{"meta":{"query_hash":"7fae06dd8c7e","filters":{"venue":"British Journal of Mathematical and Statistical Psychology"},"cohort_total":45,"direct_labels_cover":0,"predictions_cover":45,"exported":45,"export_cap":100000,"truncated":false,"label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12"},"permalink":"https://metacan.xera.ac/q/7fae06dd8c7e","api":"https://metacan.xera.ac/api/v1/cohort?venue=British+Journal+of+Mathematical+and+Statistical+Psychology"},"results":[{"id":"W1514964987","doi":"10.1348/000711010x501671","title":"Power comparisons of significance tests of location using scores, ranks, and modular ranks","year":2010,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Carleton University","funders":"","keywords":"Wilcoxon signed-rank test; Mathematics; Statistics; Type I and type II errors; Rank (graph theory); Mann–Whitney U test; Nominal level; Statistical power; Nonparametric statistics; Test (biology); Statistical hypothesis testing; Sign test; Sample size determination; Power (physics); Modular design; Parametric statistics; Combinatorics; Computer science; Confidence interval","score_opus":0.08637119769436002,"score_gpt":0.43211560470370597,"score_spread":0.34574440700934594,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1514964987","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.13488588,0.0007797937,0.84741014,0.0004324427,0.0006104093,0.0013555663,0.0006622711,0.0005939865,0.013269438],"genre_scores_gemma":[0.79740644,0.00030839935,0.1945601,0.0002942746,0.0002752293,0.004641641,0.00047717884,0.00050350063,0.0015333273],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.90797484,0.060650337,0.0039236383,0.010302016,0.015925456,0.0012237136],"domain_scores_gemma":[0.48853168,0.44476795,0.015558265,0.04059855,0.009090019,0.001453474],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07325441,0.0014217015,0.0019390687,0.0026599804,0.0009256838,0.0027811553,0.002842284,0.0021927352,0.00801883],"category_scores_gemma":[0.47313437,0.0006471993,0.0022899692,0.0023066555,0.006877111,0.0064579067,0.0034287646,0.0033246423,0.0010252154],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.009990929,0.00080452266,0.0617395,0.003464819,0.0051742764,0.0013077318,0.0044041616,0.047566164,0.024916215,0.34869474,0.009577051,0.48235983],"study_design_scores_gemma":[0.0018761597,0.010978791,0.08973908,0.0008828154,0.0016777898,0.0017445971,0.0018808506,0.175756,0.04692356,0.6504841,0.017568676,0.00048750974],"about_ca_topic_score_codex":0.00042805847,"about_ca_topic_score_gemma":0.0003089581,"teacher_disagreement_score":0.07325441,"about_ca_system_score_codex":0.0010731448,"about_ca_system_score_gemma":0.001547223,"threshold_uncertainty_score":0.38741094},"labels":[],"label_agreement":null},{"id":"W1561373080","doi":"10.1111/j.2044-8317.2012.02049.x","title":"The performance of robust test statistics with categorical data","year":2012,"lang":"en","type":"review","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":162,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia; University of British Columbia Hospital","funders":"","keywords":"Statistics; Categorical variable; Mathematics; Statistic; Estimator; Standard error; Test statistic; Robust statistics; Type I and type II errors; Sample size determination; Statistical hypothesis testing; Variance (accounting)","score_opus":0.3136864516944847,"score_gpt":0.4807478002530468,"score_spread":0.16706134855856208,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1561373080","genre_codex":"methods","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.01398529,0.079260334,0.8938299,0.002705927,0.001045346,0.0005091094,0.0009140019,0.0010592284,0.0066907797],"genre_scores_gemma":[0.22083172,0.044515934,0.72624326,0.0012283319,0.0011265434,0.002232052,0.0016574772,0.0009318128,0.0012327973],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.8947149,0.077945314,0.0054403814,0.00552532,0.01594289,0.00043117642],"domain_scores_gemma":[0.36212304,0.5939381,0.013736037,0.015574475,0.014021704,0.00060665276],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.11301835,0.0013782451,0.0031866764,0.006660007,0.000543486,0.002712308,0.004579699,0.002823907,0.0041701356],"category_scores_gemma":[0.44564143,0.00084247044,0.0029353774,0.0070268875,0.0030073596,0.0046126274,0.002000602,0.003674139,0.0017434824],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00059174496,0.00018566106,0.013861242,0.00685537,0.0022797044,0.00025081664,0.000618469,0.02442619,0.00082658115,0.074811906,0.009714002,0.86557835],"study_design_scores_gemma":[0.00083454757,0.0043562567,0.04163879,0.0127333775,0.0025317227,0.0045227176,0.0012256351,0.21690927,0.0105526345,0.60036886,0.103549734,0.0007765696],"about_ca_topic_score_codex":0.001479561,"about_ca_topic_score_gemma":0.0010525094,"teacher_disagreement_score":0.11301835,"about_ca_system_score_codex":0.001410517,"about_ca_system_score_gemma":0.00246109,"threshold_uncertainty_score":0.5977053},"labels":[],"label_agreement":null},{"id":"W1572657933","doi":"10.1111/j.2044-8317.2011.02014.x","title":"Effect of non‐normality on test statistics for one‐way independent groups designs","year":2011,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":56,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Manitoba; York University","funders":"","keywords":"Statistics; Normality; Estimator; Truncated mean; Type I and type II errors; Mathematics; Test statistic; Nominal level; Parametric statistics; Statistic; Statistical hypothesis testing; Normal distribution; Variance (accounting); Econometrics; Nonparametric statistics; Asymptotic distribution; Confidence interval","score_opus":0.4357774667779384,"score_gpt":0.5329337513487912,"score_spread":0.09715628457085279,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1572657933","genre_codex":"methods","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.075256236,0.0032563328,0.89715815,0.0040107793,0.0024286078,0.0041755163,0.00073783816,0.0020427518,0.0109337615],"genre_scores_gemma":[0.4255878,0.00099541,0.5546073,0.0036229638,0.00052065327,0.010529082,0.00053885306,0.0016071271,0.0019908827],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.3372015,0.5184656,0.0409755,0.025767708,0.07453962,0.0030501795],"domain_scores_gemma":[0.046255026,0.9011414,0.015186869,0.025743179,0.010854807,0.00081871747],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4830267,0.002197025,0.004558305,0.0034326154,0.0027151161,0.0055341595,0.0052844062,0.004633602,0.009696848],"category_scores_gemma":[0.81352895,0.0018453144,0.004216443,0.00601883,0.013113833,0.00875107,0.004946493,0.011207606,0.0020462526],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.018734958,0.0022079092,0.06634997,0.010901072,0.0056959507,0.0026374215,0.021039011,0.03323927,0.015481546,0.21449977,0.03032857,0.57888454],"study_design_scores_gemma":[0.0050654397,0.020723665,0.15849917,0.008199779,0.003237636,0.003677329,0.004603592,0.22399752,0.08753853,0.37090173,0.11190482,0.0016507833],"about_ca_topic_score_codex":0.0013831885,"about_ca_topic_score_gemma":0.001625097,"teacher_disagreement_score":0.51697326,"about_ca_system_score_codex":0.005789003,"about_ca_system_score_gemma":0.006432535,"threshold_uncertainty_score":0.6375201},"labels":[],"label_agreement":null},{"id":"W1594740803","doi":"10.1111/j.2044-8317.2012.02038.x","title":"Evaluating the performance of different procedures for constructing confidence intervals for coefficient alpha: A simulation study","year":2012,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":16,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Confidence interval; Statistics; Reliability (semiconductor); Parametric statistics; Mathematics; Rubric; Robust confidence intervals; Population; Sample (material); CDF-based nonparametric confidence interval; Arithmetic; Medicine","score_opus":0.5157300170116348,"score_gpt":0.571068175661356,"score_spread":0.05533815864972125,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1594740803","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6410308,0.0006190917,0.35408983,0.00022689738,0.000056045043,0.001024198,0.0002850399,0.0003965137,0.002271612],"genre_scores_gemma":[0.80647737,0.00020581669,0.19158812,0.000055042616,0.0000132682935,0.0011322903,0.00026046616,0.0000849394,0.00018277141],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9361617,0.055808052,0.0022021062,0.0016714766,0.003498809,0.00065778964],"domain_scores_gemma":[0.30375126,0.65822166,0.008723916,0.018885415,0.009747865,0.0006698996],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.07491292,0.0012198264,0.0010942166,0.0022335355,0.00066074746,0.0013606773,0.0021511407,0.0020475236,0.0014322324],"category_scores_gemma":[0.33658245,0.0006779252,0.0018151964,0.0020192435,0.0013136363,0.0019262909,0.0021929366,0.0021667907,0.00020641164],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.01836703,0.0042287414,0.06291115,0.0017142082,0.0014886738,0.00047653032,0.005934827,0.60038495,0.016095696,0.034938,0.0024534708,0.25100666],"study_design_scores_gemma":[0.0009973911,0.00846024,0.02621778,0.00049924775,0.00043806614,0.0005940395,0.001070736,0.92629594,0.019983854,0.012503802,0.002637761,0.00030115197],"about_ca_topic_score_codex":0.0020423098,"about_ca_topic_score_gemma":0.001245208,"teacher_disagreement_score":0.9250871,"about_ca_system_score_codex":0.0011541476,"about_ca_system_score_gemma":0.0012330877,"threshold_uncertainty_score":0.39618212},"labels":[],"label_agreement":null},{"id":"W1602526694","doi":"10.1111/j.2044-8317.2012.02052.x","title":"A unified approach to multiple‐set canonical correlation analysis and principal components analysis","year":2012,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Sensory Analysis and Statistical Methods","field":"Agricultural and Biological Sciences","cited_by":23,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre for Addiction and Mental Health; University of British Columbia; McGill University","funders":"","keywords":"Canonical correlation; Principal component analysis; Set (abstract data type); Data set; Variance (accounting); Reduction (mathematics); Mathematics; Data mining; Computer science; Function (biology); Data reduction; Canonical analysis; Variance reduction; Algorithm; Statistics","score_opus":0.08086236820783665,"score_gpt":0.34052989123659283,"score_spread":0.2596675230287562,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1602526694","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0002466561,0.00013749942,0.9987968,0.00012536632,0.000033369586,0.00002926052,0.00002229926,0.000056436365,0.00055231986],"genre_scores_gemma":[0.016475132,0.00057985744,0.9807916,0.00013119022,0.00020625108,0.00043529205,0.00013356954,0.00015705485,0.0010900166],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99227446,0.003942751,0.00047316527,0.0011137193,0.001952857,0.00024304267],"domain_scores_gemma":[0.995061,0.0017455345,0.00046706575,0.0008346955,0.0017198054,0.00017204508],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0073343995,0.0022700247,0.002081948,0.0035558022,0.0013271827,0.0040472797,0.0028283389,0.0019762246,0.0036859787],"category_scores_gemma":[0.015460104,0.0011290476,0.0031952974,0.0055310763,0.0027443082,0.0035074563,0.004303744,0.004816591,0.0018690678],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000037834263,0.000076710196,0.0006725806,0.00037181456,0.00028883526,0.00017534356,0.00041636173,0.09129787,0.004690757,0.67085844,0.007896453,0.22321703],"study_design_scores_gemma":[0.000026435746,0.00008496323,0.0009895733,0.0001144526,0.000088267436,0.0002684561,0.00012124013,0.49372005,0.0022557534,0.4657951,0.036387146,0.00014850912],"about_ca_topic_score_codex":0.0028428328,"about_ca_topic_score_gemma":0.0035468694,"teacher_disagreement_score":0.0073343995,"about_ca_system_score_codex":0.0013224266,"about_ca_system_score_gemma":0.0040314435,"threshold_uncertainty_score":0.038788438},"labels":[],"label_agreement":null},{"id":"W1607630103","doi":"10.1348/2044-8317.002007","title":"Bootstrap standard error and confidence intervals for the correlations corrected for indirect range restriction","year":2010,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":22,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Confidence interval; Statistics; Standard error; Estimator; Range (aeronautics); Mathematics; Monte Carlo method; Correlation coefficient; Sample size determination; Correlation; Robust confidence intervals; CDF-based nonparametric confidence interval; Coverage