178442,0.9530067,0.006714799,0.0333934],"study_design_scores_gemma":[0.00001457233,0.00003388017,0.000253686,0.000234511,0.000009642955,0.00003465766,0.0005712118,0.01207656,0.0001249253,0.9654593,0.02115938,0.00002763482],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.005973547,0.007611197,0.8209217,0.1387645,0.001692507,0.0001479627,0.0003435817,0.0002513969,0.02429369],"genre_scores_gemma":[0.2600171,0.01826705,0.684487,0.02215453,0.003738298,0.001706643,0.0002926857,0.0006243467,0.008712257],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.1035089,"threshold_uncertainty_score":0.5474138,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2760364374","doi":"10.3102/0034654317726529","title":"Developing, Analyzing, and Using Distractors for Multiple-Choice Tests in Education: A Comprehensive Review","year":2017,"lang":"en","type":"review","venue":"Review of Educational Research","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":217,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"","keywords":"Task (project management); Quality (philosophy); Process (computing); Psychology; Cognitive psychology; Multiple choice; Computer science; Applied psychology; Medicine","authors":[{"name":"Mark J. Gierl","is_ca":true},{"name":"Okan Bulut","is_ca":true},{"name":"Qi Guo","is_ca":true},{"name":"Xinxin Zhang","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9466681700442642,"gpt":0.7483343693634743,"spread":0.1983338006807899,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01750243,0.001710907,0.003064554,0.008125543,0.0005182126,0.002582664,0.002520105,0.001909873,0.003294997],"category_scores_gemma":[0.04834219,0.000737999,0.002303149,0.008366144,0.001660925,0.003310092,0.001323319,0.001862786,0.00117783],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001843823,"about_ca_system_score_gemma":0.007643739,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004528125,"about_ca_topic_score_gemma":0.009332346,"domain_scores_codex":[0.9865631,0.005324154,0.002952297,0.0008768407,0.004091548,0.0001920877],"domain_scores_gemma":[0.9229996,0.06711262,0.003674079,0.0008094588,0.005064957,0.0003392966],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00004491158,0.00007602288,0.0005907072,0.08480968,0.0002352328,0.00005912361,0.0002198515,0.0002855082,0.0002179695,0.002130728,0.004329993,0.9070003],"study_design_scores_gemma":[0.0001044696,0.0005575147,0.00884002,0.2961251,0.003484318,0.001868617,0.001113846,0.0009779353,0.002338011,0.01197102,0.6723614,0.0002577849],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0002482248,0.9974374,0.001021503,0.0004880391,0.0001021361,0.00005667032,0.00003884082,0.00001474836,0.0005923814],"genre_scores_gemma":[0.002155871,0.9933705,0.003885901,0.0002404659,0.0000924074,0.00007831497,0.00004749577,0.000007136405,0.0001219368],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9824976,"threshold_uncertainty_score":0.09256279,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W2142004159","doi":"10.3899/jrheum.130813","title":"Item Response Theory, Computerized Adaptive Testing, and PROMIS: Assessment of Physical Function","year":2013,"lang":"en","type":"article","venue":"The Journal of Rheumatology","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":208,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"National Institute of Arthritis and Musculoskeletal and Skin Diseases","keywords":"Computerized adaptive testing; Patient-Reported Outcomes Measurement Information System; Item response theory; Item bank; CLARITY; Medicine; Reliability (semiconductor); Standardization; Perspective (graphical); Short Forms; Construct (python library); Applied psychology; Function (biology); Psychometrics; Computer science; Psychology; Clinical psychology; Artificial intelligence","authors":[{"name":"James F. Fries","is_ca":false},{"name":"James Witter","is_ca":false},{"name":"Matthias Rose","is_ca":false},{"name":"David Cella","is_ca":false},{"name":"Dinesh Khanna","is_ca":false},{"name":"Esi Morgan-DeWitt","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2335860100349148,"gpt":0.4218876205134344,"spread":0.1883016104785196,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03893904,0.001956327,0.002096681,0.006277141,0.0005139171,0.001547497,0.002088175,0.001750572,0.005746262],"category_scores_gemma":[0.1074202,0.0004529819,0.002025729,0.01152929,0.001423411,0.002077831,0.002520027,0.003374055,0.002693571],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001877078,"about_ca_system_score_gemma":0.003768194,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001639657,"about_ca_topic_score_gemma":0.001493077,"domain_scores_codex":[0.8934825,0.08685353,0.004762859,0.002287758,0.01205563,0.0005576941],"domain_scores_gemma":[0.9454875,0.03565212,0.008493402,0.002819315,0.006859671,0.000687935],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.001166531,0.001322663,0.06975468,0.007402937,0.001879806,0.0002238467,0.001471587,0.01072439,0.0009973865,0.03147014,0.04650845,0.8270776],"study_design_scores_gemma":[0.002620512,0.00622469,0.5763922,0.01189532,0.002263198,0.005749926,0.002112772,0.06208725,0.002787787,0.146112,0.1810233,0.0007311305],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.09679815,0.06298154,0.6988117,0.0168528,0.001993426,0.0356941,0.02186958,0.004347635,0.06065103],"genre_scores_gemma":[0.380141,0.02061686,0.5076527,0.00548115,0.001137567,0.06560244,0.01348241,0.0002615754,0.005624328],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.03893904,"threshold_uncertainty_score":0.2059318,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2073056556","doi":"10.1016/j.jclinepi.2005.09.005","title":"“Precision” and “Accuracy”: Two Terms That Are Neither","year":2006,"lang":"en","type":"review","venue":"Journal of Clinical Epidemiology","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":194,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Baycrest Hospital; McMaster University; University of Toronto","funders":"","keywords":"Reliability (semiconductor); Conceptualization; Inter-rater reliability; Scale (ratio); Test (biology); Computer science; Synonym (taxonomy); Validity; Criterion validity; Psychology; Process (computing); Cognitive psychology; Psychometrics; Artificial intelligence; Statistics; Mathematics; Rating scale; Construct validity; Geography","authors":[{"name":"David L. Streiner","is_ca":true},{"name":"Geoffrey R. Norman","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9346514191644626,"gpt":0.7216304917470712,"spread":0.2130209274173914,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.08440655,0.003306514,0.01289451,0.01468502,0.001677286,0.01400916,0.006889733,0.01176646,0.004662692],"category_scores_gemma":[0.2399666,0.001902039,0.003642855,0.01617365,0.03032742,0.02654573,0.007876617,0.01430784,0.003336005],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003495447,"about_ca_system_score_gemma":0.01107837,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001778671,"about_ca_topic_score_gemma":0.002015358,"domain_scores_codex":[0.8563383,0.06882838,0.02596654,0.006863461,0.04044553,0.001557832],"domain_scores_gemma":[0.610328,0.3351598,0.02501561,0.009708659,0.01755492,0.002233133],"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":[0.0003798386,0.0001282789,0.002162852,0.04512649,0.001834178,0.0001744826,0.001925149,0.001050344,0.0004769255,0.2417904,0.04069464,0.6642565],"study_design_scores_gemma":[0.0004336022,0.000448141,0.005033592,0.06269553,0.003618574,0.003334086,0.002380859,0.001780316,0.001142711,0.547768,0.3708443,0.0005203561],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0008315027,0.8730987,0.05673955,0.05375074,0.009579377,0.0001012941,0.0002120639,0.0001218141,0.005564966],"genre_scores_gemma":[0.07494909,0.6804658,0.1148896,0.07850616,0.04336712,0.001682391,0.0003660358,0.0002913956,0.005482451],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9155934,"threshold_uncertainty_score":0.4463899,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2431662475","doi":"10.1371/journal.pone.0157795","title":"Examining