{"meta":{"query_hash":"b6f3f6bdb720","filters":{"venue":"Psychological Methods"},"cohort_total":95,"direct_labels_cover":1,"predictions_cover":95,"exported":95,"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/b6f3f6bdb720","api":"https://metacan.xera.ac/api/v1/cohort?venue=Psychological+Methods"},"results":[{"id":"W1964275428","doi":"10.1037/1082-989x.7.2.210","title":"When constraints interact: A caution about reference variables, identification constraints, and scale dependencies in structural equation modeling.","year":2002,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Cognitive Abilities and Testing","field":"Psychology","cited_by":66,"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":"Structural equation modeling; Latent variable; Identification (biology); Variable (mathematics); Computer science; Econometrics; Scale (ratio); Latent variable model; Instrumental variable; Mathematical optimization; Software; Mathematics; Artificial intelligence; Machine learning","score_opus":0.21359218176728356,"score_gpt":0.4363520711109367,"score_spread":0.22275988934365315,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1964275428","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":[0.009646862,0.01293162,0.7955468,0.13383552,0.008525855,0.00058678316,0.0005652437,0.0020453807,0.036315918],"genre_scores_gemma":[0.1346819,0.00564274,0.74022585,0.09593133,0.0054826387,0.0026285623,0.00039921197,0.002461338,0.012546485],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8026214,0.1484913,0.009660525,0.010517824,0.026896032,0.0018129151],"domain_scores_gemma":[0.6612702,0.29168543,0.0115068,0.019888094,0.014271432,0.0013780134],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1453964,0.0034664744,0.003547031,0.0034517385,0.008341491,0.012694265,0.0081969295,0.012036284,0.0060685133],"category_scores_gemma":[0.4124532,0.0029557394,0.002605572,0.0071893143,0.021527724,0.022968903,0.011910481,0.045011986,0.0031652702],"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":[0.00016768966,0.00012616687,0.0038821998,0.0012814498,0.00043541126,0.0012036097,0.02947932,0.0018553246,0.0009700211,0.6360111,0.17925112,0.14533658],"study_design_scores_gemma":[0.000110926296,0.000110450004,0.0023960469,0.0026229187,0.00020737184,0.00088589627,0.006241935,0.0063654464,0.002453408,0.75334024,0.22501051,0.00025476006],"about_ca_topic_score_codex":0.00705278,"about_ca_topic_score_gemma":0.014028471,"teacher_disagreement_score":0.8546036,"about_ca_system_score_codex":0.0034630564,"about_ca_system_score_gemma":0.007108645,"threshold_uncertainty_score":0.76893884},"labels":[],"label_agreement":null},{"id":"W1966359789","doi":"10.1037/1082-989x.12.4.399","title":"Toward using confidence intervals to compare correlations.","year":2007,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":758,"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":"Natural Sciences and Engineering Research Council of Canada","keywords":"Confidence interval; Statistics; Robust confidence intervals; Mathematics; Sample size determination; Dependency (UML); Correlation; Nominal level; Sampling (signal processing); Confidence distribution; Sample (material); Econometrics; Computer science; Artificial intelligence","score_opus":0.915936087266207,"score_gpt":0.7443180110581955,"score_spread":0.1716180762080115,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1966359789","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":[0.0007366174,0.0021666407,0.99329114,0.0006576921,0.00024896167,0.00019196572,0.000117068295,0.000518439,0.0020713778],"genre_scores_gemma":[0.030456716,0.0014228014,0.96511644,0.00050695904,0.000372998,0.0012235115,0.00022623774,0.00027639748,0.00039795067],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.66789854,0.2856264,0.009899836,0.00941344,0.02624428,0.0009175248],"domain_scores_gemma":[0.32381424,0.6005755,0.020143168,0.03532253,0.01874742,0.0013971835],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.23903453,0.0034908678,0.0045507015,0.0157368,0.0014408234,0.011093044,0.009213609,0.0049004713,0.0054645925],"category_scores_gemma":[0.6294144,0.0019415591,0.00354942,0.012227186,0.00829594,0.011716662,0.008167849,0.015247483,0.0024460615],"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":[0.0004113586,0.00020370042,0.0036848478,0.0020901423,0.0014855242,0.0003616612,0.0023400995,0.030219244,0.00080860575,0.5715664,0.012351857,0.37447652],"study_design_scores_gemma":[0.00026703608,0.00043687044,0.0023400253,0.002241748,0.00038695458,0.0007398527,0.0007861437,0.08251865,0.0020525677,0.8614325,0.04651105,0.00028667238],"about_ca_topic_score_codex":0.0022706718,"about_ca_topic_score_gemma":0.0013746591,"teacher_disagreement_score":0.76096547,"about_ca_system_score_codex":0.0034119582,"about_ca_system_score_gemma":0.0047077537,"threshold_uncertainty_score":0.9384059},"labels":[],"label_agreement":null},{"id":"W1966967512","doi":"10.1037/1082-989x.13.2.110","title":"A generally robust approach for testing hypotheses and setting confidence intervals for effect sizes.","year":2008,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":125,"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 British Columbia; Manitoba Health; University of Manitoba","funders":"Social Sciences and Humanities Research Council of Canada; Canadian Institutes of Health Research; Manitoba Health Research Council","keywords":"Nonparametric statistics; Type I and type II errors; Heteroscedasticity; Statistics; Statistical hypothesis testing; Normality; Confidence interval; Robust statistics; Robustness (evolution); Mathematics; Statistic; Statistical power; Standard error; Variance (accounting); Nominal level; Econometrics; Sample size determination; Degrees of freedom (physics and chemistry); Multiple comparisons problem; Outlier","score_opus":0.5411151333206106,"score_gpt":0.547425561088569,"score_spread":0.006310427767958426,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1966967512","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":[0.00096950773,0.0017744025,0.9869538,0.0007531752,0.00042669202,0.0010963568,0.0011242003,0.0009110717,0.0059907767],"genre_scores_gemma":[0.039050277,0.0019368833,0.9451697,0.0013153396,0.0004178502,0.008404836,0.0010673894,0.00049652514,0.002141192],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8766762,0.08412058,0.0063508293,0.012055929,0.020055443,0.00074094866],"domain_scores_gemma":[0.73735076,0.20132014,0.01688362,0.028200496,0.015224614,0.0010203264],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.087261915,0.002687519,0.004246311,0.0074191005,0.0015714555,0.0051234346,0.005255323,0.0042262464,0.022524381],"category_scores_gemma":[0.3537436,0.0016102861,0.004529857,0.006656958,0.003980364,0.0041173976,0.004726139,0.008073249,0.0073288362],"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":[0.0006503986,0.00044942196,0.0035464852,0.0071724514,0.0041064336,0.00043473355,0.0016659376,0.012587545,0.0049774833,0.22596508,0.062157977,0.67628604],"study_design_scores_gemma":[0.0010189523,0.0025991474,0.011088288,0.0046156766,0.002313118,0.0022328035,0.0011471072,0.04582865,0.01270483,0.6963742,0.21945201,0.0006251917],"about_ca_topic_score_codex":0.0018468616,"about_ca_topic_score_gemma":0.0018154944,"teacher_disagreement_score":0.9127381,"about_ca_system_score_codex":0.0019442175,"about_ca_system_score_gemma":0.0047895084,"threshold_uncertainty_score":0.46149063},"labels":[],"label_agreement":null},{"id":"W1970325005","doi":"10.1037/a0029315","title":"When can categorical variables be treated as continuous? A comparison of robust continuous and categorical SEM estimation methods under suboptimal conditions.","year":2012,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":2534,"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 British Columbia","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Categorical variable; Statistics; Mathematics; Normality; Continuous variable; Sample size determination; Ordinal data; Sample (material); Latent variable; Confirmatory factor analysis; Econometrics; Structural equation modeling","score_opus":0.35917610160414,"score_gpt":0.5393551820295689,"score_spread":0.1801790804254289,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1970325005","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":[0.021141026,0.00081698404,0.9742245,0.0011790813,0.000119162,0.00022701826,0.00020054152,0.00040262134,0.0016889924],"genre_scores_gemma":[0.31972712,0.00036485106,0.6776032,0.00043273784,0.00006936018,0.00086116715,0.00031661798,0.00026544757,0.0003595275],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.7808656,0.19783688,0.0041539418,0.0059556486,0.010490451,0.0006974708],"domain_scores_gemma":[0.5048312,0.44588098,0.012231586,0.023324842,0.012968239,0.00076321606],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.16756894,0.0007757712,0.0015717668,0.0021138478,0.0008731722,0.0034273556,0.0021813004,0.002006901,0.0027879674],"category_scores_gemma":[0.5875014,0.0008818818,0.0018426786,0.003849228,0.0037483932,0.0044853473,0.0024771956,0.0024174834,0.0006994708],"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":[0.0022826034,0.00026824774,0.040434245,0.0026935723,0.002570498,0.0004127681,0.005495862,0.053921618,0.0036387406,0.24708107,0.010106775,0.631094],"study_design_scores_gemma":[0.0005078312,0.00095821125,0.02687798,0.0015137085,0.0005347121,0.00070125493,0.0027753033,0.44512802,0.0058605936,0.50112164,0.013703029,0.00031775204],"about_ca_topic_score_codex":0.0022630824,"about_ca_topic_score_gemma":0.002712318,"teacher_disagreement_score":0.8324311,"about_ca_system_score_codex":0.0015315914,"about_ca_system_score_gemma":0.0031041438,"threshold_uncertainty_score":0.88619983},"labels":[],"label_agreement":null},{"id":"W1973926374","doi":"10.1037/a0016955","title":"Killeen's (2005) prep coefficient: Logical and mathematical problems.","year":2010,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Advanced Statistical Modeling Techniques","field":"Computer Science","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University","funders":"","keywords":"Null hypothesis; Replication (statistics); Statistical hypothesis testing; Sign (mathematics); Statistical inference; Statistics; Statistical power; p-value; Noise (video); Psychology; Mathematics; Computer science; Artificial intelligence","score_opus":0.06215015672492949,"score_gpt":0.44347826398017864,"score_spread":0.3813281072552491,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1973926374","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":[0.0048348582,0.0072696162,0.92306143,0.03317251,0.00445494,0.00041079303,0.00036792114,0.00032750526,0.02610033],"genre_scores_gemma":[0.20163888,0.004987569,0.7400494,0.03361081,0.0048072045,0.0041081184,0.00016253199,0.0006411104,0.009994421],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.91078424,0.05194994,0.0065564415,0.010256764,0.01966092,0.00079168414],"domain_scores_gemma":[0.55425256,0.38859314,0.01183073,0.027520902,0.016679684,0.0011229186],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.11571991,0.001869801,0.0021445295,0.004031258,0.0033424988,0.0074213445,0.00573717,0.00646851,0.0054564816],"category_scores_gemma":[0.4236227,0.0020746102,0.0022041176,0.004388239,0.028942637,0.021249713,0.007375496,0.023894668,0.0021221603],"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":[0.00008969721,0.000024925648,0.0010098871,0.0003087884,0.00007634773,0.00013561478,0.0008874175,0.0004115071,0.00029310765,0.9517656,0.011796303,0.033200733],"study_design_scores_gemma":[0.00006625713,0.000068124114,0.00058996835,0.00028865834,0.00006556563,0.00052109704,0.0001258575,0.002954473,0.0012800966,0.9578608,0.03610432,0.00007467599],"about_ca_topic_score_codex":0.0015278894,"about_ca_topic_score_gemma":0.0011180041,"teacher_disagreement_score":0.8842801,"about_ca_system_score_codex":0.0041012513,"about_ca_system_score_gemma":0.004335096,"threshold_uncertainty_score":0.6119927},"labels":[],"label_agreement":null},{"id":"W1976410238","doi":"10.1037/a0029314","title":"Asking sensitive questions: A statistical power analysis of randomized response models.","year":2012,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Survey Sampling and Estimation Techniques","field":"Mathematics","cited_by":70,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"World Anti-Doping Agency","keywords":"Randomized response; Statistical power; Statistics; Statistical model; Sample size determination; Wald test; Goodness of fit; Statistical hypothesis testing; Power (physics); Item response theory; Sample (material); Econometrics; Computer science; Statistical analysis; Mathematics; Psychology; Psychometrics","score_opus":0.2830794992700697,"score_gpt":0.5610199716537944,"score_spread":0.27794047238372466,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1976410238","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":[0.0030584345,0.0004959358,0.99231887,0.00070587255,0.00011427205,0.00069902017,0.00011901466,0.0001534142,0.0023352064],"genre_scores_gemma":[0.21304703,0.0012287603,0.77255255,0.0014243827,0.0005053515,0.009033543,0.0003350401,0.00034897515,0.0015244156],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.5543028,0.41488937,0.004305012,0.00858814,0.016873611,0.0010410193],"domain_scores_gemma":[0.22586891,0.73150694,0.010357273,0.026342021,0.0052774604,0.00064740545],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2755057,0.0025067506,0.0036846288,0.006426738,0.0016927304,0.0047375606,0.0039927484,0.003828015,0.010561097],"category_scores_gemma":[0.69462687,0.001787536,0.0048305034,0.0060455417,0.009682613,0.0107063865,0.0070145186,0.0071949633,0.0015614876],"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":[0.00045152457,0.00023952953,0.005489983,0.0013944046,0.0012529211,0.00022233349,0.0031057138,0.019219263,0.0009280211,0.7469597,0.006049796,0.21468686],"study_design_scores_gemma":[0.0002500035,0.0009899767,0.0034793813,0.00069083006,0.00046772906,0.00049837603,0.00071118685,0.13979638,0.0019403164,0.8355834,0.015434237,0.0001581086],"about_ca_topic_score_codex":0.00081886083,"about_ca_topic_score_gemma":0.0004636892,"teacher_disagreement_score":0.2755057,"about_ca_system_score_codex":0.0027920152,"about_ca_system_score_gemma":0.0027638425,"threshold_uncertainty_score":0.8934305},"labels":[],"label_agreement":null},{"id":"W1980718559","doi":"10.1037/1082-989x.10.3.317","title":"An Alternative to Cohen's Standardized Mean Difference Effect Size: A Robust Parameter and Confidence Interval in the Two Independent Groups Case.","year":2005,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":292,"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; Percentile; Confidence interval; Mathematics; Robust confidence intervals; Coverage probability; Truncated mean; Confidence distribution; CDF-based nonparametric confidence interval; Population; Standard deviation; Tolerance interval; Range (aeronautics); Measure (data warehouse); Medicine","score_opus":0.4903035578928938,"score_gpt":0.6284214039490661,"score_spread":0.13811784605617228,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1980718559","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":[0.005138597,0.023231005,0.9401139,0.0074794027,0.005053491,0.0009789307,0.0010043145,0.0008151158,0.016185297],"genre_scores_gemma":[0.20211224,0.005037784,0.7731439,0.006877162,0.0025924426,0.0064084227,0.00056193816,0.0006126289,0.0026535753],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.78326374,0.15120818,0.014056884,0.017371133,0.032770436,0.0013296427],"domain_scores_gemma":[0.49029607,0.42555285,0.027591344,0.031165242,0.023641149,0.0017533367],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.15619795,0.0019024778,0.0037628764,0.011788593,0.0014876244,0.0061135376,0.007917622,0.0084135225,0.0063816896],"category_scores_gemma":[0.52033836,0.0010250376,0.00417984,0.010300811,0.007579938,0.010140146,0.00530811,0.009762796,0.0021082098],"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":[0.0010189961,0.00024791167,0.006050284,0.0056254705,0.0032226695,0.0003878167,0.0028571426,0.0029974626,0.0011935277,0.49899173,0.03654748,0.4408594],"study_design_scores_gemma":[0.00088564464,0.0013601081,0.010298495,0.0032947029,0.0017543258,0.002868542,0.0010774076,0.015051687,0.002992973,0.8485402,0.11139276,0.0004831542],"about_ca_topic_score_codex":0.0013594882,"about_ca_topic_score_gemma":0.0009684995,"teacher_disagreement_score":0.15619795,"about_ca_system_score_codex":0.0029686547,"about_ca_system_score_gemma":0.0035701578,"threshold_uncertainty_score":0.8260636},"labels":[],"label_agreement":null},{"id":"W1987144197","doi":"10.1037/1082-989x.10.2.139","title":"Structural Equation Models for Interchangeable Dyads: Being the Same Makes a Difference.","year":2005,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Attachment and Relationship Dynamics","field":"Psychology","cited_by":65,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Wilfrid Laurier University; University of Waterloo","funders":"","keywords":"Structural equation modeling; Psychology; Latent variable; Personality; Social psychology; Range (aeronautics); Latent variable model; Econometrics; Cognitive psychology; Developmental psychology; Statistics; Mathematics","score_opus":0.27351794498441073,"score_gpt":0.540779147405514,"score_spread":0.2672612024211033,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1987144197","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":[0.017844668,0.009540954,0.9245577,0.0138513055,0.0012726162,0.0006374761,0.0028923189,0.00072203716,0.028681038],"genre_scores_gemma":[0.34889916,0.0114060715,0.61719865,0.002831773,0.00056633353,0.0035856313,0.0054884464,0.00029859762,0.009725298],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9850766,0.011381717,0.00056961714,0.0012812176,0.0014084639,0.00028233422],"domain_scores_gemma":[0.97264975,0.02097397,0.0022277336,0.0020840578,0.0016712132,0.0003933221],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01293912,0.0015979634,0.0014422314,0.0025739437,0.0011393212,0.0040488946,0.0026288007,0.0024337254,0.011145474],"category_scores_gemma":[0.04796938,0.0010072793,0.001990563,0.006273718,0.0019737668,0.007541832,0.0032834997,0.0063330852,0.0019643847],"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":[0.000039330476,0.0001723168,0.011900905,0.0006871919,0.0007392174,0.00022509399,0.004227905,0.015705159,0.00026923014,0.8473045,0.022348238,0.09638092],"study_design_scores_gemma":[0.00004942061,0.000099119454,0.0048635257,0.00066221517,0.00030100573,0.00022971425,0.0014287024,0.06350366,0.000166569,0.85558856,0.073026575,0.00008093428],"about_ca_topic_score_codex":0.0073255207,"about_ca_topic_score_gemma":0.011536222,"teacher_disagreement_score":0.01293912,"about_ca_system_score_codex":0.0025634447,"about_ca_system_score_gemma":0.0034361097,"threshold_uncertainty_score":0.06842947},"labels":[],"label_agreement":null},{"id":"W1993260957","doi":"10.1037/a0024918","title":"Marginal mean weighting through stratification: A generalized method for evaluating multivalued and multiple treatments with nonexperimental data.","year":2011,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":106,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"National Academy of Education; Spencer Foundation; Social Sciences and Humanities Research Council of Canada; William T. Grant Foundation","keywords":"Causal inference; Covariate; Propensity score matching; Weighting; Average treatment effect; Inverse probability weighting; Statistics; Mathematics; Randomized experiment; Econometrics; Nonparametric statistics; Inference; Binary data; Computer science; Artificial intelligence; Binary number","score_opus":0.793516695385926,"score_gpt":0.6567447400249953,"score_spread":0.13677195536093067,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1993260957","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":[0.0008678468,0.00077454903,0.99500376,0.00029506916,0.00016641893,0.0012580244,0.0002252099,0.0003133252,0.0010957641],"genre_scores_gemma":[0.018492635,0.0008521231,0.9743406,0.00033669273,0.00009976841,0.0047049527,0.00029017392,0.00019301086,0.0006900327],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8507547,0.13168994,0.0037170965,0.0041177482,0.009042163,0.00067834626],"domain_scores_gemma":[0.83727866,0.13022807,0.0082661295,0.016095951,0.0070015574,0.0011296945],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1238361,0.0024136512,0.0038901397,0.0074761673,0.0017559455,0.0037977102,0.0036098734,0.0026289558,0.011373541],"category_scores_gemma":[0.2568608,0.0015689647,0.0060199485,0.0073901345,0.00573038,0.0046860417,0.007082033,0.005841863,0.0014945406],"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":[0.00045908327,0.00019125358,0.00535722,0.0025550576,0.0029950824,0.00023847312,0.0021952705,0.023756951,0.0012111122,0.3658085,0.014599153,0.58063275],"study_design_scores_gemma":[0.00040968476,0.00059201673,0.0040582046,0.001697769,0.0012413079,0.00039923,0.00048581793,0.10758884,0.0019328446,0.8190166,0.06230559,0.00027205813],"about_ca_topic_score_codex":0.004315569,"about_ca_topic_score_gemma":0.005233668,"teacher_disagreement_score":0.1238361,"about_ca_system_score_codex":0.0039730337,"about_ca_system_score_gemma":0.009438946,"threshold_uncertainty_score":0.6549157},"labels":[],"label_agreement":null},{"id":"W1995508717","doi":"10.1037/1082-989x.6.3.258","title":"Testing parameters in structural equation modeling: Every \"one\" matters.","year":2001,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":152,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Structural equation modeling; Identification (biology); Standard error; Statistical model; Statistics; Set (abstract data type); Statistical hypothesis testing; Mathematics; Invariant (physics); Econometrics; Likelihood-ratio test; Scale (ratio); Applied mathematics; Computer science; Physics","score_opus":0.8579973400321181,"score_gpt":0.6038590198508541,"score_spread":0.25413832018126403,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1995508717","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":[0.024144448,0.0077685723,0.92566425,0.027323777,0.0013783873,0.0005429653,0.00037294303,0.000847201,0.011957492],"genre_scores_gemma":[0.43060222,0.0061761294,0.5508698,0.0063937213,0.00082968146,0.002471504,0.00052193215,0.0004967376,0.0016383082],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.87463725,0.10102915,0.0045011854,0.004955043,0.01422124,0.0006562084],"domain_scores_gemma":[0.5480332,0.39937586,0.013121466,0.02703453,0.0108755855,0.0015593834],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1279731,0.001912369,0.0021528148,0.0044964645,0.00232094,0.006927496,0.0031461706,0.0048489035,0.0043021487],"category_scores_gemma":[0.5113057,0.0011903322,0.0020086332,0.007395243,0.009646317,0.019310135,0.005317865,0.010670255,0.0016022559],"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":[0.00018822384,0.00033183367,0.03304484,0.0015505546,0.0013095462,0.00021801253,0.00541467,0.00783158,0.0008451301,0.5795932,0.014656778,0.35501567],"study_design_scores_gemma":[0.000109881374,0.00035181124,0.007954169,0.001477082,0.0003611017,0.00032536927,0.0027404155,0.021029172,0.0013254925,0.94530743,0.018847724,0.0001703368],"about_ca_topic_score_codex":0.0018267756,"about_ca_topic_score_gemma":0.0019428966,"teacher_disagreement_score":0.1279731,"about_ca_system_score_codex":0.0021691334,"about_ca_system_score_gemma":0.0051596626,"threshold_uncertainty_score":0.6767945},"labels":[],"label_agreement":null},{"id":"W2004921315","doi":"10.1037/1082-989x.8.3.254","title":"Modern Robust Data Analysis Methods: Measures of Central Tendency.","year":2003,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":504,"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":"Contrast (vision); Computer science; Variance (accounting); Econometrics; Range (aeronautics); Statistical hypothesis testing; Statistical power; Statistics; Value (mathematics); Mathematics; Artificial intelligence; Machine learning; Engineering","score_opus":0.6806927522206976,"score_gpt":0.6148759361173601,"score_spread":0.06581681610333745,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2004921315","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":[0.0020749024,0.019943612,0.9636545,0.0019011054,0.0010301864,0.00044385248,0.0012953685,0.0013783532,0.00827814],"genre_scores_gemma":[0.07344596,0.02443843,0.88071287,0.002639154,0.0028128687,0.003899565,0.002306827,0.001222689,0.008521535],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9800702,0.009680931,0.0014797691,0.002470221,0.006004567,0.00029433199],"domain_scores_gemma":[0.9611399,0.024846727,0.0046647666,0.004771974,0.0041253828,0.00045124593],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.018426793,0.0022931483,0.0025841724,0.006812479,0.0010310899,0.0053963326,0.0035274606,0.0030328552,0.009218939],"category_scores_gemma":[0.0921786,0.0010449975,0.0024993003,0.008270741,0.0035798857,0.006488199,0.0041345297,0.00547948,0.006239091],"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":[0.0001992452,0.00020197428,0.0047759986,0.0042627812,0.0013445255,0.00022632562,0.0019359082,0.010210601,0.004906012,0.32499737,0.06654276,0.5803966],"study_design_scores_gemma":[0.0000923821,0.0003178954,0.011156057,0.0013931625,0.00052507705,0.0014453395,0.00057987135,0.03900081,0.0055539585,0.67886454,0.26071733,0.00035365784],"about_ca_topic_score_codex":0.0012063908,"about_ca_topic_score_gemma":0.0010534197,"teacher_disagreement_score":0.018426793,"about_ca_system_score_codex":0.0013532534,"about_ca_system_score_gemma":0.0024172843,"threshold_uncertainty_score":0.09745133},"labels":[],"label_agreement":null},{"id":"W2005615110","doi":"10.1037/1082-989x.11.1.112","title":"Paper and plastic in daily diary research: Comment on Green, Rafaeli, Bolger, Shrout, and Reis (2006).","year":2006,"lang":"en","type":"letter","venue":"Psychological Methods","topic":"Digital Mental Health Interventions","field":"Psychology","cited_by":89,"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 British Columbia","funders":"National Institute of Mental Health; National Institute on Alcohol Abuse and Alcoholism; Social Sciences and Humanities Research Council of Canada; Arthritis Foundation","keywords":"Psychology","score_opus":0.2598997714356185,"score_gpt":0.5457960284064142,"score_spread":0.2858962569707957,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2005615110","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0002807696,0.0013949471,0.00014024648,0.9781051,0.019299434,0.000022704313,0.00008303412,0.00004106698,0.00063278765],"genre_scores_gemma":[0.0017036942,0.00074236037,0.00034429226,0.9857999,0.009091691,0.000084625,0.000021821514,0.000023534072,0.0021880926],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9842935,0.0045257313,0.0024908523,0.002047921,0.0058386624,0.00080330664],"domain_scores_gemma":[0.94939053,0.0296814,0.003246726,0.0017514722,0.012908809,0.003021131],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01860032,0.001283382,0.0017141347,0.0011811086,0.00564212,0.004996869,0.005578095,0.045670684,0.005433103],"category_scores_gemma":[0.0806748,0.001233809,0.0014978788,0.0022305546,0.0053848177,0.007959227,0.0026846589,0.04641806,0.00613895],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00004785653,0.000020450803,0.00034400163,0.00008169524,0.000007174671,0.00020764528,0.00031588267,0.000029314937,0.00005613616,0.0011858355,0.9921753,0.0055287285],"study_design_scores_gemma":[0.0001239546,0.00011455969,0.0031900578,0.0008753321,0.000039742845,0.0012437835,0.0028559,0.000386427,0.00042945202,0.006077542,0.98452723,0.00013595494],"about_ca_topic_score_codex":0.022151113,"about_ca_topic_score_gemma":0.03288069,"teacher_disagreement_score":0.98139966,"about_ca_system_score_codex":0.006911885,"about_ca_system_score_gemma":0.006931059,"threshold_uncertainty_score":0.09836906},"labels":[],"label_agreement":null},{"id":"W2011619587","doi":"10.1037/a0029250","title":"Qualitative