{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":1326,"total_is_capped":false,"direct_labels_cover":35,"predictions_cover":1326,"direct_label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline (scores rank; they never assert a category)","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12","author_layer_release":"2026-06-26"},"query_hash":"e802a3299282","filters":{"topic":"Advanced Causal Inference Techniques"}},"results":[{"id":"W2036193982","doi":"10.1080/00273171.2011.568786","title":"An Introduction to Propensity Score Methods for Reducing the Effects of Confounding in Observational Studies","year":2011,"lang":"en","type":"article","venue":"Multivariate Behavioral Research","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":11884,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences","keywords":"Propensity score matching; Observational study; Covariate; Confounding; Statistics; Inverse probability weighting; Econometrics; Mathematics","authors":[{"name":"Peter C. Austin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8974226593536919,"gpt":0.6794649079660394,"spread":0.2179577513876525,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02565967,0.002131388,0.002260635,0.004319488,0.001078761,0.002945566,0.003186648,0.003098464,0.01781762],"category_scores_gemma":[0.07310484,0.001234833,0.003387682,0.008909905,0.002494858,0.003662108,0.002916689,0.007835088,0.006071698],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00143634,"about_ca_system_score_gemma":0.003227184,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001812271,"about_ca_topic_score_gemma":0.002077783,"domain_scores_codex":[0.976929,0.01696536,0.001146374,0.001172214,0.003548952,0.0002380372],"domain_scores_gemma":[0.9561473,0.03687255,0.001960655,0.002791513,0.001883988,0.0003440085],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00006495276,0.0001144865,0.002371027,0.001981673,0.0005878498,0.000257963,0.0004048448,0.01507332,0.0008950936,0.5595564,0.06979711,0.3488953],"study_design_scores_gemma":[0.0001195996,0.0001205266,0.001645525,0.0007699517,0.0001657954,0.0005027282,0.00007333534,0.03732543,0.0007373698,0.8075804,0.150833,0.0001263534],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0002435067,0.004306037,0.9905527,0.001477932,0.0004911736,0.0002482845,0.0003570672,0.0003007431,0.002022504],"genre_scores_gemma":[0.00894371,0.01272437,0.968685,0.001255817,0.002510169,0.001746763,0.0006401471,0.0003369577,0.003156957],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.02565967,"threshold_uncertainty_score":0.1357029,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2126049444","doi":"10.1002/sim.3697","title":"Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity‐score matched samples","year":2009,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":6487,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Clinical Evaluative Sciences; Public Health Ontario; University of Toronto","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences; Heart and Stroke Foundation of Canada","keywords":"Propensity score matching; Covariate; Statistics; Matching (statistics); Quantile; Mathematics; Medicine","authors":[{"name":"Peter C. Austin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2114497428934903,"gpt":0.4202548380162484,"spread":0.208805095122758,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09158804,0.001541054,0.002624626,0.01080528,0.00119458,0.002779097,0.003842496,0.003430333,0.01373318],"category_scores_gemma":[0.3592003,0.001031592,0.002396284,0.007544484,0.003957623,0.003772966,0.003041576,0.003488846,0.00205509],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00145139,"about_ca_system_score_gemma":0.002464339,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00191029,"about_ca_topic_score_gemma":0.001033397,"domain_scores_codex":[0.880947,0.09332233,0.006248797,0.006851892,0.0116394,0.0009905659],"domain_scores_gemma":[0.6266996,0.3223576,0.0240376,0.01971458,0.006229301,0.0009612389],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.003509816,0.0005444972,0.1619432,0.003480217,0.005085194,0.001175025,0.001895633,0.02503305,0.004060078,0.2669976,0.03370348,0.4925721],"study_design_scores_gemma":[0.001395891,0.002601458,0.1281327,0.001571769,0.001691646,0.002987967,0.001397963,0.3086506,0.01230779,0.4764229,0.06239864,0.0004406149],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01982108,0.001158412,0.9687081,0.0006563682,0.000320631,0.001308316,0.002772326,0.001411181,0.003843636],"genre_scores_gemma":[0.3184685,0.0007240424,0.6667436,0.000943055,0.0003946361,0.005830506,0.004215179,0.0007186478,0.001961944],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.09158804,"threshold_uncertainty_score":0.4843696,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2110818436","doi":"10.1002/sim.6607","title":"Moving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in