{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":34,"total_is_capped":false,"direct_labels_cover":1,"predictions_cover":34,"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":"e5913b510d38","filters":{"venue":"Applied Psychological Measurement"}},"results":[{"id":"W1975836040","doi":"10.1177/0146621602239476","title":"Determining the Significance of Correlations Corrected for Unreliability and Range Restriction","year":2003,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":103,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Thompson Rivers University","funders":"","keywords":"Statistics; Mathematics; Sampling (signal processing); Variance (accounting); Range (aeronautics); Sample size determination; Monte Carlo method; Population; Correlation coefficient; Correlation; Population variance; Sample (material); Demography; Physics","authors":[{"name":"Nambury S. Raju","is_ca":false},{"name":"Paul A. Brand","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3235615786368179,"gpt":0.4344033213096365,"spread":0.1108417426728185,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09475825,0.00137344,0.002111945,0.003787035,0.001268139,0.002606831,0.003274509,0.001746706,0.002581832],"category_scores_gemma":[0.5681546,0.0009817764,0.002091921,0.003091001,0.005656102,0.004366454,0.003121507,0.004287876,0.0005523044],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001180758,"about_ca_system_score_gemma":0.002786571,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002290088,"about_ca_topic_score_gemma":0.001155504,"domain_scores_codex":[0.8982627,0.06633466,0.004971558,0.01205974,0.01656773,0.001803584],"domain_scores_gemma":[0.3712195,0.5483487,0.01454726,0.04661589,0.01805794,0.001210765],"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":[0.001238366,0.0004100893,0.178503,0.001048455,0.002813788,0.003106719,0.004631654,0.09788832,0.01566696,0.1245638,0.003781748,0.5663471],"study_design_scores_gemma":[0.0003592266,0.001729678,0.1336105,0.000494974,0.001030292,0.003117064,0.001048011,0.6061851,0.02776955,0.2164078,0.007911499,0.0003362921],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.09349423,0.0004130975,0.9019857,0.0003279433,0.0001345188,0.0002968109,0.000108424,0.0006536697,0.002585673],"genre_scores_gemma":[0.7199972,0.0001878845,0.2777439,0.0001788693,0.0001226704,0.0007253013,0.0001746688,0.0002908898,0.0005785932],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.09475825,"threshold_uncertainty_score":0.5011355,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2066165827","doi":"10.1177/01466210122032046","title":"Nonparametric Item Response Function Estimation for Assessing Parametric Model Fit","year":2001,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":80,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"McGill University","keywords":"Nonparametric statistics; Item response theory; Identifiability; Resampling; Parametric statistics; Consistency (knowledge bases); Econometrics; Parametric model; Statistics; Mathematics; Differential item functioning; Computer science; Artificial intelligence; Psychometrics","authors":[{"name":"Jeffrey A. Douglas","is_ca":false},{"name":"Allan S. Cohen","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7555750990116348,"gpt":0.5231085521292439,"spread":0.2324665468823909,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.05992638,0.003592042,0.003675489,0.01063637,0.001408881,0.003024122,0.004742021,0.00342653,0.009989274],"category_scores_gemma":[0.3367467,0.001403237,0.003161358,0.01393562,0.002342806,0.005339304,0.004274278,0.006501303,0.003823268],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001309822,"about_ca_system_score_gemma":0.003150993,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002111927,"about_ca_topic_score_gemma":0.002293773,"domain_scores_codex":[0.9174658,0.06670265,0.002457011,0.003637692,0.009091212,0.0006456491],"domain_scores_gemma":[0.7379347,0.2116615,0.01184847,0.02538045,0.01253539,0.0006394714],"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":[0.0003848029,0.000809241,0.02589176,0.003036042,0.002226939,0.0005829183,0.002122414,0.1059102,0.005669839,0.1534276,0.01879728,0.681141],"study_design_scores_gemma":[0.0002365453,0.001264874,0.03371589,0.001342678,0.000679397,0.002640933,0.001650289,0.5896305,0.006495697,0.3225759,0.03897921,0.0007880956],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.002591745,0.0002827854,0.9947173,0.00007313531,0.00003773591,0.0003046503,0.000267609,0.0008141253,0.0009109896],"genre_scores_gemma":[0.07210263,0.0005145405,0.9209173,0.0001680191,0.000069038,0.003929756,0.001132447,0.0006119418,0.0005543533],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.05992638,"threshold_uncertainty_score":0.3169248,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2037069694","doi":"10.1177/01466210022031741","title":"Restriction of Range and Correlation in Outlier-Prone Distributions","year":2000,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":71,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Carleton University","funders":"","keywords":"Outlier; Statistics; Correlation; Range (aeronautics); Mathematics; Normal distribution; Gaussian; Physics","authors":[{"name":"Donald W. Zimmerman","is_ca":true},{"name":"Richard H. Williams","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2231647714868146,"gpt":0.4176716541584364,"spread":0.1945068826716219,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02014207,0.0006536433,0.001410478,0.001585592,0.0006024017,0.002394759,0.001339182,0.001046472,0.001792678],"category_scores_gemma":[0.1969924,0.0005467002,0.0009778584,0.0017467,0.00391418,0.003997777,0.003582285,0.00246226,0.0002541596],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006293859,"about_ca_system_score_gemma":0.0008566484,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004359905,"about_ca_topic_score_gemma":0.000300337,"domain_scores_codex":[0.9699208,0.01962219,0.002043364,0.003476858,0.004217788,0.0007189141],"domain_scores_gemma":[0.7091357,0.2298487,0.02160942,0.03159796,0.006390184,0.001418028],"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":[0.001200476,0.0003122943,0.1080132,0.0007792194,0.000762646,0.002924779,0.004772806,0.2024921,0.02137699,0.4640173,0.002091635,0.1912566],"study_design_scores_gemma":[0.0001040772,0.0004747852,0.04594912,0.0001963885,0.0001474603,0.002212479,0.0008070283,0.3321165,0.0105309,0.6030562,0.00417918,0.0002258883],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.2324826,0.0004653796,0.7624274,0.0003525072,0.00003953309,0.0001367135,0.0001506421,0.0002646782,0.003680559],"genre_scores_gemma":[0.9373615,0.0002649493,0.06123194,0.0001111124,0.00005040758,0.0003001954,0.0002018069,0.00006752011,0.0004105757],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.02014207,"threshold_uncertainty_score":0.1065227,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2143619291","doi":"10.1177/0146621603254799","title":"A New Look at the Influence of Guessing on the Reliability of Multiple-Choice Tests","year":2003,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":59,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Carleton University","funders":"","keywords":"Variance (accounting); Reliability (semiconductor); Test (biology); Statistics; Econometrics; Psychology; Mathematics","authors":[{"name":"Donald W. Zimmerman","is_ca":true},{"name":"Richard H. Williams","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2678450942987462,"gpt":0.4308893442137286,"spread":0.1630442499149824,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03635688,0.001126134,0.002083933,0.003438599,0.0009726277,0.005469345,0.002841339,0.002304553,0.005035478],"category_scores_gemma":[0.3355232,0.001190444,0.00196572,0.002396918,0.007089938,0.01388765,0.002943872,0.007951532,0.0008967706],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001478232,"about_ca_system_score_gemma":0.00109029,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003920957,"about_ca_topic_score_gemma":0.003524185,"domain_scores_codex":[0.96777,0.01843235,0.001300542,0.002724738,0.009016134,0.0007562079],"domain_scores_gemma":[0.498598,0.4451222,0.009669635,0.02714857,0.01808354,0.001378102],"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":[0.0007197473,0.0002786555,0.07418545,0.0009171611,0.00106798,0.001731428,0.006217902,0.03349894,0.009697593,0.2369936,0.01447639,0.6202152],"study_design_scores_gemma":[0.0001144737,0.001192348,0.1171358,0.00104245,0.0007767976,0.004062324,0.002006994,0.1799316,0.01311008,0.642949,0.03667842,0.0009996843],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2008027,0.03857173,0.6678993,0.04971242,0.001902056,0.00008438898,0.0003895318,0.001297344,0.0393407],"genre_scores_gemma":[0.8830085,0.01221034,0.08928097,0.002877403,0.00470623,0.00006965434,0.0001953022,0.0009731274,0.00667848],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.03635688,"threshold_uncertainty_score":0.1922759,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2100391776","doi":"10.1177/0146621606286206","title":"The Effect of Examinee Motivation on Test Construction Within an IRT Framework","year":2006,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":41,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Lakehead University","funders":"","keywords":"Item response theory; Test (biology); Psychology; Statistics; Econometrics; Bayesian probability; Differential item functioning; Equating; Response bias; Computerized adaptive testing; Logistic regression; Social psychology; Mathematics; Psychometrics; Rasch