research as methodical hermeneutics.","year":2012,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Language, Discourse, Communication Strategies","field":"Arts and Humanities","cited_by":200,"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":"Hermeneutics; Epistemology; Qualitative research; Exegesis; Demonstrative; Credibility; Conversation; Meaning (existential); Mainstream; Thematic analysis; Generality; Rhetoric; Sociology; Psychology; Verstehen; Interpretation (philosophy); Social psychology; Linguistics; Philosophy; Social science","score_opus":0.7749174545978643,"score_gpt":0.7295370932943747,"score_spread":0.04538036130348966,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2011619587","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":[0.004369189,0.011256832,0.8762409,0.028754853,0.0042412747,0.015246834,0.0012243083,0.0007781901,0.057887655],"genre_scores_gemma":[0.06886694,0.0064360634,0.8641749,0.0065785283,0.0007694332,0.04126413,0.0004950385,0.000402108,0.011012913],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.63965905,0.33202884,0.0076778806,0.0055947523,0.0141810505,0.00085843727],"domain_scores_gemma":[0.6503644,0.3064026,0.007809706,0.020617997,0.013037452,0.0017678499],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.25041068,0.0012090927,0.0018066034,0.0059511904,0.003956228,0.01243976,0.0028131614,0.0022622407,0.009409114],"category_scores_gemma":[0.19540659,0.0010712546,0.0011871562,0.007824576,0.028026165,0.008046771,0.008228288,0.0072671175,0.0024816166],"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":[0.00008265394,0.00009069029,0.00042226037,0.0042424346,0.00006569504,0.00013350559,0.07773993,0.0009064386,0.000898209,0.754424,0.02472658,0.13626765],"study_design_scores_gemma":[0.00015574755,0.00016035471,0.0006882516,0.0070853406,0.000043727305,0.00034589545,0.020836044,0.0027645456,0.0011278646,0.5201351,0.44652736,0.00012968965],"about_ca_topic_score_codex":0.001704251,"about_ca_topic_score_gemma":0.0026859203,"teacher_disagreement_score":0.7495893,"about_ca_system_score_codex":0.0072812634,"about_ca_system_score_gemma":0.014635963,"threshold_uncertainty_score":0.9243771},"labels":[],"label_agreement":null},{"id":"W2011843009","doi":"10.1037/1082-989x.6.1.3","title":"Empirical and hermeneutic approaches to phenomenological research in psychology: A comparison.","year":2001,"lang":"en","type":"review","venue":"Psychological Methods","topic":"Counseling, Therapy, and Family Dynamics","field":"Psychology","cited_by":230,"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":"Intuition; Phenomenology (philosophy); Psychology; Epistemology; Similarity (geometry); Social psychology; Meaning (existential); Empirical research; Cognitive psychology; Psychotherapist; Cognitive science; Philosophy; Artificial intelligence","score_opus":0.870865608223874,"score_gpt":0.6743232721351252,"score_spread":0.19654233608874883,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2011843009","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0018353205,0.9205236,0.036531616,0.010769694,0.0021758978,0.00025160442,0.000033223772,0.00007692237,0.027802067],"genre_scores_gemma":[0.038092032,0.9065055,0.042437576,0.005108667,0.0017200989,0.0017942004,0.000081074075,0.00008721419,0.004173583],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.97440636,0.019922279,0.0011851897,0.00066983496,0.0034714576,0.00034487483],"domain_scores_gemma":[0.95688784,0.035664264,0.0013754768,0.0028680936,0.0023851653,0.0008191027],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02579959,0.0013570072,0.0019527011,0.013216639,0.002152735,0.010703342,0.0023185804,0.0037884368,0.003078412],"category_scores_gemma":[0.034318686,0.0010898692,0.00081369415,0.017767953,0.020384688,0.008896349,0.005601947,0.0052353316,0.0016287899],"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":[0.00007488231,0.00026978555,0.0014987492,0.014942669,0.000104014565,0.00044054998,0.026070254,0.0006717149,0.0006608306,0.40331948,0.0143912835,0.53755575],"study_design_scores_gemma":[0.000064702705,0.00018435073,0.004881949,0.020249132,0.000064995664,0.004030676,0.028991466,0.0006720261,0.00042762022,0.20908417,0.731245,0.00010391182],"about_ca_topic_score_codex":0.00294601,"about_ca_topic_score_gemma":0.0051350514,"teacher_disagreement_score":0.9742004,"about_ca_system_score_codex":0.006041964,"about_ca_system_score_gemma":0.008179848,"threshold_uncertainty_score":0.1364429},"labels":[],"label_agreement":null},{"id":"W2029286738","doi":"10.1037/1082-989x.13.2.150","title":"Constrained versus unconstrained estimation in structural equation modeling.","year":2008,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Multi-Criteria Decision Making","field":"Decision Sciences","cited_by":105,"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":"National Institute on Drug Abuse; U.S. Social Security Administration","keywords":"Constraint (computer-aided design); Structural equation modeling; Estimation; Computer science; Boundary (topology); Estimation theory; Mathematics; Mathematical optimization; Applied mathematics; Algorithm; Econometrics; Statistics","score_opus":0.6554148748889531,"score_gpt":0.6096974645123965,"score_spread":0.04571741037655663,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2029286738","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":[0.004291513,0.0069326838,0.9824751,0.0018030852,0.00026811144,0.0002999314,0.00022038768,0.00017823538,0.0035308807],"genre_scores_gemma":[0.13598987,0.005653211,0.8533287,0.0009443987,0.00031688495,0.001709162,0.00053636933,0.00017816843,0.0013431292],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9043054,0.08509116,0.002013033,0.0034576664,0.0046769464,0.0004558936],"domain_scores_gemma":[0.86929816,0.1195292,0.003838587,0.004852864,0.0020131364,0.00046806774],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.054231856,0.0023690534,0.002443939,0.0042402963,0.0012878297,0.0036951785,0.0031996304,0.002528737,0.00593804],"category_scores_gemma":[0.16843502,0.0014974352,0.0026351595,0.006762929,0.0044680554,0.0065282397,0.00480052,0.004917375,0.0010473287],"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":[0.0001563198,0.00016332614,0.004858998,0.0013669914,0.0012064036,0.0001408784,0.0014590232,0.028667213,0.00030118818,0.7037962,0.00588019,0.25200325],"study_design_scores_gemma":[0.00007580518,0.0001272958,0.0019172496,0.00077052094,0.0002225456,0.00012415125,0.00033685748,0.13401538,0.0004004404,0.8466357,0.015284545,0.00008947354],"about_ca_topic_score_codex":0.003333553,"about_ca_topic_score_gemma":0.0058303922,"teacher_disagreement_score":0.9457681,"about_ca_system_score_codex":0.0020282161,"about_ca_system_score_gemma":0.0029528965,"threshold_uncertainty_score":0.28680885},"labels":[],"label_agreement":null},{"id":"W2030382769","doi":"10.1037/1082-989x.9.4.453","title":"Comparative Judgments as an Alternative to Ratings: Identifying the Scale Origin.","year":2004,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Multi-Criteria Decision Making","field":"Decision Sciences","cited_by":68,"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":"Psychology; Scale (ratio); Interpretation (philosophy); Value (mathematics); Econometrics; Social psychology; Cognitive psychology; Statistics; Computer science; Mathematics","score_opus":0.6618471505924736,"score_gpt":0.6779126136633634,"score_spread":0.01606546307088985,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2030382769","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":[0.03175964,0.004785122,0.91418993,0.002316348,0.0015143732,0.00067334465,0.00021185022,0.00038818785,0.04416116],"genre_scores_gemma":[0.4134102,0.0013375459,0.58016914,0.0007491649,0.0005400408,0.001087048,0.00014173824,0.00021382658,0.0023512784],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.83182025,0.13256961,0.004181996,0.005171276,0.025509203,0.00074760785],"domain_scores_gemma":[0.77109385,0.17121242,0.014033383,0.02206479,0.019391738,0.0022037914],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07342554,0.0013412379,0.0017258173,0.0054909126,0.002190784,0.008165446,0.002216762,0.00238187,0.0063551785],"category_scores_gemma":[0.31866306,0.0006542254,0.0012141151,0.007908806,0.009073892,0.011199323,0.005235143,0.004922394,0.0011143151],"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":[0.00051079196,0.00025093384,0.004776176,0.0024856897,0.00047656495,0.00031937408,0.016446803,0.0023260978,0.005466904,0.65417147,0.0061793835,0.30658993],"study_design_scores_gemma":[0.00023594812,0.000875074,0.012726869,0.0011678892,0.00020699714,0.0011423168,0.007529882,0.033626456,0.0047554793,0.89064294,0.046684097,0.00040606374],"about_ca_topic_score_codex":0.0011835994,"about_ca_topic_score_gemma":0.0021674258,"teacher_disagreement_score":0.07342554,"about_ca_system_score_codex":0.0019391665,"about_ca_system_score_gemma":0.0022960377,"threshold_uncertainty_score":0.38831604},"labels":[],"label_agreement":null},{"id":"W2040991114","doi":"10.1037/a0017737","title":"Testing multiple outcomes in repeated measures designs.","year":2010,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":27,"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 Saskatchewan","funders":"Canadian Institutes of Health Research; Manitoba Health Research Council","keywords":"Bonferroni correction; Multiple comparisons problem; Type I and type II errors; Statistics; Statistical power; Permutation (music); Monte Carlo method; Mathematics; Outcome (game theory); Variable (mathematics); Statistical hypothesis testing; Repeated measures design; Computer science","score_opus":0.8871628882891909,"score_gpt":0.695675757748921,"score_spread":0.1914871305402699,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2040991114","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":[0.027853664,0.0011739292,0.9583988,0.0004954463,0.0013484026,0.0046408246,0.00053952535,0.0009549436,0.0045944494],"genre_scores_gemma":[0.1896795,0.0008105371,0.7773188,0.0006698832,0.00059559796,0.029128047,0.00044431287,0.00033760833,0.0010157633],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.6516233,0.2836942,0.011463544,0.019191343,0.032694757,0.0013328027],"domain_scores_gemma":[0.35784274,0.5188137,0.04746712,0.05660063,0.018257244,0.0010185806],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.23086075,0.0022569087,0.002888526,0.0030376345,0.0015790993,0.0029332794,0.0033788194,0.0022810716,0.0053031044],"category_scores_gemma":[0.56200475,0.001099508,0.003227716,0.0032569757,0.004259189,0.004277585,0.002939486,0.0035426347,0.0013937446],"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":[0.0074473238,0.0026373898,0.053003713,0.009779832,0.015048828,0.00069852493,0.0117400745,0.01290304,0.012825852,0.1398421,0.010017925,0.72405535],"study_design_scores_gemma":[0.006567068,0.06417902,0.12997645,0.0062003797,0.009510139,0.00302561,0.004098683,0.13157547,0.055842817,0.5005888,0.086935356,0.0015002838],"about_ca_topic_score_codex":0.0007096217,"about_ca_topic_score_gemma":0.00075090944,"teacher_disagreement_score":0.23086075,"about_ca_system_score_codex":0.0014601501,"about_ca_system_score_gemma":0.0036767914,"threshold_uncertainty_score":0.9484856},"labels":[],"label_agreement":null},{"id":"W2042003034","doi":"10.1037/a0025810","title":"Many tests of significance: New methods for controlling type I errors.","year":2011,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Statistical Methods in Clinical Trials","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":"Multiple comparisons problem; Control (management); Computer science; Mathematics; Statistics; False discovery rate; Artificial intelligence","score_opus":0.8514252015725781,"score_gpt":0.7047830235817717,"score_spread":0.14664217799080637,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2042003034","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":[0.0010835055,0.0058499523,0.9740201,0.0040029758,0.00557271,0.0022753682,0.0009950718,0.0023725396,0.0038277816],"genre_scores_gemma":[0.01775199,0.0033828267,0.9516289,0.004049818,0.002901432,0.01645045,0.00067206123,0.0013205714,0.0018419458],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.45020154,0.4357767,0.024977038,0.024729779,0.06278576,0.001529177],"domain_scores_gemma":[0.1925948,0.68076015,0.038787633,0.060027983,0.02510109,0.0027282948],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.31011614,0.0058404426,0.0077706585,0.016924383,0.004235075,0.013140488,0.013656861,0.00850977,0.020128073],"category_scores_gemma":[0.69505656,0.0037748935,0.0071535744,0.016990859,0.020197833,0.017366182,0.012226121,0.025206735,0.0070001488],"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":[0.0015145966,0.0005093571,0.009372774,0.010874511,0.008240254,0.0007663262,0.006556671,0.007956322,0.0021759653,0.23961091,0.1059957,0.6064266],"study_design_scores_gemma":[0.0011978565,0.0013773857,0.0050722607,0.007215946,0.0027514673,0.0017359172,0.0015553337,0.061379883,0.0054858658,0.7054163,0.20600219,0.00080958055],"about_ca_topic_score_codex":0.001870553,"about_ca_topic_score_gemma":0.002172298,"teacher_disagreement_score":0.6898838,"about_ca_system_score_codex":0.005759132,"about_ca_system_score_gemma":0.016101213,"threshold_uncertainty_score":0.8507496},"labels":[],"label_agreement":null},{"id":"W2042334535","doi":"10.1037/a0014269","title":"A proposed framework for conducting data-based test analysis.","year":2008,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":40,"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","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Test (biology); Extant taxon; Computer science; Test theory; Test data; Data mining; Psychometrics; Statistics; Mathematics; Software engineering","score_opus":0.9565826883810709,"score_gpt":0.7111345330937316,"score_spread":0.2454481552873392,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2042334535","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":[0.00076772645,0.00033756052,0.987468,0.0037649844,0.00009905321,0.001833496,0.0002646689,0.00037344274,0.0050910995],"genre_scores_gemma":[0.010833326,0.00012069208,0.98560333,0.0003725494,0.00003589253,0.0023569514,0.000257442,0.000027507844,0.0003922214],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8820273,0.08053266,0.0086318115,0.008064193,0.019275095,0.0014688513],"domain_scores_gemma":[0.8628222,0.0804963,0.0073694107,0.01804485,0.026973076,0.0042942083],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1170323,0.0023314918,0.001666188,0.01088661,0.003320565,0.015680222,0.008019691,0.0050080647,0.0052648713],"category_scores_gemma":[0.15561083,0.0013623826,0.002818655,0.008583164,0.011930543,0.008933496,0.008489103,0.005636294,0.0033818749],"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":[0.000053227897,0.00018880975,0.003330164,0.0007744307,0.00015422876,0.0002642296,0.0027908557,0.0055863764,0.00072967756,0.8302676,0.008872573,0.14698774],"study_design_scores_gemma":[0.00013303517,0.00019135207,0.0017308118,0.0012993984,0.00009435198,0.00055794354,0.0016226385,0.05086049,0.0010555716,0.8654759,0.07687627,0.00010222041],"about_ca_topic_score_codex":0.007055275,"about_ca_topic_score_gemma":0.005270126,"teacher_disagreement_score":0.1170323,"about_ca_system_score_codex":0.0063524544,"about_ca_system_score_gemma":0.02269909,"threshold_uncertainty_score":0.6189333},"labels":[],"label_agreement":null},{"id":"W2048699385","doi":"10.1037/1082-989x.5.3.370","title":"How important is transient error in estimating reliability? Going beyond simulation studies.","year":2000,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Optimal Experimental Design Methods","field":"Decision Sciences","cited_by":84,"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 Winnipeg; University of Manitoba","funders":"","keywords":"Reliability (semiconductor); Transient (computer programming); Observational error; Statistics; Standard error; Range (aeronautics); Error analysis; Reliability engineering; Mathematics; Computer science; Algorithm; Applied mathematics; Engineering","score_opus":0.3949998641647077,"score_gpt":0.6121051813343663,"score_spread":0.21710531716965858,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2048699385","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":[0.06582218,0.0033889867,0.9178565,0.004499236,0.00052051205,0.00025502016,0.00014108662,0.00038232034,0.0071342257],"genre_scores_gemma":[0.87778616,0.0011729848,0.11845635,0.0009140138,0.00020819153,0.0003782096,0.00013868613,0.00012556998,0.000819833],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.90675175,0.08201126,0.0019412042,0.0031888024,0.00547489,0.00063204794],"domain_scores_gemma":[0.4223566,0.5238027,0.016960884,0.025609937,0.010412793,0.0008571279],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.07941175,0.0008020794,0.0011617993,0.00088998454,0.0005814828,0.002696261,0.0014469802,0.001967724,0.0018555605],"category_scores_gemma":[0.51974,0.0004913379,0.001211138,0.001398986,0.0033421174,0.007668643,0.0019761268,0.0030578237,0.0003884001],"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":[0.0009537895,0.00040221403,0.0992327,0.0019485493,0.002853945,0.0005229456,0.004543021,0.2175223,0.001781677,0.41919625,0.008294142,0.24274853],"study_design_scores_gemma":[0.00016199869,0.0009097041,0.018498145,0.0010613471,0.0005019572,0.00050866394,0.0015269375,0.39085117,0.003900064,0.5695575,0.012332745,0.00018982911],"about_ca_topic_score_codex":0.0033102226,"about_ca_topic_score_gemma":0.0022545026,"teacher_disagreement_score":0.92058825,"about_ca_system_score_codex":0.0013305258,"about_ca_system_score_gemma":0.0020984577,"threshold_uncertainty_score":0.4199745},"labels":[],"label_agreement":null},{"id":"W2052567024","doi":"10.1037/1082-989x.12.3.247","title":"Combining propensity score matching and group-based trajectory analysis in an observational study.","year":2007,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":285,"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":"National Institute of Mental Health","keywords":"Propensity score matching; Covariate; Observational study; Matching (statistics); Mahalanobis distance; Statistics; Trajectory; Mathematics; Econometrics","score_opus":0.7505569186862165,"score_gpt":0.6029601374272204,"score_spread":0.1475967812589961,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2052567024","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":[0.013449521,0.0012892657,0.98045105,0.00073794374,0.0002665622,0.0017484695,0.0005220608,0.00037821545,0.0011568983],"genre_scores_gemma":[0.23188546,0.0014232802,0.76051503,0.0006034064,0.00029046254,0.0032461402,0.001129962,0.00013335797,0.0007728262],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8568908,0.13051654,0.0031465457,0.0043555005,0.0046053994,0.0004852359],"domain_scores_gemma":[0.8617817,0.101223,0.010020178,0.022330243,0.003583957,0.0010609328],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.11793492,0.0012689318,0.0025489489,0.0065897065,0.0008403763,0.0025513358,0.0020115871,0.0015055721,0.003739397],"category_scores_gemma":[0.21283753,0.0008797659,0.0030862745,0.011334265,0.0013111477,0.0022668464,0.0044352617,0.0023033668,0.0006858302],"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":[0.0024342632,0.0009505604,0.15381618,0.0024143986,0.018474802,0.0005647708,0.001177857,0.06416535,0.0014386411,0.09799723,0.012510141,0.64405584],"study_design_scores_gemma":[0.002754193,0.0030985458,0.09804085,0.0009167027,0.007038114,0.0010404966,0.0006788975,0.37238145,0.002636098,0.48273063,0.028314091,0.00037000113],"about_ca_topic_score_codex":0.0035297945,"about_ca_topic_score_gemma":0.0032670954,"teacher_disagreement_score":0.11793492,"about_ca_system_score_codex":0.0011312142,"about_ca_system_score_gemma":0.0034806146,"threshold_uncertainty_score":0.62370694},"labels":[],"label_agreement":null},{"id":"W2064765903","doi":"10.1037/1082-989x.8.4.468","title":"Bidimensional Regression: Assessing the Configural Similarity and Accuracy of Cognitive Maps and Other Two-Dimensional Data Sets.","year":2003,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Spatial Cognition and Navigation","field":"Engineering","cited_by":180,"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":"Similarity (geometry); Regression; Fidelity; Regression analysis; Euclidean distance; Distortion (music); Euclidean geometry; Artificial intelligence; Mathematics; Cognition; Cognitive map; Linear regression; Pattern recognition (psychology); Statistics; Psychology; Computer science; Image (mathematics)","score_opus":0.18311456404649293,"score_gpt":0.48913956596141756,"score_spread":0.30602500191492465,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2064765903","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":[0.039767727,0.0012808242,0.94830024,0.0006676953,0.0003363664,0.0003005815,0.001055769,0.0017011993,0.0065896707],"genre_scores_gemma":[0.45220485,0.0010418355,0.53969944,0.0002589028,0.00014320212,0.00084947806,0.0015109583,0.00067050924,0.0036208916],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9917143,0.004007835,0.0008025767,0.0012881218,0.0020483993,0.00013868134],"domain_scores_gemma":[0.94994414,0.03281772,0.0050844606,0.007191424,0.0044450476,0.0005171392],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.013585281,0.0014280207,0.00092794537,0.0063768597,0.00074280653,0.0040702247,0.0016434211,0.0010205421,0.003743654],"category_scores_gemma":[0.14233907,0.0005345331,0.0011680878,0.0074275853,0.0019727948,0.004913792,0.0027374432,0.001422424,0.0012710758],"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":[0.0008793093,0.00031079838,0.06732385,0.001098487,0.0009482128,0.0003070925,0.003387554,0.039573584,0.010527578,0.14495821,0.01737103,0.71331435],"study_design_scores_gemma":[0.00013220988,0.0006052297,0.106212504,0.00041461142,0.00035640274,0.0019614398,0.0018325974,0.51411426,0.018217847,0.31196323,0.043605592,0.0005840365],"about_ca_topic_score_codex":0.0034007358,"about_ca_topic_score_gemma":0.0022640734,"teacher_disagreement_score":0.013585281,"about_ca_system_score_codex":0.0009531584,"about_ca_system_score_gemma":0.0009329599,"threshold_uncertainty_score":0.07184672},"labels":[],"label_agreement":null},{"id":"W2079540101","doi":"10.1037/1082-989x.10.1.65","title":"A Latent Markov Model for the Analysis of Longitudinal Data Collected in Continuous Time: States, Durations, and Transitions.","year":2005,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Mental Health Research Topics","field":"Psychology","cited_by":47,"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":"Generalization; Markov chain; Markov model; Computer science; Markov process; Econometrics; Latent variable; Variable-order Markov model; Sampling (signal processing); Statistics; Observable; Duration (music); Artificial intelligence; Machine learning; Mathematics","score_opus":0.2726942149715341,"score_gpt":0.5589278597371689,"score_spread":0.28623364476563473,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2079540101","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":[0.0028055422,0.0021228332,0.9912985,0.00070387224,0.00015126842,0.00011756431,0.00096716185,0.00033851367,0.0014947103],"genre_scores_gemma":[0.30579293,0.00893684,0.6651235,0.0009651773,0.0009841251,0.0030442656,0.005544154,0.00029040012,0.009318601],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99522084,0.002906947,0.0002779748,0.0008421928,0.0005142951,0.00023777806],"domain_scores_gemma":[0.9809552,0.015539395,0.0014247401,0.0011456059,0.0006505633,0.00028445196],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008060882,0.0013381616,0.0014988019,0.002418935,0.00095389795,0.0027168337,0.0027012108,0.0020855027,0.00571236],"category_scores_gemma":[0.02713326,0.00091794034,0.0023497941,0.0033986825,0.0022011036,0.004962325,0.0020906716,0.004165817,0.001228673],"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":[0.00012945845,0.00013826908,0.0055685383,0.0005676259,0.00041676668,0.0003766092,0.00085568154,0.13315348,0.001150305,0.77534944,0.009117251,0.07317651],"study_design_scores_gemma":[0.00004077357,0.000079576734,0.0019521933,0.00017659065,0.00013180081,0.00020839156,0.00011661514,0.42213267,0.0002615362,0.5616958,0.013111619,0.000092466325],"about_ca_topic_score_codex":0.011789864,"about_ca_topic_score_gemma":0.011161898,"teacher_disagreement_score":0.011789864,"about_ca_system_score_codex":0.0025750839,"about_ca_system_score_gemma":0.0031461196,"threshold_uncertainty_score":0.042630553},"labels":[],"label_agreement":null},{"id":"W2082558248","doi":"10.1037/a0020143","title":"Expected versus observed information in SEM with incomplete normal and nonnormal data.","year":2010,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Multi-Criteria Decision Making","field":"Decision Sciences","cited_by":80,"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":"Statistics; Statistic; Standard error; Fisher information; Test statistic; Missing data; Computation; Mathematics; Maximum likelihood; Information Criteria; Structural equation modeling; Computer science; Econometrics; Statistical hypothesis testing; Algorithm; Model selection","score_opus":0.6072341590302414,"score_gpt":0.5821967672574048,"score_spread":0.025037391772836592,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2082558248","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":[0.053736594,0.0031512782,0.9187319,0.0026573979,0.0005696927,0.0011772271,0.0012235921,0.0006334215,0.018118856],"genre_scores_gemma":[0.4933541,0.0012018876,0.49893132,0.0011395632,0.00022399871,0.0027820852,0.0007916381,0.00027147625,0.0013038998],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.72483647,0.23297776,0.010095154,0.008837804,0.022185626,0.0010671258],"domain_scores_gemma":[0.28864318,0.6654857,0.015772697,0.021054937,0.00830013,0.00074335176],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1311796,0.001251234,0.0022236905,0.003352824,0.0011409259,0.0049839555,0.0019145119,0.0028182585,0.007649312],"category_scores_gemma":[0.5441467,0.0009851317,0.003160847,0.004787784,0.005653372,0.00912087,0.0035178845,0.0038693193,0.0007397527],"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":[0.0021279226,0.00060145353,0.0358272,0.005648759,0.0023602662,0.00069627224,0.0052954373,0.048743255,0.0014301301,0.5241476,0.0076210164,0.36550054],"study_design_scores_gemma":[0.0005136588,0.0020242208,0.032871313,0.0028517519,0.0010337909,0.0008915576,0.0019828156,0.14783175,0.0069691376,0.782364,0.02025053,0.00041552156],"about_ca_topic_score_codex":0.0009931509,"about_ca_topic_score_gemma":0.0017523917,"teacher_disagreement_score":0.8688204,"about_ca_system_score_codex":0.0024593228,"about_ca_system_score_gemma":0.002784086,"threshold_uncertainty_score":0.6937523},"labels":[],"label_agreement":null},{"id":"W2086012744","doi":"10.1037/a0029253","title":"Individual