observational studies","year":2015,"lang":"en","type":"review","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":4309,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research; National Institute of Mental Health; Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences","keywords":"Propensity score matching; Covariate; Inverse probability weighting; Observational study; Inverse probability; Weighting; Average treatment effect; Statistics; Treatment and control groups; Econometrics; Sample size determination; Baseline (sea); Mathematics; Selection bias; Computer science; Posterior probability; Medicine; Bayesian probability","authors":[{"name":"Peter C. Austin","is_ca":true},{"name":"Elizabeth A. Stuart","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6765608631638398,"gpt":0.5702584500397571,"spread":0.1063024131240827,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3219493,0.001957789,0.00456967,0.009172669,0.001790489,0.01072376,0.007313782,0.007535535,0.002848967],"category_scores_gemma":[0.5747454,0.002086388,0.003907751,0.01045066,0.02269435,0.01663251,0.01022789,0.01856037,0.001250197],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00563736,"about_ca_system_score_gemma":0.01103963,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004402351,"about_ca_topic_score_gemma":0.003131632,"domain_scores_codex":[0.5464783,0.3953557,0.02092759,0.01266301,0.02350643,0.00106892],"domain_scores_gemma":[0.4113832,0.5122638,0.02444915,0.03574902,0.01497808,0.001176717],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00009943253,0.00008329565,0.002905237,0.006163369,0.001226492,0.0001766439,0.003135765,0.009738759,0.0003221479,0.7230092,0.01097896,0.2421608],"study_design_scores_gemma":[0.00007939994,0.00007738317,0.0005319855,0.003203576,0.0001395571,0.0001094208,0.0002324469,0.008052526,0.0003939082,0.9689416,0.01816838,0.00006975682],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.001151097,0.01156664,0.9593448,0.02429986,0.0008635535,0.0003328921,0.0001452179,0.0001886445,0.002107228],"genre_scores_gemma":[0.03870384,0.01061026,0.9379324,0.008251778,0.001469575,0.00225951,0.0001742692,0.0001940757,0.000404227],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.6780507,"threshold_uncertainty_score":0.8361572,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2111635289","doi":"10.1002/pst.433","title":"Optimal caliper widths for propensity‐score matching when estimating differences in means and differences in proportions in observational studies","year":2010,"lang":"en","type":"article","venue":"Pharmaceutical Statistics","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":3895,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Clinical Evaluative Sciences; Public Health Ontario; University of Toronto","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences","keywords":"Propensity score matching; Covariate; Calipers; Statistics; Estimator; Mathematics; Logit; Matching (statistics); Observational study; Econometrics; Standard deviation; Confidence interval; Logistic regression; Medicine","authors":[{"name":"Peter C. Austin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5850172035475715,"gpt":0.5093227869028153,"spread":0.07569441664475618,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.08407174,0.001653233,0.00219179,0.003623212,0.001245006,0.002821066,0.003149831,0.00283857,0.005573568],"category_scores_gemma":[0.3525944,0.001690412,0.002448533,0.004770826,0.002704137,0.004646594,0.003786211,0.005900076,0.00134389],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001716484,"about_ca_system_score_gemma":0.003781013,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002329814,"about_ca_topic_score_gemma":0.002309705,"domain_scores_codex":[0.9359957,0.05323295,0.002745491,0.003674112,0.003831944,0.0005197449],"domain