model","authors":[{"name":"Christina van Barneveld","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3951392929941238,"gpt":0.4304190280454949,"spread":0.03527973505137111,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1542658,0.0006963143,0.0009100254,0.001247051,0.0008873567,0.003046283,0.001102655,0.001794135,0.001824368],"category_scores_gemma":[0.5076049,0.0007908639,0.0009495441,0.0008996676,0.002951695,0.002377841,0.003264646,0.00269155,0.0003030277],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001538507,"about_ca_system_score_gemma":0.001272641,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001445298,"about_ca_topic_score_gemma":0.00147123,"domain_scores_codex":[0.8567997,0.1208176,0.003302913,0.006267493,0.01109277,0.00171951],"domain_scores_gemma":[0.170791,0.7745883,0.02319142,0.02415604,0.005506573,0.001766721],"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":[0.004397757,0.0006291412,0.9158244,0.0001564672,0.0004765476,0.0003603592,0.003940279,0.006822529,0.005235731,0.004149678,0.0003418887,0.05766531],"study_design_scores_gemma":[0.0002292401,0.003938539,0.9400472,0.0001150689,0.0003992734,0.000894251,0.001197772,0.03671583,0.007285113,0.007924582,0.001139494,0.0001137087],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9598405,0.0003531245,0.03639647,0.0005708996,0.00002133144,0.0001167255,0.00005234686,0.0000959704,0.002552625],"genre_scores_gemma":[0.9898694,0.00006733354,0.009456737,0.0001536732,0.00001598773,0.00008328701,0.00004978581,0.00004450301,0.0002592801],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1542658,"threshold_uncertainty_score":0.8158453,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2070673350","doi":"10.1177/0146621606292215","title":"Investigation of IRT-Based Equating Methods in the Presence of Outlier Common Items","year":2008,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":39,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"","keywords":"Equating; Outlier; Statistics; Item response theory; Calibration; Comparability; Mathematics; Econometrics; Computer science; Psychometrics","authors":[{"name":"Huiqin Hu","is_ca":false},{"name":"W. Todd Rogers","is_ca":true},{"name":"Zarko Vukmirovic","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8304468687797645,"gpt":0.5388212625480718,"spread":0.2916256062316926,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2246548,0.001864325,0.00182753,0.003200948,0.001232548,0.002957797,0.003533938,0.001897482,0.002831019],"category_scores_gemma":[0.5783927,0.0008860425,0.00210609,0.004959769,0.002387419,0.004695791,0.004801919,0.003196447,0.000942643],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001626488,"about_ca_system_score_gemma":0.001925302,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008502225,"about_ca_topic_score_gemma":0.0010278,"domain_scores_codex":[0.7399748,0.2243471,0.01051213,0.01098689,0.01340613,0.0007730176],"domain_scores_gemma":[0.376565,0.5199229,0.02338755,0.05294217,0.02638422,0.0007982188],"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":[0.00235677,0.0007740978,0.07198077,0.001138099,0.002573541,0.0002177803,0.00724459,0.04874792,0.007406957,0.03818785,0.001704613,0.8176669],"study_design_scores_gemma":[0.0008978567,0.004252087,0.07400165,0.001025165,0.001322688,0.001622566,0.003344226,0.7923597,0.03406962,0.07382856,0.01265728,0.0006185352],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1001273,0.0003905313,0.8948936,0.0001794077,0.0001083418,0.0008579048,0.00007739105,0.0006670575,0.002698476],"genre_scores_gemma":[0.3569662,0.0001432705,0.6404704,0.0001430763,0.00003256529,0.00127742,0.0002353478,0.0003257669,0.0004058995],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.2246548,"threshold_uncertainty_score":0.9561386,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2580371235","doi":"10.1177/0146621616684584","title":"An Evaluation of Interrater Reliability Measures on Binary Tasks Using <i>d-Prime</i>","year":2016,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":35,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Memorial University of Newfoundland","funders":"","keywords":"Inter-rater reliability; Kappa; Prime (order theory); Psychology; Statistics; Reliability (semiconductor); Cohen's kappa; Binary number; Agreement; Psychometrics; Social psychology; Mathematics; Combinatorics; Arithmetic; Rating scale; Linguistics","authors":[{"name":"Malcolm Grant","is_ca":true},{"name":"Cathryn M. Button","is_ca":true},{"name":"Brent Snook","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5204254440661797,"gpt":0.4693431790011876,"spread":0.05108226506499214,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2886595,0.001134056,0.001182251,0.004919657,0.002584416,0.002402039,0.001667546,0.001105747,0.001202315],"category_scores_gemma":[0.4241366,0.0009815239,0.00181328,0.004785256,0.003575735,0.0026757,0.003493607,0.001720036,0.0007188705],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001606336,"about_ca_system_score_gemma":0.001992717,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008587345,"about_ca_topic_score_gemma":0.002157314,"domain_scores_codex":[0.7193463,0.2144912,0.02753669,0.009409014,0.0278793,0.001337506],"domain_scores_gemma":[0.4231398,0.4529221,0.02396384,0.03081702,0.06789482,0.001262418],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.005113521,0.0009351372,0.4490764,0.005384285,0.003329968,0.0003702682,0.0527439,0.01153817,0.02583556,0.04490153,0.01013549,0.3906358],"study_design_scores_gemma":[0.0009052373,0.01175264,0.5599464,0.003018348,0.002156993,0.00355986,0.02671421,0.19321,0.08770531,0.07292041,0.03659859,0.001511905],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4082651,0.00137811,0.5665505,0.0005003331,0.0004652507,0.003672213,0.0006822516,0.000749335,0.01773684],"genre_scores_gemma":[0.7268816,0.0002713609,0.2677065,0.0001556997,0.00005822937,0.003708395,0.0002984886,0.000152713,0.0007669865],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7113405,"threshold_uncertainty_score":0.8772095,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2745464303","doi":"10.1177/0146621617726788","title":"Using Automatic Item Generation to Create Solutions and Rationales for Computerized Formative Testing","year":2017,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Intelligent Tutoring Systems and Adaptive Learning","field":"Computer Science","cited_by":33,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"","keywords":"Formative assessment; Computer science; Item bank; Computerized adaptive testing; Test (biology); Quality (philosophy); Item response theory; Psychometrics; Psychology; Mathematics education","authors":[{"name":"Mark J. Gierl","is_ca":true},{"name":"Hollis Lai","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.534223811307965,"gpt":0.385328940422152,"spread":0.148894870885813,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03824244,0.002680288,0.001050098,0.006609483,0.0009641433,0.003421476,0.003407287,0.001722362,0.008732769],"category_scores_gemma":[0.1745299,0.001339741,0.001476971,0.003413301,0.00147145,0.003256897,0.002550721,0.002802091,0.003484465],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001415725,"about_ca_system_score_gemma":0.003650214,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001046562,"about_ca_topic_score_gemma":0.001819351,"domain_scores_codex":[0.9654917,0.02233861,0.003349878,0.002096359,0.00639554,0.000327939],"domain_scores_gemma":[0.7930138,0.1456481,0.01069306,0.02288139,0.02685905,0.0009044861],"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":[0.0004369041,0.0007394668,0.003939663,0.0007960885,0.0001085711,0.0004281734,0.002001417,0.01368782,0.01278565,0.02358065,0.009504001,0.9319916],"study_design_scores_gemma":[0.001634482,0.00184272,0.005021922,0.001883539,0.0003433718,0.001926005,0.00153875,0.6633273,0.1029804,0.1355617,0.08314712,0.0007926568],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.005644991,0.00008036325,0.9853556,0.0002080049,0.0001048313,0.001533333,0.000271565,0.005596731,0.001204636],"genre_scores_gemma":[0.01960601,0.00006410777,0.9777033,0.00006526697,0.00003487245,0.00110317,0.0005033651,0.00040766,0.0005122962],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.03824244,"threshold_uncertainty_score":0.2022478,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2590325518","doi":"10.1177/0146621617692079","title":"Plausible-Value Imputation Statistics for Detecting Item Misfit","year":2017,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":31,"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":"Statistics; Statistic; Item response theory; Trait; Imputation (statistics); Econometrics; Parametric statistics; Test statistic; Mathematics; Statistical hypothesis testing; Differential item functioning; Latent variable model; Item analysis; Null hypothesis; Latent variable; Computer science; Psychometrics; Missing data","authors":[{"name":"R. Philip Chalmers","is_ca":true},{"name":"Victoria Ka Yin