the Effect of Reverse Worded Items on the Factor Structure of the Need for Cognition Scale","year":2016,"lang":"en","type":"article","venue":"PLoS ONE","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":194,"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; Confirmatory factor analysis; Scale (ratio); Covariance; Factor analysis; Scale factor (cosmology); Factor (programming language); Cognition; Statistics; Psychometrics; Structural equation modeling; Developmental psychology; Mathematics; Computer science","authors":[{"name":"Xijuan Zhang","is_ca":true},{"name":"Ramsha Noor","is_ca":true},{"name":"Victoria Savalei","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5134880551927866,"gpt":0.3872550404907291,"spread":0.1262330147020576,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.04677668,0.001427566,0.0008977913,0.0009445358,0.0008817427,0.001876984,0.0008813345,0.0009160815,0.004350483],"category_scores_gemma":[0.1889394,0.0009399116,0.002220491,0.0009207583,0.002511975,0.001752726,0.001950488,0.002323744,0.0008019017],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009945835,"about_ca_system_score_gemma":0.001457491,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002260488,"about_ca_topic_score_gemma":0.004267266,"domain_scores_codex":[0.9475862,0.0358097,0.004017279,0.004751068,0.006901817,0.0009340452],"domain_scores_gemma":[0.6723993,0.2701934,0.02031272,0.02312893,0.01269565,0.001270078],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.006294232,0.002347892,0.7004659,0.0008195693,0.002297291,0.0005669692,0.02775894,0.002750936,0.03584223,0.004303707,0.002186946,0.2143654],"study_design_scores_gemma":[0.0006941109,0.005022535,0.9461076,0.0004491807,0.000914716,0.0004379553,0.004338363,0.01607208,0.01708331,0.003648148,0.004939523,0.0002924796],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.982627,0.0001596347,0.01438365,0.0001664056,0.0001172728,0.0003383288,0.0001140308,0.00007350444,0.002020127],"genre_scores_gemma":[0.9762112,0.0001117509,0.02142325,0.0001470192,0.00002673405,0.0009102064,0.0002430664,0.0000946427,0.0008321413],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9532233,"threshold_uncertainty_score":0.2473817,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2364328457","doi":"10.1007/978-3-319-07794-9","title":"Validity and Validation in Social, Behavioral, and Health Sciences","year":2014,"lang":"en","type":"book","venue":"Social indicators research series","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":194,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Behavioural sciences; Psychology; Data science; Computer science; Psychotherapist","authors":[{"name":"Bruno D. Zumbo","is_ca":true},{"name":"Eric K. H. Chan","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8242681565438117,"gpt":0.6357213862676169,"spread":0.1885467702761948,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1473631,0.002162323,0.003189668,0.01405198,0.00193467,0.007176831,0.002311154,0.003499738,0.003471903],"category_scores_gemma":[0.3222251,0.001625605,0.001987219,0.01005796,0.01513673,0.01079248,0.005269266,0.00684711,0.002209773],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004212887,"about_ca_system_score_gemma":0.01168973,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008371486,"about_ca_topic_score_gemma":0.007463596,"domain_scores_codex":[0.8820859,0.08170141,0.01034779,0.00323835,0.02172725,0.0008992989],"domain_scores_gemma":[0.5552329,0.3910984,0.00659783,0.0192529,0.02657454,0.001243399],"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":[0.00008430265,0.000201467,0.0147168,0.001670026,0.0002324709,0.000128663,0.003631367,0.002302728,0.0005101863,0.3120616,0.0444228,0.6200376],"study_design_scores_gemma":[0.00006776948,0.0002666938,0.02624173,0.009578217,0.0002306195,0.0006563863,0.002598481,0.010142,0.00172149,0.8460587,0.1022648,0.0001731733],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0204599,0.2088825,0.572464,0.03906704,0.00752952,0.002402243,0.00141229,0.001386777,0.1463956],"genre_scores_gemma":[0.2685459,0.1138051,0.5720927,0.007484007,0.006544968,0.007302758,0.002468716,0.00139748,0.0203584],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8526369,"threshold_uncertainty_score":0.7793399,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2170853093","doi":"10.1111/j.1745-3992.2003.tb00136.x","title":"Using Multidimensional Item Response Theory to Evaluate Educational and Psychological Tests","year":2003,"lang":"en","type":"article","venue":"Educational Measurement Issues and Practice","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":192,"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":"Item response theory; Test (biology); Measure (data warehouse); Ninth; Computer science; Process (computing); Meaning (existential); Multidimensional analysis; Psychology; Mathematics education; Psychometrics; Management science; Econometrics; Mathematics; Data mining; Developmental psychology; Psychotherapist","authors":[{"name":"Terry A. Ackerman","is_ca":false},{"name":"Mark J. Gierl","is_ca":true},{"name":"Cindy M. Walker","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8128600435609681,"gpt":0.6116511900217015,"spread":0.2012088535392665,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04339649,0.001033863,0.001226441,0.009163097,0.0006829954,0.004670768,0.001471392,0.00145226,0.003420649],"category_scores_gemma":[0.17624,0.0003370776,0.001570862,0.007322065,0.001610456,0.004032248,0.002428999,0.001924938,0.001351627],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002130138,"about_ca_system_score_gemma":0.00176599,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001305096,"about_ca_topic_score_gemma":0.001607282,"domain_scores_codex":[0.9479811,0.03645249,0.0038254,0.00138038,0.009746374,0.0006141915],"domain_scores_gemma":[0.8472014,0.1199909,0.008324559,0.006960807,0.01671018,0.0008121657],"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":[0.0002041943,0.001139192,0.1803217,0.001075189,0.0007259836,0.0001570436,0.004814392,0.03045269,0.001789077,0.08034903,0.008778763,0.6901928],"study_design_scores_gemma":[0.0003561981,0.004224301,0.320914,0.002857245,0.0004318709,0.001121471,0.01565909,0.2767255,0.008473374,0.299978,0.06858903,0.0006698622],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.234611,0.002348512,0.7009992,0.003197948,0.000456961,0.004598121,0.001724204,0.001287763,0.05077626],"genre_scores_gemma":[0.4826855,0.001158635,0.506171,0.0005711629,0.0001095543,0.005266155,0.001834797,0.0001350841,0.002068012],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.04339649,"threshold_uncertainty_score":0.2295053,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2768881033","doi":"10.1037/ipp0000042","title":"Adaptation of Assessment Scales in