probability","score_opus":0.16151295514791586,"score_gpt":0.47109177901239585,"score_spread":0.30957882386448,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1607630103","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.079150975,0.0028072202,0.90655667,0.00045368826,0.0003806217,0.000522386,0.0013201237,0.001163894,0.007644502],"genre_scores_gemma":[0.57401067,0.0010235982,0.4187984,0.00024958313,0.00018215159,0.0019259539,0.0021961632,0.00045554293,0.0011578454],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.95203686,0.027472883,0.0025871885,0.004671121,0.012376585,0.00085541484],"domain_scores_gemma":[0.56482273,0.32934818,0.021923665,0.048355017,0.034548886,0.0010015746],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06973001,0.001191453,0.0022958466,0.006206425,0.00095777924,0.0027117145,0.0046692183,0.0035208713,0.0040382845],"category_scores_gemma":[0.49499997,0.00063482265,0.002303299,0.005393247,0.0026240305,0.0038428323,0.0030674606,0.0037200048,0.0012623384],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002536505,0.00063996215,0.056846205,0.0018998523,0.0023407661,0.0011443881,0.0038871272,0.16184965,0.005967365,0.3404202,0.020254942,0.40221298],"study_design_scores_gemma":[0.00057502,0.0010925618,0.08361192,0.0029778646,0.00091156806,0.0025484583,0.0016893992,0.562044,0.02119911,0.2920279,0.030573586,0.0007486447],"about_ca_topic_score_codex":0.0020646928,"about_ca_topic_score_gemma":0.0010846256,"teacher_disagreement_score":0.06973001,"about_ca_system_score_codex":0.0011804618,"about_ca_system_score_gemma":0.0013643471,"threshold_uncertainty_score":0.36877197},"labels":[],"label_agreement":null},{"id":"W1899992665","doi":"10.1348/000711010x524739","title":"A simple and effective decision rule for choosing a significance test to protect against non‐normality","year":2010,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":24,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Carleton University","funders":"","keywords":"Normality; Normality test; Statistics; Parametric statistics; Sample size determination; Wilcoxon signed-rank test; Statistical hypothesis testing; Mathematics; Test (biology); Type I and type II errors; Nominal level; Variance (accounting); Mann–Whitney U test; Sample (material); Computer science; Econometrics; Confidence interval","score_opus":0.0458017058659475,"score_gpt":0.43433415168309225,"score_spread":0.38853244581714474,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1899992665","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.010537031,0.0006619776,0.9780692,0.0028393106,0.00086591183,0.0009016406,0.00027112418,0.0015064896,0.0043474333],"genre_scores_gemma":[0.07050886,0.0002627524,0.9242436,0.0019948243,0.0006152524,0.00093611603,0.00014983279,0.00037826205,0.0009104299],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9259913,0.04412038,0.007251384,0.0068306047,0.014404324,0.0014019535],"domain_scores_gemma":[0.6773732,0.26784062,0.010581704,0.017513353,0.023061695,0.0036294875],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09972114,0.0017448739,0.0032912677,0.0065813055,0.0031124405,0.005153696,0.0044765966,0.0072937645,0.005521242],"category_scores_gemma":[0.36532173,0.0010087057,0.0022236293,0.002908217,0.005137658,0.004409776,0.0028286574,0.009018634,0.004828129],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0026883176,0.00068933837,0.013815573,0.002242125,0.0008387187,0.0018314236,0.002275817,0.010137616,0.022499317,0.16398467,0.057541426,0.72145563],"study_design_scores_gemma":[0.0023939256,0.0029897727,0.015495382,0.004121863,0.0009110252,0.008909556,0.0020046902,0.14571624,0.07146803,0.64904916,0.09531254,0.001627832],"about_ca_topic_score_codex":0.00072427647,"about_ca_topic_score_gemma":0.0008561042,"teacher_disagreement_score":0.09972114,"about_ca_system_score_codex":0.0010836781,"about_ca_system_score_gemma":0.0033881841,"threshold_uncertainty_score":0.52738214},"labels":[],"label_agreement":null},{"id":"W1968781945","doi":"10.1348/000711007x270843","title":"Nearly unbiased estimators for the three‐parameter Weibull distribution with greater efficiency than the iterative likelihood method","year":2008,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Statistical Distribution Estimation and Applications","field":"Mathematics","cited_by":42,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal","funders":"","keywords":"Mathematics; Statistics; Weibull distribution; Estimator; Monte Carlo method; Sample size determination; Maximum likelihood; Population; Estimation theory; Shape parameter; Sample (material); Applied mathematics; Physics","score_opus":0.06871545134842222,"score_gpt":0.37757502735672843,"score_spread":0.3088595760083062,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1968781945","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0034023495,0.0001268459,0.99577516,0.00004169576,0.000009003084,0.000016223134,0.000016244638,0.00015058841,0.00046191658],"genre_scores_gemma":[0.11680416,0.0004911146,0.8800215,0.00015282206,0.00006728023,0.0002560213,0.00016993553,0.00025699445,0.0017801425],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9971277,0.0013147399,0.000144175,0.00028707265,0.0010273734,0.00009899142],"domain_scores_gemma":[0.990053,0.006132751,0.0008972092,0.001556386,0.0012723341,0.00008841956],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0053810473,0.00069184013,0.001045446,0.0013935036,0.00032257303,0.0012896151,0.001054794,0.0009925568,0.002119364],"category_scores_gemma":[0.030553028,0.00066661974,0.00072802894,0.0015852124,0.0006253413,0.0022871753,0.0016463537,0.0012972264,0.0013216762],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00011831641,0.00010006835,0.005452205,0.00023264212,0.00028341167,0.0001270075,0.000423532,0.24407552,0.015889982,0.1610413,0.0030490872,0.5692069],"study_design_scores_gemma":[0.0000575575,0.000085933,0.0045902412,0.00009204612,0.00007319963,0.00043483343,0.000080011945,0.85012907,0.007941446,0.12574029,0.010642926,0.00013244963],"about_ca_topic_score_codex":0.0010465009,"about_ca_topic_score_gemma":0.0011386735,"teacher_disagreement_score":0.0053810473,"about_ca_system_score_codex":0.00039176305,"about_ca_system_score_gemma":0.0010445615,"threshold_uncertainty_score":0.028458059},"labels":[],"label_agreement":null},{"id":"W2005050546","doi":"10.1348/000711006x117853","title":"Robust step‐down tests for multivariate independent group designs","year":2006,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University; University of Manitoba","funders":"University of Manitoba","keywords":"Statistics; Estimator; Mathematics; Truncated mean; Multivariate statistics; Covariance; Statistic; Type I and type II errors; Sample size determination; Analysis of covariance; Test statistic; Statistical hypothesis testing; Econometrics","score_opus":0.48204144907285956,"score_gpt":0.527100397187587,"score_spread":0.04505894811472749,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2005050546","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.011700478,0.00023969992,0.98311806,0.00011940042,0.0001885786,0.0020577107,0.0005389994,0.0006396352,0.0013974328],"genre_scores_gemma":[0.15232874,0.00023536215,0.83082926,0.00032996436,0.00014312734,0.013389206,0.001247973,0.00041687404,0.0010796053],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8667594,0.09621794,0.0064613144,0.011885195,0.017250666,0.0014254783],"domain_scores_gemma":[0.5154081,0.4144174,0.017374415,0.035669688,0.015641162,0.0014892698],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.11052383,0.0026143792,0.00541266,0.0051857606,0.0018361029,0.0029237655,0.004920423,0.0024849358,0.014196156],"category_scores_gemma":[0.45028248,0.0013142852,0.0060295374,0.0043401,0.0036225168,0.0037311674,0.004168233,0.0052889395,0.0022334708],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.008240425,0.0010826123,0.024716998,0.0047874264,0.009379524,0.0010925689,0.0026036706,0.06401567,0.010732802,0.1985082,0.017809022,0.65703106],"study_design_scores_gemma":[0.003014096,0.015057578,0.046021428,0.0014233589,0.0030110073,0.0012824233,0.0012485093,0.5875852,0.024277166,0.28683048,0.029376384,0.00087233755],"about_ca_topic_score_codex":0.0010705736,"about_ca_topic_score_gemma":0.0007683712,"teacher_disagreement_score":0.11052383,"about_ca_system_score_codex":0.0013310855,"about_ca_system_score_gemma":0.0047374796,"threshold_uncertainty_score":0.5845128},"labels":[],"label_agreement":null},{"id":"W2006005214","doi":"10.1348/000711000159187","title":"Repeated measures ANOVA: Some new results on comparing trimmed means and means","year":2000,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":40,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Manitoba","funders":"","keywords":"Univariate; Mathematics; Percentile; Statistics; Bonferroni correction; Pairwise comparison; Type I and type II errors; Repeated measures design; Truncated mean; Multivariate statistics; Extant taxon; Confidence interval; Econometrics","score_opus":0.15745614856899287,"score_gpt":0.4242876479967463,"score_spread":0.2668314994277534,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2006005214","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0048450194,0.009169696,0.97887963,0.0008478224,0.0009929041,0.000120292316,0.00012897841,0.00050225155,0.0045134407],"genre_scores_gemma":[0.07188589,0.009932691,0.9095093,0.001436935,0.0028526166,0.0012121706,0.00028934947,0.0007318172,0.0021492487],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.97432536,0.012929859,0.0015021486,0.0033725672,0.007509543,0.00036053578],"domain_scores_gemma":[0.8852288,0.08901154,0.005414985,0.012446986,0.007151224,0.0007464649],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02683509,0.0020127408,0.003631491,0.0052122846,0.0015470425,0.002699941,0.0033440352,0.0023897635,0.0056877164],"category_scores_gemma":[0.12855582,0.0011337764,0.0041492907,0.0052411333,0.0076593193,0.00660414,0.0028705674,0.0067368713,0.0020179544],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00067731284,0.00041775178,0.004123249,0.0040650847,0.0020094872,0.0011478834,0.005046183,0.015578932,0.00912386,0.32142133,0.009793567,0.62659526],"study_design_scores_gemma":[0.0001768295,0.0024739131,0.008415529,0.0010919436,0.0008381221,0.002192763,0.0007220749,0.027444018,0.009887809,0.8637385,0.08256611,0.00045233007],"about_ca_topic_score_codex":0.0009009453,"about_ca_topic_score_gemma":0.0007850953,"teacher_disagreement_score":0.02683509,"about_ca_system_score_codex":0.0014846673,"about_ca_system_score_gemma":0.0017858826,"threshold_uncertainty_score":0.1419192},"labels":[],"label_agreement":null},{"id":"W2030349076","doi":"10.1348/000711005x66770","title":"Bootstrap scree tests: A Monte Carlo simulation and applications to published data","year":2006,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University; Alberta Advanced Education","funders":"","keywords":"Parametric statistics; Statistics; Statistical hypothesis testing; Econometrics; Monte Carlo method; Eigenvalues and eigenvectors; Nonparametric statistics; Mathematics; Principal component analysis; Goodness of fit; Computer science","score_opus":0.19945774268679498,"score_gpt":0.4912955240526531,"score_spread":0.29183778136585814,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2030349076","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.005617257,0.00032414572,0.9884653,0.00032572952,0.0001169661,0.00032685662,0.0003956978,0.002652554,0.0017754735],"genre_scores_gemma":[0.07484001,0.00046570488,0.91941285,0.00011764717,0.00014600836,0.0022423589,0.00086861994,0.0009511086,0.0009557768],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.96356744,0.030352432,0.0012476451,0.0012993986,0.0033132848,0.00021971142],"domain_scores_gemma":[0.8002772,0.17267978,0.0056526307,0.011111388,0.009657078,0.00062185613],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.038553145,0.0016026931,0.0024013768,0.008968638,0.001557026,0.0028834066,0.003041267,0.0019178268,0.012934723],"category_scores_gemma":[0.29538277,0.00092611654,0.0016080015,0.010768896,0.0021649513,0.0035138333,0.0027272396,0.0023399265,0.004070772],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003378301,0.00029363815,0.013985941,0.0012059779,0.0008705063,0.00067515235,0.00236058,0.1560237,0.0015299432,0.15567847,0.03776975,0.6292684],"study_design_scores_gemma":[0.0001384104,0.00023388599,0.0049858196,0.000377755,0.00010038362,0.0007157368,0.00054523424,0.7386566,0.0020734484,0.22858569,0.023344912,0.00024216423],"about_ca_topic_score_codex":0.0015960186,"about_ca_topic_score_gemma":0.0018924844,"teacher_disagreement_score":0.038553145,"about_ca_system_score_codex":0.0009529287,"about_ca_system_score_gemma":0.0016826263,"threshold_uncertainty_score":0.20389098},"labels":[],"label_agreement":null},{"id":"W2036701577","doi":"10.1348/000711009x475853","title":"Detecting