Ng","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.72207883373125,"gpt":0.5322352636234017,"spread":0.1898435701078484,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1168906,0.001436061,0.002673173,0.005757773,0.001440663,0.003113023,0.005699732,0.003370548,0.009745712],"category_scores_gemma":[0.5533647,0.001159718,0.002607848,0.008477095,0.003739392,0.006097394,0.003416544,0.006426449,0.001944491],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001301532,"about_ca_system_score_gemma":0.002465201,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007246162,"about_ca_topic_score_gemma":0.0009566799,"domain_scores_codex":[0.8352187,0.1378361,0.007765887,0.00634974,0.01212587,0.0007036942],"domain_scores_gemma":[0.4102183,0.5116881,0.02188965,0.04191566,0.01328535,0.001002992],"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":[0.00103557,0.0004948975,0.05186252,0.002283193,0.002399977,0.0007567042,0.003777806,0.06228556,0.001961221,0.3234849,0.0167042,0.5329533],"study_design_scores_gemma":[0.0004404788,0.001093511,0.02675035,0.001320269,0.0006042446,0.001662489,0.00102153,0.3778016,0.006255635,0.5623298,0.02028913,0.0004310801],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.00640875,0.0002236011,0.9912319,0.0002563562,0.00005826055,0.0003438239,0.0002551956,0.0004282956,0.0007938517],"genre_scores_gemma":[0.1670225,0.0002266043,0.8289458,0.0001914956,0.0000754752,0.002214528,0.0007270878,0.0002409579,0.0003555094],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.1168906,"threshold_uncertainty_score":0.6181841,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2169517221","doi":"10.1177/0146621610391777","title":"Accuracy of Person-Fit Statistics","year":2011,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Advanced Text Analysis Techniques","field":"Computer Science","cited_by":28,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Université Laval; Université de Sherbrooke","funders":"","keywords":"Statistics; Cheating; Monte Carlo method; Mathematics; Econometrics; Psychology; Social psychology","authors":[{"name":"Christina St‐Onge","is_ca":true},{"name":"Pierre Valois","is_ca":true},{"name":"Belkacem Abdous","is_ca":true},{"name":"Stéphane Germain","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2430805157256228,"gpt":0.3434006195884511,"spread":0.1003201038628284,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04380894,0.0007473846,0.001124276,0.001807728,0.0004511579,0.002905105,0.0009182058,0.001781002,0.00590012],"category_scores_gemma":[0.4940274,0.0004906363,0.001133437,0.001378092,0.00170118,0.005767066,0.002583757,0.001721802,0.001081796],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008893315,"about_ca_system_score_gemma":0.0006506518,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001335475,"about_ca_topic_score_gemma":0.0006186913,"domain_scores_codex":[0.9520745,0.02466871,0.004808754,0.008370388,0.008795697,0.001281951],"domain_scores_gemma":[0.4484861,0.4514304,0.03982987,0.04361455,0.01510066,0.001538448],"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":[0.01511028,0.001435388,0.44398,0.001011296,0.001726207,0.0003728309,0.01143897,0.06436241,0.03489734,0.04026034,0.003448627,0.3819562],"study_design_scores_gemma":[0.0004756723,0.007966646,0.6039726,0.000297732,0.0007570283,0.001400219,0.001443896,0.2467565,0.06868179,0.05792437,0.009689617,0.0006338442],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"methods","genre_scores_codex":[0.8019421,0.0004197682,0.17949,0.0003535234,0.0001461325,0.0007055679,0.0007901518,0.001085886,0.01506678],"genre_scores_gemma":[0.985611,0.00005758883,0.01262637,0.00007410775,0.00002013005,0.0001981714,0.0003122187,0.0001591084,0.0009413544],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.04380894,"threshold_uncertainty_score":0.2316866,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2171716385","doi":"10.1177/0146621614520958","title":"Maximum-Likelihood Estimation of Noncompensatory IRT Models With the MH-RM Algorithm","year":2014,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":26,"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":"Item response theory; Estimation; Maximum likelihood; Latent variable; Statistics; Computer science; Econometrics; Population; Estimation theory; Expectation–maximization algorithm; Mathematics; Algorithm; Artificial intelligence; Machine learning; Psychometrics; Engineering","authors":[{"name":"R. Philip Chalmers","is_ca":true},{"name":"David B. Flora","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3757797696663633,"gpt":0.4016023114350326,"spread":0.02582254176866927,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01134817,0.0009974251,0.00157755,0.00113049,0.0006880196,0.001109859,0.003430546,0.001641348,0.003180036],"category_scores_gemma":[0.03877758,0.0009495133,0.001277528,0.001122771,0.0009635956,0.00179572,0.001649673,0.00271752,0.0009419785],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008725596,"about_ca_system_score_gemma":0.001877937,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007185911,"about_ca_topic_score_gemma":0.00749372,"domain_scores_codex":[0.995344,0.003589031,0.000200768,0.0003796929,0.0003386847,0.0001478195],"domain_scores_gemma":[0.9670488,0.02932623,0.0008588543,0.001477416,0.001080936,0.0002076954],"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":[0.0005592451,0.0002342988,0.004123205,0.0002457647,0.0002685425,0.0002240162,0.0004470186,0.8011401,0.001880009,0.05045402,0.001478867,0.1389448],"study_design_scores_gemma":[0.00003001514,0.00002847824,0.0002419603,0.000009692762,0.000009269774,0.00002550532,0.00001607525,0.9918011,0.0002747569,0.007343192,0.0002079363,0.00001201327],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02005583,0.0001229838,0.9785966,0.0001185244,0.000009868882,0.00008596419,0.0000310385,0.0004547232,0.0005244662],"genre_scores_gemma":[0.2221319,0.00009545752,0.7751131,0.0001205592,0.0000231499,0.0003705707,0.0003313271,0.000184329,0.001629647],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.01134817,"threshold_uncertainty_score":0.06001562,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2582841787","doi":"10.1177/0146621615597894","title":"faoutlier","year":2015,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":24,"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":"Exploratory factor analysis; Computer science; Confirmatory factor analysis; Regression analysis; Software; Extant taxon; Factor (programming language); Statistics; Statistical analysis; Regression; Artificial intelligence; Machine learning; Mathematics; Structural equation modeling; Programming language; Biology","authors":[{"name":"R. Philip Chalmers","is_ca":true},{"name":"David B. Flora","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6421002834531417,"gpt":0.5022808864825751,"spread":0.1398193969705667,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.002679386,0.001369673,0.000872977,0.003526707,0.001455383,0.005014893,0.00174,0.001931375,0.5931423],"category_scores_gemma":[0.01241159,0.0005062351,0.0008261485,0.001947397,0.0007397812,0.002484211,0.002310161,0.001356864,0.3775394],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001228313,"about_ca_system_score_gemma":0.001871006,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004286341,"about_ca_topic_score_gemma":0.005273296,"domain_scores_codex":[0.9980465,0.0003654924,0.0001113277,0.0004567927,0.0008680969,0.0001518263],"domain_scores_gemma":[0.9959913,0.00105247,0.0001919907,0.001152693,0.001207956,0.0004035665],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001722361,0.00005274525,0.0006613071,0.0002259195,0.00001768354,0.000164465,0.0002438389,0.0004874014,0.001320411,0.02038068,0.5905657,0.3857077],"study_design_scores_gemma":[0.00001636007,0.00001294376,0.0002734261,0.00005668333,0.000003928768,0.0001914081,0.00003655638,0.0006726859,0.0007025664,0.003824424,0.9941987,0.0000104338],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"software","genre_scores_codex":[0.004851395,0.003273697,0.1165668,0.00670093,0.007079472,0.0004190966,0.02606467,0.06620541,0.7688386],"genre_scores_gemma":[0.04234795,0.002137428,0.08745123,0.002020441,0.0009754526,0.0004324429,0.02921102,0.01929475,0.8161293],"genre_candidate":"software","genre_consensus":null,"teacher_disagreement_score":0.4068577,"threshold_uncertainty_score":0,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2134979859","doi":"10.1177/01466210022031606","title":"Cross-Validation Sample Sizes","year":2000,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":23,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Manitoba","funders":"","keywords":"Statistics; Sample size determination; Mathematics; Correlation coefficient; Mean squared error; Correlation; Pearson product-moment correlation coefficient; Linear regression; Sample (material); Cross-validation; Coefficient of determination; Chemistry","authors":[{"name":"James Algina","is_ca":false},{"name":"H. J. Keselman","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3344098990462734,"gpt":0.4815746876227396,"spread":0.1471647885764661,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.134205,0.001428698,0.00247648,0.004117292,0.002843641,0.002450645,0.003875065,0.002702219,0.02540204],"category_scores_gemma":[0.4725113,0.001185172,0.003164393,0.002276931,0.002279284,0.002618528,0.004583098,0.003639986,0.007018775],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001105124,"about_ca_system_score_gemma":0.003251286,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002013545,"about_ca_topic_score_gemma":0.002423948,"domain_scores_codex":[0.9074318,0.05174052,0.01013065,0.00865514,0.02013059,0.001911307],"domain_scores_gemma":[0.6292523,0.2282657,0.005946336,0.05383492,0.08070657,0.001994189],"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":[0.006178437,0.002973127,0.08606475,0.002546445,0.001962546,0.0006036594,0.01324236,0.01404178,0.008594897,0.06349736,0.1076195,0.6926752],"study_design_scores_gemma":[0.005221029,0.008853916,0.2414511,0.004638466,0.004282487,0.002364971,0.006256314,0.06624738,0.04814548,0.09329955,0.5185439,0.0006953122],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.2219029,0.003689748,0.584416,0.001645925,0.005280737,0.08570094,0.01674231,0.002866722,0.07775477],"genre_scores_gemma":[0.5031244,0.001117883,0.3157665,0.002028221,0.0004635141,0.1426243,0.01287686,0.001991313,0.02000717],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.865795,"threshold_uncertainty_score":0.7097524,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2078503709","doi":"10.1177/0146621614557272","title":"Evaluating