Cross-National Research: Issues, Guidelines, and Caveats","year":2015,"lang":"en","type":"article","venue":"International Perspectives in Psychology","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":183,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Ottawa","funders":"","keywords":"Equivalence (formal languages); Criticism; Scale (ratio); Set (abstract data type); Adaptation (eye); Psychology; Computer science; Management science; Data science; Political science; Mathematics; Geography; Engineering","authors":[{"name":"Barbara M. Byrne","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9019762010793025,"gpt":0.722889276717478,"spread":0.1790869243618245,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5514816,0.002650469,0.003655323,0.008149463,0.007110244,0.01350858,0.01163782,0.008312499,0.003763509],"category_scores_gemma":[0.7486199,0.004056482,0.004704888,0.01740324,0.02168977,0.01733611,0.01201111,0.02341486,0.003255937],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.007108877,"about_ca_system_score_gemma":0.02958779,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01083931,"about_ca_topic_score_gemma":0.03280259,"domain_scores_codex":[0.3408393,0.5300882,0.07550613,0.007144206,0.04410351,0.002318574],"domain_scores_gemma":[0.1907385,0.6141142,0.01789631,0.0630304,0.112121,0.002099644],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0004067685,0.000686984,0.01540035,0.01467884,0.000550726,0.0005926862,0.05364981,0.001896254,0.001139942,0.3232871,0.1261825,0.461528],"study_design_scores_gemma":[0.000447476,0.0007516771,0.02330498,0.07993384,0.0005635657,0.002342578,0.03050078,0.01155633,0.002416611,0.3150545,0.5324154,0.0007122558],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0201714,0.03788426,0.6511277,0.1653628,0.02486353,0.03493638,0.002072659,0.002390668,0.06119068],"genre_scores_gemma":[0.03333942,0.006135121,0.8982655,0.01929305,0.001531824,0.03833994,0.0005083996,0.0005816251,0.002005117],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.4485184,"threshold_uncertainty_score":0.5531031,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2166319027","doi":"10.1177/070674370605100507","title":"Building a Better Model: An Introduction to Structural Equation Modelling","year":2006,"lang":"en","type":"article","venue":"The Canadian Journal of Psychiatry","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":183,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Structural equation modeling; Latent variable; Confirmatory factor analysis; Path analysis (statistics); Exploratory factor analysis; Path coefficient; A priori and a posteriori; Econometrics; Latent variable model; Variable (mathematics); Scale (ratio); Statistics; Psychology; Factor analysis; Variables; Construct (python library); Path (computing); Psychometrics; Mathematics; Computer science","authors":[{"name":"David L. Streiner","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2531690852912864,"gpt":0.3971817068862711,"spread":0.1440126215949847,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01762945,0.00277535,0.002473789,0.004137295,0.001272742,0.00497033,0.003621293,0.003024605,0.01668718],"category_scores_gemma":[0.0456832,0.001647153,0.003399739,0.006783809,0.002925206,0.009039386,0.003062421,0.007827298,0.004293698],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003191603,"about_ca_system_score_gemma":0.004332602,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006121951,"about_ca_topic_score_gemma":0.007680598,"domain_scores_codex":[0.9860995,0.01012184,0.0006683181,0.001021826,0.001908507,0.0001800482],"domain_scores_gemma":[0.9691558,0.0269957,0.0009391063,0.001191483,0.001492604,0.0002252815],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00003081678,0.0001364709,0.001729897,0.002364327,0.0006096013,0.0002478241,0.001604959,0.043111,0.0004074192,0.7091183,0.03698336,0.203656],"study_design_scores_gemma":[0.00003184501,0.0000657945,0.000566219,0.00166092,0.0001116314,0.0001673402,0.0004486596,0.06958611,0.0001956633,0.8249506,0.1021051,0.0001100708],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0009232505,0.008461025,0.9734908,0.007555684,0.0004433194,0.0003257045,0.0006094793,0.0005500098,0.007640731],"genre_scores_gemma":[0.01497504,0.009885233,0.970044,0.0009512038,0.0005484065,0.0009510361,0.0006381655,0.0002599411,0.001746866],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.01762945,"threshold_uncertainty_score":0.0932346,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2240102487","doi":"10.1007/s12144-015-9404-0","title":"Using the Gamma Generalized Linear Model for Modeling Continuous, Skewed and Heteroscedastic Outcomes in Psychology","year":2016,"lang":"en","type":"article","venue":"Current Psychology","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":178,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"York University","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Homoscedasticity; Heteroscedasticity; Psychology; Normality; Econometrics; Outcome (game theory); Linear regression; Linear model; Generalized linear model; Variance (accounting); Statistics; General linear model; Range (aeronautics); Mathematics; Social psychology; Mathematical economics; Engineering; Economics","authors":[{"name":"Victoria Ka Yin Ng","is_ca":true},{"name":"Robert A. Cribbie","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7652380167565765,"gpt":0.5944211459043176,"spread":0.1708168708522589,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02768444,0.00193656,0.001789903,0.002199857,0.001019133,0.00373531,0.003868656,0.002641971,0.005337352],"category_scores_gemma":[0.09174776,0.001000659,0.003237424,0.003827237,0.001998166,0.003465519,0.003126196,0.005105277,0.001500611],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002776245,"about_ca_system_score_gemma":0.005521496,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02125395,"about_ca_topic_score_gemma":0.01383419,"domain_scores_codex":[0.9811952,0.01442014,0.0005036615,0.001881018,0.001237594,0.0007624095],"domain_scores_gemma":[0.9515391,0.03934474,0.002017345,0.004199762,0.002447827,0.0004511059],"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":[0.001034711,0.0007386135,0.07842694,0.0007541919,0.002315162,0.0008796811,0.004136434,0.4269298,0.001696929,0.255434,0.01207905,0.2155745],"study_design_scores_gemma":[0.0001917703,0.000548316,0.01710925,0.0003264476,0.000407558,0.0003460406,0.0009821121,0.6923064,0.001032145,0.2793869,0.007175511,0.0001874379],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.06029372,0.0003966272,0.9347233,0.000655294,0.0001721889,0.0004842262,0.0008505623,0.000946915,0.001477151],"genre_scores_gemma":[0.6449842,0.001122005,0.3393406,0.000642646,0.0001607687,0.003294266,0.002602172,0.0005775698,0.007275622],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.02768444,"threshold_uncertainty_score":0.1464111,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2077199939","doi":"10.1007/s11336-009-9123-2","title":"When Cognitive Diagnosis Meets Computerized Adaptive Testing: CD-CAT","year":2009,"lang":"en","type":"article","venue":"Psychometrika","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":176,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Heuristics; Computerized adaptive testing; Entropy (arrow of time); Information theory; Item response theory; Computer science; Selection (genetic algorithm); Context (archaeology); Latent class