influence on model selection.","year":2012,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Statistical Methods and Bayesian Inference","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":"York University","funders":"","keywords":"Akaike information criterion; Bayesian information criterion; Model selection; Selection (genetic algorithm); Ranking (information retrieval); Generality; Information Criteria; Computer science; Bayesian probability; Deviance information criterion; Multilevel model; Econometrics; Bayesian inference; Statistics; Data mining; Machine learning; Artificial intelligence; Psychology; Mathematics","score_opus":0.3007814790311631,"score_gpt":0.5548223446057223,"score_spread":0.25404086557455924,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2086012744","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":[0.07725093,0.0025553042,0.8895614,0.0040089665,0.0005734379,0.00080039824,0.0002841616,0.0008346066,0.024130655],"genre_scores_gemma":[0.7324887,0.0008589068,0.26198888,0.0012693128,0.00038645495,0.0008661279,0.00029185732,0.00042994268,0.0014196939],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.7725892,0.18996902,0.005409789,0.010428481,0.020252325,0.0013513094],"domain_scores_gemma":[0.35322276,0.59258765,0.01076056,0.033499833,0.008301547,0.0016276005],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.14615585,0.0018190026,0.0021110545,0.0037677502,0.0018245608,0.004714647,0.0023436898,0.002227333,0.004834996],"category_scores_gemma":[0.546576,0.0010104453,0.0033254689,0.002495131,0.00533899,0.0050179656,0.0058477647,0.005408008,0.00068163057],"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":[0.00088531955,0.0003185941,0.10331663,0.0015905659,0.0051529496,0.0013091826,0.008750361,0.07748014,0.0026130106,0.45699328,0.010094955,0.3314951],"study_design_scores_gemma":[0.00012853857,0.00046521006,0.018131295,0.00056143495,0.000978865,0.00082270603,0.0010834045,0.39574823,0.004228465,0.56272453,0.014926584,0.00020079584],"about_ca_topic_score_codex":0.0031264827,"about_ca_topic_score_gemma":0.0039344635,"teacher_disagreement_score":0.14615585,"about_ca_system_score_codex":0.0026000352,"about_ca_system_score_gemma":0.0031054162,"threshold_uncertainty_score":0.77295524},"labels":[],"label_agreement":null},{"id":"W2093491439","doi":"10.1037/1082-989x.7.1.126","title":"The role of qualitative research in psychological journals.","year":2002,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":183,"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 Windsor","funders":"","keywords":"PsycINFO; Qualitative research; Qualitative analysis; Psychology; Content analysis; Applied psychology; Social psychology; MEDLINE; Social science; Sociology","score_opus":0.9375029604873701,"score_gpt":0.8324841901757505,"score_spread":0.10501877031161966,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2093491439","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":[0.2607274,0.06968929,0.44312492,0.1493303,0.005907123,0.018074468,0.0016525198,0.00069772714,0.050796293],"genre_scores_gemma":[0.8138507,0.008668195,0.14743055,0.013302271,0.0005914396,0.01440429,0.00020277953,0.0001684028,0.0013813172],"study_design_codex":"qualitative","study_design_gemma":"observational","domain_scores_codex":[0.110892646,0.8447934,0.0172627,0.0051192264,0.02033895,0.0015931325],"domain_scores_gemma":[0.017494343,0.9325496,0.017552383,0.014860805,0.01559195,0.001950845],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.70476586,0.0010357877,0.002725138,0.017615212,0.010555927,0.031475052,0.0050584716,0.004966942,0.004298286],"category_scores_gemma":[0.836419,0.0019971963,0.0011466629,0.016322296,0.035698425,0.022596562,0.0157482,0.006455934,0.0007036294],"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":[0.0006476457,0.0003801841,0.050082922,0.018767225,0.00042548764,0.0005846398,0.6659735,0.00090333243,0.0021364861,0.06440751,0.005146754,0.19054437],"study_design_scores_gemma":[0.00027375293,0.0013745506,0.022433762,0.04611215,0.000337705,0.0020838731,0.7188229,0.0062583513,0.0037708634,0.11286052,0.08526483,0.00040682178],"about_ca_topic_score_codex":0.003349968,"about_ca_topic_score_gemma":0.0040093013,"teacher_disagreement_score":0.29523414,"about_ca_system_score_codex":0.013575962,"about_ca_system_score_gemma":0.033170898,"threshold_uncertainty_score":0.36407632},"labels":[],"label_agreement":null},{"id":"W2094893424","doi":"10.1037/a0029316","title":"A multilevel simultaneous equations model for within-cluster dynamic effects, with an application to reciprocal parent–child and sibling effects.","year":2012,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Psychological and Temporal Perspectives Research","field":"Psychology","cited_by":22,"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 Toronto","funders":"Economic and Social Research Council; Medical Research Council; Social Sciences and Humanities Research Council; Wellcome Trust","keywords":"Reciprocal; Multilevel model; Psychology; Sibling; Developmental psychology; Set (abstract data type); Social psychology; Peer group; Cluster (spacecraft); Statistics; Computer science; Mathematics","score_opus":0.11546710169619348,"score_gpt":0.517022654787798,"score_spread":0.4015555530916045,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2094893424","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":[0.009899726,0.0008853216,0.9785372,0.0016699122,0.00026683745,0.0005287515,0.00420461,0.00078426243,0.0032233677],"genre_scores_gemma":[0.18780188,0.0016335899,0.7855337,0.00041521367,0.00035780756,0.004209591,0.0075481054,0.00035677818,0.012143294],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99287075,0.0050252764,0.00032560932,0.00089656346,0.0005575385,0.0003241757],"domain_scores_gemma":[0.99002445,0.0071505993,0.00092230743,0.00066671736,0.00094796694,0.00028789692],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.012382717,0.0018502886,0.0025188627,0.0022117537,0.0015396196,0.002764162,0.004830327,0.0018377423,0.019133592],"category_scores_gemma":[0.026375122,0.0017339175,0.0046395,0.005068448,0.0009058616,0.0029639278,0.0033107626,0.004624004,0.0022014931],"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":[0.00031877702,0.0003695592,0.04040321,0.00066511467,0.0039275778,0.0007897675,0.0025045353,0.24179679,0.000752287,0.5598977,0.028852163,0.1197225],"study_design_scores_gemma":[0.00024200825,0.00025138227,0.008702636,0.00024000875,0.00085379014,0.0002069493,0.00045580114,0.6651574,0.00021136296,0.28513932,0.038403496,0.00013582529],"about_ca_topic_score_codex":0.057511292,"about_ca_topic_score_gemma":0.07200588,"teacher_disagreement_score":0.057511292,"about_ca_system_score_codex":0.0032231475,"about_ca_system_score_gemma":0.0057393247,"threshold_uncertainty_score":0.11435312},"labels":[],"label_agreement":null},{"id":"W2097789390","doi":"10.1037/a0033266","title":"Why the items versus parcels controversy needn’t be one.","year":2013,"lang":"en","type":"review","venue":"Psychological Methods","topic":"Financial Reporting and Valuation Research","field":"Business, Management and Accounting","cited_by":1553,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Eunice Kennedy Shriver National Institute of Child Health and Human Development; Social Sciences and Humanities Research Council of Canada; University of Kansas; National Science Foundation","keywords":"Statistics; Econometrics; Psychology; Environmental science; Mathematics","score_opus":0.6121447167284271,"score_gpt":0.5685405495960681,"score_spread":0.04360416713235893,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2097789390","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0032637801,0.7434067,0.03122101,0.19449139,0.015449246,0.00015725105,0.00070963247,0.00024991203,0.0110511705],"genre_scores_gemma":[0.11585324,0.5606369,0.076519266,0.2114622,0.021940133,0.0011503228,0.0012758932,0.000929533,0.010232506],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9480487,0.03168993,0.0036017373,0.0044681905,0.011703254,0.0004881313],"domain_scores_gemma":[0.7973453,0.16251275,0.008553246,0.009868309,0.020263817,0.0014566352],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.091727175,0.001208495,0.003670046,0.004169492,0.0019486748,0.007804246,0.0049912087,0.005728063,0.0064523746],"category_scores_gemma":[0.17873055,0.0010344071,0.0017718814,0.012866738,0.017946936,0.020090861,0.0029667937,0.012169791,0.004381565],"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":[0.00021285926,0.00005729876,0.0019056068,0.012005125,0.00066947745,0.00015602207,0.0045521082,0.0002486551,0.00044279985,0.31005877,0.1870814,0.48260987],"study_design_scores_gemma":[0.00007449983,0.00012409827,0.004092342,0.01553648,0.00032848117,0.0005915264,0.0054447553,0.000399303,0.00052049337,0.21760616,0.75514376,0.00013806546],"about_ca_topic_score_codex":0.0048394212,"about_ca_topic_score_gemma":0.00623001,"teacher_disagreement_score":0.9082728,"about_ca_system_score_codex":0.0030846086,"about_ca_system_score_gemma":0.003974338,"threshold_uncertainty_score":0.48510545},"labels":[],"label_agreement":null},{"id":"W2101928545","doi":"10.1037/1082-989x.10.3.285","title":"Structural Equation Modeling of Paired-Comparison and Ranking Data.","year":2005,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Economic and Environmental Valuation","field":"Economics, Econometrics and Finance","cited_by":123,"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":"Thurstone scale; Structural equation modeling; Covariance; Ranking (information retrieval); Statistical model; Factor analysis; Econometrics; Model selection; Statistics; Computer science; Factor (programming language); Mathematics; Artificial intelligence","score_opus":0.6611313861810452,"score_gpt":0.43620183126280326,"score_spread":0.2249295549182419,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2101928545","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":[0.030489428,0.001305567,0.9537761,0.0017550146,0.0002843411,0.0008210775,0.0030998315,0.0005510508,0.007917642],"genre_scores_gemma":[0.42337656,0.0024357683,0.549287,0.000615647,0.00032317115,0.004593874,0.008634626,0.00022457268,0.010508788],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.98292553,0.013042164,0.00052181917,0.0013124094,0.0017854011,0.0004127559],"domain_scores_gemma":[0.95642734,0.036287874,0.0024642362,0.0027911675,0.0017438992,0.0002855097],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.021286795,0.0016616683,0.002116105,0.003370921,0.0010688942,0.002820068,0.0033302214,0.0023472793,0.014990987],"category_scores_gemma":[0.054335013,0.0009628233,0.003702782,0.0063548507,0.0014942084,0.0033123472,0.002319283,0.0036816418,0.0031112512],"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":[0.00023524048,0.0008542322,0.033152185,0.0006805144,0.0015850939,0.00059599563,0.0028923198,0.14438099,0.00059994735,0.60598034,0.01675069,0.1922924],"study_design_scores_gemma":[0.00008373175,0.00023606152,0.009074501,0.00018357219,0.00017779808,0.00023116365,0.0004381796,0.46577793,0.00035654323,0.5083473,0.015010458,0.00008281514],"about_ca_topic_score_codex":0.008820793,"about_ca_topic_score_gemma":0.00903881,"teacher_disagreement_score":0.021286795,"about_ca_system_score_codex":0.0022204227,"about_ca_system_score_gemma":0.0023206263,"threshold_uncertainty_score":0.11257672},"labels":[],"label_agreement":null},{"id":"W2102599529","doi":"10.1037/a0034177","title":"Why the resistance to statistical innovations? Bridging the communication gap.","year":2013,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Statistics Education and Methodologies","field":"Mathematics","cited_by":108,"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 Regina","funders":"","keywords":"Bridging (networking); Bridge (graph theory); Resistance (ecology); Unavailability; Publication; Computer science; Mandate; Statistical analysis; Statistical hypothesis testing; Statistical software; Publish or perish; Management science; Psychology; Data science; Statistics; Publishing; Engineering; Political science; Mathematics; Computer security","score_opus":0.46777193615161816,"score_gpt":0.5816223954573272,"score_spread":0.11385045930570903,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2102599529","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0022303613,0.03900795,0.025843438,0.9221954,0.00664751,0.0000662399,0.000037824226,0.0001438379,0.0038275565],"genre_scores_gemma":[0.23867811,0.10002232,0.087668784,0.52580494,0.038955092,0.0017005033,0.00017731467,0.00085192977,0.0061409534],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.5158572,0.37964186,0.027816439,0.01521995,0.058130924,0.0033335667],"domain_scores_gemma":[0.10396626,0.81604904,0.019852748,0.021773718,0.034081962,0.004276237],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.397875,0.00087070675,0.0022095712,0.0075797304,0.0052805296,0.020262906,0.0061077485,0.017675336,0.0040041935],"category_scores_gemma":[0.71362984,0.0015634998,0.0014404003,0.005985468,0.051357057,0.033257265,0.015917525,0.031350464,0.0022622773],"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":[0.00021173953,0.0001091387,0.0026589856,0.0052325884,0.0002631363,0.0012149535,0.07349534,0.0005357851,0.0010680453,0.5084113,0.16404173,0.24275734],"study_design_scores_gemma":[0.00010114059,0.0001865469,0.0017932369,0.009931517,0.00009573771,0.0016871883,0.03076494,0.0016100493,0.0007853124,0.58860147,0.36420408,0.00023873404],"about_ca_topic_score_codex":0.0011867174,"about_ca_topic_score_gemma":0.0009667336,"teacher_disagreement_score":0.602125,"about_ca_system_score_codex":0.007045488,"about_ca_system_score_gemma":0.020188838,"threshold_uncertainty_score":0.74252737},"labels":[],"label_agreement":null},{"id":"W2105035924","doi":"10.1037/a0015566","title":"Integrative data analysis through coordination of measurement and analysis protocol across independent longitudinal studies.","year":2009,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Cognitive Science and Mapping","field":"Computer Science","cited_by":258,"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":"National Institute on Aging; National Institutes of Health","keywords":"Comparability; Replication (statistics); Sophistication; Meta-analysis; Computer science; Protocol (science); Longitudinal study; Data science; Sample (material); Statistics; Mathematics","score_opus":0.4613856083544618,"score_gpt":0.5923953237612258,"score_spread":0.13100971540676404,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2105035924","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":[0.005779548,0.0013591753,0.94179,0.0018878669,0.0010773117,0.038674098,0.0029775144,0.0023241476,0.0041303365],"genre_scores_gemma":[0.023614366,0.00043871175,0.8827029,0.0006716321,0.00022460135,0.089301944,0.0016596965,0.00046740568,0.00091866293],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.65235466,0.24828452,0.039588027,0.027728604,0.030361414,0.0016828199],"domain_scores_gemma":[0.5247702,0.18467289,0.034010254,0.14970975,0.10319134,0.0036456292],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3828295,0.0027360376,0.0044448185,0.0088653695,0.0033212155,0.007536412,0.004794232,0.0017372097,0.007028864],"category_scores_gemma":[0.4612209,0.0023264152,0.0046559516,0.012320883,0.0024100065,0.005451009,0.010951141,0.005093619,0.0034091852],"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":[0.0016740279,0.00077185855,0.019956052,0.012319616,0.006336753,0.00028846716,0.016554281,0.003699503,0.007851438,0.042172454,0.059366766,0.82900876],"study_design_scores_gemma":[0.0053403974,0.004477848,0.10139513,0.0138179865,0.014376682,0.001202471,0.011627595,0.03198785,0.027384095,0.23525833,0.5513847,0.001746953],"about_ca_topic_score_codex":0.0036188301,"about_ca_topic_score_gemma":0.005354312,"teacher_disagreement_score":0.6171705,"about_ca_system_score_codex":0.004527245,"about_ca_system_score_gemma":0.03318331,"threshold_uncertainty_score":0.7610811},"labels":[],"label_agreement":null},{"id":"W2129748849","doi":"10.1037/a0014694","title":"Reporting practices in confirmatory factor analysis: An overview and some recommendations.","year":2009,"lang":"en","type":"review","venue":"Psychological Methods","topic":"Psychological Testing and Assessment","field":"Psychology","cited_by":2243,"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 Windsor","funders":"","keywords":"Confirmatory factor analysis; Checklist; Psychology; Normality; Structural equation modeling; Test (biology); Statistics; Applied psychology; Clinical psychology; Social psychology; Mathematics","score_opus":0.8017652952165407,"score_gpt":0.719740427808512,"score_spread":0.08202486740802872,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2129748849","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":"reporting","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"reporting","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0034774493,0.68765575,0.15730293,0.10499748,0.018382711,0.015794765,0.004984937,0.0024668535,0.004937105],"genre_scores_gemma":[0.015714753,0.3185741,0.61918926,0.0139876325,0.0029735581,0.024190223,0.0033612167,0.00052492175,0.0014843444],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.41189325,0.34581387,0.17552449,0.0077741034,0.057275034,0.0017192473],"domain_scores_gemma":[0.16464216,0.41875654,0.12270554,0.031616844,0.25893602,0.0033429218],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.52623,0.0050977473,0.0070829685,0.039027598,0.0037392357,0.010736667,0.0129669495,0.005706757,0.0024424363],"category_scores_gemma":[0.73213714,0.0048007816,0.007241349,0.052101597,0.00540907,0.0139944125,0.005580505,0.008828283,0.004324366],"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":[0.00031372326,0.00019259774,0.013471954,0.1100688,0.0024217593,0.00049012434,0.0045424276,0.0006036463,0.0009815532,0.004009651,0.18223712,0.68066674],"study_design_scores_gemma":[0.0006736406,0.00056717434,0.028469741,0.46145442,0.0065836185,0.002614635,0.008734709,0.004922215,0.0048778635,0.014039544,0.46607625,0.0009861917],"about_ca_topic_score_codex":0.01256532,"about_ca_topic_score_gemma":0.02142365,"teacher_disagreement_score":0.47377002,"about_ca_system_score_codex":0.008674033,"about_ca_system_score_gemma":0.029888371,"threshold_uncertainty_score":0.5842428},"labels":[],"label_agreement":null},{"id":"W2160348024","doi":"10.1037/a0034282","title":"Modeling continuous self-report measures of perceived emotion using generalized additive mixed models.","year":2013,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":60,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Queen's University; Queen's University Belfast","keywords":"Categorical variable; Generalized additive model; Ambiguity; Inference; Generalized linear mixed model; Autocorrelation; Computer science; Facial expression; Mixed model; Feature (linguistics); Statistical inference; Econometrics; Artificial intelligence; Statistics; Mathematics; Machine learning","score_opus":0.39491482003814043,"score_gpt":0.5170832974120531,"score_spread":0.12216847737391262,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2160348024","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":[0.026232386,0.0006516921,0.9695404,0.00059834466,0.00028069215,0.0006355408,0.0008709102,0.00043898827,0.000751088],"genre_scores_gemma":[0.27667823,0.00086797046,0.7123406,0.00038689614,0.00020887011,0.004900871,0.0017285525,0.00015738237,0.002730585],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9785841,0.016714325,0.00075678946,0.0022468783,0.0013303633,0.00036760402],"domain_scores_gemma":[0.94802505,0.04338819,0.0029824057,0.0035784245,0.0017680796,0.00025782865],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.036607496,0.0022389095,0.0018743343,0.0021608553,0.0006741706,0.0028486876,0.0043310127,0.0019486127,0.003710266],"category_scores_gemma":[0.0786607,0.0008729006,0.0057114474,0.0028642237,0.0013391408,0.0020392837,0.001958781,0.0037529438,0.00077031995],"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":[0.001511486,0.0010704721,0.074402295,0.002212507,0.008505909,0.0008068919,0.004917477,0.28842002,0.0037195089,0.31636384,0.011314336,0.2867553],"study_design_scores_gemma":[0.0001733776,0.00074768136,0.011806629,0.00024429907,0.0011076147,0.0002233183,0.0005079379,0.8214569,0.0016346568,0.15298885,0.008936601,0.0001721761],"about_ca_topic_score_codex":0.00678891,"about_ca_topic_score_gemma":0.0076170135,"teacher_disagreement_score":0.036607496,"about_ca_system_score_codex":0.0014055416,"about_ca_system_score_gemma":0.0015500926,"threshold_uncertainty_score":0.19360119},"labels":[],"label_agreement":null},{"id":"W2164150544","doi":"10.1037/a0029487","title":"The ironic effect of significant results on the credibility of multiple-study articles.","year":2012,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Behavioral Health and Interventions","field":"Psychology","cited_by":476,"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 Toronto","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Credibility; Replication (statistics); Statistical power; Sample size determination; Publication; Psychology; Extrasensory perception; Replicate; Sample (material); Power (physics); Statistics; Econometrics; Statistical hypothesis testing; Social psychology; Mathematics; Medicine; Law; Political science","score_opus":0.292115897921022,"score_gpt":0.5647337317607719,"score_spread":0.2726178338397499,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2164150544","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":"evaluation","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"evaluation","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.01814373,0.11027426,0.0838906,0.56850994,0.15406361,0.004992344,0.0024754042,0.0011193005,0.056530815],"genre_scores_gemma":[0.3690023,0.021189572,0.1715155,0.34117377,0.07278785,0.01276658,0.0008049856,0.000977954,0.009781515],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.23251088,0.42733112,0.10395549,0.039058894,0.19396634,0.003177224],"domain_scores_gemma":[0.020795174,0.865986,0.033570234,0.03363337,0.04384172,0.0021734547],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.67773104,0.0029235068,0.007367182,0.028405532,0.008292085,0.018065626,0.011963984,0.026042145,0.009819234],"category_scores_gemma":[0.9260633,0.003585297,0.0063859695,0.011676495,0.050578292,0.024640558,0.014067456,0.027231766,0.0033828074],"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":[0.00626307,0.00052445335,0.017863883,0.05722093,0.010451728,0.0034853618,0.019018525,0.0011561033,0.0022699654,0.19882101,0.4097844,0.27314046],"study_design_scores_gemma":[0.0029707486,0.0010689356,0.014486852,0.06692008,0.0054286243,0.0050455253,0.0057061166,0.008656091,0.0052738735,0.53869087,0.34486216,0.0008901521],"about_ca_topic_score_codex":0.0019988462,"about_ca_topic_score_gemma":0.0023292855,"teacher_disagreement_score":0.32226896,"about_ca_system_score_codex":0.0108751645,"about_ca_system_score_gemma":0.011154512,"threshold_uncertainty_score":0.39741504},"labels":[],"label_agreement":null},{"id":"W2167269590","doi":"10.1037/a0034176","title":"Cognitive diagnostic assessment via Bayesian evaluation of informative diagnostic hypotheses.","year":2013,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Mental Health Research Topics","field":"Psychology","cited_by":11,"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 Sherbrooke","funders":"Nederlandse Organisatie voor Wetenschappelijk Onderzoek","keywords":"Bayes factor; Bayes' theorem; Diagnostic test; Bayesian probability; Bayesian statistics; Cognition; Cognitive test; Psychology; Psychometrics; Machine learning; Artificial intelligence; Cognitive psychology; Computer science; Bayesian inference; Clinical psychology; Medicine; Psychiatry","score_opus":0.2813465974015578,"score_gpt":0.612014935139333,"score_spread":0.3306683377377752,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2167269590","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":[0.029683702,0.001482936,0.9564714,0.002144661,0.00011919561,0.0009354268,0.0004815502,0.0003614683,0.008319676],"genre_scores_gemma":[0.4016266,0.0006757697,0.5941347,0.00044828755,0.00017501498,0.0014708085,0.0007422529,0.000062669205,0.00066387],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9380829,0.047284905,0.0030262698,0.0028413427,0.008002182,0.00076238276],"domain_scores_gemma":[0.6666509,0.29201514,0.015070403,0.007143884,0.016985837,0.0021338633],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.08037516,0.0027027773,0.0020881153,0.012921482,0.0015411151,0.0046418346,0.0032088265,0.0027291337,0.0061980695],"category_scores_gemma":[0.4038501,0.0008215573,0.0021200418,0.004186548,0.0037192358,0.0064587207,0.0056693056,0.0036522576,0.0007450762],"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":[0.001295312,0.0005413524,0.044716623,0.0017301217,0.0011860916,0.0006055526,0.0037682408,0.050265823,0.0014182477,0.44968656,0.009576184,0.4352098],"study_design_scores_gemma":[0.00019339356,0.00022080383,0.0055719954,0.0005420384,0.00028283207,0.0004370474,0.00074786006,0.22744124,0.0013217767,0.7593841,0.0037235836,0.00013329231],"about_ca_topic_score_codex":0.0026457869,"about_ca_topic_score_gemma":0.002467835,"teacher_disagreement_score":0.08037516,"about_ca_system_score_codex":0.0029827622,"about_ca_system_score_gemma":0.0043944265,"threshold_uncertainty_score":0.4250695},"labels":[],"label_agreement":null},{"id":"W2193158494","doi":"10.1037/met0000059","title":"A flexible full-information approach to the modeling of response styles.","year":2015,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":120,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Social Sciences and Humanities Research Council of Canada; Institute of Education Sciences","keywords":"Item response theory; PsycINFO; Trait; Set (abstract data type); Computer science; Factor analysis; Statistics; Estimation; Structural equation modeling; Econometrics; Psychometrics; Psychology; Data mining; Machine