_scores_gemma":[0.8081051,0.1661406,0.009309796,0.01203812,0.003731029,0.0006754523],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0009411592,0.0002998363,0.008877102,0.001239966,0.0005341051,0.000424256,0.001060161,0.1359064,0.002644177,0.495127,0.009719002,0.3432268],"study_design_scores_gemma":[0.000567504,0.0004350573,0.003966538,0.0007937132,0.0002206217,0.0004044973,0.0002057705,0.3619267,0.004012359,0.6027439,0.02457988,0.000143381],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.002250815,0.0005286876,0.995418,0.0002713604,0.00007565333,0.0003410179,0.00008470524,0.000246089,0.0007836892],"genre_scores_gemma":[0.03348788,0.0004953936,0.9635423,0.0002366899,0.0001018768,0.001449892,0.0001616129,0.0001603236,0.0003640376],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9159282,"threshold_uncertainty_score":0.4446192,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3122958054","doi":"10.1016/j.jeconom.2007.05.001","title":"Regression discontinuity designs: A guide to practice","year":2007,"lang":"en","type":"article","venue":"Journal of Econometrics","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":3773,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"Social Sciences and Humanities Research Council of Canada; National Science Foundation","keywords":"Regression discontinuity design; Covariate; Econometrics; Discontinuity (linguistics); Regression; Value (mathematics); Regression analysis; Statistics; Mathematics; Computer science; Economics; Psychology","authors":[{"name":"Guido W. Imbens","is_ca":false},{"name":"Thomas Lemieux","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4194458121712126,"gpt":0.4801592340539934,"spread":0.06071342188278078,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04673511,0.002950878,0.003660113,0.004488074,0.001197855,0.005716627,0.006633234,0.005214334,0.01668956],"category_scores_gemma":[0.1446435,0.00446376,0.002587011,0.004915842,0.003172684,0.003209397,0.003226354,0.01140582,0.008301846],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001627118,"about_ca_system_score_gemma":0.006411365,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00378071,"about_ca_topic_score_gemma":0.005073522,"domain_scores_codex":[0.9579268,0.03519358,0.00210963,0.001167939,0.003378217,0.0002238622],"domain_scores_gemma":[0.8458677,0.1359931,0.001970138,0.007121723,0.008200283,0.00084704],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0004083981,0.0008483504,0.00171664,0.003876817,0.000706368,0.0006303746,0.001646398,0.02722686,0.002048156,0.3455864,0.1265098,0.4887953],"study_design_scores_gemma":[0.001004184,0.0004249224,0.001357989,0.00301529,0.000456868,0.001175065,0.0006237741,0.1303616,0.003530745,0.6651215,0.1925891,0.0003389764],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0001812599,0.001614722,0.9943152,0.0008057201,0.0001463685,0.0003941654,0.000272776,0.0008414558,0.001428357],"genre_scores_gemma":[0.00339799,0.001869414,0.9906617,0.0003039759,0.0001481832,0.00192917,0.0001815653,0.0004161349,0.00109197],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.04673511,"threshold_uncertainty_score":0,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"grok","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"},{"model":"opus","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W2264767503","doi":"10.1161/circulationaha.115.017719","title":"Introduction to the Analysis of Survival Data in the Presence of Competing Risks","year":2016,"lang":"en","type":"article","venue":"Circulation","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":2536,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto; University Health Network; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences; Heart and Stroke Foundation of Canada","keywords":"Covariate; Medicine; Proportional hazards model; Survival analysis; Cumulative incidence; Incidence (geometry); Survival function; Event (particle physics); Statistics; Hazard ratio; Outcome (game theory); Hazard; Demography; Accelerated failure time model; Econometrics; Confidence interval; Internal medicine; Cohort; Mathematics","authors":[{"name":"Peter C. Austin","is_ca":true},{"name":"Douglas S. Lee","is_ca":true},{"name":"Jason