model; Kullback–Leibler divergence; Artificial intelligence; Mathematics; Machine learning; Algorithm; Statistics; Mathematical optimization; Psychometrics","authors":[{"name":"Ying Cheng","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5850420918789638,"gpt":0.4702225324092145,"spread":0.1148195594697494,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02296867,0.0007717189,0.0016587,0.002667945,0.001316775,0.00388827,0.002449526,0.007001451,0.009817972],"category_scores_gemma":[0.1625264,0.0005807714,0.0008747557,0.002297346,0.002956237,0.00655891,0.004969101,0.003810309,0.001390355],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003032775,"about_ca_system_score_gemma":0.005682112,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005487372,"about_ca_topic_score_gemma":0.004348063,"domain_scores_codex":[0.9693515,0.01189606,0.003184639,0.005913744,0.007291699,0.002362429],"domain_scores_gemma":[0.9003233,0.07302517,0.005341008,0.00778797,0.009181382,0.004341207],"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":[0.005160226,0.001298909,0.2780442,0.0008760993,0.0003951978,0.006666537,0.002296213,0.02522545,0.004133022,0.1866186,0.03592801,0.4533575],"study_design_scores_gemma":[0.001339997,0.003156388,0.0659805,0.0007334474,0.0003574329,0.01707237,0.003309135,0.2664193,0.01147328,0.5701072,0.0596872,0.0003638589],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"methods","genre_scores_codex":[0.460125,0.004221083,0.3887239,0.06823914,0.002212047,0.003083372,0.003165041,0.002915639,0.06731487],"genre_scores_gemma":[0.9078565,0.0004012964,0.08174129,0.005998499,0.0002553339,0.0007365221,0.0006822697,0.0001095471,0.002218719],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.02296867,"threshold_uncertainty_score":0.1214715,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2135663628","doi":"10.1177/0265532208097336","title":"Cognitive diagnostic assessment of L2 reading comprehension ability: Validity arguments for Fusion Model application to <i>LanguEdge</i> assessment","year":2008,"lang":"en","type":"article","venue":"Language Testing","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":175,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Reading comprehension; Psychology; Cognition; Test (biology); Profiling (computer programming); Dependability; Comprehension; Reading (process); Cognitive psychology; Computer science; Linguistics","authors":[{"name":"Eunice Eunhee Jang","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4824119830634888,"gpt":0.5159648486704076,"spread":0.03355286560691884,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0486821,0.0007586219,0.001072731,0.004358161,0.0009653732,0.003803343,0.002019352,0.001754677,0.002137555],"category_scores_gemma":[0.2508255,0.0003459126,0.001366724,0.002192287,0.004630828,0.004936526,0.005208767,0.00225757,0.0003793943],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00236771,"about_ca_system_score_gemma":0.00219577,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003331222,"about_ca_topic_score_gemma":0.001666186,"domain_scores_codex":[0.9650055,0.02072003,0.001915497,0.002848978,0.008765366,0.0007446035],"domain_scores_gemma":[0.7637811,0.1912163,0.008795157,0.0173636,0.01699496,0.001848906],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003420413,0.0008683964,0.4591054,0.0006854163,0.000779748,0.0006137151,0.01071717,0.02013513,0.007388768,0.07253753,0.003212869,0.4205355],"study_design_scores_gemma":[0.0003997541,0.002202476,0.1764929,0.0005914565,0.0005031759,0.001437463,0.005373914,0.6001076,0.01589247,0.1918741,0.004769902,0.0003548207],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7499437,0.0008948687,0.2258897,0.005164768,0.0001694286,0.0006953829,0.0004554502,0.000531794,0.01625488],"genre_scores_gemma":[0.9653413,0.00006738762,0.03382828,0.0002036996,0.00003523198,0.0002243116,0.00009728307,0.00001957974,0.0001829591],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0486821,"threshold_uncertainty_score":0.2574586,"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":"W2039891070","doi":"10.3758/bf03192810","title":"SPSS and SAS programs for generalizability theory analyses","year":2006,"lang":"en","type":"article","venue":"Behavior Research Methods","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":166,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Lakehead University","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Generalizability theory; Computer science; Psychology; Developmental psychology","authors":[{"name":"Christopher J. Mushquash","is_ca":true},{"name":"Brian P. O’Connor","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9596454361401149,"gpt":0.7860996307771648,"spread":0.1735458053629501,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01988859,0.001789949,0.002228324,0.006665637,0.001306197,0.003127971,0.001568871,0.0007923092,0.1087888],"category_scores_gemma":[0.1347008,0.00153132,0.002311159,0.006873764,0.0009402372,0.003117946,0.001928485,0.00543518,0.02568113],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001303836,"about_ca_system_score_gemma":0.004971946,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003136103,"about_ca_topic_score_gemma":0.004010021,"domain_scores_codex":[0.9799625,0.01066444,0.002737831,0.00172618,0.004144711,0.0007644312],"domain_scores_gemma":[0.8176038,0.1321396,0.006784012,0.02571976,0.01696202,0.0007908967],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000925875,0.001797459,0.01616076,0.002965647,0.0007287257,0.000347219,0.005710913,0.005307489,0.003049587,0.07330794,0.3457654,0.543933],"study_design_scores_gemma":[0.00155664,0.001730733,0.06264229,0.002296957,0.001448732,0.0008098196,0.006435853,0.05954558,0.0223147,0.2355008,0.6053033,0.0004146054],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"software","genre_scores_codex":[0.03621095,0.0003097749,0.7378261,0.001493827,0.001370726,0.01423138,0.07030912,0.06365109,0.07459705],"genre_scores_gemma":[0.07189094,0.0004464766,0.8045235,0.0006020825,0.0003290198,0.04820314,0.02391244,0.01599695,0.03409544],"genre_candidate":"software","genre_consensus":null,"teacher_disagreement_score":0.1087888,"threshold_uncertainty_score":0.3639348,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2285879792","doi":"10.1080/00224499.2015.1137538","title":"A Methodological Review of Exploratory Factor Analysis in Sexuality Research: Used Practices, Best Practices, and Data Analysis Resources","year":2016,"lang":"en","type":"review","venue":"The Journal of Sex Research","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":161,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Human sexuality; Exploratory factor analysis; Sample (material); Psychology; Exploratory research; Set (abstract data type); Best practice; Social psychology; Sociology; Psychometrics; Computer science; Clinical psychology; Social science; Political science; Gender studies","authors":[{"name":"John Kitchener Sakaluk","is_ca":true},{"name":"Stephen