Person Fit for Cognitive Diagnostic Assessment","year":2014,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":23,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Manitoba; University of Alberta","funders":"","keywords":"Statistic; Cognition; Psychology; Item response theory; Conformity; Cognitive psychology; Context (archaeology); Statistics; Social psychology; Applied psychology; Psychometrics; Clinical psychology; Mathematics","authors":[{"name":"Ying Cui","is_ca":true},{"name":"Johnson Li","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8911900554521754,"gpt":0.6108802067461848,"spread":0.2803098487059906,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.08439425,0.001161047,0.001323027,0.006606461,0.001088741,0.003445595,0.001567478,0.002063828,0.002233198],"category_scores_gemma":[0.3578418,0.0004870988,0.002212827,0.004602216,0.00164008,0.004076266,0.003468936,0.00175262,0.0005174878],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001352366,"about_ca_system_score_gemma":0.001651726,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008718336,"about_ca_topic_score_gemma":0.0009785715,"domain_scores_codex":[0.9080693,0.05709263,0.008657043,0.00556268,0.01972795,0.0008904295],"domain_scores_gemma":[0.5832041,0.3291888,0.03606105,0.02511482,0.02450475,0.001926482],"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":[0.001909032,0.0009071669,0.6706119,0.0006053483,0.001524999,0.0002002946,0.004733304,0.01152451,0.002016098,0.01399306,0.002726951,0.2892474],"study_design_scores_gemma":[0.0006538241,0.007396821,0.4839167,0.0009120623,0.001034068,0.001970411,0.007883376,0.3789693,0.01473369,0.08851072,0.01321493,0.0008041261],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4412414,0.0006203065,0.5464425,0.0005677344,0.0001798862,0.002118087,0.0008069832,0.0007813587,0.007241731],"genre_scores_gemma":[0.8545927,0.0001184959,0.1422368,0.0001459739,0.00003644099,0.001822389,0.0006088745,0.00008380241,0.0003545378],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.08439425,"threshold_uncertainty_score":0.4463248,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2029082333","doi":"10.1177/0146621607301094","title":"Effects of Semantic Incompatibility on Rating Response","year":2008,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":18,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Psychology; Rating scale; Semantic differential; Internal consistency; Consistency (knowledge bases); Scale (ratio); Cognitive psychology; Social psychology; Statistics; Psychometrics; Developmental psychology; Artificial intelligence; Computer science; Mathematics","authors":[{"name":"Tony C. M. Lam","is_ca":true},{"name":"Mary Kolic","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6035678497989727,"gpt":0.4659181755804515,"spread":0.1376496742185212,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04382954,0.0007789284,0.0007457315,0.0009152116,0.0006480835,0.001059229,0.000613721,0.00072554,0.003756229],"category_scores_gemma":[0.2428551,0.0007323997,0.0007985233,0.000782876,0.001542595,0.001231269,0.001700642,0.001429091,0.0005865988],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006641244,"about_ca_system_score_gemma":0.0004486923,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002824568,"about_ca_topic_score_gemma":0.0004011744,"domain_scores_codex":[0.8827703,0.08423266,0.009837208,0.005300843,0.01682444,0.001034528],"domain_scores_gemma":[0.3935457,0.5504494,0.02672509,0.01924897,0.008028414,0.002002345],"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":[0.05529235,0.005244384,0.4411224,0.001401935,0.00138799,0.001053064,0.01547314,0.005777983,0.2043931,0.003672724,0.001887041,0.263294],"study_design_scores_gemma":[0.0007898977,0.01115619,0.9268277,0.0002344582,0.0006527408,0.001455149,0.001983196,0.008545023,0.0411891,0.003743603,0.003147267,0.0002755488],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9870746,0.0002028289,0.007392663,0.0001950931,0.00007511272,0.0002196073,0.00006846321,0.0001243057,0.004647487],"genre_scores_gemma":[0.9934008,0.00008346887,0.00538465,0.0001691776,0.00004778813,0.0002235079,0.0001206938,0.00009188417,0.0004779999],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.04382954,"threshold_uncertainty_score":0.2317955,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2035434565","doi":"10.1177/01466216010251006","title":"The Extra-Factor Phenomenon Revisited: Unidimensional Unfolding as Quadratic Factor Analysis","year":2001,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Sensory Analysis and Statistical Methods","field":"Agricultural and Biological Sciences","cited_by":18,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Simon Fraser University","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Phenomenon; Factor (programming language); Covariance; Quadratic equation; Metric (unit); Factor analysis; Mathematics; Set (abstract data type); Applied mathematics; Statistics; Computer science; Physics; Geometry","authors":[{"name":"Michael D. Maraun","is_ca":false},{"name":"Natasha T. Rossi","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1400589295419584,"gpt":0.3338867503694962,"spread":0.1938278208275377,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0307411,0.001553975,0.001311309,0.003093343,0.001787801,0.004978817,0.0021313,0.001565185,0.003220956],"category_scores_gemma":[0.1143771,0.0007261214,0.001550615,0.007342659,0.01230724,0.01137857,0.005565487,0.00566833,0.0004294178],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001455135,"about_ca_system_score_gemma":0.001798205,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001189615,"about_ca_topic_score_gemma":0.001000695,"domain_scores_codex":[0.9768024,0.01557828,0.001154915,0.002210146,0.003745382,0.0005089307],"domain_scores_gemma":[0.9240366,0.05410444,0.004951093,0.01227708,0.00411746,0.0005133107],"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":[0.0001269581,0.00005480192,0.003609949,0.0004231234,0.0001334637,0.0002827127,0.008362156,0.001923048,0.001611849,0.8718451,0.001764178,0.1098628],"study_design_scores_gemma":[0.00002453707,0.00009788469,0.004252256,0.0001921033,0.00003627981,0.0005157854,0.001227208,0.02102294,0.0009343385,0.9628034,0.008803538,0.00008960732],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.03163601,0.0028704,0.950754,0.002093074,0.0004940338,0.0001518217,0.00006273931,0.0002870943,0.01165075],"genre_scores_gemma":[0.6019787,0.001583512,0.3928115,0.0007651679,0.0004034483,0.0004780405,0.0001036944,0.0001772042,0.001698728],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.0307411,"threshold_uncertainty_score":0.1625765,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1942728350","doi":"10.1177/0146621608316603","title":"Computer Software Review: Conducting Automated Test Assembly Using the Premium Solver Platform Version 7.0 With Microsoft Excel and the Large-Scale LP/QP Solver Engine Add-In","year":2008,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Manufacturing Process and Optimization","field":"Engineering","cited_by":14,"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":"Solver; Computer science; Microsoft excel; Scale (ratio); Software; Test (biology); Programming language; Computational science; Software engineering; Operating system; Cartography","authors":[{"name":"Ken Cor","is_ca":true},{"name":"Cecilia Alves","is_ca":true},{"name":"Mark J. Gierl","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.04766665868336487,"gpt":0.246875605860797,"spread":0.1992089471774322,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009237908,0.000671649,0.001326209,0.007147159,0.0005495425,0.001510308,0.001788403,0.0006626453,0.05600004],"category_scores_gemma":[0.06515843,0.00059421,0.0007451447,0.006498455,0.0006348281,0.001200747,0.00090579,0.0009991551,0.0405303],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001116771,"about_ca_system_score_gemma":0.00674536,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004156727,"about_ca_topic_score_gemma":0.007083481,"domain_scores_codex":[0.9908564,0.002249602,0.001531737,0.0005276205,0.004662689,0.0001719337],"domain_scores_gemma":[0.8873472,0.02898402,0.005102083,0.006612078,0.07037739,0.001577266],"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":[0.0001223644,0.00003818909,0.0008291856,0.002190661,0.0000791165,0.0000883437,0.00006295127,0.0005168233,0.002965733,0.001321636,0.4489811,0.5428039],"study_design_scores_gemma":[0.0001555921,0.0002499729,0.009652043,0.001146471,0.0002575337,0.0009894853,0.00009932523,0.00579053,0.01068356,0.002501794,0.9683877,0.00008614117],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"review","genre_scores_codex":[0.02787815,0.07107179,0.6267645,0.02613405,0.01918058,0.005615072,0.02287115,0.07734317,0.1231414],"genre_scores_gemma":[0.1104643,0.06668252,0.5208767,0.006844281,0.009171831,0.004739448,0.03977386,0.0283619,0.2130851],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.05600004,"threshold_uncertainty_score":0.1873388,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2097506538","doi":"10.1177/01466216010251011","title":"Computer