a lack of association: An equivalence testing approach","year":2009,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":54,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"York University","funders":"","keywords":"Association (psychology); Type I and type II errors; Equivalence (formal languages); Sample size determination; Test (biology); Association test; Statistics; Statistical hypothesis testing; Econometrics; Independence (probability theory); Statistical power; Sample (material); Psychology; Computer science; Mathematics","score_opus":0.6273935671893198,"score_gpt":0.5794021805264894,"score_spread":0.047991386662830426,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2036701577","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.043631077,0.001382154,0.9422256,0.0029970417,0.00058665994,0.0015608112,0.00032605554,0.00026832853,0.007022237],"genre_scores_gemma":[0.570909,0.0007154406,0.42092016,0.00156379,0.0006784122,0.0041608545,0.0003073624,0.0001490074,0.0005959603],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.61061335,0.29433456,0.0191898,0.022576883,0.051150624,0.0021347576],"domain_scores_gemma":[0.2145181,0.7191576,0.021967512,0.028945902,0.013587895,0.0018229715],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.28656146,0.0017949637,0.004775877,0.008750044,0.0019227223,0.0050897417,0.006749189,0.004569479,0.007042953],"category_scores_gemma":[0.6846911,0.0010843384,0.004986869,0.006152449,0.013298943,0.011297093,0.008941964,0.0074145575,0.0006847212],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0040785777,0.0014058545,0.0695333,0.0035036127,0.0068619456,0.0013254898,0.006751082,0.012317366,0.0028865137,0.35076785,0.0064397007,0.5341287],"study_design_scores_gemma":[0.0010411587,0.0059813038,0.024785656,0.0009631812,0.0016882147,0.001953684,0.0017842302,0.08333107,0.0027974958,0.8645161,0.010882982,0.00027490116],"about_ca_topic_score_codex":0.0007561881,"about_ca_topic_score_gemma":0.00037738905,"teacher_disagreement_score":0.7134385,"about_ca_system_score_codex":0.0021217058,"about_ca_system_score_gemma":0.004477589,"threshold_uncertainty_score":0.87979674},"labels":[],"label_agreement":null},{"id":"W2075806409","doi":"10.1111/bmsp.12045","title":"Equivalence tests for comparing correlation and regression coefficients","year":2014,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":55,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"York University","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Equivalence (formal languages); Mathematics; Statistics; Correlation; Regression analysis; Regression; Linear regression; Applied mathematics; Null hypothesis; Econometrics; Pure mathematics","score_opus":0.4380628287343638,"score_gpt":0.5641740061112318,"score_spread":0.126111177376868,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2075806409","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.027072376,0.0018157412,0.95651054,0.00059759244,0.000788778,0.0018224122,0.0022174262,0.001106555,0.008068545],"genre_scores_gemma":[0.51333433,0.0015405918,0.46580452,0.0010792962,0.0008234092,0.010886954,0.00333562,0.0009032934,0.0022919949],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.77859557,0.1556842,0.014056776,0.0148603795,0.0349235,0.0018795788],"domain_scores_gemma":[0.36074588,0.57529193,0.02335399,0.024996774,0.014252562,0.0013588448],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.11919678,0.0024233598,0.0041824225,0.010763852,0.0014529631,0.0036582677,0.0044924687,0.0031365617,0.015391161],"category_scores_gemma":[0.5263608,0.00083619746,0.005717297,0.010124274,0.006924456,0.0068056816,0.0046581994,0.007870583,0.0024135811],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0041687037,0.00093878974,0.066041924,0.0044442597,0.009388068,0.0009791747,0.0019559462,0.043614864,0.0028249233,0.2806957,0.017312454,0.5676352],"study_design_scores_gemma":[0.0015394397,0.0072667333,0.061083555,0.0019124562,0.0026657882,0.0019287099,0.001685202,0.25276768,0.006700216,0.61474,0.047090534,0.0006196118],"about_ca_topic_score_codex":0.0016051976,"about_ca_topic_score_gemma":0.0007526754,"teacher_disagreement_score":0.11919678,"about_ca_system_score_codex":0.0018922682,"about_ca_system_score_gemma":0.0035130272,"threshold_uncertainty_score":0.6303804},"labels":[],"label_agreement":null},{"id":"W2082983435","doi":"10.1111/bmsp.12001","title":"Consequences of choosing samples in hypothesis testing to ensure homogeneity of variance","year":2013,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Carleton University","funders":"","keywords":"Statistics; Mathematics; Homogeneity (statistics); F-test of equality of variances; Type I and type II errors; Sample size determination; Variance (accounting); Levene's test; Standard deviation; Population; Econometrics; Statistical hypothesis testing; Test statistic; Demography","score_opus":0.5089277066177947,"score_gpt":0.5105138001551611,"score_spread":0.0015860935373663843,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2082983435","genre_codex":"methods","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.099536516,0.0011831003,0.8708172,0.0069330293,0.0020849977,0.011732823,0.00021651671,0.00065996405,0.006835871],"genre_scores_gemma":[0.3895895,0.0001906106,0.58946985,0.00510895,0.0005643396,0.013692871,0.00013148038,0.00034063825,0.00091184216],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.24677177,0.60850686,0.040463876,0.027663276,0.07222175,0.004372483],"domain_scores_gemma":[0.10318071,0.8234228,0.016982406,0.043684483,0.011052533,0.001677204],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.6066401,0.0022153587,0.0042022117,0.0028340493,0.0046855034,0.004429076,0.0049922075,0.0068047885,0.0031160822],"category_scores_gemma":[0.81666833,0.002175024,0.0036221035,0.0036379637,0.017231284,0.0057921736,0.005525991,0.0123735005,0.0008187941],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.034375727,0.0030570175,0.08021499,0.0041355602,0.006903916,0.009681783,0.033417027,0.032636464,0.026943764,0.30285496,0.02687322,0.4389056],"study_design_scores_gemma":[0.010395703,0.020437645,0.06919311,0.0032786438,0.0019921155,0.0047925375,0.004496744,0.16381668,0.06935993,0.61052173,0.040567536,0.0011476275],"about_ca_topic_score_codex":0.001552159,"about_ca_topic_score_gemma":0.0014720024,"teacher_disagreement_score":0.3933599,"about_ca_system_score_codex":0.004377043,"about_ca_system_score_gemma":0.005858383,"threshold_uncertainty_score":0.4850828},"labels":[],"label_agreement":null},{"id":"W2088581243","doi":"10.1348/000711005x63755","title":"Adaptive robust estimation and testing","year":2006,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":42,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Manitoba","funders":"","keywords":"Statistics; Normality; Type I and type II errors; Trimming; Mathematics; Sample size determination; Variance (accounting); Econometrics; Null hypothesis; Population; Computer science; Medicine","score_opus":0.09456399157201978,"score_gpt":0.37824000341710623,"score_spread":0.28367601184508645,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2088581243","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0020743543,0.00015783265,0.9956689,0.00025081765,0.000052424733,0.00014711155,0.000066791494,0.0003576487,0.0012240371],"genre_scores_gemma":[0.112626396,0.00027320706,0.88269705,0.00034604408,0.00013562833,0.001291118,0.0003993353,0.00038370586,0.0018475372],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.95661885,0.030455384,0.0018414591,0.004582764,0.00546042,0.0010410353],"domain_scores_gemma":[0.7957727,0.16243787,0.0074052303,0.020411657,0.012830295,0.0011423188],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.055219833,0.0022196146,0.0032505163,0.0025231072,0.0015218947,0.0033483207,0.007867158,0.0031226878,0.010926387],"category_scores_gemma":[0.30601722,0.0011963727,0.002881333,0.0026700546,0.004289446,0.0044745356,0.005531486,0.005048908,0.0026990294],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006781377,0.00020820223,0.008345185,0.00084587734,0.00090274814,0.00035174747,0.00057626783,0.20394628,0.0017531232,0.3829587,0.007746024,0.39168766],"study_design_scores_gemma":[0.00017208674,0.00026785358,0.0015348986,0.00025105698,0.00012124268,0.00021915892,0.00012947763,0.67336315,0.0025870693,0.3144946,0.0067792237,0.00008013186],"about_ca_topic_score_codex":0.0031858692,"about_ca_topic_score_gemma":0.0022779265,"teacher_disagreement_score":0.055219833,"about_ca_system_score_codex":0.0017909033,"about_ca_system_score_gemma":0.0046199546,"threshold_uncertainty_score":0.29203385},"labels":[],"label_agreement":null},{"id":"W2099457688","doi":"10.1111/bmsp.12015","title":"Evaluating clinical significance: Incorporating robust statistics with normative comparison tests","year":2013,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"York University","funders":"","keywords":"Statistics; Normative; Econometrics; Statistical hypothesis testing; Mathematics; Epistemology; Philosophy","score_opus":0.6101012297971987,"score_gpt":0.6102618176534461,"score_spread":0.00016058785624739613,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2099457688","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.022011755,0.0010183487,0.971902,0.0005792681,0.00024330716,0.0010237232,0.00015002346,0.00055024034,0.0025213158],"genre_scores_gemma":[0.37768292,0.00044269432,0.61660385,0.0005442991,0.00031148212,0.003466432,0.00027832415,0.00030252093,0.00036762547],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.6559089,0.2778592,0.015176712,0.012767205,0.03715629,0.0011316892],"domain_scores_gemma":[0.22713828,0.6956818,0.023080967,0.030240573,0.022831598,0.0010268128],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2609399,0.0017511544,0.0035269982,0.0061071753,0.0011999077,0.004741474,0.0046885554,0.0033159903,0.0035246054],"category_scores_gemma":[0.77255523,0.0009789696,0.0024621976,0.003719446,0.0056867674,0.0060383426,0.0051296023,0.0048708273,0.0005892296],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0037853199,0.0011284319,0.041107554,0.0024114526,0.0037371523,0.0010068837,0.0022608144,0.10737009,0.0027601165,0.16579685,0.0058587966,0.6627766],"study_design_scores_gemma":[0.0010390439,0.006293753,0.023093846,0.0013476177,0.0011351261,0.0012361093,0.00062417565,0.5721117,0.0072197514,0.37199637,0.0135077555,0.00039467795],"about_ca_topic_score_codex":0.0011958493,"about_ca_topic_score_gemma":0.0010022421,"teacher_disagreement_score":0.2609399,"about_ca_system_score_codex":0.0018837021,"about_ca_system_score_gemma":0.0045905765,"threshold_uncertainty_score":0.9113927},"labels":[],"label_agreement":null},{"id":"W2102817893","doi":"10.1111/j.2044-8317.2012.02056.x","title":"Robust tests of equivalence for <i>k</i> independent groups","year":2012,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":18,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"York University","funders":"","keywords":"Mathematics; Statistics; Equivalence (formal languages); Population; Null hypothesis; Analysis of variance; Psychology; Social psychology; Econometrics; Pure mathematics; Demography","score_opus":0.5007231631368919,"score_gpt":0.5401128261342657,"score_spread":0.03938966299737379,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2102817893","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.10809417,0.0020984288,0.8678611,0.001400416,0.0011285136,0.0036124804,0.0026590363,0.000840196,0.012305688],"genre_scores_gemma":[0.68992937,0.0006058228,0.2922828,0.001313323,0.00057856354,0.010929209,0.0026725389,0.00045281585,0.0012355426],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.77099675,0.12494908,0.019136777,0.036719736,0.04475302,0.0034446656],"domain_scores_gemma":[0.24367893,0.62555045,0.035544313,0.076128535,0.016644245,0.0024534604],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.17660302,0.0020405205,0.0050699417,0.0059721437,0.0022000251,0.0054081953,0.0065775714,0.004677476,0.009626767],"category_scores_gemma":[0.65368,0.0010986216,0.005651397,0.0056707417,0.014352557,0.0094046965,0.0065476066,0.006326001,0.0016420112],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.018728573,0.0026275099,0.12239864,0.009157435,0.019352749,0.0028592842,0.011170901,0.04145072,0.011261166,0.34077618,0.019453904,0.40076292],"study_design_scores_gemma":[0.0028776221,0.013391148,0.1352842,0.0026730357,0.0031179923,0.0026500295,0.0047828117,0.13582666,0.014118253,0.65694535,0.02741931,0.000913514],"about_ca_topic_score_codex":0.00070939143,"about_ca_topic_score_gemma":0.00042319574,"teacher_disagreement_score":0.17660302,"about_ca_system_score_codex":0.0021844858,"about_ca_system_score_gemma":0.0031707878,"threshold_uncertainty_score":0.9339772},"labels":[],"label_agreement":null},{"id":"W2113330311","doi":"10.1348/000711002159680","title":"Controlling