D. Short","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9901045164332588,"gpt":0.7777037022173531,"spread":0.2124008142159057,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1131326,0.001747946,0.003770166,0.02700509,0.002234141,0.005900602,0.002932213,0.002092642,0.003963319],"category_scores_gemma":[0.2769586,0.001664891,0.004051795,0.03114607,0.002992039,0.00680855,0.003634275,0.00269866,0.00133629],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.008290716,"about_ca_system_score_gemma":0.03082654,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009025148,"about_ca_topic_score_gemma":0.0202152,"domain_scores_codex":[0.8927017,0.06197308,0.0252749,0.003305104,0.01599644,0.0007487968],"domain_scores_gemma":[0.7011442,0.2254778,0.01693745,0.007132119,0.04828963,0.001018806],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0000927771,0.00005208897,0.002004093,0.1928492,0.0009204926,0.0001309379,0.003499904,0.0003403253,0.000385055,0.008552911,0.02063459,0.7705377],"study_design_scores_gemma":[0.00009059835,0.0001893051,0.006872069,0.5686455,0.003737582,0.0007769538,0.003792429,0.0004184318,0.0008095739,0.01287826,0.4016224,0.0001667433],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0008086029,0.98125,0.009887166,0.003548717,0.0005948832,0.001067626,0.0004166656,0.00004573622,0.002380578],"genre_scores_gemma":[0.007142878,0.9475569,0.03965152,0.001245328,0.0002867621,0.003302701,0.0003324252,0.00005208603,0.0004294629],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.8868673,"threshold_uncertainty_score":0.5983098,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"design_other","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W2811232920","doi":"10.1177/0272989x18776637","title":"Estimation of a Preference-Based Summary Score for the Patient-Reported Outcomes Measurement Information System: The PROMIS <sup>®</sup> -Preference (PROPr) Scoring System","year":2018,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":160,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University","funders":"National Institutes of Health; National Cancer Institute; Social Sciences and Humanities Research Council of Canada; National Center for Advancing Translational Sciences; National Institute of Arthritis and Musculoskeletal and Skin Diseases; Riksbankens Jubileumsfond","keywords":"Patient-Reported Outcomes Measurement Information System; Preference; Quality of life (healthcare); Medicine; Applied psychology; Physical therapy; Psychology; Statistics; Psychometrics; Clinical psychology; Computerized adaptive testing; Mathematics","authors":[{"name":"Barry Dewitt","is_ca":false},{"name":"David Feeny","is_ca":true},{"name":"Baruch Fischhoff","is_ca":false},{"name":"David Cella","is_ca":false},{"name":"Ron D. Hays","is_ca":false},{"name":"Rachel Hess","is_ca":false},{"name":"Paul A. Pilkonis","is_ca":false},{"name":"Dennis A. Revicki","is_ca":false},{"name":"Mark S. Roberts","is_ca":false},{"name":"Joel Tsevat","is_ca":false},{"name":"Lan Yu","is_ca":false},{"name":"Janel Hanmer","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4962130764892222,"gpt":0.4115047991206986,"spread":0.0847082773685236,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02011336,0.0007298635,0.0009702279,0.003336092,0.0002606008,0.001193418,0.0009081989,0.0007255228,0.00695418],"category_scores_gemma":[0.0612493,0.0002499207,0.002908292,0.003273527,0.0003378697,0.001027099,0.001028491,0.001105966,0.001673814],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009306574,"about_ca_system_score_gemma":0.001222088,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001169742,"about_ca_topic_score_gemma":0.001592514,"domain_scores_codex":[0.9835877,0.01062979,0.001901264,0.0007931576,0.002754976,0.0003332281],"domain_scores_gemma":[0.9748776,0.01399123,0.004907839,0.001717605,0.004094248,0.000411536],"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":[0.002853385,0.000716926,0.5422422,0.001836009,0.00198134,0.0001397345,0.0007534287,0.01115809,0.001139268,0.006817371,0.03231953,0.3980428],"study_design_scores_gemma":[0.001029649,0.004924516,0.8494917,0.001141342,0.001050409,0.0009847659,0.001577506,0.09162503,0.004235802,0.01382159,0.02981215,0.0003054593],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.57498,0.002395483,0.2940501,0.002318409,0.0002968192,0.01410946,0.08108769,0.001126758,0.02963528],"genre_scores_gemma":[0.8051679,0.0006719135,0.1581267,0.000390836,0.00006388113,0.01291974,0.02085281,0.0000840486,0.001722038],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02011336,"threshold_uncertainty_score":0.1063709,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2157620189","doi":"10.3899/jrheum.110402","title":"The PROMIS of Better Outcome Assessment: Responsiveness, Floor and Ceiling Effects, and Internet Administration","year":2011,"lang":"en","type":"article","venue":"The Journal of Rheumatology","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":156,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"National Institute of Arthritis and Musculoskeletal and Skin Diseases; National Institutes of Health; Stony Brook University; University of Pittsburgh; University of Washington; Northwestern University","keywords":"Ceiling effect; Medicine; Ceiling (cloud); Computerized adaptive testing; Item response theory; Sample size determination; Physical therapy; Rheumatoid arthritis; Physical medicine and rehabilitation; Psychometrics; Internal medicine; Statistics; Alternative medicine; Pathology; Clinical psychology","authors":[{"name":"James F. Fries","is_ca":false},{"name":"Matthias Rose","is_ca":false},{"name":"Eswar Krishnan","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2544543516444623,"gpt":0.4364932432144383,"spread":0.1820388915699759,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0610722,0.001084704,0.001799443,0.003420366,0.0004959364,0.001690991,0.0013131,0.0007852087,0.003817585],"category_scores_gemma":[0.1651222,0.000643635,0.003278488,0.005982286,0.001176224,0.001630076,0.003359651,0.002499352,0.000685559],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001156492,"about_ca_system_score_gemma":0.003489103,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009737923,"about_ca_topic_score_gemma":0.001357304,"domain_scores_codex":[0.9054979,0.07524727,0.005705742,0.001914083,0.01094688,0.0006881182],"domain_scores_gemma":[0.862746,0.09800357,0.02257716,0.007745255,0.007816464,0.001111522],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.004890011,0.001363692,0.5253822,0.003683841,0.00404295,0.0001752107,0.002129717,0.008547811,0.00176533,0.01319365,0.01492632,0.4198993],"study_design_scores_gemma":[0.001964784,0.005674725,0.9133701,0.002039323,0.001524666,0.0009579103,0.0005711688,0.0241826,0.00345861,0.02108989,0.02497044,0.0001957154],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5868615,0.02227487,0.2684026,0.01052587,0.001273005,0.01754661,0.01818166,0.002469389,0.0724646],"genre_scores_gemma":[0.8284899,0.003963834,0.1282546,0.001428126,0.0005217265,0.02825178,0.006233042,0.0002690714,0.002587848],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0610722,"threshold_uncertainty_score":0.3229845,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1995508717","doi":"10.1037/1082-989x.6.3.258","title":"Testing parameters in structural equation modeling: Every \"one\" matters.","year":2001,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":152,"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; Identification (biology); Standard error; Statistical model; Statistics; Set (abstract data type); Statistical hypothesis testing; Mathematics; Invariant (physics); Econometrics; Likelihood-ratio test; Scale (ratio); Applied mathematics; Computer science; Physics","authors":[{"name":"Richard Gonzalez","is_ca":false},{"name":"Dale