Program Exchange","year":2001,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Distributed and Parallel Computing Systems","field":"Computer Science","cited_by":11,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Lakehead University","funders":"Lakehead University","keywords":"Computer program; Computer science; Psychology; Statistics; Mathematics; Programming language","authors":[{"name":"Brian P. O’Connor","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.0944139960001945,"gpt":0.3083394133823271,"spread":0.2139254173821326,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001597226,0.0008587234,0.0007425613,0.002150355,0.001513438,0.002699099,0.001582949,0.0009666742,0.5037364],"category_scores_gemma":[0.006095225,0.0005087687,0.0006041103,0.002016391,0.0004330276,0.002865284,0.003517181,0.001378582,0.3115626],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000722299,"about_ca_system_score_gemma":0.001231891,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001261936,"about_ca_topic_score_gemma":0.001106374,"domain_scores_codex":[0.9985839,0.0003198709,0.00008007944,0.0002406845,0.0005325148,0.0002429209],"domain_scores_gemma":[0.9958854,0.0005231347,0.000091072,0.002294973,0.0007805686,0.0004248235],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0005848237,0.000445777,0.000773551,0.0001229587,0.00002187675,0.0001475529,0.0002617164,0.0009285879,0.003737369,0.05814629,0.6285829,0.3062465],"study_design_scores_gemma":[0.00009519178,0.00006992184,0.0007798176,0.00003039739,0.00001327174,0.0001387986,0.00007729043,0.003713936,0.005413864,0.01457474,0.9750741,0.00001866004],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"software","genre_scores_codex":[0.008158066,0.0001999805,0.08599443,0.001435451,0.001006647,0.0009573349,0.008636622,0.02976546,0.8638461],"genre_scores_gemma":[0.04895432,0.0002343651,0.02124692,0.0004774612,0.0002852429,0.0005847755,0.0173153,0.006669187,0.9042324],"genre_candidate":"software","genre_consensus":null,"teacher_disagreement_score":0.5037364,"threshold_uncertainty_score":0.7078598,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2055728918","doi":"10.1177/0146621608329503","title":"A Monte Carlo Study of the Effect of Item Characteristic Curve Estimation on the Accuracy of Three Person-Fit Statistics","year":2009,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":10,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Université Laval; Université de Sherbrooke","funders":"","keywords":"Nonparametric statistics; Statistics; Monte Carlo method; Logistic regression; Parametric statistics; Mathematics; Item response theory; Sample size determination; Econometrics; Psychometrics","authors":[{"name":"Christina St‐Onge","is_ca":true},{"name":"Pierre Valois","is_ca":true},{"name":"Belkacem Abdous","is_ca":true},{"name":"Stéphane Germain","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5775002634706006,"gpt":0.460701353218835,"spread":0.1167989102517655,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2860002,0.001012602,0.00170299,0.002498301,0.001232297,0.002235778,0.00160116,0.002476741,0.001464546],"category_scores_gemma":[0.7602907,0.001024936,0.002192738,0.002536174,0.003029221,0.003495897,0.002235457,0.002433323,0.0002945283],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002140573,"about_ca_system_score_gemma":0.001436138,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003365141,"about_ca_topic_score_gemma":0.002457422,"domain_scores_codex":[0.7120539,0.259611,0.007246682,0.008103768,0.01204058,0.0009441455],"domain_scores_gemma":[0.04116737,0.9274645,0.008218995,0.01698967,0.005890972,0.0002684908],"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":[0.0207854,0.00165863,0.2839131,0.001535407,0.005047232,0.0008561473,0.01107393,0.2843124,0.00583367,0.05319038,0.003879601,0.3279141],"study_design_scores_gemma":[0.000828818,0.00594932,0.0977667,0.0006317837,0.001502893,0.001668729,0.001110873,0.8550518,0.009802617,0.02143538,0.003722864,0.0005282085],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6320564,0.00178235,0.3594206,0.0005736877,0.0001383762,0.001244937,0.0002282984,0.0006014991,0.003953765],"genre_scores_gemma":[0.8871796,0.000188608,0.1111643,0.000177494,0.00003095832,0.0007179047,0.0001528786,0.0001316047,0.0002565377],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2860002,"threshold_uncertainty_score":0.8804889,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4293408802","doi":"10.1177/01466216221124089","title":"Item Selection With Collaborative Filtering in On-The-Fly Multistage Adaptive Testing","year":2022,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"","keywords":"Computerized adaptive testing; Item bank; Selection (genetic algorithm); Computer science; Item response theory; Collaborative filtering; On the fly; Feature selection; Test (biology); Machine learning; Data mining; Artificial intelligence; Statistics; Recommender system; Mathematics; Psychometrics","authors":[{"name":"Jiaying Xiao","is_ca":false},{"name":"Okan Bulut","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6812333426754545,"gpt":0.4329920465484254,"spread":0.2482412961270291,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03210677,0.00147571,0.002591869,0.004818678,0.001120065,0.001360172,0.003086034,0.001821515,0.0030142],"category_scores_gemma":[0.07920349,0.0009090815,0.002295027,0.003652702,0.001305954,0.002017346,0.002172514,0.001849524,0.0007884575],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001127541,"about_ca_system_score_gemma":0.00192796,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007533535,"about_ca_topic_score_gemma":0.008993585,"domain_scores_codex":[0.9602072,0.03106329,0.001591883,0.00254804,0.004082534,0.000507108],"domain_scores_gemma":[0.9283035,0.05895305,0.002139696,0.004619695,0.00543072,0.000553369],"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":[0.0009384935,0.001077836,0.0134124,0.0006530724,0.0009294539,0.0002481549,0.001042986,0.07481388,0.005763336,0.006857583,0.001497255,0.8927656],"study_design_scores_gemma":[0.0003253664,0.001560526,0.0158884,0.000179749,0.0003498522,0.0003912414,0.0002226171,0.9535317,0.008527355,0.01566453,0.003172243,0.0001864358],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01706706,0.0001768804,0.9809979,0.00006572378,0.0000382857,0.0007097546,0.00003309865,0.0004527823,0.0004585442],"genre_scores_gemma":[0.1641603,0.0001423597,0.8336399,0.0001179524,0.00005118258,0.001296445,0.0001378486,0.00005547606,0.0003986157],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.03210677,"threshold_uncertainty_score":0.1697989,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3016903275","doi":"10.1177/0146621620909897","title":"The Monotonic Polynomial Graded Response Model: Implementation and a Comparative Study","year":2020,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University","funders":"Fonds de recherche du Québec – Nature et technologies","keywords":"Monotonic function; Mathematics; Categorical variable; Probit; Probit model; Heteroscedasticity; Polynomial; Applied mathematics; Logistic regression; Function (biology); Statistics; Econometrics","authors":[{"name":"Carl F. Falk","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8339036917039917,"gpt":0.5454023379535109,"spread":0.2885013537504808,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01439453,0.0006651503,0.0007542479,0.0008172605,0.0003797819,0.001670964,0.002708568,0.00128771,0.009166408],"category_scores_gemma":[0.05191952,0.0003472061,0.0009198419,0.001664758,0.0005549758,0.002663107,0.001381324,0.001861165,0.001979758],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001622381,"about_ca_system_score_gemma":0.001696208,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01917642,"about_ca_topic_score_gemma":0.01002732,"domain_scores_codex":[0.9939483,0.004793915,0.0001537415,0.0003566475,0.0006120965,0.0001352245],"domain_scores_gemma":[0.9779479,0.01648792,0.0004804542,0.002561999,0.00231206,0.0002097124],"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":[0.001779714,0.001936595,0.02184641,0.0007952817,0.0002424699,0.0004722833,0.001476237,0.1763367,0.003283425,0.08687114,0.008371662,0.6965882],"study_design_scores_gemma":[0.0001583082,0.000520749,0.004560205,0.0000701331,0.00005727898,0.0002066026,0.0003831415,0.9691421,0.001126051,0.01873575,0.004959899,0.0000798631],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.202211,0.0006973566,0.7752358,0.001661912,0.00008444696,0.001123333,0.001022324,0.00210919,0.01585471],"genre_scores_gemma":[0.4764812,0.0007441873,0.5165635,0.0002648996,0.00003827003,0.0008515225,0.0008149195,0.0004525877,0.003788838],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01917642,"threshold_uncertainty_score":0.0761264,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2151525783","doi":"10.1177/0146621610378289","title":"A Test-Length Correction to the Estimation of Extreme Proficiency Levels","year":2010,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Université du Québec à Montréal","funders":"","keywords":"Maximum a posteriori estimation; Estimator; Statistics; Rasch model; Mathematics; Bayes estimator; Item response theory; Bayesian probability; Estimation; Bias of an estimator; Scale (ratio); Econometrics; Maximum likelihood; Minimum-variance unbiased estimator; Psychometrics","authors":[{"name":"David Magis","is_ca":false},{"name":"Sébastien Béland","is_ca":true},{"name":"Gilles