learning; Mathematics","score_opus":0.8157017601333325,"score_gpt":0.6109273046144487,"score_spread":0.20477445551888385,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2193158494","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":[0.003958419,0.00009030028,0.9934255,0.00041263603,0.000021305348,0.00022500775,0.00048891,0.0002873656,0.0010906955],"genre_scores_gemma":[0.25298667,0.00031540717,0.73765755,0.0005179692,0.00010803968,0.002716414,0.0015119322,0.0002164175,0.003969541],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.95305204,0.03833431,0.0015251607,0.0032549533,0.0032754585,0.00055803783],"domain_scores_gemma":[0.8505494,0.1228176,0.0067992825,0.015163149,0.0038558065,0.00081478636],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.043734923,0.0020843996,0.0017592584,0.002558127,0.0009045346,0.004357758,0.0049928473,0.0027642641,0.010086858],"category_scores_gemma":[0.16295229,0.0021889617,0.004036611,0.0033500495,0.002401322,0.0060600834,0.0042327107,0.0044843126,0.0027918865],"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":[0.0008095373,0.00055510324,0.024837134,0.0008009702,0.0014525637,0.0007264131,0.0049056956,0.2742106,0.002050739,0.37226582,0.00799463,0.3093908],"study_design_scores_gemma":[0.00011970232,0.00022434705,0.004351168,0.00017285942,0.00016933329,0.0003822227,0.00017835696,0.6715987,0.0006147769,0.3159868,0.0060784756,0.00012325017],"about_ca_topic_score_codex":0.0044751284,"about_ca_topic_score_gemma":0.005083468,"teacher_disagreement_score":0.043734923,"about_ca_system_score_codex":0.0025429947,"about_ca_system_score_gemma":0.0025933785,"threshold_uncertainty_score":0.23129517},"labels":[],"label_agreement":null},{"id":"W2320131044","doi":"10.1037/a0034524","title":"Longitudinal design considerations to optimize power to detect variances and covariances among rates of change: Simulation results based on actual longitudinal studies.","year":2013,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Cognitive Abilities and Testing","field":"Psychology","cited_by":98,"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":"National Institute on Aging; National Institutes of Health; Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung","keywords":"Sample size determination; Statistics; Statistical power; Context (archaeology); Variance (accounting); Growth curve (statistics); Bivariate analysis; Monte Carlo method; Econometrics; Reliability (semiconductor); Mathematics; Covariance; Power (physics); Sample (material)","score_opus":0.5016745736120142,"score_gpt":0.5277738604236346,"score_spread":0.026099286811620437,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2320131044","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":[0.16660903,0.0016346555,0.81608987,0.0028368153,0.00021794646,0.0025504727,0.0010731159,0.00031097588,0.008677083],"genre_scores_gemma":[0.662828,0.0005940126,0.32939264,0.00096503133,0.000051729705,0.0044873953,0.00053793716,0.00008112506,0.001062058],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9526438,0.043787573,0.0008614239,0.000962969,0.0012344819,0.0005097882],"domain_scores_gemma":[0.497413,0.478622,0.0065175733,0.008620102,0.007973903,0.0008533365],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.101272054,0.0008838708,0.0012010095,0.0008228606,0.0005839799,0.0015328834,0.0015179788,0.001775673,0.00386861],"category_scores_gemma":[0.3701179,0.0007092279,0.0019204664,0.00096591737,0.001101045,0.0027572762,0.0013539721,0.0026558947,0.0003002666],"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":[0.002625559,0.00080992107,0.04688545,0.0015121957,0.0014614007,0.0005424589,0.0010956979,0.78665173,0.0039619184,0.087692305,0.0053818664,0.061379496],"study_design_scores_gemma":[0.0014048694,0.0028902118,0.011115143,0.00068082206,0.0009973927,0.00041278295,0.00035643295,0.8524935,0.00573448,0.11526738,0.008512971,0.00013402755],"about_ca_topic_score_codex":0.0030283474,"about_ca_topic_score_gemma":0.0029647818,"teacher_disagreement_score":0.89872795,"about_ca_system_score_codex":0.0013199804,"about_ca_system_score_gemma":0.0026368017,"threshold_uncertainty_score":0.5355842},"labels":[],"label_agreement":null},{"id":"W2329543180","doi":"10.1037/met0000015","title":"An affirmation of the phenomenological psychological descriptive method: A response to Rennie (2012).","year":2014,"lang":"en","type":"letter","venue":"Psychological Methods","topic":"Social Representations and Identity","field":"Psychology","cited_by":32,"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; Université du Québec à Montréal","funders":"","keywords":"Phenomenology (philosophy); Phenomenological method; Epistemology; Interpretation (philosophy); Psychology; Perspective (graphical); Object (grammar); Value (mathematics); Descriptive statistics; Core (optical fiber); Social psychology; Philosophy; Computer science; Mathematics; Statistics; Artificial intelligence","score_opus":0.19200363662915845,"score_gpt":0.5393592691486196,"score_spread":0.3473556325194612,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2329543180","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00035454927,0.0017589801,0.0008233689,0.98765045,0.008660543,0.000011003602,0.000018588253,0.000031261556,0.00069121725],"genre_scores_gemma":[0.006066272,0.00085653004,0.0018040563,0.9827901,0.0063716928,0.00010209131,0.000011654496,0.000051906034,0.0019457024],"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.94624096,0.029435692,0.0035257183,0.006496509,0.012656283,0.0016448814],"domain_scores_gemma":[0.8948787,0.079938345,0.0038826938,0.0041071884,0.013712751,0.0034803082],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.043954905,0.0020006048,0.002035712,0.0015315542,0.01021858,0.01173566,0.008564171,0.06880956,0.0031364828],"category_scores_gemma":[0.14766392,0.0013303269,0.0022685155,0.0016589715,0.03984795,0.033287082,0.012028712,0.11515231,0.0033934696],"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":[0.00015616382,0.00010213918,0.00062707975,0.0004746886,0.000046425772,0.0015068649,0.0146250725,0.00023338657,0.0006393795,0.14228632,0.8220017,0.017300833],"study_design_scores_gemma":[0.00010328244,0.0000986377,0.0007417736,0.0013858736,0.000023390043,0.0030279069,0.013932749,0.0011295713,0.00068003504,0.13055475,0.8480646,0.00025743936],"about_ca_topic_score_codex":0.008542779,"about_ca_topic_score_gemma":0.007829372,"teacher_disagreement_score":0.9560451,"about_ca_system_score_codex":0.0115508195,"about_ca_system_score_gemma":0.01083916,"threshold_uncertainty_score":0.23245853},"labels":[],"label_agreement":null},{"id":"W2395889112","doi":"10.1037/met0000075","title":"Testing measurement invariance in longitudinal data with ordered-categorical measures.","year":2016,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Advanced Statistical Modeling Techniques","field":"Computer Science","cited_by":344,"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 Health Network","funders":"National Institute of Mental Health","keywords":"Categorical variable; Measurement invariance; Item response theory; Latent variable; Metric (unit); Latent variable model; Psychology; Econometrics; Statistics; Structural equation modeling; Psychometrics; Computer science; Confirmatory factor analysis; Mathematics","score_opus":0.49908626373203135,"score_gpt":0.48438075925113383,"score_spread":0.014705504480897524,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2395889112","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":[0.16224727,0.000861543,0.80743754,0.002836795,0.00090388046,0.0033541399,0.0071764784,0.0014693958,0.013712951],"genre_scores_gemma":[0.62694645,0.0005687795,0.35027745,0.0009721556,0.0002001441,0.0122660855,0.0076674037,0.00036045216,0.0007411172],"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.88900506,0.08155185,0.0069135977,0.008971732,0.012720438,0.00083741534],"domain_scores_gemma":[0.48117352,0.40881035,0.03490122,0.0513782,0.022388972,0.0013476883],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.11413068,0.00096830295,0.001438473,0.002968744,0.0015753036,0.0038002657,0.0022049053,0.0011252664,0.0071332455],"category_scores_gemma":[0.50450915,0.0005852869,0.0025798157,0.007468833,0.0028923438,0.005390749,0.00370057,0.0038877144,0.0012711206],"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":[0.0005832986,0.0007830009,0.6240197,0.0018070206,0.0044132937,0.0003236156,0.0068087718,0.0070076287,0.0016764734,0.07410872,0.018056631,0.26041195],"study_design_scores_gemma":[0.0006885064,0.0034992078,0.5513957,0.0022733286,0.0022595439,0.00091596396,0.008674911,0.13636313,0.005400844,0.24998236,0.03811364,0.00043286203],"about_ca_topic_score_codex":0.004783126,"about_ca_topic_score_gemma":0.005046391,"teacher_disagreement_score":0.11413068,"about_ca_system_score_codex":0.0017638963,"about_ca_system_score_gemma":0.0046552094,"threshold_uncertainty_score":0.603588},"labels":[],"label_agreement":null},{"id":"W2412046924","doi":"10.1037/met0000071","title":"Dynamical correlation: A new method for quantifying synchrony with multivariate intensive longitudinal data.","year":2016,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Functional Brain Connectivity Studies","field":"Neuroscience","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":"Health Sciences North","funders":"","keywords":"Correlation; Nonparametric statistics; Bivariate analysis; Homogeneity (statistics); Multivariate statistics; Dynamical systems theory; Statistical hypothesis testing; Computer science; Population; Statistics; Econometrics; Mathematics; Data mining","score_opus":0.39901170294233734,"score_gpt":0.525208746155382,"score_spread":0.1261970432130447,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2412046924","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":[0.00644501,0.00061869767,0.98806775,0.00034722494,0.00020464433,0.00017763975,0.0011839466,0.00063100836,0.002324088],"genre_scores_gemma":[0.22368425,0.0013423905,0.7644071,0.00051820587,0.00069442485,0.0020033289,0.0030322468,0.0008198211,0.003498287],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99502736,0.002097197,0.0004865053,0.0010932055,0.0011478711,0.00014798598],"domain_scores_gemma":[0.97998834,0.011595119,0.0034948855,0.0026646487,0.0016967228,0.00056029594],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009609547,0.0010282143,0.0009224077,0.006648402,0.0009776475,0.0020937857,0.0014370093,0.0010320285,0.007076727],"category_scores_gemma":[0.0538359,0.0007056466,0.0015095313,0.0070798504,0.0021267596,0.005278444,0.0032221824,0.0021350272,0.0013619869],"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":[0.00038406902,0.00017724783,0.050864805,0.0015156416,0.0011959488,0.0006903255,0.003001816,0.036242362,0.015128422,0.25878337,0.031209635,0.60080636],"study_design_scores_gemma":[0.000113278824,0.00035511062,0.062391773,0.0005795955,0.00037222393,0.0021499814,0.0011644884,0.39288384,0.007583034,0.44435027,0.087612055,0.00044440865],"about_ca_topic_score_codex":0.003133675,"about_ca_topic_score_gemma":0.0036388382,"teacher_disagreement_score":0.009609547,"about_ca_system_score_codex":0.001081705,"about_ca_system_score_gemma":0.002028881,"threshold_uncertainty_score":0.050820768},"labels":[],"label_agreement":null},{"id":"W2417673837","doi":"10.1037/met0000053","title":"Partial least squares correspondence analysis: A framework to simultaneously analyze behavioral and genetic data.","year":2015,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Sensory Analysis and Statistical Methods","field":"Agricultural and Biological Sciences","cited_by":36,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Canadian Institutes of Health Research; National Institute on Aging; National Institutes of Health; National Institute on Drug Abuse; Agence Nationale de la Recherche; U.S. Department of Defense","keywords":"Partial least squares regression; Categorical variable; Computer science; Correspondence analysis; Multivariate statistics; Artificial intelligence; Multivariate analysis; Statistics; Machine learning; Data mining; Psychology; Mathematics","score_opus":0.2672039613477072,"score_gpt":0.4999134662670226,"score_spread":0.2327095049193154,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2417673837","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":[0.00038124475,0.0003473692,0.998064,0.00017321044,0.00003814884,0.00006209104,0.0002470797,0.00027137637,0.0004154243],"genre_scores_gemma":[0.018424386,0.000857368,0.9771597,0.00027045183,0.00018693478,0.00087816967,0.0009631015,0.00030542686,0.00095449685],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9788032,0.013760751,0.0010268892,0.0029113693,0.003205814,0.0002919144],"domain_scores_gemma":[0.9668026,0.025222944,0.0021078514,0.0030752164,0.002397536,0.00039378012],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.022820622,0.003125698,0.0022170015,0.0063277464,0.0011783425,0.0041514845,0.004038975,0.0019974587,0.0067111133],"category_scores_gemma":[0.053162027,0.0014495343,0.004643038,0.008577133,0.0035426228,0.0034958974,0.0054769367,0.004936268,0.00266761],"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":[0.00021774787,0.00025897636,0.00882715,0.0017384132,0.0023239711,0.0005142605,0.0011555974,0.10879262,0.004239293,0.3943595,0.020022336,0.45755014],"study_design_scores_gemma":[0.000045142573,0.00014646686,0.0031619105,0.0001884573,0.00019027787,0.00034551084,0.00023688063,0.3050198,0.0016222432,0.65220547,0.03670119,0.0001365875],"about_ca_topic_score_codex":0.0059907474,"about_ca_topic_score_gemma":0.006655224,"teacher_disagreement_score":0.022820622,"about_ca_system_score_codex":0.001583983,"about_ca_system_score_gemma":0.005782982,"threshold_uncertainty_score":0.12068844},"labels":[],"label_agreement":null},{"id":"W2460913564","doi":"10.1037/met0000055","title":"A correction factor for the impact of cluster randomized sampling and its applications.","year":2015,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Bayesian Methods and Mixture Models","field":"Computer Science","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":"Université du Québec à Trois-Rivières; University of Ottawa","funders":"","keywords":"Statistics; Spurious relationship; Cluster sampling; Sample size determination; Statistical power; Sampling (signal processing); Covariate; Cluster analysis; Variance (accounting); Population; Cluster (spacecraft); Mathematics; Standard error; Confidence interval; Sample (material); Population variance; Statistical hypothesis testing; Type I and type II errors; Econometrics; Computer science; Demography; Filter (signal processing)","score_opus":0.19036396902259523,"score_gpt":0.5067937905813413,"score_spread":0.3164298215587461,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2460913564","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":[0.0021809712,0.00063503085,0.9869797,0.0014535794,0.002449679,0.0012230821,0.00053105265,0.0011424127,0.0034043686],"genre_scores_gemma":[0.051184967,0.00045948516,0.93277395,0.0020626632,0.00078631967,0.006470138,0.0004049899,0.0011666252,0.004690834],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.83519274,0.111043066,0.008141727,0.016094489,0.027672924,0.0018550571],"domain_scores_gemma":[0.5233366,0.3270287,0.021465791,0.08061668,0.045650553,0.0019016393],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.12116329,0.0023223262,0.0023533236,0.0031691527,0.003378823,0.0039500436,0.004137399,0.0044275117,0.01938888],"category_scores_gemma":[0.52523726,0.0015863796,0.0037944068,0.0050427867,0.0046345177,0.0045258338,0.0054981243,0.008050119,0.0040601087],"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":[0.0013612472,0.0002723954,0.014485314,0.0025844895,0.00150706,0.0005040762,0.0027265255,0.012190583,0.004066766,0.32722962,0.08791722,0.5451547],"study_design_scores_gemma":[0.001114556,0.0017384546,0.020006286,0.0030827732,0.0021684943,0.0025622954,0.00067442725,0.10429809,0.017573692,0.3782539,0.4677607,0.00076637854],"about_ca_topic_score_codex":0.006085152,"about_ca_topic_score_gemma":0.0056098714,"teacher_disagreement_score":0.8788367,"about_ca_system_score_codex":0.003652391,"about_ca_system_score_gemma":0.009971328,"threshold_uncertainty_score":0.6407804},"labels":[],"label_agreement":null},{"id":"W2470860284","doi":"10.1037/met0000036","title":"Planning sample sizes when effect sizes are uncertain: The power-calibrated effect size approach.","year":2015,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","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":"Kellogg's (Canada)","funders":"","keywords":"Sample size determination; Statistical power; Sample (material); Statistics; Replication (statistics); Computer science; Power (physics); Confidence interval; Econometrics; Mathematics","score_opus":0.6449695789679178,"score_gpt":0.6157682207415001,"score_spread":0.029201358226417784,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2470860284","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":[0.0013353291,0.0021747153,0.98356134,0.0036666032,0.00046685836,0.002630195,0.00025702335,0.0004160355,0.0054919254],"genre_scores_gemma":[0.052182313,0.0012593479,0.92996544,0.0032352402,0.00034542978,0.011617229,0.0001428103,0.0003033981,0.0009488703],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.7102823,0.2523579,0.008150291,0.010369266,0.01795614,0.00088411814],"domain_scores_gemma":[0.41068426,0.5354333,0.013685231,0.027810413,0.010839737,0.0015470491],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.24429561,0.0029360196,0.003113913,0.0058916886,0.0015591808,0.0050571524,0.0071155815,0.006301742,0.009131426],"category_scores_gemma":[0.61765707,0.0027050816,0.0035484636,0.004555026,0.009136894,0.011186915,0.0072039813,0.013169694,0.0024133092],"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":[0.0010566475,0.00026731982,0.0029795617,0.0039687045,0.0014323872,0.000284898,0.0040536127,0.01777584,0.0016809889,0.40713814,0.020153582,0.5392083],"study_design_scores_gemma":[0.00084967376,0.0006346955,0.0024536597,0.002214359,0.00063481054,0.00040768122,0.00039087413,0.033146124,0.0033257222,0.91518325,0.040523272,0.00023578393],"about_ca_topic_score_codex":0.0021519186,"about_ca_topic_score_gemma":0.0029832772,"teacher_disagreement_score":0.7557044,"about_ca_system_score_codex":0.0051351706,"about_ca_system_score_gemma":0.007909727,"threshold_uncertainty_score":0.931918},"labels":[],"label_agreement":null},{"id":"W2474654233","doi":"10.1037/met0000048","title":"Modeling intensive longitudinal data with mixtures of nonparametric trajectories and time-varying effects.","year":2015,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":48,"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 Health Centre","funders":"National Cancer Institute; National Institute on Drug Abuse","keywords":"Covariate; Semiparametric regression; Nonparametric statistics; Econometrics; Population; Regression analysis; Regression; Statistics; Mathematics; Latent variable; Linear regression; Semiparametric model; Demography","score_opus":0.2803564483775052,"score_gpt":0.498751203955735,"score_spread":0.21839475557822985,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2474654233","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":[0.009116186,0.00043790042,0.98890555,0.00034544212,0.00006141482,0.00009662,0.00034450187,0.00023482837,0.00045764213],"genre_scores_gemma":[0.2886594,0.0016092595,0.7005462,0.0005221119,0.00030686,0.0014045666,0.0023175813,0.00018236194,0.0044515873],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99312544,0.004580028,0.00030432493,0.0012288085,0.00053835684,0.00022299272],"domain_scores_gemma":[0.9685259,0.026167072,0.002203976,0.002080811,0.0006850178,0.00033722294],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.018806582,0.001473221,0.0019521697,0.0024121923,0.0008924847,0.002520875,0.003461538,0.0025364151,0.0029717898],"category_scores_gemma":[0.050460454,0.0014860549,0.0034857308,0.0024502454,0.0019431332,0.0033627495,0.003189915,0.0036518604,0.00055912416],"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":[0.0003048482,0.00021990901,0.02664401,0.0004432792,0.0013218679,0.00043241042,0.0013288646,0.41473535,0.0013033099,0.45406428,0.00446708,0.09473471],"study_design_scores_gemma":[0.000045954705,0.00007842337,0.0029206523,0.000075189215,0.00014729703,0.00015783779,0.00008395054,0.7838272,0.00027300182,0.20774487,0.0045935824,0.000051940286],"about_ca_topic_score_codex":0.012253273,"about_ca_topic_score_gemma":0.015285906,"teacher_disagreement_score":0.018806582,"about_ca_system_score_codex":0.0016084426,"about_ca_system_score_gemma":0.001651188,"threshold_uncertainty_score":0.09945989},"labels":[],"label_agreement":null},{"id":"W2522861822","doi":"10.1037/met0000084","title":"The impact of total and partial inclusion or exclusion of active and inactive time invariant covariates in growth mixture models.","year":2016,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":145,"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 Sherbrooke","funders":"","keywords":"Covariate; Statistics; Mathematics; Bayesian information criterion; Akaike information criterion; Econometrics; Population; Sample size determination; Latent class model; Demography","score_opus":0.08867699414247705,"score_gpt":0.46423015804126466,"score_spread":0.3755531638987876,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2522861822","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":[0.3133751,0.006089495,0.6691847,0.0019688958,0.0004045394,0.0006469709,0.0012436527,0.0013954634,0.0056911693],"genre_scores_gemma":[0.8689276,0.0010079845,0.12515137,0.0005046226,0.00016204575,0.00045242917,0.002274244,0.00024210644,0.0012776542],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9060982,0.077596195,0.0038781005,0.006834638,0.004309989,0.0012828807],"domain_scores_gemma":[0.5130673,0.4471374,0.012806916,0.019530255,0.005849234,0.0016089298],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.11404058,0.0024686756,0.0027449618,0.0017859642,0.0020071408,0.004803291,0.002271265,0.00236025,0.0030515755],"category_scores_gemma":[0.34834903,0.0012069753,0.0052996296,0.0023536975,0.0032049625,0.0059349723,0.0054319436,0.00403807,0.0006687887],"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":[0.004807894,0.0004358901,0.37641165,0.0017478359,0.009253886,0.00081699085,0.003127393,0.21926951,0.0027090856,0.047689337,0.0033688555,0.33036175],"study_design_scores_gemma":[0.00044071843,0.0027580021,0.14125986,0.0010505654,0.008096762,0.00077382,0.0016188544,0.7364724,0.0086297905,0.08381944,0.014655947,0.00042381356],"about_ca_topic_score_codex":0.008959607,"about_ca_topic_score_gemma":0.008789447,"teacher_disagreement_score":0.11404058,"about_ca_system_score_codex":0.0013962694,"about_ca_system_score_gemma":0.00336668,"threshold_uncertainty_score":0.60311145},"labels":[],"label_agreement":null},{"id":"W2558952171","doi":"10.1037/met0000097","title":"Type I error rates and power of several versions of scaled chi-square difference tests in investigations of measurement invariance.","year":2016,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":17,"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":"Statistics; Type I and type II errors; Mathematics; Measurement invariance; Sample size determination; Goodness of fit; Monte Carlo method; Confirmatory factor analysis; Population; Statistical power; Chi-square test; Econometrics; Structural equation modeling; Demography","score_opus":0.679952449318081,"score_gpt":0.5593365285098917,"score_spread":0.12061592080818928,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2558952171","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":[0.3825369,0.023933833,0.5344028,0.0032878444,0.004745046,0.0072577917,0.016069666,0.002837443,0.024928682],"genre_scores_gemma":[0.83856654,0.0017602536,0.14522326,0.0007639899,0.0005058563,0.0067415866,0.00378629,0.0008438286,0.0018084926],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.69052863,0.23107815,0.024008222,0.026488421,0.02639495,0.0015016537],"domain_scores_gemma":[0.098274045,0.836546,0.019555898,0.034464996,0.010435286,0.00072376546],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.28984085,0.0014265229,0.0019087791,0.007848727,0.0012957397,0.002980422,0.0043368544,0.0032542679,0.007474812],"category_scores_gemma":[0.7706445,0.0010831429,0.0057505514,0.008439787,0.0046499805,0.0050787525,0.0027372711,0.0042676483,0.0011450442],"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":[0.008206665,0.0007304953,0.5164793,0.005257129,0.015927978,0.00069719943,0.0061792512,0.013327665,0.0015787295,0.027847547,0.030619979,0.37314814],"study_design_scores_gemma":[0.0029584244,0.0076005436,0.6812434,0.007043049,0.015758855,0.0033457926,0.005123903,0.15100706,0.010592463,0.086642586,0.02785224,0.0008316538],"about_ca_topic_score_codex":0.00280071,"about_ca_topic_score_gemma":0.0028974488,"teacher_disagreement_score":0.7101592,"about_ca_system_score_codex":0.0020208305,"about_ca_system_score_gemma":0.00258038,"threshold_uncertainty_score":0.8757527},"labels":[],"label_agreement":null},{"id":"W2593944763","doi":"10.1037/met0000126","title":"Reporting effect sizes in original psychological research: A discussion and tutorial.","year":2017,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Mental Health Research Topics","field":"Psychology","cited_by":325,"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; Natural Sciences and Engineering Research Council of Canada","keywords":"Categorical variable; PsycINFO; Psychology; Mediation; Variety (cybernetics); Interpretation (philosophy); Statistical hypothesis testing; Regression analysis; Statistics; MEDLINE; Computer science; Mathematics","score_opus":0.6435478184034068,"score_gpt":0.7307775730395879,"score_spread":0.0872297546361811,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2593944763","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"reporting","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"reporting","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00067451515,0.15772803,0.65220666,0.14143248,0.03400782,0.0020839034,0.0015214223,0.0025614465,0.0077837864],"genre_scores_gemma":[0.008268639,0.12973748,0.7846033,0.0321231,0.023591869,0.013139342,0.001069103,0.0019716674,0.0054954854],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.76237935,0.18959576,0.024572695,0.0049383473,0.017856931,0.00065696065],"domain_scores_gemma":[0.2868371,0.6572662,0.015664928,0.017834622,0.020730088,0.0016671204],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.29442337,0.0038127278,0.004024295,0.012926252,0.0021889398,0.009672746,0.009009919,0.007910505,0.014186931],"category_scores_gemma":[0.5957249,0.0030302224,0.0042258627,0.016708925,0.014637064,0.019066822,0.006172422,0.016896764,0.010375454],"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":[0.00014303981,0.00012875577,0.0014022207,0.016871803,0.0004456548,0.00041138008,0.0037478057,0.0008966654,0.00093251915,0.12787244,0.3734044,0.47374338],"study_design_scores_gemma":[0.000077588025,0.0002520122,0.0027336692,0.019912278,0.0001825587,0.0017712807,0.0016382239,0.002623204,0.0013344776,0.23407395,0.73519015,0.00021070328],"about_ca_topic_score_codex":0.0014667993,"about_ca_topic_score_gemma":0.0025840201,"teacher_disagreement_score":0.70557666,"about_ca_system_score_codex":0.004716332,"about_ca_system_score_gemma":0.009093597,"threshold_uncertainty_score":0.8701016},"labels":[],"label_agreement":null},{"id":"W2624920789","doi":"10.1037/met0000151","title":"ExperienceSampler: An open-source scaffold for building smartphone apps for experience sampling.","year":2017,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Mental Health Research Topics","field":"Psychology","cited_by":87,"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 Toronto","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Experience sampling method; Computer science; The