W. Griffin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8579973400321181,"gpt":0.6038590198508541,"spread":0.254138320181264,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1279731,0.001912369,0.002152815,0.004496465,0.00232094,0.006927496,0.003146171,0.004848904,0.004302149],"category_scores_gemma":[0.5113057,0.001190332,0.002008633,0.007395243,0.009646317,0.01931014,0.005317865,0.01067025,0.001602256],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002169133,"about_ca_system_score_gemma":0.005159663,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001826776,"about_ca_topic_score_gemma":0.001942897,"domain_scores_codex":[0.8746372,0.1010292,0.004501185,0.004955043,0.01422124,0.0006562084],"domain_scores_gemma":[0.5480332,0.3993759,0.01312147,0.02703453,0.01087559,0.001559383],"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":[0.0001882238,0.0003318337,0.03304484,0.001550555,0.001309546,0.0002180125,0.00541467,0.00783158,0.0008451301,0.5795932,0.01465678,0.3550157],"study_design_scores_gemma":[0.0001098814,0.0003518112,0.007954169,0.001477082,0.0003611017,0.0003253693,0.002740416,0.02102917,0.001325493,0.9453074,0.01884772,0.0001703368],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02414445,0.007768572,0.9256642,0.02732378,0.001378387,0.0005429653,0.000372943,0.000847201,0.01195749],"genre_scores_gemma":[0.4306022,0.006176129,0.5508698,0.006393721,0.0008296815,0.002471504,0.0005219321,0.0004967376,0.001638308],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.1279731,"threshold_uncertainty_score":0.6767945,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2005249146","doi":"10.1007/s11205-011-9843-4","title":"Validity and the Consequences of Test Interpretation and Use","year":2011,"lang":"en","type":"article","venue":"Social Indicators Research","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":149,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Cognitive reframing; Interpretation (philosophy); Construct validity; Psychology; Test (biology); Construct (python library); Social psychology; Quality of Life Research; Value (mathematics); Meaning (existential); Test validity; External validity; Psychometrics; Applied psychology; Public health; Developmental psychology; Computer science; Medicine; Statistics; Mathematics","authors":[{"name":"Anita M. Hubley","is_ca":true},{"name":"Bruno D. Zumbo","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8223478813702115,"gpt":0.579589150324663,"spread":0.2427587310455486,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.4885646,0.001905187,0.002509027,0.01015488,0.004654189,0.01227103,0.005414493,0.006554813,0.0022529],"category_scores_gemma":[0.8837844,0.001780528,0.003437825,0.007097366,0.03764729,0.01532551,0.008032512,0.007499733,0.0006784843],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.008068381,"about_ca_system_score_gemma":0.009902392,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006687575,"about_ca_topic_score_gemma":0.005938133,"domain_scores_codex":[0.3673476,0.4662299,0.03506152,0.02385939,0.1035404,0.003961232],"domain_scores_gemma":[0.03665493,0.8891597,0.01965638,0.03204976,0.02167418,0.0008051973],"domain_codex":"methods","domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.001894839,0.0007380336,0.5364481,0.001792981,0.003140507,0.0006085646,0.02480683,0.006134736,0.001133125,0.1983328,0.00709326,0.2178762],"study_design_scores_gemma":[0.0006336313,0.001092004,0.2415487,0.004277037,0.001871279,0.001621081,0.01146041,0.05166855,0.01123918,0.6597077,0.01436218,0.0005182429],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6296203,0.01363306,0.2295671,0.05377368,0.002603462,0.00192013,0.001006445,0.0005246787,0.0673512],"genre_scores_gemma":[0.9563724,0.0008689804,0.03718042,0.002528077,0.0006738863,0.0009948666,0.0002328516,0.0001695119,0.0009789806],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4885646,"threshold_uncertainty_score":0.6306908,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4407088834","doi":"10.3390/encyclopedia5010018","title":"Likert-Type Scale","year":2025,"lang":"en","type":"article","venue":"Encyclopedia","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":149,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Public Health Ontario; University of Toronto","funders":"","keywords":"Likert scale; Scale (ratio); Psychology; Statistics; Mathematics; Geography; Cartography","authors":[{"name":"Malcolm Koo","is_ca":true},{"name":"Shih-Wei Yang","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2298581624155352,"gpt":0.4757040976475932,"spread":0.2458459352320581,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01185761,0.0007650076,0.001060393,0.001863436,0.0007507417,0.002350076,0.001749066,0.0009056433,0.0863695],"category_scores_gemma":[0.04918903,0.0002658194,0.0007108973,0.002646264,0.0005697905,0.001672629,0.001404887,0.002000242,0.03401623],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001213785,"about_ca_system_score_gemma":0.002380294,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001268939,"about_ca_topic_score_gemma":0.002442766,"domain_scores_codex":[0.9759266,0.008697736,0.001901264,0.00121069,0.01160055,0.0006632516],"domain_scores_gemma":[0.9674208,0.01272433,0.002607587,0.001620221,0.01462277,0.001004226],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0009303562,0.00155965,0.0162603,0.003270854,0.000228308,0.0002597042,0.003945792,0.0008856211,0.004073394,0.01697915,0.4428941,0.5087128],"study_design_scores_gemma":[0.0001627754,0.0007794675,0.02429464,0.001900107,0.0001402567,0.0003766513,0.003586706,0.001024657,0.001783495,0.007202154,0.9585965,0.0001525357],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"other","genre_gemma":"methods","genre_scores_codex":[0.090952,0.006901539,0.08576988,0.006563869,0.004352589,0.0260209,0.04643743,0.002288784,0.730713],"genre_scores_gemma":[0.4061804,0.01268934,0.222107,0.01069237,0.001686433,0.03434418,0.06277187,0.001076294,0.248452],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.0863695,"threshold_uncertainty_score":0.2889348,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3006236708","doi":"10.1080/00273171.2020.1717922","title":"Improving Fit Indices in Structural Equation Modeling with Categorical Data","year":2020,"lang":"en","type":"article","venue":"Multivariate Behavioral Research","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":148,"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":"Categorical variable; Structural equation modeling; Estimator; Statistics; Computation; Mathematics; Binary data; Binary number; Econometrics; Algorithm","authors":[{"name":"Victoria Savalei","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9579372916956201,"gpt":0.654397403620915,"spread":0.3035398880747051,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07878893,0.001737368,0.00178567,0.00564356,0.001626256,0.004441587,0.002367914,0.00215324,0.002357239],"category_scores_gemma":[0.3266593,0.001124763,0.003106975,0.009475463,0.00241813,0.006332848,0.003627177,0.005935724,0.0007493547],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001840833,"about_ca_system_score_gemma":0.003239329,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002230435,"about_ca_topic_score_gemma":0.003510748,"domain_scores_codex":[0.9081886,0.0745894,0.003569521,0.004011941,0.009120503,0.0005200948],"domain_scores_gemma":[0.7600685,0.2047669,0.007134627,0.01582126,0.01136433,0.0008444689],"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":[0.0003892511,0.000648996,0.128715,0.001673945,0.002116571,0.0002954784,0.005188219,0.08530961,0.001897515,0.1751173,0.006461318,0.5921867],"study_design_scores_gemma":[0.0002153539,0.00134031,0.04039998,0.001318175,0.0005875578,0.000371959,0.002278362,0.5664134,0.00281513,0.3760105,0.007931188,0.000317987],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.05536526,0.001108275,0.9377327,0.00127037,0.0001854942,0.0003468024,0.0003002293,0.0004381394,0.003252689],"genre_scores_gemma":[0.3074358,0.0007035364,0.6893826,0.0002435545,0.00008712285,0.001191757,0.0005469765,0.0001468199,0.0002617078],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.07878893,"threshold_uncertainty_score":0.4166806,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2180735259","doi":"10.1177/2158244014545964","title":"The