Raîche","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6476178436147106,"gpt":0.4734027822600229,"spread":0.1742150613546877,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01446431,0.0008176924,0.0007674324,0.001312271,0.0006828071,0.001224328,0.001716632,0.001227463,0.002947042],"category_scores_gemma":[0.2430987,0.0003956406,0.0006780004,0.001754723,0.001189028,0.001676918,0.00197374,0.002678626,0.001032487],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006759015,"about_ca_system_score_gemma":0.001372811,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001404628,"about_ca_topic_score_gemma":0.001822068,"domain_scores_codex":[0.9875668,0.006707657,0.0008936035,0.001908498,0.002709961,0.0002135283],"domain_scores_gemma":[0.8425143,0.1203461,0.0104194,0.01441097,0.0114989,0.0008103704],"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":[0.0009957389,0.0002775807,0.04179763,0.0008441962,0.0003946045,0.0007523396,0.001510243,0.02809228,0.02635899,0.03676597,0.005724751,0.8564857],"study_design_scores_gemma":[0.0003108663,0.002510324,0.1596428,0.0007902439,0.0006759887,0.004976157,0.000880156,0.5830349,0.1069017,0.1039703,0.03572455,0.0005820056],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.05025891,0.0004284943,0.9453156,0.0007198913,0.0003772746,0.0001790286,0.0001759555,0.0009690083,0.001575877],"genre_scores_gemma":[0.4238849,0.0002490913,0.5707935,0.0007073468,0.0002254221,0.0004310469,0.0003421603,0.000398887,0.002967649],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.01446431,"threshold_uncertainty_score":0.07649553,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2128248024","doi":"10.1177/0146621609336540","title":"An Iterative Maximum a Posteriori Estimation of Proficiency Level to Detect Multiple Local Likelihood Maxima","year":2009,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Université du Québec à Montréal","funders":"","keywords":"Maximum a posteriori estimation; Estimator; Maxima; Maximum likelihood; Mathematics; Statistics; Focus (optics); A priori and a posteriori; Restricted maximum likelihood; M-estimator; Maximum likelihood sequence estimation; Estimation theory; Computer science","authors":[{"name":"David Magis","is_ca":false},{"name":"Gilles Raîche","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4601168525429778,"gpt":0.4576214468831099,"spread":0.002495405659867944,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006674437,0.0009878488,0.001322133,0.001448376,0.0005666669,0.001336707,0.001644128,0.001431099,0.002733935],"category_scores_gemma":[0.03765738,0.000876271,0.001279864,0.0010583,0.001084974,0.002081954,0.002242671,0.001930487,0.0007718939],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000451914,"about_ca_system_score_gemma":0.001571901,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001337651,"about_ca_topic_score_gemma":0.00126078,"domain_scores_codex":[0.9966418,0.002140512,0.0001318006,0.0004017243,0.0005755988,0.0001085846],"domain_scores_gemma":[0.983138,0.01364146,0.0007304851,0.0008238151,0.001493442,0.0001728349],"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":[0.0003621545,0.0002331563,0.01028391,0.0006262779,0.0006411395,0.0002308548,0.001068059,0.2239833,0.01848689,0.04795353,0.00290626,0.6932246],"study_design_scores_gemma":[0.00005843401,0.0002872484,0.004499644,0.00009489853,0.00009685184,0.0004116507,0.0001370844,0.9408177,0.00837616,0.04149178,0.003634105,0.0000944062],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.004156306,0.00007315559,0.995086,0.00006994295,0.000009262442,0.00003095795,0.00001294848,0.0001224252,0.0004390202],"genre_scores_gemma":[0.1207994,0.0001248358,0.8773136,0.00008245512,0.00003048664,0.0002819022,0.00008733316,0.0001280478,0.001151847],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.006674437,"threshold_uncertainty_score":0.03529823,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3037823530","doi":"10.1177/0146621620931190","title":"An Exploratory Strategy to Identify and Define Sources of Differential Item Functioning","year":2020,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Advanced Statistical Modeling Techniques","field":"Computer Science","cited_by":6,"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":"Ministry of Science and Technology","keywords":"Differential item functioning; Item response theory; Differential (mechanical device); Set (abstract data type); Computer science; Dimension (graph theory); Psychology; Process (computing); Data mining; Cognitive psychology; Psychometrics; Mathematics; Developmental psychology","authors":[{"name":"Chung‐Ping Cheng","is_ca":false},{"name":"Chi‐Chen Chen","is_ca":false},{"name":"Ching‐Lin Shih","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1582989830909169,"gpt":0.3541875478248856,"spread":0.1958885647339687,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04363774,0.003023774,0.001588853,0.009674067,0.00268419,0.00321924,0.002086602,0.001961188,0.006185487],"category_scores_gemma":[0.1244925,0.001224169,0.002346109,0.005810292,0.002230081,0.003813637,0.00524972,0.001909663,0.001374113],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001521116,"about_ca_system_score_gemma":0.004388422,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001457097,"about_ca_topic_score_gemma":0.003073179,"domain_scores_codex":[0.9596214,0.03179994,0.002111998,0.002500365,0.003377488,0.0005887546],"domain_scores_gemma":[0.9013348,0.0722547,0.006121399,0.01081225,0.008865034,0.0006118379],"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":[0.001468773,0.001380774,0.03781297,0.004059387,0.001063177,0.001324481,0.0817733,0.00873835,0.03387816,0.1168149,0.01139399,0.7002918],"study_design_scores_gemma":[0.001567387,0.006304012,0.05819171,0.003291522,0.001703526,0.006008894,0.07040374,0.3178892,0.08321841,0.3393129,0.1106769,0.001431654],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.05359657,0.0001559062,0.9334416,0.0003705474,0.00006226802,0.007156582,0.0005041555,0.0009364516,0.00377588],"genre_scores_gemma":[0.1322532,0.0001052117,0.852722,0.0002275545,0.00001754869,0.01284592,0.0004513127,0.000108864,0.0012684],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.04363774,"threshold_uncertainty_score":0.2307812,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2144652829","doi":"10.1177/0146621606288556","title":"Book Review: Adapting Educational and Psychological Tests for Cross-Cultural Assessment","year":2006,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Educational and Psychological Assessments","field":"Psychology","cited_by":4,"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; Cross-cultural; Psychological testing; Applied psychology; Social psychology; Clinical psychology; Sociology; Anthropology","authors":[{"name":"Mark J. Gierl","is_ca":true},{"name":"Samira ElAtia","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1884424090147324,"gpt":0.4865119053793811,"spread":0.2980694963646486,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005962428,0.001404755,0.002380666,0.004528864,0.0006603861,0.00175443,0.001784206,0.002753009,0.009936193],"category_scores_gemma":[0.02480399,0.0006920046,0.001057241,0.004433666,0.001258857,0.002353719,0.00108877,0.004845984,0.01395194],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001700314,"about_ca_system_score_gemma":0.002444275,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007930885,"about_ca_topic_score_gemma":0.02570596,"domain_scores_codex":[0.9954898,0.001599984,0.0004093812,0.0003045675,0.002086206,0.0001100336],"domain_scores_gemma":[0.9726663,0.01497519,0.000917151,0.0007932694,0.01003153,0.0006165644],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00002627824,0.00004797113,0.000275436,0.0007609316,0.00002622026,0.00005650531,0.00002687856,0.0001020293,0.0002461481,0.0005468036,0.7961593,0.2017254],"study_design_scores_gemma":[0.0000570319,0.0001090379,0.004351216,0.002171773,0.0001001877,0.001160507,0.00006500787,0.0003524374,0.0005443008,0.002248369,0.9887838,0.0000562586],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.001494765,0.6201566,0.02135246,0.1311476,0.1864709,0.0009094111,0.001152134,0.001170743,0.03614528],"genre_scores_gemma":[0.009206559,0.5100559,0.04243768,0.1568629,0.08346388,0.001399821,0.001841072,0.001760505,0.1929718],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.009936193,"threshold_uncertainty_score":0.03323984,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2019817087","doi":"10.1177/01466216045280142","title":"SIMCAT 1.0: A SAS Computer Program for Simulating Computer Adaptive Testing","year":2006,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Software Reliability and Analysis Research","field":"Computer Science","cited_by":4,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Université de Montréal; Université du Québec à Montréal","funders":"","keywords":"Computerized adaptive testing; Computer science; Computer program; Programming language; Statistics; Psychometrics; Mathematics","authors":[{"name":"Gilles Raîche","is_ca":true},{"name":"Jean‐Guy