Internet; Android (operating system); Internet access; PsycINFO; Sampling (signal processing); World Wide Web; Multimedia; Usage data; Operating system; Psychology; Telecommunications","score_opus":0.6325151645478696,"score_gpt":0.6907879567918435,"score_spread":0.05827279224397386,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2624920789","genre_codex":"methods","genre_gemma":"software","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"software","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.047622357,0.0014579331,0.7682179,0.0011167364,0.0008662543,0.017469596,0.027696462,0.10453324,0.0310195],"genre_scores_gemma":[0.07141829,0.0011760175,0.8434019,0.0006931546,0.0002726447,0.037486758,0.013071187,0.013847158,0.018632866],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.99698025,0.0013841263,0.00041537508,0.0004065551,0.0006777893,0.00013582339],"domain_scores_gemma":[0.9497098,0.039842054,0.0017436334,0.004912678,0.0027636255,0.0010281979],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0075000073,0.0016436973,0.0006836369,0.003107556,0.0006082794,0.0016293499,0.0015399642,0.00073785946,0.036054973],"category_scores_gemma":[0.044496547,0.00082286453,0.0012790011,0.0012240503,0.0007205012,0.003185927,0.0049774046,0.0011640132,0.0114747025],"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":[0.0011983409,0.00086261367,0.0072090095,0.0067559625,0.00019369746,0.0009338079,0.014760319,0.0010236345,0.00991029,0.006002709,0.13516451,0.8159851],"study_design_scores_gemma":[0.0011712778,0.002015685,0.042712912,0.0049274135,0.00035854348,0.001973036,0.006235818,0.013432231,0.022085179,0.033522148,0.87106305,0.00050263637],"about_ca_topic_score_codex":0.00038336287,"about_ca_topic_score_gemma":0.0012318797,"teacher_disagreement_score":0.036054973,"about_ca_system_score_codex":0.0003476607,"about_ca_system_score_gemma":0.0014658371,"threshold_uncertainty_score":0.1206159},"labels":[],"label_agreement":null},{"id":"W2778345604","doi":"10.1037/met0000112","title":"An alternative to post hoc model modification in confirmatory factor analysis: The Bayesian lasso.","year":2017,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Mental Health Research Topics","field":"Psychology","cited_by":73,"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":"Wake Forest Clinical and Translational Science Institute, Wake Forest School of Medicine; National Center for Advancing Translational Sciences; China Scholarship Council; National Natural Science Foundation of China","keywords":"Residual; Covariance; Computer science; Bayesian probability; Covariance matrix; Overfitting; Mathematics; Statistics; Robustness (evolution); Algorithm; Artificial intelligence","score_opus":0.42000440021908697,"score_gpt":0.6572029316560931,"score_spread":0.23719853143700614,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2778345604","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":[0.0021127078,0.00024621363,0.9956227,0.0003798966,0.00009772764,0.00030934397,0.000105521154,0.0004407889,0.0006850195],"genre_scores_gemma":[0.06681933,0.00035289774,0.9279747,0.00052955514,0.00015194641,0.0022649602,0.000505608,0.00044722293,0.0009537653],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9470978,0.044626914,0.0015296149,0.0027843488,0.0035586308,0.00040277315],"domain_scores_gemma":[0.9038624,0.06863236,0.010271584,0.010293843,0.006115637,0.00082420476],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.058573853,0.0032317778,0.0025177463,0.0029019779,0.0020454698,0.0028780485,0.0033816902,0.0020827816,0.005261381],"category_scores_gemma":[0.1819431,0.0015714931,0.0024525053,0.0038812472,0.0028897994,0.0032374216,0.0042161485,0.0060040853,0.001549962],"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":[0.0009429601,0.0005841743,0.016899237,0.003008222,0.0041660643,0.00035369064,0.0030796113,0.05482221,0.0047921822,0.17626844,0.02899706,0.70608604],"study_design_scores_gemma":[0.0005308677,0.0007926364,0.007962856,0.0009230242,0.00073378056,0.0004001733,0.0006564384,0.5693686,0.0069431816,0.3742411,0.03705487,0.00039243564],"about_ca_topic_score_codex":0.0035126284,"about_ca_topic_score_gemma":0.008175795,"teacher_disagreement_score":0.058573853,"about_ca_system_score_codex":0.0014989168,"about_ca_system_score_gemma":0.007058647,"threshold_uncertainty_score":0.3097719},"labels":[],"label_agreement":null},{"id":"W2803568088","doi":"10.1037/met0000181","title":"A comparison of several approaches for controlling measurement error in small samples.","year":2018,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":106,"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":"Structural equation modeling; Reliability (semiconductor); Statistics; Latent variable; Standard error; Mathematics; Inference; Observational error; Variable (mathematics); Econometrics; Computer science; Power (physics); Artificial intelligence","score_opus":0.9584692213383208,"score_gpt":0.6539327846799784,"score_spread":0.3045364366583424,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2803568088","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":[0.01975757,0.020703485,0.9308168,0.0038449673,0.0011953479,0.0047774627,0.00084643305,0.0006648787,0.01739304],"genre_scores_gemma":[0.09205,0.0078056,0.88953984,0.0007477168,0.000212532,0.0077387914,0.00039083348,0.00022650251,0.001288179],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.6419871,0.30319107,0.010379946,0.008647386,0.034868848,0.00092571444],"domain_scores_gemma":[0.5349643,0.41317686,0.012616517,0.019612275,0.018150473,0.0014795549],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.23524809,0.002204517,0.0021975373,0.009489047,0.0018156612,0.0039008674,0.0036787167,0.0019579476,0.006837807],"category_scores_gemma":[0.4035976,0.0011813095,0.0043956074,0.013707124,0.0035147218,0.006489579,0.0064393226,0.0031980155,0.0009689351],"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":[0.0008297589,0.0003742175,0.022124473,0.005017464,0.003335426,0.00010913815,0.0038387333,0.0046666465,0.00047851072,0.06885314,0.00940158,0.88097095],"study_design_scores_gemma":[0.0038179113,0.006620201,0.19595183,0.023254821,0.010291971,0.0030014035,0.015333968,0.14389607,0.0059856437,0.40213394,0.18823346,0.001478836],"about_ca_topic_score_codex":0.006319522,"about_ca_topic_score_gemma":0.009333594,"teacher_disagreement_score":0.7647519,"about_ca_system_score_codex":0.004526904,"about_ca_system_score_gemma":0.007350077,"threshold_uncertainty_score":0.94307524},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"simulation_or_modeling","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"simulation_or_modeling","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W2888454927","doi":"10.1037/met0000176","title":"A cautionary note on the finite sample behavior of maximal reliability.","year":2018,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Reliability and Maintenance Optimization","field":"Engineering","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":"Public Safety Canada","funders":"Aalto-Yliopisto; Academy of Finland","keywords":"Reliability (semiconductor); Sample (material); Statistics; Mathematics; Psychology; Econometrics; Applied mathematics; Physics; Thermodynamics","score_opus":0.056199717784466405,"score_gpt":0.38422726487056996,"score_spread":0.32802754708610354,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2888454927","genre_codex":"commentary","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":[0.014798485,0.031099075,0.2870837,0.5910544,0.038748596,0.00047812727,0.0020880415,0.002204279,0.032445394],"genre_scores_gemma":[0.24595408,0.009911564,0.34478068,0.34181845,0.028271606,0.0020998556,0.00048935047,0.0013562769,0.02531808],"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9012196,0.057953358,0.008918181,0.012886186,0.018111574,0.0009110235],"domain_scores_gemma":[0.55040866,0.3915319,0.008944341,0.023502985,0.024127632,0.0014844898],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.11036038,0.0016746034,0.0023432516,0.0024215162,0.0032296802,0.005442529,0.008788242,0.005244973,0.004645648],"category_scores_gemma":[0.41949606,0.00080005766,0.002185091,0.0038837106,0.015580269,0.007876874,0.0037400688,0.030800791,0.003149368],"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":[0.0004725113,0.00011614779,0.011995209,0.0016835985,0.00053242786,0.0016483658,0.010631501,0.0023544438,0.0015318466,0.371765,0.48155272,0.11571614],"study_design_scores_gemma":[0.00023197503,0.0003297259,0.015769992,0.0032028263,0.00031083636,0.0027755648,0.002921083,0.013360972,0.004752382,0.5926485,0.3631836,0.00051245635],"about_ca_topic_score_codex":0.010138965,"about_ca_topic_score_gemma":0.013518292,"teacher_disagreement_score":0.8896396,"about_ca_system_score_codex":0.0026600016,"about_ca_system_score_gemma":0.0035743758,"threshold_uncertainty_score":0.58364844},"labels":[],"label_agreement":null},{"id":"W2914981656","doi":"10.1037/met0000337","title":"Power contours: Optimising sample size and precision in experimental psychology and human neuroscience.","year":2020,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Face Recognition and Perception","field":"Neuroscience","cited_by":212,"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":"Biotechnology and Biological Sciences Research Council; Directorate for Biological Sciences; Medical Research Council; Centre for Chronic Diseases and Disorders; University of Cambridge; Wellcome Trust","keywords":"Sample size determination; Statistical power; Variance (accounting); Sample (material); Design of experiments; Power (physics); Meaning (existential); Statistical hypothesis testing; Experimental data","score_opus":0.27881685704017695,"score_gpt":0.5244552156108684,"score_spread":0.24563835857069144,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2914981656","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":[0.0027795376,0.00094664516,0.9866209,0.00043959697,0.00033667774,0.0026148309,0.0004491522,0.00302738,0.0027853423],"genre_scores_gemma":[0.037448067,0.00048578694,0.94514126,0.00031786898,0.00015888053,0.013687453,0.00026536753,0.0018663758,0.0006289329],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.86226183,0.10619654,0.0072069434,0.009496872,0.013952018,0.0008857652],"domain_scores_gemma":[0.51159483,0.42646804,0.015741795,0.034314714,0.010645643,0.0012350298],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2029443,0.0030523094,0.0031430721,0.006373567,0.0018085385,0.0066913897,0.0053477944,0.0065885517,0.015409967],"category_scores_gemma":[0.55945706,0.0035996996,0.0036291727,0.006421406,0.007753952,0.010970551,0.008475124,0.006498171,0.0029586216],"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":[0.004116461,0.0004835415,0.0057732267,0.008316548,0.001592688,0.0005401296,0.004656588,0.033313777,0.021813076,0.14763317,0.029684488,0.7420762],"study_design_scores_gemma":[0.0037245974,0.0029731733,0.019249607,0.00456873,0.0016012138,0.0011034224,0.0007786041,0.22720574,0.0469666,0.57258046,0.11835872,0.0008891145],"about_ca_topic_score_codex":0.0013888413,"about_ca_topic_score_gemma":0.0014795194,"teacher_disagreement_score":0.7970557,"about_ca_system_score_codex":0.0029119346,"about_ca_system_score_gemma":0.003439328,"threshold_uncertainty_score":0.9829116},"labels":[],"label_agreement":null},{"id":"W2963126549","doi":"10.1037/met0000163","title":"Parameter uncertainty in structural equation models: Confidence sets and fungible estimates.","year":2018,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Mental Health Research Topics","field":"Psychology","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Ontario Ministry of Research and Innovation; Natural Sciences and Engineering Research Council of Canada","keywords":"Thriving; Structural equation modeling; PsycINFO; Context (archaeology); Proxy (statistics); Psychology; Statistics; Mathematics; Political science; MEDLINE; History; Law","score_opus":0.40841950830771173,"score_gpt":0.5997492743696715,"score_spread":0.19132976606195978,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2963126549","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":[0.036333665,0.0066777975,0.9407762,0.007824004,0.0005741637,0.0008286519,0.0006672975,0.0003301422,0.0059879944],"genre_scores_gemma":[0.6740376,0.00342777,0.31536838,0.0014432585,0.00062351563,0.003464942,0.00074484624,0.00015779083,0.00073192624],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.6602821,0.29061627,0.010823168,0.01730078,0.01965735,0.0013203233],"domain_scores_gemma":[0.1358357,0.8178175,0.018750947,0.022817174,0.004048472,0.00073011],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.24111472,0.0031424833,0.0042000813,0.00885691,0.0023124118,0.014699158,0.007352415,0.0066816416,0.0055778828],"category_scores_gemma":[0.7218752,0.002140192,0.00502322,0.011366161,0.023057912,0.020975528,0.010265928,0.016116388,0.00061271345],"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":[0.0003898413,0.00014931486,0.03721713,0.0019882675,0.003093901,0.00046505078,0.005981484,0.0467281,0.00023763163,0.75363535,0.002917106,0.14719686],"study_design_scores_gemma":[0.00008008966,0.00018100417,0.0055563087,0.0013637157,0.00041917263,0.00023443371,0.0011495011,0.10483567,0.00033977523,0.8824757,0.0032001422,0.0001645433],"about_ca_topic_score_codex":0.0044454075,"about_ca_topic_score_gemma":0.0037712336,"teacher_disagreement_score":0.24111472,"about_ca_system_score_codex":0.0050942325,"about_ca_system_score_gemma":0.004096387,"threshold_uncertainty_score":0.93584067},"labels":[],"label_agreement":null},{"id":"W2963672622","doi":"10.1037/met0000175","title":"Alternating optimization for G × E modelling with weighted genetic and environmental scores: Examples from the MAVAN study.","year":2018,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Mental Health Research Topics","field":"Psychology","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Jewish General Hospital","funders":"","keywords":"Linear model; Statistics; Genetic algorithm; Computer science; Construct (python library); Mixed model; R package; Econometrics; Mathematics; Machine learning","score_opus":0.257036098063129,"score_gpt":0.5134929462929112,"score_spread":0.2564568482297822,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2963672622","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":[0.059512205,0.0005863991,0.9266604,0.0016542443,0.00010930394,0.00019365194,0.00053485535,0.0002730904,0.010475851],"genre_scores_gemma":[0.52279127,0.0005796976,0.46242127,0.00063056924,0.00007677346,0.0012570721,0.00063410035,0.00023761851,0.011371604],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99802244,0.0016517368,0.000041717307,0.00015489323,0.0000666818,0.00006257035],"domain_scores_gemma":[0.98603696,0.012723409,0.0003845531,0.0003914043,0.0002679467,0.00019564925],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0045251274,0.001108284,0.0011756114,0.0005801138,0.0007856738,0.0010236177,0.0019201098,0.0020835595,0.010255279],"category_scores_gemma":[0.024804503,0.00056987855,0.0012986498,0.0012010038,0.0010091177,0.0013680467,0.0017929145,0.002235031,0.00074897415],"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":[0.000103095044,0.000060850914,0.0021288737,0.00008359321,0.00011509664,0.00016302387,0.00019662594,0.8979198,0.0001822691,0.08327396,0.0019432101,0.0138296345],"study_design_scores_gemma":[0.000042395353,0.000028072225,0.00034604393,0.00001894075,0.000022301592,0.000029011713,0.00003831888,0.9255276,0.000060364655,0.07238569,0.0014868849,0.00001437838],"about_ca_topic_score_codex":0.018183004,"about_ca_topic_score_gemma":0.028101765,"teacher_disagreement_score":0.018183004,"about_ca_system_score_codex":0.0011064059,"about_ca_system_score_gemma":0.001426525,"threshold_uncertainty_score":0.03615433},"labels":[],"label_agreement":null},{"id":"W3006757787","doi":"10.1037/met0000254","title":"Comparing Gaussian graphical models with the posterior predictive distribution and Bayesian model selection.","year":2020,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Mental Health Research Topics","field":"Psychology","cited_by":86,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Toronto Metropolitan University","funders":"National Institute of General Medical Sciences; National Academies of Sciences, Engineering, and Medicine; National Institute on Aging; National Institutes of Health; National Science Foundation","keywords":"Bayesian probability; Posterior probability; Statistics; Model selection; Selection (genetic algorithm); Graphical model; Bayesian linear regression; Posterior predictive distribution; Gaussian; Mathematics; Econometrics; Computer science; Artificial intelligence; Bayesian inference","score_opus":0.21833325465925038,"score_gpt":0.4943070645323446,"score_spread":0.27597380987309417,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3006757787","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":[0.010909269,0.0007010056,0.98471075,0.0006578748,0.00009944214,0.00024844692,0.00053928676,0.00049313105,0.0016407156],"genre_scores_gemma":[0.26221845,0.0014042044,0.7289654,0.00067078264,0.00029474174,0.0020747818,0.0023096153,0.00045546275,0.0016065403],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9630454,0.031518415,0.00077474234,0.0024312795,0.0018931774,0.00033698155],"domain_scores_gemma":[0.7966413,0.1884166,0.005234547,0.005960328,0.002948278,0.0007990101],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0420909,0.0020790861,0.0021731963,0.0052298266,0.0011255635,0.004052269,0.0033522919,0.0027917218,0.0072974805],"category_scores_gemma":[0.1947501,0.0009630192,0.0038949337,0.004364447,0.002967931,0.0042833304,0.0037377314,0.003994281,0.0010355647],"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":[0.00053682626,0.00020010342,0.017224867,0.0010702438,0.002522024,0.00030478887,0.0009942526,0.40784466,0.00060358044,0.36253452,0.009343538,0.19682065],"study_design_scores_gemma":[0.000093330695,0.000116687734,0.002576567,0.00016648724,0.00019650564,0.00010085123,0.0001270203,0.6414238,0.00032798405,0.35062924,0.004159578,0.000081864666],"about_ca_topic_score_codex":0.010752085,"about_ca_topic_score_gemma":0.00967298,"teacher_disagreement_score":0.0420909,"about_ca_system_score_codex":0.0028686011,"about_ca_system_score_gemma":0.0037918987,"threshold_uncertainty_score":0.22260064},"labels":[],"label_agreement":null},{"id":"W3015042751","doi":"10.1037/met0000265","title":"Partitioning variation in multilevel models for count data.","year":2020,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":67,"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 Toronto","funders":"Economic and Social Research Council","keywords":"Overdispersion; Categorical variable; Intraclass correlation; Statistics; Count data; Multilevel model; Mathematics; Cluster analysis; Negative binomial distribution; Poisson distribution; Item response theory; Variance (accounting); Binary data; Econometrics; Correlation; Binary number; Psychometrics","score_opus":0.6438736940352537,"score_gpt":0.5849986390061367,"score_spread":0.05887505502911705,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3015042751","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":[0.0015669089,0.0010638121,0.9928732,0.00069407537,0.0001526909,0.00021781497,0.00035406416,0.00027036743,0.002807109],"genre_scores_gemma":[0.1296848,0.0027389554,0.85379833,0.0014202826,0.0005400441,0.0030980804,0.0017560761,0.0008089494,0.0061544534],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9376839,0.04519148,0.0021045667,0.006140953,0.007735788,0.0011432568],"domain_scores_gemma":[0.9300846,0.05235997,0.0053360052,0.008387857,0.0033043234,0.00052736053],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04017115,0.0018408552,0.0029104634,0.0040908023,0.0016448833,0.0058139013,0.0068141343,0.0033279306,0.011714089],"category_scores_gemma":[0.14277917,0.0017341427,0.0063726767,0.0072244653,0.0041843834,0.007236723,0.007154553,0.0077463174,0.003491536],"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":[0.000030178095,0.000030359955,0.0026820612,0.00054318423,0.0005326602,0.00020955381,0.0013496962,0.03314344,0.00021945623,0.91411555,0.0065630525,0.040580746],"study_design_scores_gemma":[0.000019232055,0.00005860321,0.0014869476,0.00032195915,0.00013942528,0.0002327171,0.00024812468,0.1557351,0.0001819373,0.8233667,0.018137285,0.00007197286],"about_ca_topic_score_codex":0.009594896,"about_ca_topic_score_gemma":0.011835363,"teacher_disagreement_score":0.04017115,"about_ca_system_score_codex":0.00493773,"about_ca_system_score_gemma":0.0034013677,"threshold_uncertainty_score":0.21244788},"labels":[],"label_agreement":null},{"id":"W3041079116","doi":"10.1037/met0000422","title":"Parsimony in model selection: Tools for assessing fit propensity.","year":2021,"lang":"en","type":"preprint","venue":"Psychological Methods","topic":"Mental Health Research Topics","field":"Psychology","cited_by":9,"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":"Model selection; Occam's razor; Computer science; Range (aeronautics); Econometrics; Equating; Statistical model; Statistical inference; Selection (genetic algorithm); Counterfactual conditional; Inference; Statistics; Mathematics; Machine learning; Artificial intelligence; Counterfactual thinking; Psychology","score_opus":0.7544020784369739,"score_gpt":0.6733340687936658,"score_spread":0.08106800964330818,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3041079116","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":[0.0055711926,0.001124883,0.9837573,0.0013028296,0.0001320887,0.0006531931,0.0017673533,0.0030636813,0.0026273972],"genre_scores_gemma":[0.104940124,0.0009242592,0.8843394,0.00065434806,0.00019273936,0.0038010355,0.0023172759,0.0021308723,0.000700002],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.93441325,0.050955884,0.00426575,0.0036807726,0.0061480445,0.00053641817],"domain_scores_gemma":[0.5170145,0.44405293,0.011826674,0.018683568,0.0069866376,0.0014357293],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.10035762,0.005691072,0.004204773,0.020226313,0.0025500318,0.008657505,0.0047638123,0.0041943276,0.017349169],"category_scores_gemma":[0.44817844,0.0022266337,0.007271533,0.01806249,0.0045569767,0.009117061,0.00962619,0.009362597,0.0034028713],"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":[0.0006684912,0.00048735304,0.058328137,0.006157413,0.009102545,0.0013386045,0.0065167923,0.07859029,0.0014532278,0.2753175,0.06953854,0.49250105],"study_design_scores_gemma":[0.00025606912,0.00024757738,0.008125499,0.0017026522,0.0011605273,0.00070647785,0.0012200279,0.297855,0.0010750303,0.6649706,0.022385981,0.00029463565],"about_ca_topic_score_codex":0.0024781772,"about_ca_topic_score_gemma":0.003570151,"teacher_disagreement_score":0.10035762,"about_ca_system_score_codex":0.0022192376,"about_ca_system_score_gemma":0.0044608074,"threshold_uncertainty_score":0.5307482},"labels":[],"label_agreement":null},{"id":"W3042438175","doi":"10.1037/met0000336","title":"An approach to structural equation modeling with both factors and components: Integrated generalized structured component analysis.","year":2020,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Advanced Statistical Modeling Techniques","field":"Computer Science","cited_by":78,"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 Research Foundation of Korea","keywords":"Structural equation modeling; Component (thermodynamics); Path analysis (statistics); Component analysis; Factor analysis; Factor (programming language); Econometrics; Computer science; Data mining; Statistics; Mathematics","score_opus":0.19276253301681595,"score_gpt":0.43470238270303097,"score_spread":0.241939849686215,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3042438175","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":[0.0031343233,0.0027090071,0.9869423,0.0021331797,0.00023615053,0.0005381744,0.0006979462,0.0004294303,0.0031794412],"genre_scores_gemma":[0.08891933,0.003549823,0.90094155,0.0008050272,0.00026075594,0.0026860216,0.0014273465,0.0001512377,0.0012590096],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.98013365,0.015488152,0.0006558956,0.0015624781,0.0019386287,0.00022111188],"domain_scores_gemma":[0.98669994,0.008783266,0.0014955993,0.0012427975,0.0014706558,0.00030777155],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0154333515,0.0023539213,0.0018910816,0.0067637134,0.0010660854,0.0034574857,0.0030691628,0.002104419,0.0053764046],"category_scores_gemma":[0.0443062,0.0010888993,0.0032354742,0.012322396,0.002791804,0.005052373,0.0041314443,0.0051271357,0.0015650429],"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":[0.00007535944,0.00023857722,0.018274352,0.002016549,0.0032204273,0.00023967802,0.0031043985,0.04828563,0.0009433096,0.49727827,0.023350382,0.4029731],"study_design_scores_gemma":[0.00007560266,0.0002630504,0.008927808,0.0011649705,0.00079527387,0.0002718491,0.0010780336,0.20903745,0.0006222533,0.7293556,0.048210926,0.00019727937],"about_ca_topic_score_codex":0.008062708,"about_ca_topic_score_gemma":0.013935941,"teacher_disagreement_score":0.0154333515,"about_ca_system_score_codex":0.003179582,"about_ca_system_score_gemma":0.008241829,"threshold_uncertainty_score":0.081620336},"labels":[],"label_agreement":null},{"id":"W3132372564","doi":"10.1037/met0000269","title":"Determining the number of factors using parallel analysis and its recent variants: Comment on Lim and Jahng (2019).","year":2021,"lang":"en","type":"letter","venue":"Psychological Methods","topic":"Advanced Statistical Modeling Techniques","field":"Computer Science","cited_by":8,"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 du Québec à Montréal","funders":"","keywords":"Mathematics; Statistics; Ranking (information retrieval); Rank (graph theory); Context (archaeology); Dimension (graph theory); Eigenvalues and eigenvectors; Factor analysis; Population; Algorithm; Computer science; Combinatorics; Artificial intelligence","score_opus":0.21819536965108474,"score_gpt":0.489994029440274,"score_spread":0.27179865978918927,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3132372564","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0004090314,0.0100781815,0.0031611386,0.96686447,0.017113898,0.000029668448,0.00032645834,0.00011930427,0.0018978522],"genre_scores_gemma":[0.008660592,0.0055447267,0.008466172,0.94608945,0.027696073,0.00022935322,0.000089000016,0.00013181126,0.0030927963],"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9803217,0.0058261375,0.0023757517,0.0037408057,0.007126953,0.000608674],"domain_scores_gemma":[0.8811979,0.08238791,0.0043959497,0.004732182,0.02549784,0.0017881844],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.032335892,0.0015653237,0.0016879684,0.001938753,0.003562299,0.0035162936,0.008801386,0.021478852,0.004625198],"category_scores_gemma":[0.13105173,0.0009563313,0.0022121556,0.0023249558,0.015508271,0.012569954,0.0034082136,0.038635332,0.007359929],"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":[0.00014353935,0.00004457894,0.0011539537,0.00041740658,0.00005276405,0.00042232726,0.0007030629,0.00026798222,0.0002337153,0.052577376,0.910893,0.033090267],"study_design_scores_gemma":[0.00020067458,0.00013692467,0.004279833,0.0014381862,0.00009327181,0.0017231352,0.00087457214,0.0020783155,0.0010902608,0.18628578,0.80140007,0.00039893342],"about_ca_topic_score_codex":0.02279153,"about_ca_topic_score_gemma":0.021405295,"teacher_disagreement_score":0.9676641,"about_ca_system_score_codex":0.0046922285,"about_ca_system_score_gemma":0.0052881595,"threshold_uncertainty_score":0.17101061},"labels":[],"label_agreement":null},{"id":"W3190043208","doi":"10.1037/met0000678","title":"Why