Conscientious Responders Scale","year":2014,"lang":"en","type":"article","venue":"SAGE Open","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":146,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"York University; Thompson Rivers University","funders":"","keywords":"Scale (ratio); Reliability (semiconductor); Item response theory; Psychology; Odds; Clinical psychology; Psychometrics; Statistics; Social psychology; Medicine; Applied psychology; Logistic regression; Mathematics","authors":[{"name":"Zdravko Marjanovic","is_ca":true},{"name":"C. Ward Struthers","is_ca":true},{"name":"Robert A. Cribbie","is_ca":true},{"name":"Esther R. Greenglass","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4342120929580178,"gpt":0.5144347334499795,"spread":0.08022264049196165,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003394471,0.0003098912,0.0003079388,0.001917265,0.0003900414,0.0005991934,0.0005201537,0.000408816,0.006589697],"category_scores_gemma":[0.02394484,0.0001375911,0.0004821943,0.0005569256,0.0005271324,0.0005987276,0.001023199,0.0008611567,0.001595923],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002818747,"about_ca_system_score_gemma":0.000413727,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006534122,"about_ca_topic_score_gemma":0.001405501,"domain_scores_codex":[0.9969088,0.001105401,0.0003439971,0.0001996082,0.001274108,0.0001679682],"domain_scores_gemma":[0.9880697,0.004833661,0.003437106,0.001168106,0.001697265,0.0007940767],"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":[0.001127506,0.00108084,0.6270331,0.0005601062,0.0003248083,0.0004916431,0.007280859,0.001230866,0.007842388,0.005944518,0.01574708,0.3313364],"study_design_scores_gemma":[0.000194929,0.001395441,0.9636592,0.0002188358,0.0001043439,0.001174853,0.003515094,0.002890246,0.002494243,0.004960454,0.0192803,0.0001120964],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9515358,0.0004242691,0.01124703,0.00066288,0.0002069518,0.001101217,0.0009963214,0.0002664985,0.03355907],"genre_scores_gemma":[0.981963,0.0002116867,0.0109054,0.000228918,0.00007697121,0.000981953,0.0007517165,0.00002368422,0.004856555],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.006589697,"threshold_uncertainty_score":0.02204478,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2150705942","doi":"10.1016/j.ijpsycho.2003.12.009","title":"Analysis of covariance (ANCOVA) with difference scores","year":2004,"lang":"en","type":"article","venue":"International Journal of Psychophysiology","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":144,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Lakehead University","funders":"","keywords":"Analysis of covariance; Covariate; Statistics; Significant difference; Mathematics; Analysis of variance; Covariance; Variable (mathematics); Psychology","authors":[{"name":"John Jamieson","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2276957627367072,"gpt":0.4703214309101297,"spread":0.2426256681734224,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01680005,0.003502184,0.003652653,0.002562981,0.002427348,0.002661056,0.00326911,0.001371892,0.0170882],"category_scores_gemma":[0.08761902,0.0008877605,0.003011533,0.003804521,0.001817108,0.003252686,0.002800209,0.004487564,0.002792224],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009803169,"about_ca_system_score_gemma":0.00550134,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006466439,"about_ca_topic_score_gemma":0.005545061,"domain_scores_codex":[0.9709563,0.0107235,0.002392925,0.008602276,0.005358321,0.001966691],"domain_scores_gemma":[0.927736,0.04979169,0.003032968,0.01276349,0.005676276,0.0009995592],"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":[0.04826976,0.01109566,0.1980656,0.004082323,0.01565556,0.002693779,0.01352192,0.01470477,0.06978693,0.04321574,0.04664935,0.5322586],"study_design_scores_gemma":[0.003909398,0.03747332,0.5750967,0.0008544218,0.007989507,0.00277468,0.004757368,0.1517468,0.07845186,0.06957099,0.06594698,0.001427977],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.4449978,0.0005610825,0.5166252,0.0004894768,0.003023445,0.004400634,0.006521772,0.008363977,0.01501662],"genre_scores_gemma":[0.7386796,0.0002763052,0.2322088,0.0003913383,0.0002701269,0.01086702,0.003504443,0.003693056,0.01010922],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.0170882,"threshold_uncertainty_score":0.08884817,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2148623326","doi":"10.1111/j.1365-2923.2012.04289.x","title":"Using automatic item generation to create multiple‐choice test items","year":2012,"lang":"en","type":"article","venue":"Medical Education","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":142,"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":"Computer science; Multiple choice; Licensure; Item bank; Context (archaeology); Test (biology); Certification; Item response theory; Process (computing); Artificial intelligence; Machine learning; Information retrieval; Psychometrics; Psychology; Medical education; Medicine","authors":[{"name":"Mark J. Gierl","is_ca":true},{"name":"Hollis Lai","is_ca":true},{"name":"Simon R. Turner","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6054844956655376,"gpt":0.5574866617688159,"spread":0.04799783389672174,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02882681,0.002079056,0.001063337,0.004488348,0.0007457681,0.001679938,0.002496212,0.001283766,0.00591727],"category_scores_gemma":[0.1386919,0.0008829209,0.001288673,0.003050711,0.001023258,0.001710881,0.002170047,0.00171965,0.002137759],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001131322,"about_ca_system_score_gemma":0.001745244,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00110765,"about_ca_topic_score_gemma":0.001401192,"domain_scores_codex":[0.9722748,0.02075277,0.001501842,0.00174854,0.003472572,0.000249363],"domain_scores_gemma":[0.7962414,0.167231,0.005303172,0.01403792,0.01650837,0.0006781085],"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":[0.000603206,0.001101071,0.02049997,0.0009683241,0.0003681341,0.0003720138,0.00279919,0.03598256,0.01460681,0.01118275,0.00764716,0.9038687],"study_design_scores_gemma":[0.001186638,0.00209946,0.02678665,0.0006175482,0.0003696245,0.001354717,0.0009979833,0.837029,0.05436263,0.05035126,0.02446008,0.0003844641],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02397892,0.00006151485,0.9699323,0.0001153794,0.00004481472,0.0017455,0.0002239232,0.002674722,0.001222827],"genre_scores_gemma":[0.0853012,0.00005168257,0.9102274,0.00007521699,0.00002405439,0.002911716,0.0006728244,0.0003122934,0.0004235649],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.02882681,"threshold_uncertainty_score":0.1524525,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2041006315","doi":"10.1002/jhbs.1098","title":"Measurement