the rate of Type I error over a large set of statistical tests","year":2002,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":160,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Manitoba","funders":"","keywords":"False discovery rate; Type I and type II errors; Multiple comparisons problem; Statistical power; Null hypothesis; Statistics; Statistical hypothesis testing; p-value; Mathematics; Word error rate; Null (SQL); Set (abstract data type); Statistical process control; Statistical significance; Sample size determination; False positive rate; Computer science; Data mining; Artificial intelligence; Process (computing); Biology","score_opus":0.39724289716766226,"score_gpt":0.5422999661469321,"score_spread":0.14505706897926984,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2113330311","genre_codex":"methods","genre_gemma":"methods","domain_codex":"methods","domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.022017216,0.0021301296,0.9527689,0.0016193554,0.0024168892,0.012260213,0.00037398256,0.0015211087,0.0048921453],"genre_scores_gemma":[0.14856283,0.00090149295,0.81539756,0.002903935,0.0006864138,0.029000176,0.000341535,0.00066642347,0.0015395386],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.36207607,0.440835,0.048604563,0.04983001,0.095247656,0.0034066595],"domain_scores_gemma":[0.12334668,0.731044,0.043774217,0.06515575,0.03472632,0.0019530827],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.5507762,0.00414736,0.007030083,0.007288325,0.0031711871,0.00734006,0.008346199,0.0075068963,0.0042295447],"category_scores_gemma":[0.7638771,0.0024926423,0.0052566165,0.0081691155,0.0104147745,0.008171902,0.0088915145,0.011434504,0.0018853579],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.017215539,0.003832608,0.11604199,0.016292319,0.02234209,0.0040032,0.020583592,0.032391533,0.038025554,0.13921858,0.025678607,0.5643744],"study_design_scores_gemma":[0.005432101,0.018135095,0.1061317,0.010632201,0.009907185,0.0062749973,0.0044408175,0.2338651,0.11401053,0.36057934,0.12822263,0.0023683386],"about_ca_topic_score_codex":0.0012421777,"about_ca_topic_score_gemma":0.0016905032,"teacher_disagreement_score":0.5507762,"about_ca_system_score_codex":0.0038332064,"about_ca_system_score_gemma":0.01033951,"threshold_uncertainty_score":0.55397296},"labels":[],"label_agreement":null},{"id":"W2125118056","doi":"10.1348/000711001159357","title":"The analysis of repeated measures designs: A review","year":2001,"lang":"en","type":"review","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":299,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Manitoba","funders":"","keywords":"Univariate; Strengths and weaknesses; Statistic; Repeated measures design; Statistics; Multivariate statistics; Multivariate analysis; Type I and type II errors; Statistical analysis; Multivariate analysis of variance; Statistical hypothesis testing; Analysis of variance; Term (time); Econometrics; Mathematics; Computer science; Psychology; Social psychology","score_opus":0.35337936081111715,"score_gpt":0.5400826959220477,"score_spread":0.18670333511093057,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2125118056","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000114042596,0.98882586,0.008828032,0.00051708875,0.00056026044,0.00008257876,0.000051122803,0.00008549416,0.0009354668],"genre_scores_gemma":[0.0012985982,0.9843873,0.012511761,0.0003328021,0.000540772,0.0002580975,0.000078528225,0.000039558385,0.0005526342],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.98882294,0.00470182,0.0014235978,0.0012793011,0.0036431334,0.00012922668],"domain_scores_gemma":[0.94587654,0.0428858,0.0024544944,0.0015698323,0.006845815,0.00036755565],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01848821,0.0019899283,0.0052107666,0.0064333947,0.0005880853,0.0020879316,0.0053448123,0.002636324,0.0048194397],"category_scores_gemma":[0.035763435,0.0011535458,0.0019634024,0.007824099,0.0029435195,0.0032457504,0.0011369811,0.0022871667,0.004988938],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000075797914,0.00009208147,0.00026454582,0.035729498,0.00028868022,0.000116456074,0.00017931932,0.00037620796,0.0005978918,0.0040884335,0.017405959,0.94078517],"study_design_scores_gemma":[0.00011399578,0.00077782164,0.005437768,0.03585501,0.0011559374,0.0027928208,0.0003231156,0.0008422007,0.002713501,0.023528006,0.92624384,0.00021602456],"about_ca_topic_score_codex":0.0022152753,"about_ca_topic_score_gemma":0.0022704222,"teacher_disagreement_score":0.01848821,"about_ca_system_score_codex":0.0019219685,"about_ca_system_score_gemma":0.0043538804,"threshold_uncertainty_score":0.097776175},"labels":[],"label_agreement":null},{"id":"W2131040314","doi":"10.1348/000711000159178","title":"An examination of the robustness of the empirical Bayes and other approaches for testing main and interaction effects in repeated measures designs","year":2000,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":18,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Manitoba","funders":"","keywords":"Univariate; Multivariate statistics; Statistics; Mathematics; Covariance; Type I and type II errors; Estimator; Statistic; Multivariate analysis; Robustness (evolution); Bayes' theorem; Covariance matrix; Statistical hypothesis testing; Econometrics; Bayesian probability","score_opus":0.37346727652047623,"score_gpt":0.45220365936991447,"score_spread":0.07873638284943824,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2131040314","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.024430726,0.0030576414,0.9647047,0.0013502892,0.00029037884,0.0011898392,0.00018956244,0.00030063998,0.0044862307],"genre_scores_gemma":[0.31836745,0.001684017,0.6728512,0.0014068454,0.0003273554,0.0034072804,0.00025038485,0.00035948394,0.0013459844],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.6358912,0.3033245,0.0125368405,0.015605683,0.03139818,0.0012436572],"domain_scores_gemma":[0.16863246,0.76049113,0.020508735,0.035062153,0.014741294,0.0005641982],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.35612607,0.0024782054,0.0026155943,0.0043396777,0.0019026742,0.003203898,0.004673047,0.0033194134,0.004128385],"category_scores_gemma":[0.7234297,0.0015621687,0.0056885956,0.0038749638,0.0054082773,0.0055468045,0.002631211,0.0050660996,0.0007706291],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0048465575,0.00088765525,0.077618346,0.005665941,0.023640385,0.0010727961,0.008951839,0.067211986,0.008281242,0.2131377,0.004071401,0.5846141],"study_design_scores_gemma":[0.0018410882,0.010679499,0.13412805,0.004595234,0.0119144935,0.003690906,0.0031452943,0.35961753,0.026095325,0.4010222,0.042091098,0.0011793347],"about_ca_topic_score_codex":0.003844322,"about_ca_topic_score_gemma":0.002959059,"teacher_disagreement_score":0.64387393,"about_ca_system_score_codex":0.0022493086,"about_ca_system_score_gemma":0.0028417,"threshold_uncertainty_score":0.7940112},"labels":[],"label_agreement":null},{"id":"W2140217706","doi":"10.1348/000711003321645412","title":"Pairwise multiple comparisons: A model comparison approach versus stepwise procedures","year":2003,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Optimal Experimental Design Methods","field":"Decision Sciences","cited_by":28,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Manitoba; York University","funders":"","keywords":"Pairwise comparison; Selection (genetic algorithm); Normality; Model selection; Multiple comparisons problem; Set (abstract data type); Variance (accounting); Mathematics; Statistics; Type I and type II errors; Computer science; Econometrics; Machine learning","score_opus":0.2715206547783738,"score_gpt":0.48669607144257343,"score_spread":0.21517541666419965,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2140217706","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.010269405,0.00040186505,0.9839436,0.0005578181,0.0004892607,0.0026226155,0.00023845256,0.0006039778,0.0008730654],"genre_scores_gemma":[0.05799088,0.00035858463,0.93199146,0.00026668722,0.00016513538,0.00842938,0.0001482284,0.00020235607,0.00044734302],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.738834,0.22896932,0.0069575016,0.011503419,0.012423693,0.0013119782],"domain_scores_gemma":[0.63954717,0.3136663,0.013780629,0.025276238,0.0063741775,0.0013554868],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.16662304,0.0041916836,0.005051565,0.0049515627,0.001600414,0.0025783877,0.007465878,0.003272175,0.008000341],"category_scores_gemma":[0.35133398,0.001497655,0.0053083966,0.0069822194,0.0055063227,0.0061197956,0.004459914,0.009498944,0.000989716],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.007816106,0.0025267478,0.008332752,0.01017028,0.015723227,0.0013316162,0.005784046,0.061750494,0.013291865,0.25107217,0.019524096,0.6026767],"study_design_scores_gemma":[0.0025752431,0.023238705,0.00987979,0.0016195392,0.0029138867,0.0012128042,0.0012432237,0.3451964,0.016278092,0.56724215,0.027886007,0.000714209],"about_ca_topic_score_codex":0.0006687517,"about_ca_topic_score_gemma":0.0010466195,"teacher_disagreement_score":0.16662304,"about_ca_system_score_codex":0.0014675369,"about_ca_system_score_gemma":0.0037342242,"threshold_uncertainty_score":0.8811974},"labels":[],"label_agreement":null},{"id":"W2151152712","doi":"10.1348/000711002760554516","title":"Skill set analysis in knowledge structures","year":2002,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Computability, Logic, AI Algorithms","field":"Computer Science","cited_by":58,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Brock University","funders":"","keywords":"Consistency (knowledge bases); Set (abstract data type); Interpretation (philosophy); Computer science; Test (biology); Cognitive psychology; Empirical research; Test theory; Artificial intelligence; Mathematics; Psychology; Statistics; Psychometrics","score_opus":0.03577935715460232,"score_gpt":0.328755334554809,"score_spread":0.29297597740020664,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2151152712","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.097569786,0.00056704343,0.8789157,0.0012314324,0.00003709207,0.000107598804,0.0004253976,0.00025447184,0.020891525],"genre_scores_gemma":[0.80352026,0.00040420642,0.19067357,0.0002419786,0.000137773,0.0002859505,0.0005816058,0.00007008841,0.004084507],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99632496,0.0011701415,0.00022128897,0.00068356143,0.0012883708,0.00031159286],"domain_scores_gemma":[0.97830576,0.015898613,0.0016626802,0.0023100753,0.0012722869,0.0005506248],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003633058,0.00048129336,0.00068325945,0.0061722463,0.0012294961,0.003378823,0.0017201689,0.0011791335,0.0072552077],"category_scores_gemma":[0.028320674,0.00043864187,0.0014399489,0.003498887,0.0039705494,0.008511789,0.0037541955,0.0020706297,0.00056108204],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000037858998,0.00005251221,0.0037076795,0.000106866944,0.00005007521,0.00007279427,0.0006909799,0.01782112,0.0006385058,0.92282486,0.00056535203,0.053431407],"study_design_scores_gemma":[0.000006767238,0.000020897067,0.0016597328,0.000025607189,0.000012930248,0.000028365532,0.0000850856,0.06223182,0.00037770713,0.9340693,0.0014711636,0.000010556762],"about_ca_topic_score_codex":0.0028946344,"about_ca_topic_score_gemma":0.0018213954,"teacher_disagreement_score":0.0072552077,"about_ca_system_score_codex":0.0023498456,"about_ca_system_score_gemma":0.0010966319,"threshold_uncertainty_score":0.024271071},"labels":[],"label_agreement":null},{"id":"W2158319724","doi":"10.1348/000711002760554606","title":"The comparative sensitivity of ordinal multiple regression and least squares regression to departures from interval scaling","year":2002,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Advanced Decision-Making Techniques","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Victoria","funders":"","keywords":"Mathematics; Statistics; Ordinal regression; Raw score; Ordinal data; Regression; Regression analysis; Discretization; Optimal distinctiveness theory; Scaling; Raw data; Ordinal Scale; Interval (graph