multiple hypothesis test corrections provide poor control of false positives in the real world.","year":2024,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Prioris.ai (Canada)","funders":"","keywords":"False positive paradox; Statistical hypothesis testing; False positives and false negatives; Statistics; Test (biology); Econometrics; Psychology; Multiple comparisons problem; Statistical analysis; Control (management); Mathematics; Artificial intelligence; Computer science","score_opus":0.7374936027261868,"score_gpt":0.6119179092144343,"score_spread":0.12557569351175246,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3190043208","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":[0.024448412,0.040728383,0.78627163,0.07832686,0.035103165,0.004280677,0.007652974,0.0068262476,0.016361672],"genre_scores_gemma":[0.3649772,0.0052401912,0.5814423,0.025206732,0.005343996,0.009131646,0.0018672124,0.002378675,0.0044120965],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.35887063,0.5588945,0.03073232,0.024415651,0.025315834,0.0017710528],"domain_scores_gemma":[0.10934628,0.8302451,0.017222043,0.027914615,0.014158343,0.0011136929],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.31882522,0.0026071984,0.0039024837,0.0078044934,0.0018067574,0.005136908,0.009031539,0.004747279,0.013960508],"category_scores_gemma":[0.8352908,0.0019987482,0.005044354,0.010793418,0.008544682,0.008909599,0.00443808,0.011109638,0.0031526107],"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":[0.0040309336,0.00032761443,0.04070875,0.029150603,0.031590648,0.002000489,0.005886373,0.00714062,0.0028473898,0.07724044,0.36754784,0.4315283],"study_design_scores_gemma":[0.004708773,0.0019158539,0.04881249,0.02138567,0.017468723,0.0026624645,0.0030633395,0.08793867,0.013414999,0.5831066,0.21449657,0.001025896],"about_ca_topic_score_codex":0.0032925897,"about_ca_topic_score_gemma":0.0034492577,"teacher_disagreement_score":0.68117476,"about_ca_system_score_codex":0.0023002,"about_ca_system_score_gemma":0.005274585,"threshold_uncertainty_score":0.8400098},"labels":[],"label_agreement":null},{"id":"W3206560547","doi":"10.1037/met0000414","title":"Using copulas to enable causal inference from nonexperimental data: Tutorial and simulation studies.","year":2021,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Mental Health Research Topics","field":"Psychology","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":"Employment and Social Development Canada","funders":"","keywords":"Skewness; Copula (linguistics); Confounding; Econometrics; Statistics; Sample size determination; Covariate; Psychology; Inference; Causal inference; Variables; Personality; Statistical inference; Computer science; Mathematics; Social psychology; Artificial intelligence","score_opus":0.8249197191980218,"score_gpt":0.748911382878569,"score_spread":0.07600833631945281,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3206560547","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":[0.0015048541,0.0018846482,0.9929113,0.00068753795,0.00012332195,0.00020619362,0.00014018203,0.00041889778,0.0021230893],"genre_scores_gemma":[0.03211374,0.004767119,0.9594219,0.00033387265,0.00015999003,0.0013336465,0.00021886989,0.00024975758,0.0014011286],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99004376,0.008625277,0.0002861403,0.0003450038,0.00060253765,0.00009743208],"domain_scores_gemma":[0.8007333,0.19091167,0.0022239035,0.0035158433,0.0022042391,0.00041104667],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.035704505,0.0020159963,0.0016595481,0.002594975,0.0006117716,0.0021086652,0.0024262122,0.0019336141,0.012457728],"category_scores_gemma":[0.13155653,0.0016796364,0.002580552,0.003061901,0.0015071385,0.0039930632,0.0027096793,0.0037562307,0.0013927898],"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":[0.00012597496,0.00031668835,0.0045748083,0.0023504905,0.0007538047,0.00073128176,0.0012626224,0.19000776,0.0007727758,0.5041732,0.022973077,0.27195746],"study_design_scores_gemma":[0.00008554471,0.00012996886,0.0013769884,0.00076132827,0.00011827718,0.0004457587,0.00021681513,0.6571501,0.0007797098,0.31200382,0.026830388,0.00010114332],"about_ca_topic_score_codex":0.005197397,"about_ca_topic_score_gemma":0.004860456,"teacher_disagreement_score":0.035704505,"about_ca_system_score_codex":0.0013544449,"about_ca_system_score_gemma":0.0021781859,"threshold_uncertainty_score":0.18882573},"labels":[],"label_agreement":null},{"id":"W34647757","doi":"10.1037/met0000422","title":"Juan Ramón Barat - Selection of Poems Translated into English","year":2010,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Comparative Literary Analysis and Criticism","field":"Arts and Humanities","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Poetry; Selection (genetic algorithm); Art; Media studies; Library science; Sociology; Literature; Computer science; Artificial intelligence","score_opus":0.09039564540262292,"score_gpt":0.43068538723789135,"score_spread":0.34028974183526844,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W34647757","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.035619818,0.033790328,0.017933365,0.042303476,0.016077105,0.00012632848,0.0014727943,0.00072099303,0.8519558],"genre_scores_gemma":[0.42807156,0.021476414,0.012982759,0.0043542627,0.0038050374,0.00017103745,0.0011854159,0.0015597192,0.52639383],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","domain_scores_codex":[0.9989255,0.00065666717,0.000032024418,0.00016680802,0.00019170743,0.000027188597],"domain_scores_gemma":[0.9980361,0.0014008872,0.00007875907,0.00011071677,0.000308682,0.000064849235],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0014240362,0.0003844271,0.00040389638,0.0013700391,0.0019241944,0.0024290702,0.0003491965,0.00045366614,0.031649034],"category_scores_gemma":[0.008300794,0.00015661068,0.00020171973,0.0010280664,0.0017902256,0.0021846842,0.0010449816,0.001092777,0.006236567],"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":[0.00016330829,0.000039280632,0.0036249969,0.00048401073,0.000023453478,0.00045502625,0.015483759,0.00036449815,0.00056860904,0.37752485,0.3644299,0.23683834],"study_design_scores_gemma":[0.000009269866,0.000014984356,0.0025408142,0.0003394875,0.0000062807626,0.0002132612,0.0030517988,0.00049205974,0.00030040395,0.030356994,0.96266294,0.000011740064],"about_ca_topic_score_codex":0.005024605,"about_ca_topic_score_gemma":0.008467438,"teacher_disagreement_score":0.031649034,"about_ca_system_score_codex":0.00110674,"about_ca_system_score_gemma":0.00056105905,"threshold_uncertainty_score":0.105876505},"labels":[],"label_agreement":null},{"id":"W4205105826","doi":"10.1037/met0000295","title":"Intermittent faking of personality profiles in high-stakes assessments: A grade of membership analysis.","year":2022,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Kellogg's (Canada)","funders":"","keywords":"PsycINFO; Psychology; Item response theory; Test (biology); Scale (ratio); Psychometrics; Personality; Social psychology; Big Five personality traits; Ideal (ethics); Test validity; Control (management); Statistics; Applied psychology; Computer science; Artificial intelligence; Clinical psychology; Mathematics","score_opus":0.7410462818435025,"score_gpt":0.6411131628585449,"score_spread":0.0999331189849576,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4205105826","genre_codex":"empirical","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.94917744,0.0001562328,0.04539687,0.00017933067,0.000031924832,0.00037690924,0.00046198827,0.00010308411,0.0041161864],"genre_scores_gemma":[0.9881287,0.000026132508,0.010816657,0.00001686717,0.0000069079556,0.00011312798,0.00022590635,0.000010058084,0.0006555002],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9911015,0.005646792,0.00061866787,0.0008244654,0.0015235762,0.00028510133],"domain_scores_gemma":[0.94261575,0.03163179,0.011877227,0.007904046,0.0048462385,0.0011249213],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.018106857,0.00044984452,0.00041450257,0.0025142604,0.00081018347,0.0023031926,0.0009946347,0.00064323825,0.002193118],"category_scores_gemma":[0.07727131,0.00022031649,0.00080446026,0.0012861626,0.001012372,0.0021386996,0.0027407855,0.001181731,0.0004594414],"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":[0.0005896358,0.00019242738,0.87911695,0.00013128329,0.00021663021,0.000119991535,0.003204338,0.0014503817,0.00080744876,0.005126709,0.0010167407,0.10802744],"study_design_scores_gemma":[0.000022890226,0.000443173,0.9315326,0.00011900014,0.0001627886,0.00040182337,0.0033727963,0.052303035,0.0016103275,0.007891837,0.0020899412,0.000049772683],"about_ca_topic_score_codex":0.0015524611,"about_ca_topic_score_gemma":0.0032885377,"teacher_disagreement_score":0.018106857,"about_ca_system_score_codex":0.0008508851,"about_ca_system_score_gemma":0.00049606175,"threshold_uncertainty_score":0.09575939},"labels":[],"label_agreement":null},{"id":"W4205329525","doi":"10.1037/met0000445","title":"New computations for RMSEA and CFI following FIML and TS estimation with missing data.","year":2022,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Blind Source Separation Techniques","field":"Computer Science","cited_by":20,"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 British Columbia; York University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Structural equation modeling; Missing data; Statistics; Computation; Estimation; Econometrics; Mathematics; Algorithm; Economics","score_opus":0.1610506147015826,"score_gpt":0.5012766047666443,"score_spread":0.34022599006506166,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4205329525","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":[0.0067563727,0.0011621147,0.9853941,0.0007036107,0.0005712221,0.00041916058,0.001080999,0.0010748891,0.002837457],"genre_scores_gemma":[0.09013913,0.0010154884,0.8993604,0.00062668603,0.00027555955,0.0019065884,0.0033380743,0.0006764257,0.0026615886],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9658719,0.015527169,0.0037422152,0.004702978,0.0093689095,0.00078685046],"domain_scores_gemma":[0.86635494,0.08127187,0.013878138,0.017412856,0.020242201,0.0008400775],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.032124367,0.0031359238,0.0020337782,0.0070434227,0.0015329869,0.0031094563,0.0045713554,0.002733408,0.008306213],"category_scores_gemma":[0.2315511,0.0014562252,0.004430895,0.01148463,0.0020415878,0.0062988135,0.0033934794,0.0085469,0.0031494698],"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":[0.0002841559,0.00046230745,0.038995955,0.0025662824,0.0027515003,0.0008742027,0.0028591705,0.16033271,0.0046445364,0.10482052,0.049676947,0.6317317],"study_design_scores_gemma":[0.00020924809,0.00049862335,0.031562384,0.0014683468,0.0008947672,0.0014628915,0.0014158522,0.7470121,0.010142044,0.13845736,0.06607278,0.00080363866],"about_ca_topic_score_codex":0.014288245,"about_ca_topic_score_gemma":0.022431562,"teacher_disagreement_score":0.032124367,"about_ca_system_score_codex":0.0027972802,"about_ca_system_score_gemma":0.0076317657,"threshold_uncertainty_score":0.16989195},"labels":[],"label_agreement":null},{"id":"W4220760233","doi":"10.1037/met0000448","title":"How to think clearly about the central limit theorem.","year":2022,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Statistics Education and Methodologies","field":"Mathematics","cited_by":44,"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 British Columbia; York University","funders":"University of British Columbia","keywords":"Sample (material); Central limit theorem; PsycINFO; Sample size determination; Psychology; Mathematics education; Statistics; Mathematics; Political science; Law; MEDLINE","score_opus":0.43218665808245893,"score_gpt":0.5531479969564407,"score_spread":0.12096133887398175,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4220760233","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":[0.0014461582,0.06587698,0.47468653,0.40681508,0.01963266,0.00034560548,0.0010326535,0.0007444866,0.029419895],"genre_scores_gemma":[0.10993032,0.057875983,0.57212394,0.21560867,0.023676297,0.0029220919,0.0010268813,0.001462042,0.015373715],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.93082726,0.041033484,0.005645779,0.0070471857,0.014392425,0.0010538197],"domain_scores_gemma":[0.74240303,0.21288674,0.010015016,0.012735803,0.01953736,0.002421998],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.08691233,0.001903424,0.0029592742,0.00544852,0.0032053385,0.016613055,0.007432477,0.015332666,0.009779147],"category_scores_gemma":[0.3092961,0.0014539569,0.0027910275,0.0054045646,0.027546369,0.038543742,0.0052122106,0.028422693,0.007804025],"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":[0.000054001583,0.00005000826,0.0008856745,0.001212669,0.00014861947,0.00025070293,0.0023075761,0.0014687113,0.0002008485,0.86260897,0.07807524,0.052737024],"study_design_scores_gemma":[0.000023003393,0.000023530181,0.00022941381,0.00080409006,0.000021961801,0.0002067847,0.0003836979,0.0011992006,0.00022640616,0.8916639,0.10515757,0.000060570273],"about_ca_topic_score_codex":0.007031737,"about_ca_topic_score_gemma":0.0033704257,"teacher_disagreement_score":0.08691233,"about_ca_system_score_codex":0.006077698,"about_ca_system_score_gemma":0.010841061,"threshold_uncertainty_score":0.4596418},"labels":[],"label_agreement":null},{"id":"W4236539520","doi":"10.1037/1082-989x.10.3.285.supp","title":"Supplemental Material for Structural Equation Modeling of Paired-Comparison and Ranking Data","year":2005,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Impact of AI and Big Data on Business and Society","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Structural equation modeling; Ranking (information retrieval); Statistics; Econometrics; Mathematics; Psychology; Computer science; Artificial intelligence","score_opus":0.6691282874472614,"score_gpt":0.6089437633112126,"score_spread":0.0601845241360488,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4236539520","genre_codex":"dataset","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.024245251,0.00028990326,0.2886216,0.0025599895,0.0009574525,0.0052207205,0.64868474,0.007880496,0.0215399],"genre_scores_gemma":[0.1593658,0.00057431275,0.39416328,0.0012111176,0.0005108969,0.03041427,0.34126365,0.0056908936,0.06680577],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99790454,0.0012926743,0.0001465802,0.00020610218,0.00033349206,0.000116738505],"domain_scores_gemma":[0.92387503,0.06745393,0.0010865154,0.0031958786,0.004043295,0.00034541485],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0059372773,0.0016751247,0.0013843958,0.0017638131,0.0011022183,0.0015700758,0.0030297858,0.0019185496,0.53375626],"category_scores_gemma":[0.05828729,0.0010833384,0.0016892819,0.0023811336,0.00029185574,0.0016786645,0.00089185557,0.002812867,0.066723965],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00046852336,0.002601802,0.010339739,0.00090472563,0.0004219464,0.0002533281,0.0008602135,0.00966944,0.0003890197,0.035121005,0.83657545,0.102394715],"study_design_scores_gemma":[0.002744424,0.0009787403,0.08549475,0.00092394353,0.001333965,0.0010595237,0.0019295057,0.14539379,0.0032841193,0.2578322,0.49856457,0.00046056104],"about_ca_topic_score_codex":0.018552268,"about_ca_topic_score_gemma":0.026770854,"teacher_disagreement_score":0.53375626,"about_ca_system_score_codex":0.0012215746,"about_ca_system_score_gemma":0.0021195998,"threshold_uncertainty_score":0.66504014},"labels":[],"label_agreement":null},{"id":"W4280565590","doi":"10.1037/met0000441","title":"Measurement invariance testing using confirmatory factor analysis and alignment optimization: A tutorial for transparent analysis planning and reporting.","year":2022,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Mental Health Research Topics","field":"Psychology","cited_by":170,"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":"Measurement invariance; Computer science; PsycINFO; Confirmatory factor analysis; Key (lock); Field (mathematics); Data science; Data mining; Management science; Machine learning; Structural equation modeling; Mathematics","score_opus":0.728377638129674,"score_gpt":0.6079763110290402,"score_spread":0.12040132710063389,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4280565590","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":[0.00027495562,0.0055156783,0.96595687,0.0032953152,0.0009373831,0.0025904498,0.0054955077,0.010698174,0.0052356734],"genre_scores_gemma":[0.0013948813,0.0040178085,0.9786179,0.000833104,0.00030861972,0.008137698,0.002189359,0.0023391757,0.0021615],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.96046287,0.026939502,0.0040426366,0.002048695,0.00607442,0.00043184956],"domain_scores_gemma":[0.8474704,0.115272716,0.00907293,0.011008608,0.01630497,0.0008702792],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0689704,0.00793692,0.002857992,0.014011549,0.0019975624,0.005257562,0.005776954,0.003762519,0.09016835],"category_scores_gemma":[0.19696787,0.0048093963,0.00516816,0.012941853,0.002804068,0.008414067,0.006483694,0.008074753,0.058884516],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00010107847,0.00017174853,0.0009116386,0.006365982,0.0003433302,0.000447629,0.0016000649,0.0027846224,0.0015535087,0.065742396,0.46019512,0.45978284],"study_design_scores_gemma":[0.00017541902,0.00015857797,0.0027382274,0.0065173,0.00020371312,0.0011817068,0.0010496774,0.019365817,0.0035019137,0.21331477,0.75133026,0.00046261383],"about_ca_topic_score_codex":0.0055592414,"about_ca_topic_score_gemma":0.008730284,"teacher_disagreement_score":0.9310296,"about_ca_system_score_codex":0.0029650973,"about_ca_system_score_gemma":0.009449416,"threshold_uncertainty_score":0.36475468},"labels":[],"label_agreement":null},{"id":"W4280596915","doi":"10.1037/met0000497","title":"Causal mediation effects in single case experimental designs.","year":2022,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Behavioral and Psychological Studies","field":"Psychology","cited_by":19,"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 Institute on Drug Abuse","keywords":"Mediation; Type I and type II errors; Statistics; Confidence interval; Econometrics; Estimator; Statistical power; Monte Carlo method; Causal model; Psychology; Standard error; Outcome (game theory); Statistical hypothesis testing; Mathematics","score_opus":0.551208401110505,"score_gpt":0.5270712642831462,"score_spread":0.02413713682735874,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4280596915","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":[0.07955147,0.003981174,0.70536923,0.0024158051,0.005454976,0.13699797,0.0065181595,0.0021817791,0.057529468],"genre_scores_gemma":[0.21346869,0.0006989199,0.499655,0.0011376931,0.00043650492,0.27912164,0.0009903079,0.00033940517,0.004151937],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.5279834,0.39418188,0.021248791,0.025768684,0.028307766,0.002509447],"domain_scores_gemma":[0.2539995,0.6622868,0.027816527,0.044174507,0.010547577,0.0011751023],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.31654316,0.0022985996,0.0033974953,0.0038754868,0.0027159876,0.0058231414,0.004977851,0.0063688126,0.049940538],"category_scores_gemma":[0.6501177,0.002972547,0.0038813576,0.004527022,0.007192034,0.0068607978,0.0048840987,0.0061570755,0.0035074994],"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":[0.033684548,0.007457565,0.04476142,0.023407632,0.0061884224,0.0011383104,0.014971199,0.013006215,0.003931523,0.2899746,0.066313125,0.49516547],"study_design_scores_gemma":[0.048547033,0.026128868,0.15013348,0.01017901,0.0092565045,0.0020143515,0.0039784345,0.12004092,0.013568163,0.47719374,0.13800803,0.0009514992],"about_ca_topic_score_codex":0.0013041018,"about_ca_topic_score_gemma":0.0022126914,"teacher_disagreement_score":0.31654316,"about_ca_system_score_codex":0.0032411183,"about_ca_system_score_gemma":0.0049426244,"threshold_uncertainty_score":0.842824},"labels":[],"label_agreement":null},{"id":"W4281687743","doi":"10.1037/met0000493","title":"Spatial analysis for psychologists: How to use individual-level data for research at the geographically aggregated level.","year":2022,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Urban Transport and Accessibility","field":"Social Sciences","cited_by":36,"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":"PsycINFO; Data science; Variation (astronomy); Computer science; Spatial analysis; Variety (cybernetics); Cluster analysis; Information retrieval; Data mining; Statistics; Artificial intelligence; MEDLINE; Mathematics","score_opus":0.7344584376473206,"score_gpt":0.5972475058072007,"score_spread":0.1372109318401199,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4281687743","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":[0.0013644483,0.006542127,0.9454645,0.023861635,0.0019810535,0.00052852836,0.003129937,0.0037695158,0.013358276],"genre_scores_gemma":[0.01560713,0.005603057,0.9701808,0.0020766417,0.000555873,0.0015699872,0.0013355125,0.001061755,0.002009369],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9683293,0.02490023,0.0018019011,0.0016168922,0.0031365831,0.00021503531],"domain_scores_gemma":[0.8539773,0.11240581,0.0043115024,0.017382652,0.009947604,0.0019751661],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.046927348,0.001494071,0.001893005,0.0077909604,0.0023968676,0.008091483,0.0034273118,0.0023988765,0.01955438],"category_scores_gemma":[0.2087995,0.0013449488,0.0023932753,0.013141892,0.006389398,0.0140340505,0.0072497525,0.0064543127,0.009601854],"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":[0.000053938602,0.00008732843,0.006036417,0.002702864,0.00045294894,0.00019605938,0.008143714,0.0024977983,0.0006184165,0.16351491,0.27991676,0.53577894],"study_design_scores_gemma":[0.00005280161,0.00006793822,0.006532982,0.0025949234,0.00012664842,0.00032858434,0.0054245694,0.008015831,0.00066711823,0.60131705,0.37467986,0.00019169482],"about_ca_topic_score_codex":0.006929461,"about_ca_topic_score_gemma":0.011839418,"teacher_disagreement_score":0.046927348,"about_ca_system_score_codex":0.0017907524,"about_ca_system_score_gemma":0.005642847,"threshold_uncertainty_score":0.24817848},"labels":[],"label_agreement":null},{"id":"W4283816293","doi":"10.1037/met0000512","title":"Regression with reduced rank predictor matrices: A model of trade-offs.","year":2022,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Multi-Criteria Decision Making","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Learning Partnership","funders":"","keywords":"Statistics; Mathematics; Linear regression; Regression analysis; Constant (computer programming); Regression; Rank (graph theory); Regression diagnostic; Variable (mathematics); Variables; Econometrics; Combinatorics; Polynomial regression; Computer science","score_opus":0.41604150202304757,"score_gpt":0.556736739686867,"score_spread":0.14069523766381942,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4283816293","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":[0.059101786,0.0011036432,0.9253974,0.0021232255,0.0001677427,0.00032528286,0.0009200624,0.0011415449,0.009719307],"genre_scores_gemma":[0.74140453,0.0017298468,0.20008884,0.0006027141,0.0002791368,0.0013544485,0.0015892431,0.00064287015,0.05230849],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9838643,0.01228536,0.00029801184,0.0016285791,0.00091712124,0.0010065333],"domain_scores_gemma":[0.9460876,0.044435795,0.0035266534,0.0029157237,0.0022217263,0.0008123995],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.025959643,0.0023572429,0.002125572,0.0021282528,0.0009519971,0.004807343,0.0036723223,0.0027148537,0.01896547],"category_scores_gemma":[0.077012174,0.0021243829,0.002059653,0.0031307465,0.0024371974,0.004582515,0.0024419054,0.0042988206,0.007808356],"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":[0.0007637267,0.0004298233,0.014514186,0.00028156795,0.0007905864,0.00048310574,0.00073431816,0.6101508,0.0005618686,0.2597773,0.010184528,0.101328164],"study_design_scores_gemma":[0.0000657625,0.0001266315,0.0015420446,0.00006309197,0.00007451377,0.00012028109,0.000056072793,0.8948931,0.000104769555,0.10073866,0.002174951,0.00004018026],"about_ca_topic_score_codex":0.011733769,"about_ca_topic_score_gemma":0.0104525555,"teacher_disagreement_score":0.025959643,"about_ca_system_score_codex":0.0022960652,"about_ca_system_score_gemma":0.0024198983,"threshold_uncertainty_score":0.13728935},"labels":[],"label_agreement":null},{"id":"W4284894098","doi":"10.1037/met0000487","title":"Tutorial: Artificial neural networks to analyze single-case experimental designs.","year":2022,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Advanced Statistical Modeling Techniques","field":"Computer Science","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal; Institut universitaire en santé mentale de Montréal; Institut Universitaire en Santé Mentale de Québec","funders":"Fonds de Recherche du Québec - Santé","keywords":"Artificial neural network; Computer science; Artificial intelligence; Replication (statistics); PsycINFO; Machine learning; Monte Carlo method; Statistics; MEDLINE; Mathematics","score_opus":0.301411646461321,"score_gpt":0.49046597917977713,"score_spread":0.18905433271845612,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4284894098","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":[0.00045543828,0.013176426,0.9724458,0.002229614,0.0009378243,0.00064617494,0.002154466,0.0030001735,0.004954098],"genre_scores_gemma":[0.0062909075,0.01885826,0.9595345,0.0014421602,0.0007509467,0.0045597814,0.0022577946,0.000637557,0.0056680464],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.99521625,0.0032784005,0.00036078677,0.00032638476,0.00076161005,0.000056510056],"domain_scores_gemma":[0.9676636,0.029270865,0.0008735981,0.00073797547,0.001237108,0.00021687835],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.013864516,0.0031088013,0.0013735258,0.0034424085,0.0002995335,0.0013624162,0.0022704948,0.0020530098,0.029577691],"category_scores_gemma":[0.05389495,0.0011778645,0.0017392634,0.00307146,0.0006828887,0.0025064775,0.0013952618,0.0043345015,0.008855274],"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":[0.00008399041,0.00019332068,0.0013301602,0.0059634456,0.0007939281,0.00027636954,0.000298467,0.036130495,0.0011622073,0.06440684,0.18454683,0.704814],"study_design_scores_gemma":[0.00021038974,0.000250872,0.0043354174,0.0049285674,0.00035906615,0.00075637305,0.00015236593,0.21250021,0.0024741988,0.3290735,0.44476208,0.00019690632],"about_ca_topic_score_codex":0.0024361336,"about_ca_topic_score_gemma":0.004646305,"teacher_disagreement_score":0.029577691,"about_ca_system_score_codex":0.0012391177,"about_ca_system_score_gemma":0.001954306,"threshold_uncertainty_score":0.09894729},"labels":[],"label_agreement":null},{"id":"W4285730034","doi":"10.1037/met0000508","title":"Correcting