in psychology: A critical history of a methodological concept","year":2002,"lang":"en","type":"article","venue":"Journal of the History of the Behavioral Sciences","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":135,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Calgary","funders":"","keywords":"History of psychology; Citation; Psychology; Library science; Sociology; Psychoanalysis; Computer science","authors":[{"name":"Tim Rogers","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9088941765489551,"gpt":0.5587217495193746,"spread":0.3501724270295805,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch","sts"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2386604,0.002131816,0.004648617,0.01424063,0.004868152,0.01417279,0.005800442,0.01236352,0.001666487],"category_scores_gemma":[0.3051578,0.002022036,0.001887127,0.00805358,0.1234997,0.02908728,0.007326912,0.03200493,0.000560305],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01547593,"about_ca_system_score_gemma":0.02619797,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01168121,"about_ca_topic_score_gemma":0.008381337,"domain_scores_codex":[0.8759763,0.09028783,0.009063504,0.005230069,0.01816506,0.001277197],"domain_scores_gemma":[0.4328281,0.5173598,0.005697444,0.01608853,0.02586538,0.002160687],"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":[0.00009768581,0.0001489427,0.001486559,0.002745833,0.0001291073,0.0001117557,0.004353346,0.00105066,0.000325148,0.896779,0.01902818,0.07374381],"study_design_scores_gemma":[0.00009232899,0.0001565007,0.002322788,0.0109886,0.00009864973,0.0002986906,0.002103047,0.002232551,0.0006102955,0.8195623,0.1613414,0.0001928784],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00192622,0.6543078,0.1351787,0.1896647,0.01121052,0.0002404744,0.0000892048,0.0001202562,0.007262141],"genre_scores_gemma":[0.1546936,0.40971,0.2619306,0.1223772,0.04650752,0.001985378,0.0001321728,0.000597697,0.002065788],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9951319,"threshold_uncertainty_score":0.9388673,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2024532747","doi":"10.1080/10401330701542685","title":"How to Use Structural Equation Modeling in Medical Education Research: A Brief Guide","year":2007,"lang":"en","type":"article","venue":"Teaching and Learning in Medicine","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":135,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Calgary","funders":"","keywords":"Structural equation modeling; Latent variable; Test (biology); Statistical model; Computer science; Psychology; Measure (data warehouse); Statistical hypothesis testing; Medical education; Management science; Medicine; Artificial intelligence; Mathematics; Data mining; Statistics; Machine learning; Engineering","authors":[{"name":"Claudio Violato","is_ca":true},{"name":"Kent G. Hecker","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6828362269846211,"gpt":0.5820779485981239,"spread":0.1007582783864972,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02240597,0.002671575,0.00224857,0.008977242,0.001278803,0.004121981,0.003238982,0.003715102,0.0224355],"category_scores_gemma":[0.05973827,0.001988651,0.002319049,0.006727945,0.002583883,0.005427857,0.002593713,0.008237001,0.013824],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001692367,"about_ca_system_score_gemma":0.005510222,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004427541,"about_ca_topic_score_gemma":0.008498363,"domain_scores_codex":[0.9824357,0.01358194,0.001343959,0.0005472378,0.001944124,0.000146945],"domain_scores_gemma":[0.9379736,0.05484277,0.001331118,0.001801367,0.003527088,0.0005240603],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00005362565,0.0006128726,0.002081492,0.00459113,0.0002528355,0.001039615,0.003903378,0.00918699,0.001675471,0.1728766,0.2113181,0.5924079],"study_design_scores_gemma":[0.0001475266,0.0001763934,0.002056355,0.005493496,0.00008977653,0.00154579,0.001693167,0.0262843,0.001242936,0.3545522,0.6064705,0.0002476398],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0008543568,0.009809697,0.9543115,0.01573106,0.0009981364,0.002776816,0.002297564,0.002063253,0.01115766],"genre_scores_gemma":[0.003438629,0.0129874,0.9737707,0.00197869,0.0003128655,0.003741399,0.001004372,0.0002556216,0.002510374],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.0224355,"threshold_uncertainty_score":0.1184955,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1921155735","doi":"10.1023/a:1023488915557","title":"Use of differential item functioning analysis to assess the equivalence of translations of a questionnaire","year":2003,"lang":"en","type":"article","venue":"Quality of Life Research","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":131,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Centre d'expertise et de recherche en infrastructures urbaines","funders":"","keywords":"Differential item functioning; Contingency table; Psychology; Scale (ratio); Item response theory; Statistics; Logistic regression; Test (biology); Cross-cultural studies; Linguistics; Psychometrics; Mathematics; Social psychology; Developmental psychology; Geography","authors":[{"name":"Morten Aagaard Petersen","is_ca":false},{"name":"Mogens Grøenvold","is_ca":false},{"name":"Jakob Bue Bjørner","is_ca":false},{"name":"Neil K. Aaronson","is_ca":false},{"name":"Thierry Conroy","is_ca":true},{"name":"Ann Cull","is_ca":false},{"name":"Peter Fayers","is_ca":false},{"name":"Marianne Jensen Hjermstad","is_ca":false},{"name":"Mirjam A. G. Sprangers","is_ca":false},{"name":"Marianne Sullivan","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9498982239111976,"gpt":0.6535681604742923,"spread":0.2963300634369053,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07184409,0.001872866,0.001565874,0.008545676,0.001511113,0.002947632,0.001245405,0.001050787,0.003283081],"category_scores_gemma":[0.2860516,0.0005623137,0.003661974,0.006920613,0.002257284,0.002706115,0.003897467,0.002086884,0.0007213067],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001547828,"about_ca_system_score_gemma":0.001760364,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001009664,"about_ca_topic_score_gemma":0.0009380092,"domain_scores_codex":[0.9138985,0.05818952,0.01090031,0.005221856,0.01077377,0.001016102],"domain_scores_gemma":[0.6659682,0.2726131,0.009736266,0.02016958,0.03057849,0.000934276],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.003867123,0.001617985,0.4377673,0.001381288,0.003613559,0.0003857533,0.01665814,0.003951278,0.01280854,0.02636806,0.004283598,0.4872974],"study_design_scores_gemma":[0.001960627,0.007789871,0.7925599,0.0008879072,0.003339706,0.002382356,0.01097615,0.0727883,0.02695496,0.05987783,0.01998016,0.0005022861],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7018283,0.0005519592,0.2791156,0.0005607693,0.0004921423,0.003187519,0.001404623,0.0005451489,0.01231388],"genre_scores_gemma":[0.8542942,0.0001983847,0.1349521,0.0001687127,0.00005761433,0.007084083,0.002126096,0.0002302774,0.0008884938],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07184409,"threshold_uncertainty_score":0.3799524,"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":"high"},{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"}]}