Blais","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1397537546690987,"gpt":0.3452510035012429,"spread":0.2054972488321442,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003713722,0.0009240267,0.0009472751,0.001034494,0.0003421109,0.0009881398,0.002022883,0.0006919773,0.05130713],"category_scores_gemma":[0.01546601,0.0006942251,0.0008001332,0.00121923,0.0002829061,0.0007267922,0.0007829514,0.002594603,0.006644823],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004828829,"about_ca_system_score_gemma":0.001466839,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007185446,"about_ca_topic_score_gemma":0.006165637,"domain_scores_codex":[0.9982437,0.001009731,0.0001622177,0.0001578397,0.0002775399,0.0001490858],"domain_scores_gemma":[0.9879726,0.009414878,0.0004725984,0.0009922256,0.0009758868,0.0001718074],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.003009406,0.001653475,0.0330772,0.00225329,0.001261388,0.0005448892,0.001425883,0.1706779,0.01567413,0.06108665,0.3924336,0.3169023],"study_design_scores_gemma":[0.00124997,0.001165579,0.01913882,0.0002256189,0.000427332,0.0004351449,0.000272597,0.7467606,0.02388241,0.0427086,0.1635111,0.0002223355],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"software","genre_scores_codex":[0.05017071,0.0001881455,0.7960635,0.0003296829,0.0003836624,0.00161396,0.03198331,0.1011588,0.01810819],"genre_scores_gemma":[0.18119,0.000351972,0.7649874,0.0002601543,0.000164229,0.00890279,0.01758524,0.01652056,0.0100376],"genre_candidate":"software","genre_consensus":null,"teacher_disagreement_score":0.05130713,"threshold_uncertainty_score":0.1716395,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4327910164","doi":"10.1177/01466216231165299","title":"Confidence Screening Detector: A New Method for Detecting Test Collusion","year":2023,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":4,"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":"National Natural Science Foundation of China","keywords":"Collusion; Cheating; Clique; Computer science; Test (biology); Detector; Item response theory; Scale (ratio); Statistical hypothesis testing; Variable (mathematics); Similarity (geometry); Selection (genetic algorithm); Statistics; Data mining; Algorithm; Machine learning; Artificial intelligence; Mathematics; Psychology; Psychometrics; Social psychology","authors":[{"name":"Yongze Xu","is_ca":false},{"name":"Ying Cui","is_ca":true},{"name":"Xinyi Wang","is_ca":false},{"name":"Meiwei Huang","is_ca":false},{"name":"Fang Luo","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7210023253769497,"gpt":0.531043618285663,"spread":0.1899587070912867,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007594911,0.0009653773,0.001491888,0.004691035,0.0008421253,0.001552616,0.002776377,0.001654523,0.00238728],"category_scores_gemma":[0.04269982,0.0005114629,0.0009504157,0.002315692,0.001248643,0.001839786,0.002968826,0.001660425,0.0006566626],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008657331,"about_ca_system_score_gemma":0.002068877,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001883643,"about_ca_topic_score_gemma":0.001837345,"domain_scores_codex":[0.993534,0.002475905,0.0004363583,0.001064892,0.002192798,0.0002960799],"domain_scores_gemma":[0.9699812,0.02120539,0.002454741,0.001966643,0.003597509,0.0007945159],"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":[0.0008666341,0.0005146192,0.03174175,0.000289828,0.0003988839,0.0003215641,0.0004132217,0.04259134,0.01544987,0.02067313,0.004297736,0.8824415],"study_design_scores_gemma":[0.0001217502,0.0003573005,0.008672311,0.00004149633,0.00008734166,0.0006861259,0.00008110157,0.963706,0.008776341,0.01452353,0.002849716,0.00009697982],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02439375,0.0001615797,0.9734673,0.0001699518,0.0000653062,0.0002458928,0.0001074317,0.000677355,0.0007114435],"genre_scores_gemma":[0.279286,0.0001150824,0.718148,0.0001781129,0.00008714262,0.0004804078,0.0003327878,0.000095018,0.001277454],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.007594911,"threshold_uncertainty_score":0.0401662,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3035107461","doi":"10.1177/0146621620929431","title":"OpenMx: A Modular Research Environment for Item Response Theory Method Development","year":2020,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University","funders":"National Institute on Drug Abuse; National Institutes of Health","keywords":"Modular design; Computer science; Implementation; Item response theory; Source code; Code (set theory); Selection (genetic algorithm); Software; Software engineering; Theoretical computer science; Programming language; Machine learning; Statistics; Mathematics; Psychometrics","authors":[{"name":"Joshua N. Pritikin","is_ca":false},{"name":"Carl F. Falk","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9218223146811826,"gpt":0.5923631288612036,"spread":0.329459185819979,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02880075,0.003217607,0.002255056,0.004408705,0.0008845819,0.003797815,0.003863147,0.001432644,0.191054],"category_scores_gemma":[0.09500965,0.003119263,0.002844665,0.003527382,0.0009204316,0.003772859,0.006680813,0.003779792,0.07626463],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007549652,"about_ca_system_score_gemma":0.002839179,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008198289,"about_ca_topic_score_gemma":0.001478296,"domain_scores_codex":[0.9844635,0.009105291,0.002166183,0.001446777,0.002421096,0.0003972553],"domain_scores_gemma":[0.9067127,0.07257129,0.003128331,0.008035748,0.008319978,0.001231976],"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":[0.001596041,0.0005942238,0.003314163,0.006150199,0.0006171274,0.0003127138,0.002637252,0.002951776,0.006268101,0.01542836,0.3420572,0.6180728],"study_design_scores_gemma":[0.002561999,0.001047395,0.01310594,0.003282875,0.0005353376,0.0008250903,0.000610843,0.02514863,0.01875839,0.0818213,0.8516812,0.0006211305],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"software","genre_scores_codex":[0.002038744,0.0004398882,0.8400437,0.0005524766,0.0004026003,0.005002023,0.0208498,0.1237013,0.006969558],"genre_scores_gemma":[0.00762141,0.0004455828,0.9188676,0.0004001409,0.0001918898,0.02522054,0.01129646,0.0281039,0.007852394],"genre_candidate":"software","genre_consensus":null,"teacher_disagreement_score":0.191054,"threshold_uncertainty_score":0.6391394,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"software","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"software","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W4317423393","doi":"10.1177/01466216231151704","title":"A New Approach to Desirable Responding: Multidimensional Item Response Model of Overclaiming Data","year":2023,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Social and Intergroup Psychology","field":"Social Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"Social Sciences and Humanities Research Council of Canada; Ministry of Science and Technology, Taiwan","keywords":"Item response theory; Variety (cybernetics); Set (abstract data type); Computer science; Selection (genetic algorithm); Response bias; Empirical research; Artificial intelligence; Machine learning; Psychology; Econometrics; Cognitive psychology; Statistics; Psychometrics; Social psychology; Mathematics","authors":[{"name":"Kuan‐Yu Jin","is_ca":false},{"name":"Delroy L. Paulhus","is_ca":true},{"name":"Ching‐Lin Shih","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4958442413042139,"gpt":0.4335076058483659,"spread":0.062336635455848,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.06895237,0.002895951,0.00257652,0.00626072,0.001103466,0.005087432,0.005808744,0.002788966,0.003901975],"category_scores_gemma":[0.1480864,0.001160195,0.00372815,0.007501207,0.00443503,0.007238453,0.003540368,0.005809187,0.001298509],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003390392,"about_ca_system_score_gemma":0.002857936,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002225881,"about_ca_topic_score_gemma":0.001609494,"domain_scores_codex":[0.9207672,0.06183648,0.00282394,0.006349879,0.00754886,0.000673705],"domain_scores_gemma":[0.91065,0.06435066,0.005719066,0.01182526,0.006911328,0.0005437523],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0003400622,0.0006499801,0.02335443,0.001005977,0.001284163,0.0003677364,0.007618237,0.07721433,0.003124573,0.6976553,0.006127176,0.1812581],"study_design_scores_gemma":[0.0001275971,0.0005073203,0.005358212,0.0002175458,0.0001044867,0.0004774996,0.0007002801,0.5708858,0.0009135424,0.4114362,0.009069922,0.0002015798],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.004143841,0.0001144375,0.9937829,0.0004275456,0.00004625196,0.0003161254,0.0001724254,0.000252508,0.0007439771],"genre_scores_gemma":[0.1381454,0.0002448494,0.8560314,0.0005050736,0.0001223713,0.0034602,0.0004827387,0.0001375969,0.0008703853],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9310476,"threshold_uncertainty_score":0.3646593,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3022866891","doi":"10.1177/0146621620909898","title":"Partially and Fully Noncompensatory Response Models for Dichotomous and Polytomous Items","year":2020,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":3,"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":"Polytomous Rasch model; Item response theory; Econometrics; Set (abstract data type); Statistics; Computer science; Psychometrics; Mathematics","authors":[{"name":"R. Philip