bias in extreme groups design using a missing data approach.","year":2022,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University; University of British Columbia","funders":"","keywords":"Missing data; PsycINFO; Statistics; Data collection; Statistical power; Econometrics; Computer science; Psychology; Data mining; Mathematics; MEDLINE; Political science","score_opus":0.8796631252223159,"score_gpt":0.5764435918938456,"score_spread":0.3032195333284703,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4285730034","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":[0.0015982513,0.00032567902,0.9932092,0.0006233011,0.00028356723,0.0012986388,0.00072640576,0.0009734279,0.0009615135],"genre_scores_gemma":[0.018068792,0.00018805107,0.97407997,0.00039880179,0.00011093606,0.0057725064,0.00044804058,0.00025718837,0.0006757608],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.68013895,0.2881325,0.008308144,0.010825424,0.011540971,0.0010540361],"domain_scores_gemma":[0.51918316,0.41816938,0.014951053,0.037949592,0.008776949,0.00096991495],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.21943875,0.0026667123,0.003240548,0.00488132,0.0026403163,0.00417852,0.007266161,0.0045255865,0.032599907],"category_scores_gemma":[0.48033747,0.0022285136,0.0057503884,0.0058014346,0.0051625855,0.0046146573,0.006716487,0.0072443527,0.0045477734],"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":[0.0014287801,0.00038396334,0.010948137,0.0042973426,0.005046571,0.00046894385,0.0035450524,0.015828697,0.0015271571,0.16923356,0.03663296,0.7506588],"study_design_scores_gemma":[0.0018288398,0.0022499566,0.014820216,0.003581533,0.0031212005,0.000946323,0.0010489316,0.14681943,0.009314166,0.73994,0.075854756,0.0004747376],"about_ca_topic_score_codex":0.0021858697,"about_ca_topic_score_gemma":0.0029837282,"teacher_disagreement_score":0.78056127,"about_ca_system_score_codex":0.0019635884,"about_ca_system_score_gemma":0.0055469414,"threshold_uncertainty_score":0.96257097},"labels":[],"label_agreement":null},{"id":"W4303188607","doi":"10.1037/met0000514","title":"Aberrant distortion of variance components in multilevel models under conflation of level-specific effects.","year":2022,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Mental Health Research Topics","field":"Psychology","cited_by":7,"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; Variance (accounting); Random effects model; Inference; Multilevel model; Statistics; Econometrics; Distortion (music); Cluster (spacecraft); Mathematics; Psychology; Computer science; Artificial intelligence; Meta-analysis; Epistemology; Economics","score_opus":0.5737664376662643,"score_gpt":0.5591809436304295,"score_spread":0.014585494035834867,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4303188607","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":[0.028722145,0.0010070695,0.95688033,0.0038199374,0.00026458807,0.00034724348,0.00035824502,0.00048187625,0.008118563],"genre_scores_gemma":[0.7138359,0.00073440105,0.27880964,0.0018463816,0.00023108434,0.001131737,0.00043253548,0.0003682684,0.0026100606],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.8655443,0.1064801,0.0057114833,0.010801158,0.010123861,0.0013390777],"domain_scores_gemma":[0.71571845,0.22009484,0.021144532,0.037585475,0.004752366,0.0007043813],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1033514,0.0011223439,0.0012160306,0.0017958567,0.0011327422,0.0055415253,0.0036148073,0.002139672,0.006109471],"category_scores_gemma":[0.38391364,0.0013693509,0.0031309908,0.0041649565,0.008901194,0.0073211254,0.007867627,0.00829902,0.0010722338],"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":[0.00027915463,0.00005648091,0.0284346,0.00057100056,0.0008342296,0.0007928911,0.007424757,0.027751796,0.0007608942,0.8274737,0.0049446295,0.10067585],"study_design_scores_gemma":[0.000060384344,0.00014920502,0.010084359,0.00046955582,0.00026704738,0.0007123416,0.00091498886,0.12136616,0.002281309,0.85391265,0.009649741,0.00013223648],"about_ca_topic_score_codex":0.0041192123,"about_ca_topic_score_gemma":0.0042712837,"teacher_disagreement_score":0.1033514,"about_ca_system_score_codex":0.0039342768,"about_ca_system_score_gemma":0.0026710301,"threshold_uncertainty_score":0.546581},"labels":[],"label_agreement":null},{"id":"W4313857752","doi":"10.1037/met0000537","title":"We need to change how we compute RMSEA for nested model comparisons in structural equation modeling.","year":2023,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Bayesian Modeling and Causal Inference","field":"Computer Science","cited_by":74,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University; University of British Columbia","funders":"","keywords":"Structural equation modeling; Nested set model; Mathematics; Statistics; Applied mathematics; Computer science; Data mining","score_opus":0.5169415510137839,"score_gpt":0.49881246533518586,"score_spread":0.018129085678598056,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4313857752","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":[0.009584147,0.008525519,0.88981384,0.06417385,0.012494469,0.0008534028,0.0010003543,0.0023592245,0.01119509],"genre_scores_gemma":[0.11200675,0.0023027612,0.85901237,0.017026838,0.0025493754,0.0027730851,0.0006747816,0.002055615,0.0015983279],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.6652672,0.27457258,0.023375265,0.013583685,0.021822281,0.0013789337],"domain_scores_gemma":[0.28367674,0.5970461,0.021094007,0.07023473,0.024868593,0.0030798416],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2832322,0.0024404805,0.004081482,0.009282061,0.0041133636,0.011587205,0.009246082,0.0049536964,0.009289659],"category_scores_gemma":[0.7561615,0.0017090889,0.006525139,0.0118818125,0.011049153,0.02384656,0.0077452427,0.022076955,0.0037878193],"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":[0.00038027862,0.0004413009,0.048588935,0.004366934,0.0050261323,0.00041405333,0.009375252,0.0067042266,0.0011762358,0.2255061,0.123384334,0.57463616],"study_design_scores_gemma":[0.00032499948,0.0008459389,0.028440893,0.008397565,0.0012594576,0.00090156845,0.0046491735,0.04251271,0.0032991418,0.7953766,0.113198385,0.0007934718],"about_ca_topic_score_codex":0.007620031,"about_ca_topic_score_gemma":0.01541703,"teacher_disagreement_score":0.7167678,"about_ca_system_score_codex":0.0041705514,"about_ca_system_score_gemma":0.00689297,"threshold_uncertainty_score":0.8839023},"labels":[],"label_agreement":null},{"id":"W4321004017","doi":"10.1037/met0000540","title":"Let the algorithm speak: How to use neural networks for automatic item generation in psychological scale development.","year":2023,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Explainable Artificial Intelligence (XAI)","field":"Computer Science","cited_by":63,"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":"Economic and Social Research Council; Cambridge Trust","keywords":"Computer science; Scale (ratio); Generator (circuit theory); Process (computing); Code (set theory); Artificial neural network; Generative grammar; Artificial intelligence; Machine learning; Algorithm; Natural language processing; Programming language","score_opus":0.23907500105885265,"score_gpt":0.46337447013331007,"score_spread":0.2242994690744574,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4321004017","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":[0.003534078,0.0002941179,0.9572025,0.0010561141,0.00032793827,0.00044839666,0.0012015526,0.030977871,0.0049574943],"genre_scores_gemma":[0.028492717,0.00018015735,0.9608816,0.00051841274,0.000056499208,0.0015366956,0.001289914,0.00196359,0.0050804988],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99892634,0.00054573,0.00010319681,0.00023619238,0.00014073124,0.00004772707],"domain_scores_gemma":[0.9950724,0.0035366581,0.0001711971,0.0005441785,0.0005782726,0.00009731863],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005567326,0.0014698788,0.0004928563,0.0012138949,0.00048481653,0.0019158517,0.0016622262,0.0014606724,0.05053993],"category_scores_gemma":[0.028612603,0.00084965926,0.0007441988,0.0012596564,0.00061496906,0.0021655364,0.001834025,0.0022426718,0.03177642],"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":[0.00017492988,0.0000698631,0.0020537784,0.00022419475,0.00008768346,0.00008882659,0.00017149294,0.0043749954,0.0016001624,0.005117249,0.0857147,0.9003221],"study_design_scores_gemma":[0.0005486223,0.0002161521,0.0065213074,0.0005768664,0.00016150321,0.00057183916,0.00052731816,0.69150966,0.016206974,0.143074,0.13993071,0.00015504072],"about_ca_topic_score_codex":0.0026493375,"about_ca_topic_score_gemma":0.0046155993,"teacher_disagreement_score":0.05053993,"about_ca_system_score_codex":0.0004903668,"about_ca_system_score_gemma":0.0011190429,"threshold_uncertainty_score":0.16907287},"labels":[],"label_agreement":null},{"id":"W4385715273","doi":"10.1037/met0000551","title":"Mixture multilevel vector-autoregressive modeling.","year":2023,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Health disparities and outcomes","field":"Social Sciences","cited_by":11,"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 Toronto","funders":"Max-Planck-Gesellschaft; Nederlandse Organisatie voor Wetenschappelijk Onderzoek","keywords":"Autoregressive model; STAR model; Multilevel model; Statistics; Econometrics; Mathematics; Nonlinear autoregressive exogenous model; Applied mathematics; Autoregressive integrated moving average; Time series","score_opus":0.2471882376413364,"score_gpt":0.5709472765805209,"score_spread":0.3237590389391845,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4385715273","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":[0.010546552,0.0011392286,0.9811947,0.00096878473,0.00016489117,0.0002799063,0.0019817904,0.00085616123,0.0028679846],"genre_scores_gemma":[0.32618418,0.002174712,0.6542864,0.00038246965,0.0002652666,0.0016042105,0.0050597214,0.00039794127,0.00964506],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99500644,0.003548447,0.00020945193,0.000630786,0.00040592052,0.00019896813],"domain_scores_gemma":[0.99332905,0.0047817808,0.0006285016,0.00059128704,0.000542598,0.00012681844],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009120262,0.0013243001,0.0017057372,0.0019981898,0.0006576693,0.0024422875,0.0032183607,0.001509816,0.011080853],"category_scores_gemma":[0.023424275,0.0007357768,0.0035307792,0.0033075477,0.0006436016,0.0022962466,0.0021444785,0.0026065898,0.0021691436],"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":[0.00028690387,0.00022055543,0.034607023,0.0008136392,0.002095892,0.0002479152,0.0013265801,0.22589563,0.0008769785,0.45027697,0.021830816,0.26152104],"study_design_scores_gemma":[0.000040525996,0.00011203115,0.005277525,0.00017104179,0.0002942965,0.00008756596,0.00018338351,0.813699,0.0002783698,0.16351897,0.016277563,0.000059696176],"about_ca_topic_score_codex":0.013311733,"about_ca_topic_score_gemma":0.016962865,"teacher_disagreement_score":0.013311733,"about_ca_system_score_codex":0.0012894677,"about_ca_system_score_gemma":0.0016890906,"threshold_uncertainty_score":0.04823309},"labels":[],"label_agreement":null},{"id":"W4388023864","doi":"10.1037/met0000596.supp","title":"Supplemental Material for Equivalence Testing for Linear Regression","year":2023,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Fault Detection and Control Systems","field":"Engineering","cited_by":1,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Equivalence (formal languages); Statistics; Regression analysis; Linear regression; Mathematics; Econometrics; Statistical hypothesis testing; Psychology; Regression; Discrete mathematics","score_opus":0.15170678181754713,"score_gpt":0.4677150525645851,"score_spread":0.31600827074703797,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4388023864","genre_codex":"methods","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"dataset","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0050098724,0.0005227693,0.6526238,0.0016439355,0.0014468784,0.00111913,0.25346178,0.032270096,0.051901802],"genre_scores_gemma":[0.06499884,0.0006725977,0.60178286,0.0016944844,0.0010648803,0.0059528532,0.24744341,0.017868351,0.058521785],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9984762,0.0005844175,0.00015317931,0.00023760177,0.00045552436,0.00009303412],"domain_scores_gemma":[0.9743001,0.020741437,0.00040164488,0.002001246,0.002367382,0.00018815312],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.003034964,0.0014170989,0.00082082447,0.0022453621,0.0006193231,0.0010361025,0.0021078326,0.0012171923,0.65845835],"category_scores_gemma":[0.03801846,0.0007833615,0.00093273335,0.0016639971,0.00028418694,0.0013505599,0.0012301444,0.0012513531,0.1272845],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00023051599,0.00068046915,0.0017488156,0.0008380185,0.00010407366,0.00014042435,0.0001097295,0.0017848326,0.001477319,0.029174766,0.7948678,0.16884327],"study_design_scores_gemma":[0.0014091358,0.00046012917,0.018428564,0.0007505735,0.0001792737,0.0014013629,0.00025968952,0.04816954,0.008958114,0.2769997,0.64281297,0.0001709205],"about_ca_topic_score_codex":0.0023459755,"about_ca_topic_score_gemma":0.004167299,"teacher_disagreement_score":0.65845835,"about_ca_system_score_codex":0.0006024718,"about_ca_system_score_gemma":0.0009967962,"threshold_uncertainty_score":0.48716772},"labels":[],"label_agreement":null},{"id":"W4388628386","doi":"10.1037/met0000596","title":"Equivalence testing for linear regression.","year":2023,"lang":"en","type":"article","venue":"Psychological Methods","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 British Columbia","funders":"","keywords":"Equivalence (formal languages); Linear regression; Statistics; PsycINFO; Mathematics; Regression analysis; Linear model; Regression; Multilevel model; Econometrics; Discrete mathematics; MEDLINE","score_opus":0.725563039131135,"score_gpt":0.6745870036714725,"score_spread":0.05097603545966245,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4388628386","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":[0.023808375,0.0054952065,0.9318801,0.004893113,0.0011821042,0.0006290859,0.0011402457,0.0007709191,0.030200897],"genre_scores_gemma":[0.59132665,0.0019563201,0.3937155,0.0016688645,0.001558453,0.0031731643,0.0018936755,0.0005301039,0.004177423],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8528596,0.12413702,0.0034403706,0.006651464,0.011989277,0.0009222511],"domain_scores_gemma":[0.57107365,0.39595315,0.007483269,0.013551814,0.010315802,0.001622269],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.06573515,0.0016218973,0.0023903763,0.0047510155,0.0017249319,0.0035980889,0.0040362044,0.0028381743,0.02250471],"category_scores_gemma":[0.454562,0.0005996124,0.0028252958,0.004718508,0.008823424,0.008438438,0.004343823,0.006048815,0.0020813795],"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":[0.00076188135,0.0002897918,0.009072073,0.0008835023,0.0009027616,0.0002675774,0.00071827037,0.009798627,0.00038417763,0.62415165,0.013888074,0.33888167],"study_design_scores_gemma":[0.00016407973,0.0003194756,0.004359657,0.00027374085,0.00013129346,0.00025433145,0.00027946994,0.03711337,0.0003744878,0.9516044,0.0050798473,0.000045884928],"about_ca_topic_score_codex":0.0020703026,"about_ca_topic_score_gemma":0.0013260305,"teacher_disagreement_score":0.93426484,"about_ca_system_score_codex":0.0021714566,"about_ca_system_score_gemma":0.0026457377,"threshold_uncertainty_score":0.34764487},"labels":[],"label_agreement":null},{"id":"W4388730146","doi":"10.1037/met0000616","title":"Tutorial: Assessing the impact of nonignorable missingness on regression analysis using Index of Local Sensitivity to Nonignorability.","year":2023,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Advanced Statistical Modeling Techniques","field":"Computer Science","cited_by":4,"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","funders":"Social Sciences and Humanities Research Council of Canada; Natural Sciences and Engineering Research Council of Canada; National Institutes of Health","keywords":"Missing data; Credibility; Sensitivity (control systems); Statistics; Computer science; Econometrics; Robustness (evolution); Data mining; Regression; Syntax; Regression analysis; Spurious relationship; Mathematics; Artificial intelligence","score_opus":0.16370209868771823,"score_gpt":0.5713806165324973,"score_spread":0.40767851784477904,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4388730146","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":[0.0023321419,0.007657117,0.96760386,0.0047849403,0.0015230551,0.00063833286,0.0019215387,0.0073625064,0.006176417],"genre_scores_gemma":[0.02635708,0.010781183,0.9415608,0.0030866386,0.0018605682,0.0021755276,0.0022981479,0.0042337454,0.007646242],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.98014534,0.0140696475,0.00090109435,0.001173535,0.0035347585,0.0001757053],"domain_scores_gemma":[0.738439,0.23834056,0.006623703,0.005833304,0.009477629,0.0012858574],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.042365626,0.002967416,0.0019704388,0.006028037,0.0008020034,0.0025262549,0.002592043,0.0020643263,0.0290132],"category_scores_gemma":[0.2722649,0.0013570156,0.0036299992,0.0049671377,0.0017658133,0.0043172617,0.0029979085,0.0058302595,0.009966585],"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":[0.00024816874,0.00022490033,0.007412891,0.0048363637,0.0019184102,0.00038943693,0.0008259423,0.020021921,0.0029768522,0.03534193,0.2520511,0.67375207],"study_design_scores_gemma":[0.0003243033,0.0011406237,0.031509485,0.0046812734,0.0017790882,0.0033844297,0.0008485187,0.18298934,0.011607845,0.36134788,0.3995706,0.00081658305],"about_ca_topic_score_codex":0.0020419871,"about_ca_topic_score_gemma":0.0038457282,"teacher_disagreement_score":0.042365626,"about_ca_system_score_codex":0.0012359298,"about_ca_system_score_gemma":0.0020724267,"threshold_uncertainty_score":0.2240535},"labels":[],"label_agreement":null},{"id":"W4389724851","doi":"10.1037/met0000629","title":"The case for the curve: Parametric regression with second- and third-order polynomial functions of predictors should be routine.","year":2023,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Data Analysis with R","field":"Computer Science","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":"Polynomial regression; Spurious relationship; Polynomial; Regression analysis; Nonparametric regression; Regression; Computer science; Local regression; Smoothing; Nonparametric statistics; Econometrics; Mathematics; Statistics","score_opus":0.13247304336046892,"score_gpt":0.4390515366897398,"score_spread":0.3065784933292709,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4389724851","genre_codex":"commentary","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0033183352,0.01495459,0.4605372,0.48147234,0.012057148,0.00024420116,0.00049477647,0.0019402638,0.024981072],"genre_scores_gemma":[0.11724552,0.01970236,0.6207831,0.19910249,0.015974065,0.0017050673,0.00076504925,0.0038612916,0.020861113],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.94197726,0.031143175,0.00332818,0.0103387935,0.012033246,0.0011792644],"domain_scores_gemma":[0.7031436,0.21303466,0.012434507,0.046233967,0.020230217,0.0049229753],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.11699569,0.0017345136,0.0034839704,0.0024089748,0.0038401715,0.012512333,0.0068761175,0.015481509,0.024406508],"category_scores_gemma":[0.4040792,0.0013593088,0.0031116104,0.005696163,0.031965148,0.03354392,0.008317238,0.038764082,0.016223453],"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":[0.00030255714,0.000079051555,0.0040967083,0.0011509599,0.00019450288,0.000287304,0.0023894908,0.0016521413,0.00046269453,0.6119384,0.19464454,0.18280162],"study_design_scores_gemma":[0.00007966805,0.000110707755,0.001386434,0.0017845614,0.00007915187,0.0003350154,0.00074173114,0.005276492,0.0006383934,0.782563,0.20687509,0.0001297367],"about_ca_topic_score_codex":0.006722098,"about_ca_topic_score_gemma":0.0046812203,"teacher_disagreement_score":0.8830043,"about_ca_system_score_codex":0.003881341,"about_ca_system_score_gemma":0.011273555,"threshold_uncertainty_score":0.6187397},"labels":[],"label_agreement":null},{"id":"W4393039792","doi":"10.1037/met0000507","title":"The Bayes factor, HDI-ROPE, and frequentist equivalence tests can all be reverse engineered—Almost exactly—From one another: Reply to Linde et al. (2021).","year":2024,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Statistical Methods in Clinical Trials","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 British Columbia","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Frequentist inference; Bayes factor; Rope; Bayesian probability; Bayes' theorem; Statistical hypothesis testing; Statistics; Equivalence (formal languages); Null hypothesis; Econometrics; Frequentist probability; Bayes' rule; Type I and type II errors; Mathematics; Bayesian inference; Computer science; Algorithm; Discrete mathematics","score_opus":0.6313980920682659,"score_gpt":0.6237437826978459,"score_spread":0.007654309370419998,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4393039792","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00025902063,0.0050067334,0.0024995198,0.98473614,0.00667829,0.000016086598,0.00008271646,0.00004775483,0.00067365565],"genre_scores_gemma":[0.011993256,0.004118799,0.0066851624,0.95878285,0.016642924,0.00027571485,0.00007833599,0.00006664349,0.0013562528],"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9267853,0.049760804,0.006118964,0.006301685,0.010154804,0.0008784107],"domain_scores_gemma":[0.53585535,0.42343912,0.011512885,0.011315699,0.015397063,0.0024798817],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.10715435,0.0011941008,0.0025744666,0.0025408985,0.002691622,0.0064403447,0.0073976303,0.037369385,0.0062702084],"category_scores_gemma":[0.46097443,0.0011555237,0.0021017361,0.0037230453,0.025792075,0.01572123,0.0042320057,0.051979393,0.005826052],"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":[0.00024174497,0.000053561063,0.00090001617,0.00049011863,0.0001182807,0.0002012437,0.001225984,0.0004970556,0.000072689465,0.09798967,0.8403773,0.057832345],"study_design_scores_gemma":[0.00068882917,0.0001980044,0.003024799,0.0024839295,0.00016247711,0.00073666056,0.001788651,0.00331459,0.00050477206,0.69822866,0.28841126,0.0004573986],"about_ca_topic_score_codex":0.0058075683,"about_ca_topic_score_gemma":0.0036217868,"teacher_disagreement_score":0.89284563,"about_ca_system_score_codex":0.0052247033,"about_ca_system_score_gemma":0.0064600054,"threshold_uncertainty_score":0.5666931},"labels":[],"label_agreement":null},{"id":"W4396673092","doi":"10.1037/met0000645","title":"A deep learning method for comparing Bayesian hierarchical models.","year":2024,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Gaussian Processes and Bayesian Inference","field":"Computer Science","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":"Innovation Cluster (Canada)","funders":"Deutsche Forschungsgemeinschaft; Google","keywords":"Computer science; Benchmark (surveying); Machine learning; Artificial intelligence; Inference; Bayesian inference; Model selection; Bayesian probability; Set (abstract data type); Hierarchical database model; Probabilistic logic; Code (set theory); Approximate Bayesian computation; Data mining","score_opus":0.08995888785594917,"score_gpt":0.442734058898519,"score_spread":0.35277517104256984,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4396673092","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":[0.001697578,0.00020879041,0.9958703,0.00018934211,0.000032741744,0.00011887129,0.00028107397,0.0006552815,0.0009459358],"genre_scores_gemma":[0.087944746,0.00028404838,0.9070962,0.00038713633,0.00008666846,0.00097619166,0.0012869451,0.00044550706,0.0014926447],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99416417,0.0034115221,0.00031127722,0.0009412748,0.001015034,0.0001566734],"domain_scores_gemma":[0.98141956,0.014699862,0.0010825986,0.0014938585,0.00091924134,0.000385017],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.012866945,0.0015552235,0.0014621099,0.0041964254,0.0012359739,0.002478777,0.0042963717,0.002020248,0.012635081],"category_scores_gemma":[0.058827277,0.001111009,0.002501203,0.0029575203,0.0019069758,0.0034746625,0.0052472325,0.00554209,0.0015696596],"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":[0.00025476306,0.0001975395,0.005117612,0.0007203849,0.0010801301,0.00013602793,0.00035957686,0.24982159,0.0015213317,0.2989211,0.017764535,0.4241054],"study_design_scores_gemma":[0.000052018095,0.00004250771,0.0005669311,0.00010464722,0.00007225587,0.00007052897,0.000034637447,0.7180902,0.0007478817,0.27546704,0.0047243717,0.000026979658],"about_ca_topic_score_codex":0.0075664846,"about_ca_topic_score_gemma":0.01170012,"teacher_disagreement_score":0.012866945,"about_ca_system_score_codex":0.0030258663,"about_ca_system_score_gemma":0.0052345344,"threshold_uncertainty_score":0.0680477},"labels":[],"label_agreement":null},{"id":"W4402642427","doi":"10.1037/met0000689","title":"Solving variables with Monte Carlo simulation experiments: A stochastic root-solving approach.","year":2024,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Advanced Database Systems and Queries","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":"York University","funders":"","keywords":"Monte Carlo method; Monte Carlo method in statistical physics; Statistical physics; Applied mathematics; Computer science; Statistics; Econometrics; Mathematics; Hybrid Monte Carlo; Markov chain Monte Carlo; Physics","score_opus":0.07748759442895121,"score_gpt":0.4205089106345848,"score_spread":0.3430213162056336,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4402642427","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":[0.029586636,0.00026912565,0.95853627,0.0006010007,0.00012694339,0.001818524,0.00044333926,0.0006514596,0.007966701],"genre_scores_gemma":[0.27001408,0.0003278002,0.7202843,0.00035428043,0.00007334251,0.006511974,0.00050409493,0.0001865606,0.0017435632],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9606832,0.035300653,0.0006928362,0.0017687848,0.001214756,0.00033971437],"domain_scores_gemma":[0.61430955,0.36554906,0.008272648,0.009171704,0.0018392937,0.00085780915],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04517084,0.0014999462,0.001903445,0.0015956599,0.0009329124,0.0030398532,0.003984951,0.0028615773,0.021196207],"category_scores_gemma":[0.29434055,0.0011133803,0.0012691788,0.0027367899,0.0026505913,0.004427531,0.002286328,0.0042414363,0.0009723989],"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":[0.0014441333,0.0012589832,0.005058652,0.00061446,0.00084630435,0.0001138354,0.001000873,0.43376508,0.0004130033,0.45103985,0.0071214265,0.09732338],"study_design_scores_gemma":[0.0005010419,0.00026999527,0.0006217112,0.0001334957,0.0001393473,0.000030464635,0.00010767658,0.7453506,0.00042426607,0.25001284,0.0023667938,0.000041795538],"about_ca_topic_score_codex":0.004843419,"about_ca_topic_score_gemma":0.0064152996,"teacher_disagreement_score":0.04517084,"about_ca_system_score_codex":0.0020854957,"about_ca_system_score_gemma":0.0029553107,"threshold_uncertainty_score":0.2388891},"labels":[],"label_agreement":null},{"id":"W4402730880","doi":"10.1037/met0000631","title":"An