P. Fine","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2664686297936992,"gpt":0.4426291158539977,"spread":0.1761604860602985,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03189606,0.001869967,0.001718444,0.003986382,0.0008456346,0.003089857,0.003653634,0.003057933,0.0267117],"category_scores_gemma":[0.09730981,0.001524248,0.004224772,0.00630443,0.00373912,0.002975173,0.003005893,0.01098289,0.00729861],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001336205,"about_ca_system_score_gemma":0.003676947,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002594965,"about_ca_topic_score_gemma":0.002237865,"domain_scores_codex":[0.9630908,0.02883231,0.001978118,0.001864233,0.003928177,0.0003062704],"domain_scores_gemma":[0.7887508,0.1962314,0.004274825,0.006289776,0.003561327,0.0008918426],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0001900003,0.0001815747,0.004430406,0.004368296,0.0007682825,0.001165595,0.001293277,0.01867008,0.00164295,0.5426522,0.09931553,0.3253218],"study_design_scores_gemma":[0.0001265042,0.0002474451,0.003533118,0.001977228,0.0001948417,0.002198645,0.0001788563,0.05276324,0.0007437778,0.6568273,0.2810144,0.0001945532],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0004863753,0.004737605,0.9862852,0.002550392,0.0005880888,0.0002748402,0.001176654,0.000993361,0.002907445],"genre_scores_gemma":[0.01207006,0.008027072,0.9697362,0.001822015,0.0024321,0.002063712,0.001161794,0.0006487923,0.00203818],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.03189606,"threshold_uncertainty_score":0.1686845,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2004722883","doi":"10.1097/ede.0b013e3181a819a1","title":"Overadjustment Bias and Unnecessary Adjustment in Epidemiologic Studies","year":2009,"lang":"en","type":"article","venue":"Epidemiology","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":1947,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University","funders":"National Institute of Allergy and Infectious Diseases; National Institutes of Health","keywords":"Selection bias; Omitted-variable bias; Confounding; Information bias; Econometrics; Statistics; Proxy (statistics); Causal structure; Sampling bias; Non-response bias; Outcome (game theory); Causal inference; Sample size determination; Mathematics","authors":[{"name":"Enrique F. Schisterman","is_ca":false},{"name":"Stephen R. Cole","is_ca":false},{"name":"Robert W. Platt","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6176038876181804,"gpt":0.5397161510959915,"spread":0.07788773652218894,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1786922,0.001209476,0.002302097,0.004016202,0.001849464,0.004640779,0.003415974,0.002691605,0.00187738],"category_scores_gemma":[0.5292621,0.001250007,0.002602454,0.007780286,0.01186959,0.005489334,0.006130153,0.003505712,0.000333335],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003491125,"about_ca_system_score_gemma":0.004403013,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003586757,"about_ca_topic_score_gemma":0.002899771,"domain_scores_codex":[0.7199392,0.2270248,0.01648987,0.01425966,0.02036373,0.001922771],"domain_scores_gemma":[0.4040785,0.4968787,0.040064,0.04869763,0.009363513,0.0009177378],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0003154188,0.00006677947,0.0429871,0.001484967,0.001379624,0.0004115975,0.003069644,0.02596791,0.0005930499,0.7975662,0.003518073,0.1226397],"study_design_scores_gemma":[0.000113699,0.000098788,0.007090492,0.000573914,0.0003814797,0.0004669182,0.0002645318,0.02902266,0.001101117,0.9449137,0.01588888,0.00008391789],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02828315,0.009278389,0.9467807,0.008163262,0.0007159318,0.0004783426,0.0003622227,0.0002634015,0.005674624],"genre_scores_gemma":[0.5414419,0.004976959,0.443339,0.004845949,0.001345058,0.001523151,0.0002834737,0.0002125967,0.002031929],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8213078,"threshold_uncertainty_score":0.945026,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2029059557","doi":"10.1080/03610911003650383","title":"How Big is a Big Odds Ratio? Interpreting the Magnitudes of Odds Ratios in Epidemiological Studies","year":2010,"lang":"en","type":"article","venue":"Communications in Statistics - Simulation and Computation","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":1932,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Health Canada; Columbia College","funders":"","keywords":"Odds ratio; Odds; Epidemiology; Index (typography); Statistics; Medicine; Demography; Psychology; Mathematics; Computer science; Internal medicine; Sociology; Logistic regression","authors":[{"name":"Henian Chen","is_ca":true},{"name":"Patricia Cohen","is_ca":true},{"name":"Sophie Chen","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5103286940819939,"gpt":0.559314319333483,"spread":0.04898562525148908,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1769799,0.001541856,0.003228707,0.008016573,0.001489058,0.009343868,0.003176679,0.004328647,0.002347552],"category_scores_gemma":[0.724739,0.001082082,0.002292549,0.00775061,0.01414063,0.01141796,0.004747482,0.006986049,0.0007205678],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002041,"about_ca_system_score_gemma":0.00255749,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001141683,"about_ca_topic_score_gemma":0.0009331218,"domain_scores_codex":[0.7458931,0.2155462,0.01270806,0.008113209,0.01703048,0.0007090129],"domain_scores_gemma":[0.2111917,0.7446097,0.01947299,0.01519928,0.008146442,0.001379743],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001246505,0.0001685741,0.07830749,0.008932225,0.005150931,0.001405782,0.008349321,0.00789358,0.003125768,0.345249,0.03228123,0.5078896],"study_design_scores_gemma":[0.0001930386,0.0003373337,0.01573539,0.001803318,0.00098857,0.001771519,0.002079171,0.006739135,0.00164557,0.9465638,0.02193118,0.0002120442],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.03525531,0.05287406,0.8177851,0.0706789,0.008446748,0.0005517156,0.0008271143,0.001082584,0.01249856],"genre_scores_gemma":[0.4991066,0.01089162,0.4678124,0.01394852,0.006178108,0.0008056323,0.0001765511,0.0003285058,0.0007521639],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8230201,"threshold_uncertainty_score":0.9359704,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2056621649","doi":"10.1214/10-sts321","title":"Identification, Inference and Sensitivity Analysis for Causal Mediation Effects","year":2010,"lang":"en","type":"article","venue":"Statistical Science","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":1558,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"York University; University of Wisconsin-Madison; University of Pennsylvania; Harvard University; National Science Foundation","keywords":"Causal inference; Estimator; Robustness (evolution); Identification (biology); Sensitivity (control systems); Randomized experiment; Inference; Causal model; Parametric statistics","authors":[],"retraction":null,"screen_n_in":null,"score":{"opus":0.05207382361030993,"gpt":0.4371451219746848,"spread":0.3850712983643749,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1851711,0.002129163,0.004133695,0.005751994,0.002170707,0.00338388,0.003480107,0.004098886,0.007821467],"category_scores_gemma":[0.4741857,0.001342715,0.005837895,0.003891356,0.006409148,0.006861071,0.00748228,0.007115659,0.000420116],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003788546,"about_ca_system_score_gemma":0.003709895,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002068268,"about_ca_topic_score_gemma":0.001037189,"domain_scores_codex":[0.8109351,0.1719837,0.003070384,0.006066945,0.006718392,0.001225491],"domain_scores_gemma":[0.3751533,0.5904462,0.009655176,0.02031597,0.004000045,0.0004293502],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0004576323,0.0002557609,0.005671546,0.001642518,0.002060618,0.0005954509,0.0009614325,0.1107203,0.000849895,0.8128421,0.001831893,0.062111],"study_design_scores_gemma":[0.0001492889,0.0002055117,0.00157335,0.0002683953,0.0005821406,0.0002448139,0.0002553664,0.1918924,0.001269597,0.8005385,0.002939982,0.00008069586],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.008993672,0.0008936294,0.9843215,0.00108685,0.0001088561,0.0008159914,0.0002155646,0.0001611089,0.003402901],"genre_scores_gemma":[0.4950179,0.001806658,0.4922136,0.001310742,0.0003164785,0.007036163,0.0003290318,0.0001324099,0.001836939],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.1851711,"threshold_uncertainty_score":0.97929,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2144168223","doi":"10.1002/sim.5984","title":"The use of propensity score methods with survival or time‐to‐event outcomes: reporting measures of effect similar to those used in randomized experiments","year":2013,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":1441,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Clinical