theory); Psychology; Combinatorics","score_opus":0.05512543054606262,"score_gpt":0.36836296659184314,"score_spread":0.31323753604578053,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2158319724","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.46410576,0.0016388847,0.52471983,0.00073743274,0.00019425678,0.00045243872,0.00045090826,0.0027245542,0.004975953],"genre_scores_gemma":[0.8883372,0.0004164523,0.10869741,0.00019695672,0.00004618016,0.0002942415,0.0004701441,0.0006632282,0.00087818404],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.92429537,0.060253054,0.003205246,0.004544348,0.006935448,0.0007664865],"domain_scores_gemma":[0.41447338,0.53201723,0.015618053,0.028196992,0.008725675,0.0009686092],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.08687162,0.0012114329,0.0015276581,0.0022510397,0.00052182155,0.0024616297,0.001398964,0.0014001456,0.0015519143],"category_scores_gemma":[0.4633445,0.00096778636,0.0016939418,0.0021761036,0.0027122821,0.003110083,0.0030638725,0.0025055623,0.0007992294],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.004551376,0.00022235898,0.030700691,0.0012685102,0.0014015463,0.00033868244,0.0016490228,0.78449696,0.016979089,0.013678169,0.0012046319,0.14350891],"study_design_scores_gemma":[0.00015544087,0.001968892,0.035750933,0.00017382437,0.00018256344,0.0005514027,0.0004291074,0.9274827,0.015403015,0.015971422,0.0016246923,0.0003060428],"about_ca_topic_score_codex":0.002663507,"about_ca_topic_score_gemma":0.0015083072,"teacher_disagreement_score":0.08687162,"about_ca_system_score_codex":0.0010462835,"about_ca_system_score_gemma":0.0007951683,"threshold_uncertainty_score":0.45942652},"labels":[],"label_agreement":null},{"id":"W2163998574","doi":"10.1111/j.2044-8317.2011.02017.x","title":"A note on consistency of non‐parametric rank tests and related rank transformations","year":2011,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":26,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Carleton University","funders":"","keywords":"Wilcoxon signed-rank test; Mathematics; Rank (graph theory); Consistency (knowledge bases); Statistics; Null hypothesis; Nonparametric statistics; Mann–Whitney U test; Parametric statistics; Statistical hypothesis testing; Kruskal–Wallis one-way analysis of variance; Type I and type II errors; Sample size determination; Null (SQL); Statistical significance; Sample (material); Econometrics; Combinatorics; Computer science; Discrete mathematics; Data mining","score_opus":0.09940994877516365,"score_gpt":0.415083829253759,"score_spread":0.31567388047859535,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2163998574","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.038397387,0.0010966902,0.94546574,0.0024267768,0.0005334033,0.00039172408,0.00025187226,0.0006016578,0.010834654],"genre_scores_gemma":[0.5680057,0.0008569462,0.4223594,0.0015558685,0.00072678574,0.0016131399,0.0004364234,0.0013424564,0.003103217],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.7838391,0.14814794,0.0127203,0.01448427,0.038726114,0.0020822212],"domain_scores_gemma":[0.20501733,0.68326706,0.022563705,0.068881035,0.019428512,0.00084238587],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1980346,0.0011544478,0.0021267587,0.0025169908,0.0018432601,0.0042982697,0.0041904934,0.0029668645,0.0052912557],"category_scores_gemma":[0.6896602,0.00086815015,0.002307538,0.003922755,0.012084054,0.006897678,0.005102463,0.007537326,0.0013433242],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0034351565,0.0003777665,0.045052476,0.0013771619,0.0015372321,0.0021335098,0.006877235,0.053573016,0.010685057,0.61696994,0.010592092,0.24738935],"study_design_scores_gemma":[0.0004363919,0.0015359963,0.019326244,0.0005449003,0.0003705297,0.002911475,0.0016717281,0.11703446,0.020443982,0.82059896,0.014756468,0.00036892644],"about_ca_topic_score_codex":0.00086817495,"about_ca_topic_score_gemma":0.0005589654,"teacher_disagreement_score":0.1980346,"about_ca_system_score_codex":0.0011569347,"about_ca_system_score_gemma":0.0025849815,"threshold_uncertainty_score":0.9889661},"labels":[],"label_agreement":null},{"id":"W2164006511","doi":"10.1348/000711008x299742","title":"A comparative study of robust tests for spread: Asymmetric trimming strategies","year":2008,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":18,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Manitoba","funders":"","keywords":"Trimming; Estimator; Mathematics; A priori and a posteriori; Statistics; Robustness (evolution); Type I and type II errors; Levene's test; F-test of equality of variances; Variance (accounting); Analysis of variance; Econometrics; Adaptive estimator; Statistical hypothesis testing; Computer science; Test statistic; Homogeneity (statistics)","score_opus":0.2892882943145064,"score_gpt":0.4905955814410748,"score_spread":0.20130728712656837,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2164006511","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.10787106,0.0018497106,0.88430065,0.0003229911,0.00013002231,0.0006570733,0.00032909843,0.00083141733,0.0037078664],"genre_scores_gemma":[0.56470764,0.00049135665,0.4310157,0.0001944995,0.00011758746,0.001139925,0.00058742554,0.0009458808,0.0007999989],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.8180908,0.14743975,0.0073197274,0.010019997,0.015853938,0.0012758583],"domain_scores_gemma":[0.18596496,0.74730456,0.020146398,0.030591113,0.01448474,0.0015082215],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.19091725,0.002429037,0.0023711799,0.006665353,0.0012685305,0.0028405765,0.0046235123,0.0027226536,0.0046650143],"category_scores_gemma":[0.6522364,0.0010580906,0.0038796747,0.005419356,0.0048362766,0.0077660754,0.0043598586,0.0032643012,0.00089607097],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0071236957,0.0010099267,0.056653433,0.0024996009,0.011351296,0.0010645321,0.005800326,0.10659221,0.012924202,0.1358469,0.0048429696,0.654291],"study_design_scores_gemma":[0.0014036034,0.009995886,0.10847889,0.0015331075,0.0035480259,0.0018441669,0.0027501273,0.6128081,0.025060698,0.22134596,0.01006473,0.0011666237],"about_ca_topic_score_codex":0.0012795732,"about_ca_topic_score_gemma":0.0011367477,"teacher_disagreement_score":0.19091725,"about_ca_system_score_codex":0.0017749378,"about_ca_system_score_gemma":0.0023598773,"threshold_uncertainty_score":0.99774307},"labels":[],"label_agreement":null},{"id":"W2168460233","doi":"10.1348/000711009x480640","title":"Coping with stress through decisional control: Quantification of negotiating the environment","year":2010,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Decision-Making and Behavioral Economics","field":"Decision Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Coping (psychology); Negotiation; Cognition; Computer science; Cognitive psychology; Psychology; Hierarchy; Social psychology; Clinical psychology; Economics","score_opus":0.08798309826525118,"score_gpt":0.40092374668028197,"score_spread":0.3129406484150308,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2168460233","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.17084393,0.0012496287,0.7756175,0.002389506,0.00008751292,0.00010815393,0.00016102691,0.000112960515,0.049429756],"genre_scores_gemma":[0.9622625,0.00041027484,0.035889756,0.000047770736,0.00003576031,0.00008766128,0.000038638354,0.000016222266,0.0012113004],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","domain_scores_codex":[0.998201,0.0009299904,0.000081864324,0.00021949038,0.00036381665,0.00020387545],"domain_scores_gemma":[0.9959925,0.0021841887,0.0007391108,0.0004790822,0.00030173268,0.00030348555],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0034666164,0.0008853101,0.00047840487,0.00093335065,0.0007527798,0.004788863,0.0012634781,0.0010241786,0.0030719186],"category_scores_gemma":[0.010100103,0.0003114099,0.00072901166,0.0009015494,0.005702251,0.006078977,0.0027770228,0.0013593956,0.00021576666],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00007890553,0.00008638501,0.0026064538,0.000122052865,0.000061019186,0.00007961318,0.0008529333,0.12157788,0.0017717961,0.8523217,0.00040109342,0.020040153],"study_design_scores_gemma":[0.000009263795,0.00008829393,0.0015457891,0.000043837026,0.000021259242,0.000038172377,0.0004549108,0.25877988,0.0005559658,0.7367322,0.0016959832,0.00003445428],"about_ca_topic_score_codex":0.0019583362,"about_ca_topic_score_gemma":0.0012714948,"teacher_disagreement_score":0.004788863,"about_ca_system_score_codex":0.0020396945,"about_ca_system_score_gemma":0.001323105,"threshold_uncertainty_score":0.018333435},"labels":[],"label_agreement":null},{"id":"W2168627095","doi":"10.1348/000711004849222","title":"A note on preliminary tests of equality of variances","year":2004,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":313,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Carleton University","funders":"","keywords":"F-test of equality of variances; Statistics; Test (biology); Type I and type II errors; Levene's test; Mathematics; Sample size determination; Econometrics; Nominal level; Statistical hypothesis testing; Test statistic","score_opus":0.38388052115437615,"score_gpt":0.5674083538348488,"score_spread":0.18352783268047262,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2168627095","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.007456451,0.0037762974,0.9513534,0.014054727,0.007327001,0.0014894211,0.0006107238,0.001332912,0.012599128],"genre_scores_gemma":[0.09574228,0.0023488323,0.87602174,0.010543193,0.0040088724,0.005508857,0.00052879954,0.001129901,0.004167534],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.5201839,0.3841525,0.02878077,0.01539795,0.04871037,0.0027743983],"domain_scores_gemma":[0.07748074,0.84343356,0.011343653,0.042564116,0.023914834,0.0012630296],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.3721899,0.0032068582,0.0056241066,0.005257437,0.0037128278,0.00754857,0.0098010395,0.008205086,0.011057176],"category_scores_gemma":[0.808348,0.002686398,0.0072886874,0.00681569,0.018003752,0.015507499,0.007128354,0.03231106,0.0043726107],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.003434019,0.0006565785,0.01825583,0.004752213,0.001759979,0.0048613213,0.011061739,0.014837211,0.006110269,0.4475262,0.1042025,0.3825422],"study_design_scores_gemma":[0.00073019665,0.003981824,0.012969322,0.004971502,0.0006074524,0.002808628,0.0022977262,0.042010915,0.015557892,0.73571366,0.17741379,0.0009370472],"about_ca_topic_score_codex":0.0027170402,"about_ca_topic_score_gemma":0.0020150219,"teacher_disagreement_score":0.3721899,"about_ca_system_score_codex":0.003227923,"about_ca_system_score_gemma":0.008891927,"threshold_uncertainty_score":0.77420163},"labels":[],"label_agreement":null},{"id":"W2171530386","doi":"10.1348/000711000159286","title":"Testing treatment effects in repeated measures designs: Trimmed means and bootstrapping","year":2000,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Advanced Statistical Modeling Techniques","field":"Computer Science","cited_by":45,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Saskatchewan Health; Health Canada; University of Manitoba","funders":"","keywords":"Estimator; Type I and type II errors; Statistics; Mathematics; Covariance; Sphericity; Bootstrapping (finance); Normality; Statistical hypothesis testing; Repeated measures design; Econometrics","score_opus":0.06834616108167536,"score_gpt":0.3423298604087666,"score_spread":0.27398369932709127,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2171530386","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.02343567,0.00053591304,0.9719133,0.00019550393,0.00020638111,0.0020597763,0.000081502905,0.00052215933,0.0010497483],"genre_scores_gemma":[0.085749425,0.00025239278,0.90657926,0.0002032713,0.000084154875,0.0065106023,0.000103104874,0.0001708358,0.00034692642],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.76030666,0.18929401,0.011088727,0.012429815,0.025300782,0.0015800071],"domain_scores_gemma":[0.58452016,0.32677004,0.024566924,0.044242017,0.01866855,0.0012322151],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.19604385,0.0023431713,0.004683096,0.0049273334,0.001879198,0.0026737167,0.0064108334,0.0035537332,0.0036722405],"category_scores_gemma":[0.4994042,0.0015816084,0.004270148,0.005297172,0.00619513,0.0043069185,0.0034392069,0.005205616,0.00072284223],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0059281625,0.0024694977,0.027935455,0.0050300816,0.0059456043,0.0013606976,0.012539977,0.032138664,0.021003626,0.104675494,0.004890504,0.7760823],"study_design_scores_gemma":[0.0033732408,0.024644291,0.09461439,0.0034100087,0.0042964523,0.0017691627,0.0036014519,0.4311406,0.055206146,0.34889516,0.027821133,0.0012279656],"about_ca_topic_score_codex":0.0017334851,"about_ca_topic_score_gemma":0.0021300705,"teacher_disagreement_score":0.19604385,"about_ca_system_score_codex":0.0017132276,"about_ca_system_score_gemma":0.0043517537,"threshold_uncertainty_score":0.99142104},"labels":[],"label_agreement":null},{"id":"W2330570680","doi":"10.1111/bmsp.12066","title":"Testing