overview of alternative formats to the Likert format: A comment on Wilson et al. (2022).","year":2024,"lang":"en","type":"review","venue":"Psychological Methods","topic":"Innovative Teaching Methods","field":"Social 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":"University of British Columbia; York University","funders":"","keywords":"Likert scale; Psychology; Statistics; Computer science; Mathematics","score_opus":0.5365356742093091,"score_gpt":0.6770973150293158,"score_spread":0.14056164082000666,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4402730880","genre_codex":"commentary","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00027240565,0.17556806,0.003719832,0.69773346,0.11576087,0.00025386785,0.0007766149,0.00017907769,0.005735812],"genre_scores_gemma":[0.0031853577,0.08023449,0.0069840755,0.8622584,0.033624854,0.00120616,0.00041026497,0.00018572788,0.011910738],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.96770513,0.018589623,0.003854609,0.0014846267,0.0076093646,0.00075660984],"domain_scores_gemma":[0.90208775,0.050342172,0.0047098016,0.0015903563,0.040068723,0.0012011591],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.032352168,0.0019796772,0.0014523732,0.003929074,0.0026683905,0.0045143827,0.0053553125,0.013708999,0.009258513],"category_scores_gemma":[0.12244821,0.00077416614,0.0033385584,0.0066479268,0.0039239433,0.011255923,0.0031768808,0.016995989,0.010002969],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000027309829,0.00001439525,0.00015012272,0.0016784249,0.000019892308,0.000122058336,0.00074895367,0.000029246352,0.00015505514,0.0043867338,0.9651349,0.027532814],"study_design_scores_gemma":[0.000018288714,0.000022936345,0.00052604685,0.0028748657,0.000015891119,0.0002656113,0.00062103913,0.0000359862,0.000110443805,0.0015786487,0.9938897,0.000040471914],"about_ca_topic_score_codex":0.01599687,"about_ca_topic_score_gemma":0.019897297,"teacher_disagreement_score":0.96764785,"about_ca_system_score_codex":0.004891522,"about_ca_system_score_gemma":0.0062477416,"threshold_uncertainty_score":0.17109662},"labels":[],"label_agreement":null},{"id":"W4403303551","doi":"10.1037/met0000691","title":"Lagged multidimensional recurrence quantification analysis for determining leader–follower relationships within multidimensional time series.","year":2024,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Software System Performance and Reliability","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":"McMaster University","funders":"Deutsche Forschungsgemeinschaft","keywords":"Series (stratigraphy); Recurrence quantification analysis; Multidimensional analysis; Statistics; Statistical analysis; Econometrics; Time series; Psychology; Mathematics; Geology; Nonlinear system; Physics","score_opus":0.14224758821313951,"score_gpt":0.427926050321445,"score_spread":0.2856784621083055,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4403303551","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":[0.010275731,0.0008330393,0.9851376,0.00014107283,0.0001245992,0.00017559578,0.0012899935,0.0007998962,0.0012224372],"genre_scores_gemma":[0.25570446,0.00076207606,0.73680353,0.00012489395,0.00021267433,0.0013206807,0.0030912615,0.00046665478,0.0015137627],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99710387,0.001470989,0.00027953263,0.00058528275,0.00046907607,0.00009125938],"domain_scores_gemma":[0.9836363,0.011068646,0.0018417648,0.0018540905,0.0013434042,0.00025572215],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0071536386,0.001020512,0.0006939936,0.0038327947,0.00054058706,0.0015823231,0.0009911687,0.0005767934,0.0073646884],"category_scores_gemma":[0.042704348,0.0003584737,0.0014607329,0.0041808086,0.00063032634,0.002134113,0.0013629464,0.0015407241,0.0015572498],"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":[0.00039881602,0.00018498213,0.051285475,0.0023013433,0.0014753817,0.00045657853,0.0023035596,0.0479187,0.015770946,0.115346365,0.026600916,0.735957],"study_design_scores_gemma":[0.00006350914,0.00039090205,0.06994163,0.0005807825,0.00046566268,0.00077648944,0.0009256142,0.75253385,0.0072589098,0.12625815,0.040457875,0.00034657776],"about_ca_topic_score_codex":0.002867884,"about_ca_topic_score_gemma":0.0026229636,"teacher_disagreement_score":0.0073646884,"about_ca_system_score_codex":0.00060191914,"about_ca_system_score_gemma":0.0011168784,"threshold_uncertainty_score":0.03783256},"labels":[],"label_agreement":null},{"id":"W4405320068","doi":"10.1037/met0000707","title":"Testing bipolarity.","year":2024,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Mental Health Research Topics","field":"Psychology","cited_by":4,"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 British Columbia","funders":"Social Sciences and Humanities Research Council of Canada; University of Nevada, Las Vegas","keywords":"Psychology; Pessimism; Econometrics; Dimension (graph theory); Test (biology); Social psychology; Extraversion and introversion; Statistics; PsycINFO; Optimism; Type I and type II errors; Mathematics; Big Five personality traits; Personality; Epistemology","score_opus":0.5204120153975895,"score_gpt":0.6753057951375634,"score_spread":0.15489377973997387,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4405320068","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.69298613,0.0011224671,0.12333988,0.008701572,0.0017529168,0.003741725,0.010115233,0.0007094542,0.15753064],"genre_scores_gemma":[0.95168644,0.00024976174,0.03733483,0.0014763323,0.00014415893,0.0035521684,0.003265015,0.00009568153,0.002195645],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.98054296,0.00982501,0.0018031184,0.004007033,0.0033256747,0.0004962358],"domain_scores_gemma":[0.8916944,0.056857903,0.013995048,0.020628497,0.014137698,0.0026865366],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.020611972,0.0009341207,0.00067919306,0.0016319925,0.0016126304,0.0026174546,0.0011204883,0.0011324611,0.012522223],"category_scores_gemma":[0.12458919,0.00046417522,0.0013767547,0.0023171976,0.002256827,0.004541166,0.0035910066,0.0020701233,0.00166332],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0021587473,0.0010364235,0.5238136,0.0013754811,0.0011171284,0.0005979146,0.014664741,0.0014601105,0.0033913704,0.19765691,0.043401517,0.20932606],"study_design_scores_gemma":[0.00073549186,0.0016884335,0.51265174,0.0012634723,0.0007572596,0.002167515,0.01569659,0.024218058,0.0036327702,0.38390493,0.05303811,0.00024545332],"about_ca_topic_score_codex":0.0015763614,"about_ca_topic_score_gemma":0.0017102765,"teacher_disagreement_score":0.020611972,"about_ca_system_score_codex":0.0010273341,"about_ca_system_score_gemma":0.0013282734,"threshold_uncertainty_score":0.109007835},"labels":[],"label_agreement":null},{"id":"W4407299576","doi":"10.1037/met0000735","title":"Reassessing the fitting propensity of factor models.","year":2025,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Bayesian Modeling and Causal Inference","field":"Computer Science","cited_by":3,"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":"Institute of Education Sciences; University of Missouri; U.S. Department of Education; National Science Foundation","keywords":"Statistics; Econometrics; Factor (programming language); Propensity score matching; Psychology; Mathematics; Computer science","score_opus":0.2529208410047625,"score_gpt":0.48388635618993864,"score_spread":0.23096551518517616,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4407299576","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":[0.06640717,0.0019129505,0.90594417,0.010122002,0.00039946724,0.00048735895,0.00038853608,0.0005448494,0.013793414],"genre_scores_gemma":[0.6555313,0.0010254413,0.33867583,0.0015597726,0.00035684367,0.00085374794,0.0005007919,0.00035162372,0.0011446265],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.8590153,0.10924559,0.006026708,0.010463809,0.014187639,0.0010609262],"domain_scores_gemma":[0.26978874,0.63541883,0.027811367,0.051513005,0.013534433,0.00193369],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.17072503,0.0016396146,0.0020566236,0.0060269716,0.0032953138,0.011934702,0.003379255,0.0037580675,0.006933245],"category_scores_gemma":[0.69573087,0.0012769871,0.0034263674,0.0069776224,0.012690291,0.023862598,0.007390967,0.009559422,0.00089458755],"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":[0.00034454465,0.00014330017,0.0633754,0.0010348822,0.0010108558,0.0005611301,0.01194494,0.033035316,0.0007997389,0.6639763,0.0065162764,0.21725737],"study_design_scores_gemma":[0.00003678119,0.00016457828,0.0093262,0.000609455,0.00015792147,0.0005723905,0.00213622,0.11001546,0.0006717301,0.8687601,0.0074256095,0.00012348864],"about_ca_topic_score_codex":0.0047689807,"about_ca_topic_score_gemma":0.004791646,"teacher_disagreement_score":0.17072503,"about_ca_system_score_codex":0.005138368,"about_ca_system_score_gemma":0.005126206,"threshold_uncertainty_score":0.90289104},"labels":[],"label_agreement":null},{"id":"W4407299679","doi":"10.1037/met0000739","title":"Reliability in unidimensional ordinal data: A comparison of continuous and ordinal estimators.","year":2025,"lang":"en","type":"article","venue":"Psychological Methods","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":"Université de Montréal","funders":"National Research Foundation of Korea; Kwangwoon University","keywords":"Ordinal data; Estimator; Statistics; Ordinal optimization; Reliability (semiconductor); Ordinal regression; Mathematics; Econometrics; Ordered logit; Psychology","score_opus":0.3421491523690011,"score_gpt":0.6311549700243662,"score_spread":0.28900581765536504,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4407299679","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":[0.09445639,0.021672407,0.8570301,0.0049873875,0.0014296944,0.0012734714,0.00081185426,0.0005860782,0.017752651],"genre_scores_gemma":[0.6882111,0.0031682486,0.3029779,0.001338459,0.0004295346,0.002562757,0.0005252399,0.00021600869,0.0005707369],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.57446223,0.35991743,0.014315253,0.012814165,0.037390485,0.0011004425],"domain_scores_gemma":[0.21792443,0.700796,0.031048572,0.03071018,0.018336438,0.001184469],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.30728677,0.0008723676,0.002547509,0.010017646,0.0017279908,0.0071395943,0.0026004233,0.0020117632,0.0025264134],"category_scores_gemma":[0.6546917,0.00089628017,0.0024369739,0.014356687,0.008681093,0.009353261,0.0067096194,0.005173878,0.0006215788],"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":[0.00079173147,0.00041824768,0.20333722,0.007320458,0.0066308565,0.00036158683,0.02406214,0.011905958,0.0010724791,0.32575467,0.010512582,0.40783212],"study_design_scores_gemma":[0.00037994847,0.0022127805,0.16631924,0.010031515,0.002846558,0.0021859228,0.017490568,0.087010935,0.0035637177,0.6351279,0.07200811,0.0008227395],"about_ca_topic_score_codex":0.0010547332,"about_ca_topic_score_gemma":0.0011443975,"teacher_disagreement_score":0.30728677,"about_ca_system_score_codex":0.0024887742,"about_ca_system_score_gemma":0.0032597573,"threshold_uncertainty_score":0.85423875},"labels":[],"label_agreement":null},{"id":"W4407505237","doi":"10.1037/met0000718","title":"Fit indices are insensitive to multiple minor violations of perfect simple structure in confirmatory factor analysis.","year":2025,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Multi-Criteria Decision Making","field":"Decision Sciences","cited_by":3,"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":"Confirmatory factor analysis; Minor (academic); Statistics; Simple (philosophy); Statistical analysis; Factor (programming language); Psychology; Econometrics; Mathematics; Structural equation modeling; Computer science; Humanities","score_opus":0.3580454211727005,"score_gpt":0.6116049977086426,"score_spread":0.25355957653594213,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4407505237","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":[0.041317474,0.0019718143,0.8942832,0.003299602,0.0013013993,0.002500304,0.0028224955,0.003548738,0.04895498],"genre_scores_gemma":[0.4029063,0.0011665344,0.58054197,0.0025093236,0.00039201378,0.0059769605,0.0028017727,0.00134082,0.0023642324],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.691937,0.19123061,0.02145258,0.024664963,0.068712145,0.002002625],"domain_scores_gemma":[0.29934058,0.5077388,0.049188446,0.10724636,0.03414997,0.0023358166],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.25428706,0.0042247917,0.004283784,0.010543811,0.0038210372,0.0089519,0.0057941917,0.0050139343,0.011355802],"category_scores_gemma":[0.7229806,0.004565492,0.004422227,0.016855111,0.010759619,0.015062029,0.008406305,0.011416644,0.00826574],"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":[0.00112994,0.0005897333,0.16270171,0.0050659864,0.006655146,0.0012346655,0.005916362,0.025186837,0.0035198217,0.24353515,0.056029886,0.48843476],"study_design_scores_gemma":[0.00052204827,0.001992758,0.1100715,0.0036311934,0.0015420801,0.00282081,0.0036053914,0.19712116,0.008524164,0.6084038,0.060539663,0.0012254904],"about_ca_topic_score_codex":0.002426495,"about_ca_topic_score_gemma":0.0048240987,"teacher_disagreement_score":0.74571294,"about_ca_system_score_codex":0.004239493,"about_ca_system_score_gemma":0.00920603,"threshold_uncertainty_score":0.9195968},"labels":[],"label_agreement":null},{"id":"W4407762147","doi":"10.1037/met0000730","title":"Evaluating statistical fit of confirmatory bifactor models: Updated recommendations and a review of current practice.","year":2025,"lang":"en","type":"review","venue":"Psychological Methods","topic":"Optimal Experimental Design Methods","field":"Decision Sciences","cited_by":10,"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":"Current (fluid); Confirmatory factor analysis; Psychology; Econometrics; Statistics; Statistical analysis; Psychometrics; Statistical hypothesis testing; Clinical psychology; Structural equation modeling; Mathematics; Engineering","score_opus":0.8475355186061376,"score_gpt":0.7654928065091957,"score_spread":0.08204271209694192,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4407762147","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00015096138,0.98501325,0.0048871622,0.007223485,0.0009242595,0.0002753437,0.0003180868,0.00012323436,0.0010842644],"genre_scores_gemma":[0.0025780625,0.95597434,0.036549103,0.002498843,0.00051779865,0.0008770864,0.00059674756,0.00007391107,0.00033408232],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9483415,0.027921245,0.010437397,0.0022195287,0.010554087,0.0005262042],"domain_scores_gemma":[0.67228997,0.23601925,0.021162268,0.007020057,0.061280284,0.0022282065],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.13766028,0.0034420404,0.005865119,0.030811891,0.0013476176,0.006539445,0.007032696,0.0045272023,0.0071676937],"category_scores_gemma":[0.29093593,0.0020986553,0.007042063,0.025560148,0.0033267,0.01023048,0.0035150074,0.005448502,0.0040695174],"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":[0.00007773474,0.00006564814,0.0017640892,0.089693844,0.0010295396,0.00010238611,0.000409063,0.0004299191,0.000116283634,0.006322838,0.063405275,0.83658344],"study_design_scores_gemma":[0.00016838223,0.00019978234,0.0069037178,0.44492808,0.00529206,0.0006280825,0.0011052233,0.0019198192,0.00052861735,0.02504937,0.5130259,0.00025099338],"about_ca_topic_score_codex":0.016474703,"about_ca_topic_score_gemma":0.030647406,"teacher_disagreement_score":0.86233974,"about_ca_system_score_codex":0.008547854,"about_ca_system_score_gemma":0.03403855,"threshold_uncertainty_score":0.72802585},"labels":[],"label_agreement":null},{"id":"W4408293278","doi":"10.1037/met0000746","title":"Unsupervised [randomly responding] survey bot detection: In search of high classification accuracy.","year":2025,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Spam and Phishing Detection","field":"Computer Science","cited_by":4,"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":"Fonds de Recherche du Québec - Santé; Natural Sciences and Engineering Research Council of Canada","keywords":"Statistics; Artificial intelligence; Pattern recognition (psychology); Statistical analysis; Computer science; Psychology; Mathematics","score_opus":0.16687389522439644,"score_gpt":0.4706631522301324,"score_spread":0.30378925700573595,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4408293278","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":[0.33678412,0.00054488645,0.64844483,0.0018726286,0.00023772112,0.0014668797,0.001089843,0.004231828,0.005327243],"genre_scores_gemma":[0.7839919,0.0000988308,0.21146965,0.0009587432,0.000090634654,0.0006120915,0.0011936349,0.0001257667,0.0014587516],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9748639,0.017128216,0.0008850446,0.0039244713,0.0027423443,0.00045601028],"domain_scores_gemma":[0.8431159,0.09433659,0.014774822,0.036508065,0.010061198,0.0012033841],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.028238202,0.0010022976,0.001254417,0.0018183923,0.0010292424,0.001692645,0.0025606956,0.0026042787,0.0010591383],"category_scores_gemma":[0.14343514,0.0005366796,0.0009767291,0.0016369006,0.0017256694,0.0027821155,0.0017157823,0.0023154414,0.0012428133],"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":[0.0021164792,0.0019109731,0.36174774,0.0009219027,0.0011176688,0.00030701354,0.0017890693,0.120827354,0.009676894,0.020725792,0.027504219,0.451355],"study_design_scores_gemma":[0.00014977081,0.0004983698,0.034383036,0.000106099134,0.00008479419,0.00035475421,0.00035267763,0.9241081,0.006441019,0.029834142,0.0036184222,0.00006879763],"about_ca_topic_score_codex":0.0018864494,"about_ca_topic_score_gemma":0.0032301808,"teacher_disagreement_score":0.9717618,"about_ca_system_score_codex":0.0012310997,"about_ca_system_score_gemma":0.0016766238,"threshold_uncertainty_score":0.14933968},"labels":[],"label_agreement":null},{"id":"W4414082743","doi":"10.1037/met0000797","title":"On the uncanny relationship between nonnormality and moderated multiple regression.","year":2025,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Advanced Statistical Methods and Models","field":"Mathematics","cited_by":0,"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; York University","funders":"","keywords":"Multivariate statistics; Generalization; Regression; Regression analysis; Linear regression; Multivariate analysis; Asymmetry; Association (psychology)","score_opus":0.4858594828500195,"score_gpt":0.6028049536091047,"score_spread":0.11694547075908518,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4414082743","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":[0.044073038,0.02304162,0.8009607,0.07291197,0.0022845843,0.00046141795,0.0005796067,0.00049450406,0.055192623],"genre_scores_gemma":[0.6277719,0.019553253,0.316673,0.021075767,0.0034240643,0.0014729007,0.0006088534,0.00045963362,0.008960631],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.92322266,0.05502218,0.0035884536,0.007972601,0.009599346,0.0005947746],"domain_scores_gemma":[0.3392807,0.61104757,0.01949303,0.021592302,0.007295368,0.0012910308],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.11410109,0.0011824691,0.0016363395,0.0026048003,0.0023347083,0.0045980564,0.003270575,0.0035134878,0.011160778],"category_scores_gemma":[0.41534084,0.0009387982,0.0015824082,0.0049070492,0.023912251,0.013984053,0.0077693714,0.009006462,0.0016621335],"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":[0.00032578883,0.00004352412,0.019084478,0.0008189358,0.0004730321,0.0010313255,0.0026366569,0.004519456,0.00035653816,0.85629296,0.01187266,0.10254479],"study_design_scores_gemma":[0.00004363232,0.00014937008,0.007049543,0.0007603141,0.00016541222,0.00086896046,0.00061455864,0.0137631325,0.00044290003,0.9605509,0.015509344,0.00008195053],"about_ca_topic_score_codex":0.0030455757,"about_ca_topic_score_gemma":0.0041818526,"teacher_disagreement_score":0.8858989,"about_ca_system_score_codex":0.0033349427,"about_ca_system_score_gemma":0.0031196123,"threshold_uncertainty_score":0.60343146},"labels":[],"label_agreement":null},{"id":"W4414341611","doi":"10.1037/met0000770","title":"Crowdsourcing multiverse analyses to explore the impact of different data-processing and analysis decisions: A tutorial.","year":2025,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Forecasting Techniques and Applications","field":"Decision 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":"Université du Québec à Montréal","funders":"","keywords":"Crowdsourcing; Generalizability theory; Objectivity (philosophy); Focus (optics); Outcome (game theory); Heuristics; Conflation","score_opus":0.757406363254655,"score_gpt":0.6943019283092987,"score_spread":0.06310443494535634,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4414341611","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":[0.00045939663,0.0046081487,0.98134494,0.00281599,0.0004253614,0.0005660875,0.0014706245,0.0031885118,0.0051209265],"genre_scores_gemma":[0.005553861,0.004638246,0.97964716,0.00092690054,0.000397444,0.0018403257,0.0010894105,0.0010974591,0.004809165],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9953435,0.0028633457,0.00030053555,0.00048138291,0.00092018617,0.00009101975],"domain_scores_gemma":[0.9661656,0.029574228,0.00080492406,0.0015452266,0.0014743655,0.0004356182],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.018054515,0.0019586224,0.00093287724,0.0047707283,0.0010819935,0.0032891287,0.002731802,0.0020164184,0.030739456],"category_scores_gemma":[0.046880376,0.0012728253,0.002307189,0.00428009,0.0013564244,0.004260557,0.003960443,0.0032276472,0.012852471],"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":[0.00007826528,0.0001791363,0.0015324752,0.0033825,0.0004994986,0.00046000507,0.0018310978,0.011476303,0.004086704,0.13532418,0.23074195,0.61040795],"study_design_scores_gemma":[0.00007352546,0.000091298716,0.0024187912,0.0016998776,0.000096052616,0.0005825094,0.00058399624,0.051567897,0.0026834866,0.46185166,0.47814062,0.00021031144],"about_ca_topic_score_codex":0.0030428967,"about_ca_topic_score_gemma":0.009074603,"teacher_disagreement_score":0.9819455,"about_ca_system_score_codex":0.0017638223,"about_ca_system_score_gemma":0.0030266044,"threshold_uncertainty_score":0.10283375},"labels":[],"label_agreement":null},{"id":"W4414753970","doi":"10.1037/met0000781.supp","title":"Supplemental Material for Causal Mediation Analysis With Two Mediators: A Comprehensive Guide to Estimating Total and Natural Effects Across Various Multiple Mediators Setups","year":2025,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Information and Cyber Security","field":"Computer Science","cited_by":0,"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":"Mediation; Natural (archaeology); Causal model; Causal analysis; Causal inference; Causality (physics)","score_opus":0.014206761890993121,"score_gpt":0.4123835212198742,"score_spread":0.3981767593288811,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4414753970","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":[0.0044585527,0.0014955865,0.54932505,0.001423434,0.0006764972,0.016523452,0.39175618,0.010674875,0.023666438],"genre_scores_gemma":[0.018270658,0.0017127513,0.78405803,0.0011695855,0.0001995981,0.09252086,0.07703054,0.004544221,0.020493673],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9931378,0.004492868,0.0006870854,0.00061217684,0.00080686744,0.0002631303],"domain_scores_gemma":[0.91884506,0.07268083,0.0016436022,0.0030188118,0.0034411335,0.00037053545],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.019125856,0.0032385928,0.0023154246,0.0043640393,0.0017409638,0.0027794773,0.004271097,0.0020030758,0.5625764],"category_scores_gemma":[0.07822914,0.0024536967,0.004320364,0.00466115,0.00057157467,0.0021504858,0.002773022,0.0036890989,0.058706056],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00038944144,0.0013127503,0.008403653,0.008320783,0.0007273788,0.000546436,0.0017939438,0.0063687777,0.0007919506,0.059732243,0.7192089,0.19240387],"study_design_scores_gemma":[0.0014846271,0.00063569966,0.020271655,0.007466541,0.0010695066,0.0009296422,0.002058277,0.030620504,0.0017891182,0.17186423,0.76139593,0.0004143447],"about_ca_topic_score_codex":0.010950412,"about_ca_topic_score_gemma":0.019722775,"teacher_disagreement_score":0.5625764,"about_ca_system_score_codex":0.0019567655,"about_ca_system_score_gemma":0.007245591,"threshold_uncertainty_score":0.62393165},"labels":[],"label_agreement":null},{"id":"W4415247091","doi":"10.1037/met0000781","title":"Causal mediation analysis with two mediators: A comprehensive guide to estimating total and natural effects across various multiple mediators setups.","year":2025,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Qualitative Comparative Analysis Research","field":"Social Sciences","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; Université du Québec à Montréal","funders":"Fonds de Recherche du Québec - Santé; Natural Sciences and Engineering Research Council of Canada; Fonds de recherche du Québec","keywords":"Mediation; Causal inference; Causal model; Imputation (statistics); Causal analysis; Natural (archaeology); Structural equation modeling; Causality (physics)","score_opus":0.0642801776188526,"score_gpt":0.5832066838773483,"score_spread":0.5189265062584957,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4415247091","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":[0.00060034974,0.010856084,0.9581988,0.0041694655,0.00037274556,0.0030836486,0.006179935,0.00093828945,0.015600659],"genre_scores_gemma":[0.005859329,0.011619021,0.9607041,0.0015242045,0.00014733018,0.012301173,0.0023592883,0.00045574587,0.0050297817],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.97607213,0.019424357,0.0012953605,0.0009767009,0.0020177914,0.00021366816],"domain_scores_gemma":[0.9227527,0.06914281,0.0022972773,0.0026349407,0.0027534433,0.00041888928],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.04000489,0.0026579837,0.002184307,0.005654067,0.0016539539,0.0037947728,0.005005387,0.002948427,0.059286766],"category_scores_gemma":[0.0821754,0.0022302628,0.0034371107,0.007998619,0.0025233342,0.0060123606,0.004555852,0.006830195,0.013434258],"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":[0.000050978342,0.00020675156,0.0018543423,0.009761782,0.00068204757,0.00036584138,0.004847383,0.006066433,0.0007853457,0.44873533,0.18986845,0.33677533],"study_design_scores_gemma":[0.000063112326,0.00013223985,0.0023805955,0.0058501693,0.00027003506,0.0004947865,0.0018060968,0.010471274,0.0007267735,0.41587377,0.5617521,0.00017909925],"about_ca_topic_score_codex":0.0066894256,"about_ca_topic_score_gemma":0.014627081,"teacher_disagreement_score":0.9599951,"about_ca_system_score_codex":0.003079791,"about_ca_system_score_gemma":0.010453118,"threshold_uncertainty_score":0.21156865},"labels":[],"label_agreement":null}]}