Evaluative Sciences; Public Health Ontario; University of Toronto","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences; Heart and Stroke Foundation of Canada","keywords":"Propensity score matching; Observational study; Marginal structural model; Inverse probability weighting; Medicine; Covariate; Randomized controlled trial; Confounding; Average treatment effect; Statistics; Hazard ratio; Selection bias; Population; Survival analysis; Matching (statistics); Proportional hazards model; Confidence interval; Internal medicine; Mathematics","authors":[{"name":"Peter C. Austin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4544700176042719,"gpt":0.5290019207742545,"spread":0.0745319031699826,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2174702,0.004157696,0.004999245,0.01026048,0.001575541,0.008248786,0.006004136,0.008134787,0.01001388],"category_scores_gemma":[0.5252035,0.001889699,0.008818037,0.01824831,0.006105338,0.01237269,0.008749089,0.01232288,0.003443172],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003038803,"about_ca_system_score_gemma":0.008339749,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002994108,"about_ca_topic_score_gemma":0.001900269,"domain_scores_codex":[0.6394306,0.2924809,0.02367798,0.0108429,0.03267649,0.0008911327],"domain_scores_gemma":[0.429904,0.4315079,0.05637686,0.06826769,0.01300956,0.0009340929],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0004632436,0.0002206709,0.007671661,0.01134305,0.006081732,0.0002169762,0.001714931,0.01972799,0.001002432,0.5535533,0.05548134,0.3425226],"study_design_scores_gemma":[0.0006025329,0.0006282846,0.004716466,0.005001329,0.001816547,0.0007356432,0.0003969387,0.03477466,0.003419492,0.7842777,0.1631424,0.0004880102],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.0007856977,0.007410959,0.9824929,0.002335476,0.001091803,0.001315961,0.001651181,0.000509948,0.002405997],"genre_scores_gemma":[0.03131308,0.01598787,0.9277407,0.004357662,0.002733326,0.01266883,0.002271846,0.0008063878,0.00212038],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7825298,"threshold_uncertainty_score":0.9649985,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2089198100","doi":"10.1186/1471-2288-8-70","title":"Reducing bias through directed acyclic graphs","year":2008,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":1410,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University; Jewish General Hospital","funders":"McGill University Health Centre; McGill University","keywords":"Directed acyclic graph; Confounding; Causal inference; Selection bias; Computer science; Outcome (game theory); Econometrics; Simple (philosophy); Causality (physics); Statistics; Mathematics; Algorithm","authors":[{"name":"Ian Shrier","is_ca":true},{"name":"Robert W. Platt","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9122827611924192,"gpt":0.6583154278917019,"spread":0.2539673333007173,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.03703995,0.001662157,0.001906058,0.006052916,0.001784208,0.002774684,0.003622494,0.002630305,0.007855822],"category_scores_gemma":[0.1719292,0.00148333,0.002484325,0.005784775,0.003535888,0.005335319,0.004162367,0.003786029,0.00129684],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002270475,"about_ca_system_score_gemma":0.00378701,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003274582,"about_ca_topic_score_gemma":0.003461995,"domain_scores_codex":[0.9499737,0.04094205,0.00146085,0.003475646,0.003581854,0.0005658538],"domain_scores_gemma":[0.6278923,0.3456967,0.007997833,0.01207709,0.005485676,0.0008503661],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0002926853,0.0001233308,0.005089389,0.001433803,0.0006265669,0.0004162618,0.001011164,0.1250056,0.0006653754,0.6188703,0.01036382,0.2361018],"study_design_scores_gemma":[0.00007767991,0.00003756237,0.0003225404,0.0001731952,0.0001208596,0.0001196256,0.00005991811,0.1184654,0.0003778402,0.8723541,0.007864072,0.0000272429],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.002645175,0.0005367111,0.9929096,0.001104882,0.00009985557,0.0001860576,0.0003406481,0.0003730884,0.00180395],"genre_scores_gemma":[0.1215719,0.001768062,0.8703969,0.001194147,0.0003767448,0.001411021,0.0009015774,0.0002098495,0.002169794],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9629601,"threshold_uncertainty_score":0.1958883,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1964475341","doi":"10.1002/sim.3150","title":"A