for negligible interaction: A coherent and robust approach","year":2016,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"York University","funders":"","keywords":"Categorical variable; Contrast (vision); Type I and type II errors; Normality; Econometrics; Homoscedasticity; Statistical hypothesis testing; Interaction; Homogeneity (statistics); Statistics; Variance (accounting); Null hypothesis; Statistical power; Bootstrapping (finance); Computer science; Mathematics; Heteroscedasticity; Artificial intelligence; Economics","score_opus":0.5714294779352754,"score_gpt":0.5466511481811501,"score_spread":0.024778329754125306,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2330570680","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.005855678,0.00033446794,0.99052304,0.0005557053,0.00012631134,0.0008498931,0.000098041986,0.00032398241,0.0013329717],"genre_scores_gemma":[0.12842853,0.00022361272,0.865645,0.0006555098,0.00021508358,0.0040866835,0.00020002906,0.00021237433,0.0003331982],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.5778459,0.3188587,0.019697128,0.032682244,0.04914054,0.0017755696],"domain_scores_gemma":[0.30434248,0.59843194,0.02693899,0.050574232,0.0177178,0.0019945228],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.26758814,0.0027203604,0.006475872,0.0059612296,0.0027009156,0.0067285863,0.006531767,0.0057074795,0.006339961],"category_scores_gemma":[0.65231806,0.0019455065,0.0060884804,0.004662245,0.010584312,0.008202677,0.009326665,0.008462206,0.001543969],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0034787806,0.0009184699,0.02316154,0.0029299187,0.0078736935,0.001482213,0.0042076474,0.023492316,0.0073755467,0.3767678,0.008536944,0.5397752],"study_design_scores_gemma":[0.0017391292,0.0072793346,0.016128497,0.0013915787,0.0028603843,0.0016419047,0.0011950039,0.2259261,0.012812264,0.70140177,0.026975373,0.00064864097],"about_ca_topic_score_codex":0.0008493682,"about_ca_topic_score_gemma":0.0005231944,"teacher_disagreement_score":0.73241186,"about_ca_system_score_codex":0.0022233382,"about_ca_system_score_gemma":0.0055587743,"threshold_uncertainty_score":0.9031942},"labels":[],"label_agreement":null},{"id":"W2584362917","doi":"10.1111/bmsp.12085","title":"Population models and simulation methods: The case of the Spearman rank correlation","year":2017,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":76,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Spearman's rank correlation coefficient; Rank correlation; Correlation; Statistic; Copula (linguistics); Statistics; Rank (graph theory); Population; Sample size determination; Mathematics; Nonparametric statistics; Econometrics; Test statistic; Statistical hypothesis testing; Demography","score_opus":0.12336615640232285,"score_gpt":0.48009945331021286,"score_spread":0.35673329690789,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2584362917","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.004519532,0.005914594,0.9711019,0.0065247505,0.00045597056,0.000066461966,0.000084478175,0.00009952776,0.011232756],"genre_scores_gemma":[0.39086863,0.014520099,0.58273786,0.002802189,0.002097622,0.0010208203,0.00017199077,0.00031433662,0.005466385],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9550154,0.03897704,0.00079067796,0.0016540999,0.003124074,0.0004386905],"domain_scores_gemma":[0.846248,0.13884321,0.0044670617,0.0060290066,0.003616999,0.0007957325],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0440298,0.0012151287,0.0019944287,0.002989142,0.0012244134,0.00488032,0.0022707991,0.004354761,0.0037334014],"category_scores_gemma":[0.15136415,0.00089821656,0.0015419892,0.0042149937,0.009297171,0.007504661,0.004326064,0.0063170576,0.0006646744],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000015331625,0.000013661025,0.0007772573,0.00012882156,0.00007151486,0.000095995536,0.0002570738,0.03734398,0.000045049088,0.94647896,0.001345226,0.013427068],"study_design_scores_gemma":[0.000013003097,0.000028072123,0.0002668799,0.0001239445,0.000019001578,0.000089733,0.00008556808,0.095150985,0.000078479155,0.8960834,0.008034235,0.000026681666],"about_ca_topic_score_codex":0.0056865406,"about_ca_topic_score_gemma":0.002803516,"teacher_disagreement_score":0.0440298,"about_ca_system_score_codex":0.0024868005,"about_ca_system_score_gemma":0.0036062647,"threshold_uncertainty_score":0.2328546},"labels":[],"label_agreement":null},{"id":"W2620799017","doi":"10.1111/bmsp.12103","title":"<scp>ANOVA</scp> and the variance homogeneity assumption: Exploring a better gatekeeper","year":2017,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":81,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"York University","funders":"","keywords":"Homogeneity (statistics); Analysis of variance; F-test of equality of variances; Null hypothesis; Statistics; Mathematics; Levene's test; One-way analysis of variance; Equivalence (formal languages); Statistical hypothesis testing; Variance (accounting); Econometrics; Mixed-design analysis of variance; Population variance; Population; Test statistic; Demography; Discrete mathematics","score_opus":0.4557394537833383,"score_gpt":0.5102854334071317,"score_spread":0.05454597962379343,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2620799017","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0347263,0.00032624352,0.95045215,0.0035065506,0.00030243868,0.00084957556,0.00016262587,0.0010417108,0.008632471],"genre_scores_gemma":[0.49267864,0.00020269911,0.5009387,0.0024123376,0.00024711335,0.0017785562,0.00017369386,0.00041435825,0.0011539891],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8774519,0.090359055,0.004716338,0.008097313,0.018037636,0.0013378029],"domain_scores_gemma":[0.5493031,0.36614373,0.030679578,0.035128757,0.016931685,0.0018131686],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.118957445,0.0011359325,0.002245675,0.0015918987,0.0019856186,0.0045678783,0.0037535303,0.0027064919,0.0053722854],"category_scores_gemma":[0.39081383,0.00074289925,0.0021718235,0.0024573854,0.008348247,0.008259993,0.0035712554,0.0056147096,0.001019835],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002877736,0.0007253173,0.019426063,0.001241772,0.00079414225,0.00084487087,0.0046516913,0.030553006,0.006968772,0.50633216,0.0123336315,0.41325083],"study_design_scores_gemma":[0.0006893619,0.0052165603,0.023380479,0.0008390378,0.00037884995,0.00066629384,0.0010746864,0.25316346,0.019414626,0.6616282,0.033074938,0.00047354485],"about_ca_topic_score_codex":0.003515632,"about_ca_topic_score_gemma":0.0024320798,"teacher_disagreement_score":0.88104254,"about_ca_system_score_codex":0.003116956,"about_ca_system_score_gemma":0.0065744603,"threshold_uncertainty_score":0.6291146},"labels":[],"label_agreement":null},{"id":"W2782608753","doi":"10.1111/bmsp.12126","title":"On the solution multiplicity of the Fleishman method and its impact in simulation studies","year":2018,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Normality; Monte Carlo method; Multiplicity (mathematics); Population; Applied mathematics; Mathematics; Computer science; Algorithm; Statistics; Mathematical analysis","score_opus":0.13750207654009666,"score_gpt":0.50524852138227,"score_spread":0.3677464448421734,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2782608753","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.04462082,0.0040239915,0.9299486,0.0068777907,0.00042241468,0.00032587885,0.0001079325,0.0003376555,0.0133348275],"genre_scores_gemma":[0.3983811,0.00227613,0.59474903,0.0010630232,0.00029810218,0.0009255906,0.00012389699,0.00035197052,0.0018312437],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.88945055,0.09661645,0.0019879453,0.0034942834,0.007684547,0.000766176],"domain_scores_gemma":[0.27488032,0.6924819,0.009860682,0.011178855,0.010225617,0.0013725642],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.11367284,0.001301416,0.0020619014,0.003948549,0.001956622,0.0049510035,0.0026006622,0.0042483937,0.0071453126],"category_scores_gemma":[0.599131,0.0009525251,0.0017775232,0.0037471377,0.007052567,0.008906875,0.0058272895,0.0058739167,0.0007345711],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007694045,0.00023901883,0.011534074,0.0009961355,0.00030192704,0.0002538748,0.0022501135,0.14596939,0.0010051659,0.68750405,0.004286917,0.1448899],"study_design_scores_gemma":[0.00018186765,0.00050934806,0.001873652,0.001110963,0.00012288599,0.00026906581,0.00052994007,0.5088646,0.0016604861,0.47696525,0.007754896,0.00015710795],"about_ca_topic_score_codex":0.0038383426,"about_ca_topic_score_gemma":0.0033875047,"teacher_disagreement_score":0.11367284,"about_ca_system_score_codex":0.0030725903,"about_ca_system_score_gemma":0.004322305,"threshold_uncertainty_score":0.6011666},"labels":[],"label_agreement":null},{"id":"W2942651266","doi":"10.1111/bmsp.12166","title":"Bayesian generalized structured component analysis","year":2019,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"National University of Singapore","keywords":"Component (thermodynamics); Bayesian probability; Component analysis; Econometrics; Mathematics; Statistics; Computer science; Applied mathematics","score_opus":0.0576921882517152,"score_gpt":0.41720794336886785,"score_spread":0.35951575511715267,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2942651266","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.002865864,0.00048537427,0.9936698,0.00025586892,0.000078776466,0.00019687854,0.00043707062,0.00045069944,0.00155969],"genre_scores_gemma":[0.14351568,0.0015203842,0.84346545,0.00038863227,0.00028117062,0.0015136842,0.0034337833,0.0005101292,0.005371041],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99222845,0.004781129,0.00031692523,0.001254315,0.0010857309,0.00033350373],"domain_scores_gemma":[0.9896774,0.0057863123,0.0007462106,0.0012164969,0.0023382683,0.00023527944],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008764805,0.0019519597,0.0025067816,0.0031115972,0.0012137057,0.0030172206,0.0030719303,0.0018633849,0.008353111],"category_scores_gemma":[0.03563248,0.0008867326,0.0025451886,0.0045315325,0.0014435325,0.0024129818,0.0026885567,0.0029249904,0.0027233816],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000278528,0.00015365575,0.005742321,0.0005646863,0.0008993614,0.0002001438,0.0005874211,0.24006231,0.0013584348,0.32014102,0.022655124,0.40735704],"study_design_scores_gemma":[0.000042407723,0.000043737076,0.001675648,0.000109347646,0.00010247232,0.00009480141,0.00007125559,0.7793291,0.00032910286,0.20665975,0.011476762,0.00006555477],"about_ca_topic_score_codex":0.012175342,"about_ca_topic_score_gemma":0.010900549,"teacher_disagreement_score":0.012175342,"about_ca_system_score_codex":0.0015629833,"about_ca_system_score_gemma":0.0043336106,"threshold_uncertainty_score":0.04635328},"labels":[],"label_agreement":null},{"id":"W2995591168","doi":"10.1111/bmsp.12194","title":"The danger of conflating level‐specific effects of control variables when primary interest lies in level‐2 effects","year":2019,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":38,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Conflation; Cluster (spacecraft); Variable (mathematics); Control variable; Econometrics; Multilevel model; Estimation; Control (management); Statistics; Variables; Psychology; Computer science; Mathematics; Economics; Epistemology; Artificial intelligence","score_opus":0.0946494624905834,"score_gpt":0.3728073056028315,"score_spread":0.2781578431122481,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2995591168","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.018651726,0.0023765655,0.955196,0.010593482,0.0017242417,0.00032411327,0.00045386815,0.0009470012,0.009732953],"genre_scores_gemma":[0.36678034,0.0017161584,0.6093718,0.013263452,0.0013066146,0.0017703363,0.00053379487,0.0011831993,0.0040743896],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8167135,0.13480504,0.011473543,0.012319761,0.023198655,0.0014895357],"domain_scores_gemma":[0.5868557,0.2912759,0.028999822,0.07641584,0.015358238,0.0010944951],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.14596398,0.00183239,0.0031974732,0.0027506968,0.002662772,0.0055341646,0.003775885,0.0032166003,0.0042164368],"category_scores_gemma":[0.46376657,0.0016742726,0.002321546,0.0052447096,0.010593778,0.007612426,0.00851299,0.013037918,0.001926974],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00062986684,0.0001630341,0.026452232,0.0019807424,0.0011365394,0.00064418756,0.019155564,0.008385325,0.0043511204,0.6821377,0.029954378,0.22500928],"study_design_scores_gemma":[0.000109081804,0.00029006563,0.013016979,0.001385972,0.000345155,0.00086290936,0.0025371595,0.02616815,0.010009819,0.8825261,0.06246445,0.00028427926],"about_ca_topic_score_codex":0.0055482117,"about_ca_topic_score_gemma":0.0047724107,"teacher_disagreement_score":0.85403603,"about_ca_system_score_codex":0.0026656168,"about_ca_system_score_gemma":0.0030529827,"threshold_uncertainty_score":0.7719405},"labels":[],"label_agreement":null},{"id":"W3005846352","doi":"10.1111/bmsp.12201","title":"Can