Chalmers","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.700374042316265,"gpt":0.4526023993021298,"spread":0.2477716430141352,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02729426,0.002049045,0.001715029,0.002182194,0.0009547389,0.002366508,0.005360984,0.002318057,0.009326401],"category_scores_gemma":[0.09530693,0.001203048,0.003057,0.002184137,0.002537825,0.004310145,0.003256677,0.003768581,0.001986398],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001606472,"about_ca_system_score_gemma":0.001831283,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002260014,"about_ca_topic_score_gemma":0.003266378,"domain_scores_codex":[0.9787758,0.01449477,0.001252001,0.002687821,0.002258518,0.0005310928],"domain_scores_gemma":[0.9032019,0.07369729,0.0052516,0.01318026,0.004032298,0.0006366596],"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":[0.001907238,0.001165304,0.04500903,0.001028193,0.0009576252,0.001222751,0.004764824,0.2930966,0.006496598,0.3555328,0.004077533,0.2847414],"study_design_scores_gemma":[0.0001565225,0.0007186498,0.01615435,0.0001302256,0.00014856,0.00123972,0.0004267542,0.7765836,0.001526542,0.1998352,0.002878798,0.0002011003],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1172788,0.0001530669,0.8773322,0.0005400858,0.0000710805,0.0008222497,0.0005672378,0.0004052413,0.002830112],"genre_scores_gemma":[0.6122205,0.0002595301,0.3778217,0.0004322395,0.00008279874,0.002190914,0.002053836,0.0002018614,0.004736674],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.02729426,"threshold_uncertainty_score":0.1443476,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4223957752","doi":"10.1177/01466216221084210","title":"A Comparison of Modern and Popular Approaches to Calculating Reliability for Dichotomously Scored Items","year":2022,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Multi-Criteria Decision Making","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University; Université de Montréal","funders":"","keywords":"Cronbach's alpha; Reliability (semiconductor); Monte Carlo method; Statistics; Mathematics; Econometrics; Psychology; Computer science; Psychometrics","authors":[{"name":"Sébastien Béland","is_ca":true},{"name":"Carl F. Falk","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7494451842918853,"gpt":0.4901284049147518,"spread":0.2593167793771335,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0769159,0.001092385,0.0008528571,0.006223049,0.001117828,0.002708596,0.001833847,0.001084131,0.00260749],"category_scores_gemma":[0.221654,0.0005023901,0.001122998,0.008510004,0.002823382,0.005230124,0.003177057,0.003331412,0.0007689103],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002443132,"about_ca_system_score_gemma":0.002243564,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002059894,"about_ca_topic_score_gemma":0.004633613,"domain_scores_codex":[0.9198112,0.054617,0.003925826,0.003643673,0.01738975,0.0006125647],"domain_scores_gemma":[0.8078433,0.1424576,0.008833358,0.01234984,0.02762082,0.0008951352],"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":[0.0008294308,0.0003646787,0.0535571,0.003534798,0.001405229,0.0001175768,0.01787682,0.007321856,0.002642954,0.2913171,0.01569939,0.6053331],"study_design_scores_gemma":[0.0006205178,0.003289935,0.2665031,0.01090537,0.001532221,0.002504111,0.02134194,0.1002182,0.01058354,0.4580758,0.1233947,0.001030582],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1504417,0.02268943,0.7707635,0.006190842,0.0009948574,0.0006250883,0.0006168027,0.000496751,0.04718095],"genre_scores_gemma":[0.5249273,0.009622931,0.4600832,0.0008009142,0.0004093893,0.001531478,0.0005279533,0.0002715403,0.001825423],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.0769159,"threshold_uncertainty_score":0.406775,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2056123061","doi":"10.1177/01466210122032127","title":"Book Review","year":2001,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Behavioral and Psychological Studies","field":"Psychology","cited_by":2,"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","authors":[{"name":"Mark J. Gierl","is_ca":false},{"name":"Jeffrey Bisanz","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4091965530648787,"gpt":0.3927407493978079,"spread":0.01645580366707078,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0006429603,0.0007554556,0.0008720522,0.003272999,0.0010404,0.00336596,0.001235343,0.001481146,0.350561],"category_scores_gemma":[0.002902314,0.0003617071,0.0004534091,0.003431122,0.0005100829,0.002053529,0.001387348,0.002004422,0.3359389],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001526581,"about_ca_system_score_gemma":0.002382085,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004269444,"about_ca_topic_score_gemma":0.01235352,"domain_scores_codex":[0.9994718,0.0000493246,0.00001567621,0.00005761452,0.0003597153,0.00004583007],"domain_scores_gemma":[0.9985999,0.0002764052,0.00005922141,0.0001010053,0.0007097389,0.0002537711],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000004987595,0.000008322657,0.00001940028,0.00008130247,0.000001237976,0.0000107635,0.00000749166,0.00002743552,0.00004089436,0.001346973,0.9331899,0.06526138],"study_design_scores_gemma":[0.000001618251,0.000003064084,0.00005984343,0.00006472031,0.0000011597,0.00002248312,0.000008634619,0.0000115508,0.00001654463,0.0005116042,0.9992974,0.000001499086],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"review","genre_scores_codex":[0.0003838809,0.1264689,0.002498573,0.02452893,0.05079316,0.0001870374,0.001395864,0.0007696669,0.7929739],"genre_scores_gemma":[0.0006255953,0.02212541,0.000507275,0.003334633,0.004899929,0.00004219596,0.0004791945,0.0001611667,0.9678247],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.350561,"threshold_uncertainty_score":0.9263459,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4412156054","doi":"10.1177/01466216251358492","title":"Including Empirical Prior Information in the Reliable Change Index","year":2025,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":1,"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":"Index (typography); Statistics; Econometrics; Mathematics; Psychology; Computer science","authors":[{"name":"R. Philip Chalmers","is_ca":true},{"name":"Sarah Campbell","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6224068475520583,"gpt":0.5555954167953175,"spread":0.06681143075674079,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.05375432,0.001327041,0.002185381,0.003198641,0.0006910494,0.003163863,0.002755688,0.002370463,0.00504943],"category_scores_gemma":[0.3042504,0.0009633715,0.001896202,0.00339486,0.002435115,0.004549445,0.002817374,0.004676475,0.001345325],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001101846,"about_ca_system_score_gemma":0.001611942,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002062632,"about_ca_topic_score_gemma":0.001850926,"domain_scores_codex":[0.9666463,0.0209972,0.002004796,0.00406889,0.005661621,0.0006211557],"domain_scores_gemma":[0.694805,0.2367335,0.01190254,0.04327318,0.01248013,0.0008057153],"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":[0.001481761,0.0008751136,0.1289611,0.001363776,0.001701816,0.000651159,0.001867692,0.1541677,0.00551129,0.1255901,0.008146776,0.5696818],"study_design_scores_gemma":[0.0002847,0.001312589,0.1019059,0.0007775544,0.001083784,0.001217967,0.0004341063,0.6377127,0.01059838,0.2252585,0.01888633,0.0005275754],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.03977596,0.0006432856,0.9541922,0.0005407206,0.000133256,0.000416644,0.000650682,0.0007631676,0.002884103],"genre_scores_gemma":[0.5221643,0.0004708944,0.4723026,0.0004313485,0.0001992539,0.001129991,0.00161566,0.0003845744,0.001301318],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9462457,"threshold_uncertainty_score":0.2842834,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4405622011","doi":"10.1177/01466216241310600","title":"An Information Manifold Perspective for Analyzing Test Data","year":2024,"lang":"en","type":"article","venue":"Applied Psychological Measurement","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Ottawa Hospital; McGill University","funders":"","keywords":"Item response theory; Mathematics; Measure (data warehouse); Metric (unit); Computer science; Test (biology); Manifold (fluid mechanics); Perspective (graphical); Scale (ratio); Artificial intelligence; Data mining; Statistics; Psychometrics","authors":[{"name":"J. O. Ramsay","is_ca":true},{"name":"Juan Li","is_ca":true},{"name":"Joakim Wallmark","is_ca":false},{"name":"Marie Wiberg","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7563916288387372,"gpt":0.5558183455944733,"spread":0.2005732832442639,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01990516,0.00160652,0.001863036,0.01486438,0.001295935,0.007581825,0.002672588,0.001838903,0.003270471],"category_scores_gemma":[0.08319007,0.0007592049,0.00334435,0.01123558,0.008535625,0.010935,0.00492485,0.004466482,0.0007268071],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003347845,"about_ca_system_score_gemma":0.002060162,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002850842,"about_ca_topic_score_gemma":0.001447795,"domain_scores_codex":[0.9775336,0.01406062,0.001401599,0.0021481,0.004406584,0.0004493516],"domain_scores_gemma":[0.913956,0.0638996,0.004964052,0.01084638,0.005551019,0.0007829599],"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":[0.00005442781,0.00008626415,0.005678355,0.0003324732,0.0002698477,0.0002140508,0.001239065,0.02267304,0.000759666,0.8630878,0.001883257,0.1037218],"study_design_scores_gemma":[0.00001124488,0.00007813238,0.002495231,0.00009338139,0.00004704991,0.0001493228,0.0002516714,0.1001869,0.0004939677,0.8902158,0.005929592,0.0000476032],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.004973253,0.001125401,0.9892964,0.001151247,0.00005154523,0.00008007485,0.0002785446,0.0001614855,0.002882033],"genre_scores_gemma":[0.2916406,0.002625818,0.7000664,0.0007177786,0.000805143,0.0008181244,0.001178136,0.0002258936,0.001922084],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.01990516,"threshold_uncertainty_score":0.1052698,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}