we disregard the whole model? Omnibus non‐inferiority testing for <i>R</i> <sup>2</sup> in multi‐variable linear regression and in ANOVA","year":2020,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":23,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"Nederlandse Organisatie voor Wetenschappelijk Onderzoek","keywords":"Statistics; Null hypothesis; Statistical hypothesis testing; Outcome (game theory); Type I and type II errors; Multiple comparisons problem; Linear regression; Bayesian probability; Econometrics; Regression analysis; Linear model; Analysis of variance; Statistical power; Variable (mathematics); Computer science; Mathematics","score_opus":0.16531862255891705,"score_gpt":0.426819047784887,"score_spread":0.26150042522596995,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3005846352","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.11672719,0.0017045974,0.8656868,0.006068457,0.0012365379,0.0008783725,0.00040022147,0.0004247933,0.006873023],"genre_scores_gemma":[0.7287282,0.00079900696,0.2639321,0.0023441685,0.00067553046,0.0016600648,0.0003223421,0.000394545,0.001144073],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.77515805,0.18062486,0.008285982,0.016101483,0.017633168,0.0021964293],"domain_scores_gemma":[0.20952035,0.74693865,0.013355234,0.025052926,0.0037635518,0.0013692501],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.25696465,0.001957929,0.0052871415,0.0019751037,0.001396806,0.0051087756,0.0051036035,0.0045992243,0.008619613],"category_scores_gemma":[0.5446261,0.001174679,0.0054005287,0.0019812079,0.0114797875,0.008808415,0.004160023,0.0065269726,0.0011510185],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.016191404,0.0013976264,0.058215957,0.0042592892,0.012055244,0.0034341791,0.009159283,0.044627864,0.0057664355,0.44646597,0.012759797,0.38566706],"study_design_scores_gemma":[0.0010953015,0.0042893463,0.016848154,0.0007210572,0.0015172771,0.0012256248,0.0013496801,0.17109825,0.003754645,0.79162675,0.0061415597,0.00033227992],"about_ca_topic_score_codex":0.0012965563,"about_ca_topic_score_gemma":0.0012948422,"teacher_disagreement_score":0.25696465,"about_ca_system_score_codex":0.0023277402,"about_ca_system_score_gemma":0.0044779796,"threshold_uncertainty_score":0.9162949},"labels":[],"label_agreement":null},{"id":"W3047129708","doi":"10.1111/bmsp.12207","title":"A note on computing Louis’ observed information matrix identity for IRT and cognitive diagnostic models","year":2020,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"York University","funders":"National Taiwan Normal University","keywords":"Latent variable; Latent variable model; Computer science; Representation (politics); Variable (mathematics); Identity matrix; Monte Carlo method; Flexibility (engineering); Algorithm; Matrix (chemical analysis); Mathematics; Algebraic number; Maximization; Applied mathematics; Theoretical computer science; Artificial intelligence; Mathematical optimization; Statistics; Eigenvalues and eigenvectors","score_opus":0.42078208780786935,"score_gpt":0.49142747486818933,"score_spread":0.07064538706031998,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3047129708","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0007097138,0.00025020514,0.99583995,0.0007424399,0.00019976422,0.00002861434,0.000052213236,0.00040146965,0.001775626],"genre_scores_gemma":[0.022635592,0.00053363707,0.9724467,0.00051912974,0.00025100147,0.00023687184,0.00015437258,0.0005588867,0.0026637898],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.987581,0.0076637063,0.000816249,0.0009982386,0.0026198954,0.00032084278],"domain_scores_gemma":[0.94205046,0.043162417,0.0014959357,0.0076705907,0.0051114694,0.0005090891],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01846344,0.0018429188,0.0013704585,0.0026848668,0.0012884273,0.0043460038,0.0032842269,0.0025896663,0.014611276],"category_scores_gemma":[0.150753,0.00096260733,0.0018716939,0.0033375667,0.0033869282,0.008501317,0.004226754,0.0074116983,0.007972815],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00006983091,0.00006722762,0.0016903965,0.0002505525,0.0000701608,0.0003996108,0.00040936476,0.024710944,0.0017810831,0.73020023,0.017760955,0.22258972],"study_design_scores_gemma":[0.00003891913,0.00010516387,0.0010259759,0.00023647219,0.000042786887,0.00080092205,0.00015346541,0.26481718,0.0054681697,0.68304306,0.04409496,0.00017288394],"about_ca_topic_score_codex":0.004400261,"about_ca_topic_score_gemma":0.005348517,"teacher_disagreement_score":0.01846344,"about_ca_system_score_codex":0.0017833717,"about_ca_system_score_gemma":0.0033776732,"threshold_uncertainty_score":0.09764522},"labels":[],"label_agreement":null},{"id":"W3147659269","doi":"10.1111/bmsp.12240","title":"Model‐based recursive partitioning of extended redundancy analysis with an application to nicotine dependence among US adults","year":2021,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Sensory Analysis and Statistical Methods","field":"Agricultural and Biological Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; University of Manitoba","funders":"","keywords":"Covariate; Recursive partitioning; Partition (number theory); Econometrics; Statistics; Mathematics; Hierarchy; Set (abstract data type); Redundancy (engineering); Computer science; Combinatorics","score_opus":0.02750667536499382,"score_gpt":0.324438908424291,"score_spread":0.29693223305929717,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3147659269","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.037229236,0.00022947631,0.9612537,0.00015234126,0.000025596575,0.00013731816,0.00014494368,0.00047951192,0.00034784377],"genre_scores_gemma":[0.26248455,0.0002084041,0.73529404,0.00007427979,0.000040210743,0.00057165115,0.00052223325,0.00024383272,0.00056082435],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9936958,0.00515192,0.00018881932,0.000462571,0.0003243956,0.00017650964],"domain_scores_gemma":[0.99140817,0.0061090295,0.00049293635,0.0010698709,0.00079431816,0.00012560078],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008938575,0.0011225861,0.0014356578,0.0019967526,0.000935524,0.00081520557,0.0013234724,0.0006115303,0.0015000227],"category_scores_gemma":[0.025769403,0.0005931618,0.002985468,0.0018553725,0.00069076166,0.00085562875,0.002458476,0.0012613648,0.00040630187],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00065810804,0.0005596532,0.031946152,0.000729012,0.0024807372,0.0008201765,0.0029288882,0.41435227,0.010885663,0.057629026,0.0052562053,0.47175404],"study_design_scores_gemma":[0.000051630795,0.00019482545,0.006696838,0.000036123798,0.0001379734,0.00013101622,0.0001332299,0.9606264,0.0008929824,0.029143935,0.0018950349,0.00005994911],"about_ca_topic_score_codex":0.010069333,"about_ca_topic_score_gemma":0.012165734,"teacher_disagreement_score":0.010069333,"about_ca_system_score_codex":0.0007498322,"about_ca_system_score_gemma":0.001645921,"threshold_uncertainty_score":0.047272325},"labels":[],"label_agreement":null},{"id":"W3205145300","doi":"10.1111/bmsp.12255","title":"A comparative evaluation of factor‐ and component‐based structural equation modelling approaches under (in)correct construct representations","year":2021,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":51,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Component (thermodynamics); Construct (python library); Structural equation modeling; Misrepresentation; Population; Principal component analysis; Factor analysis; Mathematics; Econometrics; Factor (programming language); Statistics; Computer science","score_opus":0.8016671738027995,"score_gpt":0.5424559714075778,"score_spread":0.25921120239522166,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3205145300","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.15326893,0.010365038,0.80033773,0.0046561006,0.00029821828,0.0010443941,0.0007942423,0.0012898501,0.02794541],"genre_scores_gemma":[0.4111773,0.0055323658,0.57905376,0.0004413408,0.00011408404,0.0012547858,0.00069326884,0.0003099028,0.0014231845],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.886271,0.092448495,0.003471092,0.0028287424,0.014430487,0.0005502852],"domain_scores_gemma":[0.8077559,0.15662861,0.00484243,0.0091172485,0.020830011,0.000825794],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.109157026,0.001981472,0.0013026393,0.007989854,0.0010416713,0.0050281086,0.002800477,0.0014496697,0.004228466],"category_scores_gemma":[0.21122219,0.0009773285,0.0026934762,0.010970255,0.0029603029,0.008641358,0.0031268247,0.0022811312,0.0007299738],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007221734,0.00036539134,0.029547816,0.002902623,0.0019093946,0.00010638405,0.008828939,0.028651802,0.0011354955,0.23285122,0.004403048,0.6885757],"study_design_scores_gemma":[0.00056619296,0.0023122567,0.07940053,0.004366674,0.0026761314,0.00047770623,0.014636941,0.47530168,0.0057381,0.37291062,0.040902447,0.0007107017],"about_ca_topic_score_codex":0.0041600745,"about_ca_topic_score_gemma":0.0077704308,"teacher_disagreement_score":0.890843,"about_ca_system_score_codex":0.004960216,"about_ca_system_score_gemma":0.0046868916,"threshold_uncertainty_score":0.57728446},"labels":[],"label_agreement":null},{"id":"W4214686438","doi":"10.1111/bmsp.12269","title":"Reliability coefficients for multiple group item response theory models","year":2022,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Reliability (semiconductor); Item response theory; Group (periodic table); Mathematics; Statistics; Econometrics; Reliability engineering; Psychology; Psychometrics; Engineering; Physics","score_opus":0.3057821453871747,"score_gpt":0.4634622157514617,"score_spread":0.157680070364287,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4214686438","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.009707831,0.0007768097,0.98387885,0.00047842867,0.0001469976,0.00047989187,0.0008146158,0.0005211448,0.003195492],"genre_scores_gemma":[0.246539,0.0015705245,0.73848885,0.0004507006,0.0003392499,0.0060811867,0.0026632845,0.00090929226,0.002957852],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.92637783,0.054408286,0.003183649,0.0052495087,0.009955405,0.000825277],"domain_scores_gemma":[0.7610295,0.18901195,0.010814685,0.02080178,0.01779338,0.0005487977],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07864314,0.0019936773,0.002107329,0.0064278007,0.0008077191,0.0032566646,0.003379025,0.0027299465,0.008009175],"category_scores_gemma":[0.35552385,0.0014060335,0.0038303942,0.0077444008,0.002575793,0.00524166,0.0025681856,0.005468875,0.004096461],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002800782,0.00024867425,0.042286836,0.001987909,0.0013929831,0.00036678594,0.0048682643,0.12339369,0.0014189965,0.4627885,0.029583102,0.33138406],"study_design_scores_gemma":[0.00015053342,0.00047050067,0.030145997,0.0019687691,0.00066497433,0.00091532216,0.0012525684,0.33255094,0.0013709186,0.5907818,0.03935513,0.00037257455],"about_ca_topic_score_codex":0.002282486,"about_ca_topic_score_gemma":0.0019805487,"teacher_disagreement_score":0.07864314,"about_ca_system_score_codex":0.0020065429,"about_ca_system_score_gemma":0.0017154483,"threshold_uncertainty_score":0.41590965},"labels":[],"label_agreement":null},{"id":"W4281689061","doi":"10.1111/bmsp.12274","title":"Theoretical considerations when simulating data from the <i>g</i> ‐and‐ <i>h</i> family of distributions","year":2022,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Statistical Distribution Estimation and Applications","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Kurtosis; Skewness; Copula (linguistics); Multivariate statistics; Marginal distribution; Mathematics; Generalization; Multivariate normal distribution; Population; Gaussian; Statistics; Statistical physics; Applied mathematics; Econometrics; Random variable; Mathematical analysis; Physics","score_opus":0.11198058467526725,"score_gpt":0.39204246552775257,"score_spread":0.2800618808524853,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4281689061","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.04504068,0.00035097174,0.94424856,0.0028946637,0.000078620826,0.0001710454,0.00014171927,0.00019779908,0.0068758684],"genre_scores_gemma":[0.6141163,0.0007388324,0.38124308,0.0009994785,0.00014223233,0.0009285391,0.00026044264,0.00017196462,0.0013991063],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99328023,0.005331169,0.00019411345,0.00029023038,0.00076232053,0.00014197027],"domain_scores_gemma":[0.8857901,0.10764535,0.0016486191,0.0029989888,0.0015407423,0.00037613144],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.017177602,0.00058092334,0.0009139325,0.0012022268,0.000994503,0.0022453098,0.0025249226,0.0017196841,0.003868235],"category_scores_gemma":[0.11065628,0.000647722,0.0011111405,0.0014009378,0.0028246993,0.0046149706,0.0019998709,0.0032064326,0.00035954363],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00010628912,0.00008134848,0.0045891795,0.00013658247,0.000051284125,0.00028474667,0.00039297718,0.58538055,0.000701127,0.3881632,0.0011599429,0.018952793],"study_design_scores_gemma":[0.00002999848,0.00006647094,0.00065955543,0.00004742696,0.000013779779,0.00016217814,0.0001501089,0.78924316,0.00054618483,0.20778681,0.0012710384,0.000023270959],"about_ca_topic_score_codex":0.0036780285,"about_ca_topic_score_gemma":0.0030137042,"teacher_disagreement_score":0.017177602,"about_ca_system_score_codex":0.0015569779,"about_ca_system_score_gemma":0.0012336164,"threshold_uncertainty_score":0.09084493},"labels":[],"label_agreement":null},{"id":"W4283316432","doi":"10.1111/bmsp.12272","title":"An explanatory mixture <scp>IRT</scp> model for careless and insufficient effort responding in self‐report measures","year":2022,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Survey Methodology and Nonresponse","field":"Social Sciences","cited_by":51,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Respondent; Set (abstract data type); Scale (ratio); Computer science; Trait; Statistics; Econometrics; Mathematics; Data mining","score_opus":0.1287896801544608,"score_gpt":0.44059646563512533,"score_spread":0.31180678548066454,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4283316432","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.08738404,0.0010208325,0.88723207,0.0068178214,0.00039927737,0.0015278799,0.0057856417,0.0020918865,0.0077406038],"genre_scores_gemma":[0.7297482,0.0011522719,0.20960245,0.0024568345,0.0005337813,0.006093248,0.0076317526,0.0005656898,0.042215824],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.98603976,0.0088847345,0.0005041221,0.0022733912,0.0012917679,0.0010061373],"domain_scores_gemma":[0.9470778,0.04033155,0.005066937,0.0031821362,0.0034270012,0.00091454666],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02580677,0.002893538,0.0032543144,0.0051995674,0.0016103075,0.004440002,0.009530683,0.0062038964,0.025908846],"category_scores_gemma":[0.050857294,0.002292363,0.006950276,0.0047954037,0.0033249443,0.0048585744,0.0037864698,0.0068508503,0.005713116],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013634837,0.000998823,0.09347045,0.0010348846,0.0023385473,0.0022191857,0.0034076162,0.3673347,0.0013379452,0.39336386,0.024994884,0.108135544],"study_design_scores_gemma":[0.00020382422,0.00029964393,0.008811651,0.00027758215,0.0004947215,0.00048332117,0.0003874901,0.89042366,0.00025195014,0.09261976,0.0055896863,0.00015678532],"about_ca_topic_score_codex":0.023008788,"about_ca_topic_score_gemma":0.015941758,"teacher_disagreement_score":0.9741932,"about_ca_system_score_codex":0.0028499314,"about_ca_system_score_gemma":0.0029821869,"threshold_uncertainty_score":0.13648087},"labels":[],"label_agreement":null},{"id":"W4295077327","doi":"10.1111/bmsp.12285","title":"Ordinal state‐trait regression for intensive longitudinal data","year":2022,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Mental Health Research Topics","field":"Psychology","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Carleton University","funders":"Israel Science Foundation","keywords":"Ordinal regression; Trait; Statistics; Longitudinal data; Ordinal data; Econometrics; Regression; Regression analysis; Mathematics; Cross-sectional regression; State (computer science); Computer science; Data mining; Algorithm; Polynomial regression","score_opus":0.2205900365319809,"score_gpt":0.5017560123301307,"score_spread":0.2811659757981498,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4295077327","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0039107567,0.00016417023,0.99404645,0.00020019063,0.000046649184,0.000113325696,0.0003177943,0.000915681,0.00028502333],"genre_scores_gemma":[0.157946,0.00053858716,0.8323164,0.0003315264,0.00016384787,0.0020629573,0.0021768059,0.00070909463,0.0037547844],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9769198,0.017833225,0.0007542026,0.0026675488,0.0014001516,0.00042508927],"domain_scores_gemma":[0.9084165,0.06999598,0.0054896045,0.012582612,0.0027232016,0.0007920662],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04720091,0.0016808385,0.0023058327,0.003026433,0.001422628,0.0024398933,0.0042107985,0.0022879688,0.00979744],"category_scores_gemma":[0.13361776,0.001349246,0.0024160026,0.005093017,0.0021242595,0.004372658,0.0041685635,0.006190737,0.0032500373],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00057886256,0.00045700136,0.051906567,0.0009662271,0.0010331837,0.0006986158,0.0016835589,0.28174853,0.002245091,0.36839533,0.012438903,0.27784815],"study_design_scores_gemma":[0.00004097176,0.00011798521,0.004375617,0.00008398077,0.00004346514,0.00012822132,0.00013166007,0.8566689,0.00040025378,0.13081475,0.0071258503,0.00006841921],"about_ca_topic_score_codex":0.0097031,"about_ca_topic_score_gemma":0.008787055,"teacher_disagreement_score":0.04720091,"about_ca_system_score_codex":0.0018014516,"about_ca_system_score_gemma":0.002523424,"threshold_uncertainty_score":0.2496252},"labels":[],"label_agreement":null},{"id":"W4384277172","doi":"10.1111/bmsp.12317","title":"Evaluating the performance of existing and novel equivalence tests for fit indices in structural equation modelling","year":2023,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"York University","funders":"","keywords":"Equivalence (formal languages); Mathematics; Monte Carlo method; Statistics; Sample size determination; Structural equation modeling; Statistical hypothesis testing; Goodness of fit; Applied mathematics; Discrete mathematics","score_opus":0.8160285585535664,"score_gpt":0.5966260124549783,"score_spread":0.2194025460985881,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4384277172","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.15879285,0.0012194674,0.8306245,0.00093873363,0.0002434626,0.00061011844,0.00027770965,0.0005496011,0.006743613],"genre_scores_gemma":[0.71097875,0.0003382094,0.28656238,0.00020900609,0.00008883715,0.0009916326,0.00038407912,0.000185514,0.00026167554],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.8085109,0.16025804,0.0071400586,0.006721981,0.016331896,0.0010370426],"domain_scores_gemma":[0.34721023,0.60596305,0.014618763,0.01955386,0.011498757,0.0011552852],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.17712252,0.0017162814,0.0019812456,0.005786358,0.0013700579,0.0033933823,0.0031247507,0.0025120429,0.0035661417],"category_scores_gemma":[0.58374566,0.0006648268,0.0029537475,0.005667946,0.005206544,0.008819192,0.0047906637,0.004388895,0.0005510702],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002035596,0.0014622505,0.12732978,0.0017658961,0.0028494617,0.000369737,0.004302936,0.11503535,0.0026600484,0.20013437,0.0036297145,0.53842497],"study_design_scores_gemma":[0.0005498475,0.004225008,0.049998883,0.00079016236,0.0006097548,0.00050281273,0.0016289292,0.6473566,0.00446133,0.285262,0.004315276,0.0002993046],"about_ca_topic_score_codex":0.0014881515,"about_ca_topic_score_gemma":0.0016238503,"teacher_disagreement_score":0.82287747,"about_ca_system_score_codex":0.0019195682,"about_ca_system_score_gemma":0.0027575556,"threshold_uncertainty_score":0.9367246},"labels":[],"label_agreement":null},{"id":"W4406715877","doi":"10.1111/bmsp.12381","title":"Efficient and accurate variational inference for multilevel threshold autoregressive models in intensive longitudinal data","year":2025,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Manitoba","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research","keywords":"Markov chain Monte Carlo; Inference; Autoregressive model; Bayes' theorem; Bayesian inference; Bayesian probability; Computer science; Algorithm; Data set; Mathematics; Statistics; Artificial intelligence","score_opus":0.17965977955819226,"score_gpt":0.4642400169581189,"score_spread":0.28458023739992666,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4406715877","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.006703662,0.00022470136,0.99245024,0.00018901912,0.000014224808,0.00002213066,0.000052507337,0.00007452343,0.0002689813],"genre_scores_gemma":[0.24106276,0.00064247096,0.75524926,0.0002333341,0.000098548335,0.00030621956,0.00053840346,0.00019327187,0.0016757132],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9976113,0.0015899335,0.000116115814,0.00030710717,0.00026231914,0.00011325435],"domain_scores_gemma":[0.97810304,0.019887097,0.00062718475,0.0005303427,0.0006051914,0.00024717968],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008957706,0.0007046088,0.0013938594,0.0011675111,0.0006702879,0.0013058181,0.0022876705,0.0015363195,0.0019304382],"category_scores_gemma":[0.03568288,0.001105256,0.0014661449,0.0011071981,0.001218282,0.001985851,0.0019955568,0.0030646315,0.0002600198],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000068999005,0.000048299797,0.0027825658,0.0001657435,0.000176823,0.00009523079,0.00024389825,0.7971346,0.001158099,0.14459461,0.0013761872,0.052155036],"study_design_scores_gemma":[0.00000688508,0.000005621384,0.00015276289,0.000011537543,0.000007303028,0.0000090808635,0.000008387027,0.964795,0.00009627598,0.03460341,0.00029798454,0.0000057923417],"about_ca_topic_score_codex":0.016542755,"about_ca_topic_score_gemma":0.018884316,"teacher_disagreement_score":0.016542755,"about_ca_system_score_codex":0.0015636099,"about_ca_system_score_gemma":0.0024722456,"threshold_uncertainty_score":0.047373414},"labels":[],"label_agreement":null},{"id":"W7117121226","doi":"10.1111/bmsp.70025","title":"Power priors for latent variable mediation models under small sample sizes","year":2025,"lang":"en","type":"article","venue":"British Journal of Mathematical and Statistical Psychology","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Prior probability; Univariate; Bayesian probability; Sample size determination; Latent variable; Variable (mathematics); Bayes' theorem","score_opus":0.07205106244704894,"score_gpt":0.3891975518028085,"score_spread":0.31714648935575956,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7117121226","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.004757041,0.00048272437,0.9912238,0.00051472307,0.000057284517,0.00048170114,0.00023593535,0.00025253624,0.001994257],"genre_scores_gemma":[0.17744346,0.00088103994,0.8148569,0.0005851384,0.0001307468,0.003723089,0.0005715543,0.000353499,0.0014546375],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9554059,0.036603566,0.0015541551,0.003142132,0.0027645568,0.0005297228],"domain_scores_gemma":[0.741943,0.23063216,0.0058241202,0.015052097,0.005891212,0.0006573723],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09475764,0.0019180167,0.0021134727,0.00216627,0.0012662569,0.0032410133,0.00410079,0.0030091815,0.012701164],"category_scores_gemma":[0.3831127,0.0013493198,0.0026791363,0.0026167661,0.0033851005,0.0064614257,0.0042971736,0.0076816115,0.0016844702],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0009634441,0.00020986574,0.0077144178,0.0017322053,0.000714795,0.00041405277,0.0019435665,0.09888117,0.0025101441,0.58870864,0.007579058,0.28862876],"study_design_scores_gemma":[0.00043378113,0.0004067856,0.0038833083,0.0007328911,0.00044365524,0.0003027196,0.00026363178,0.17422539,0.002889575,0.80162907,0.01468269,0.000106512394],"about_ca_topic_score_codex":0.0023406837,"about_ca_topic_score_gemma":0.0030991049,"teacher_disagreement_score":0.09475764,"about_ca_system_score_codex":0.0020626187,"about_ca_system_score_gemma":0.002548144,"threshold_uncertainty_score":0.50113225},"labels":[],"label_agreement":null}]}