critical appraisal of propensity‐score matching in the medical literature between 1996 and 2003","year":2007,"lang":"en","type":"review","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":1250,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences","keywords":"Propensity score matching; Covariate; Wilcoxon signed-rank test; Statistics; Observational study; Matching (statistics); Selection bias; Logistic regression; Sample size determination; Statistical significance; Medicine; Mathematics; Mann–Whitney U test","authors":[{"name":"Peter C. Austin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2610013184506083,"gpt":0.5385518179316583,"spread":0.27755049948105,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.05833513,0.001329439,0.004062122,0.02584374,0.000761193,0.003283175,0.001759189,0.002228014,0.002259823],"category_scores_gemma":[0.2239352,0.001473965,0.003369766,0.02596256,0.001959588,0.004014051,0.001824863,0.001456791,0.0005482028],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.008626286,"about_ca_system_score_gemma":0.01494324,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003820794,"about_ca_topic_score_gemma":0.006543594,"domain_scores_codex":[0.9621062,0.01544194,0.01266201,0.001348254,0.008068171,0.0003734498],"domain_scores_gemma":[0.7730316,0.1399897,0.03943323,0.003207063,0.04342225,0.0009161732],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0007162528,0.0000383889,0.002134661,0.2777888,0.00306015,0.0003837084,0.0006293024,0.0004497331,0.0003200697,0.004256745,0.028777,0.6814451],"study_design_scores_gemma":[0.0004692006,0.0005288425,0.01708895,0.5946848,0.01622293,0.001766398,0.0008723705,0.000550853,0.001195812,0.006874116,0.3596042,0.0001415402],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0004594776,0.9956713,0.0004080421,0.00229579,0.0005183068,0.0001847842,0.00007661323,0.00000605359,0.0003796209],"genre_scores_gemma":[0.006608132,0.9885823,0.001570967,0.00185631,0.0006772203,0.0003861758,0.0001445081,0.000005954185,0.0001685134],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9416649,"threshold_uncertainty_score":0.3085093,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2026656096","doi":"10.1136/bmj.b5087","title":"Problem of immortal time bias in cohort studies: example using statins for preventing progression of diabetes","year":2010,"lang":"en","type":"article","venue":"BMJ","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":1155,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Jewish General Hospital; Kingston Health Sciences Centre; McGill University; Queen's University","funders":"Canadian Institutes of Health Research","keywords":"Medicine; Diabetes mellitus; Cohort; Internal medicine; Endocrinology","authors":[{"name":"Linda E. Lévesque","is_ca":true},{"name":"J. A Hanley","is_ca":true},{"name":"Abbas Kezouh","is_ca":true},{"name":"Samy Suissa","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2976455350419575,"gpt":0.5009101891130961,"spread":0.2032646540711386,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1767487,0.00105577,0.003502779,0.002412648,0.001785391,0.003184696,0.002684435,0.008718506,0.00302433],"category_scores_gemma":[0.4338726,0.0009828607,0.004736546,0.003186682,0.007211768,0.006371436,0.003455001,0.01078721,0.000408807],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001534206,"about_ca_system_score_gemma":0.001900682,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004728219,"about_ca_topic_score_gemma":0.003523619,"domain_scores_codex":[0.8743265,0.1105205,0.003619842,0.005941259,0.00478916,0.0008026496],"domain_scores_gemma":[0.3947338,0.5614565,0.01139445,0.02440605,0.007198845,0.0008102486],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00220828,0.0002173703,0.04088666,0.003756139,0.006990506,0.003198452,0.003457029,0.02263799,0.001474362,0.7132538,0.02936157,0.1725578],"study_design_scores_gemma":[0.0006909545,0.0001969509,0.004038032,0.0006176957,0.001366454,0.001819618,0.0003058435,0.03141654,0.0007964139,0.9452678,0.01336219,0.0001214731],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.04118169,0.03044707,0.8060591,0.1096369,0.003328102,0.0004052757,0.001097638,0.0004744086,0.007369834],"genre_scores_gemma":[0.656632,0.01424389,0.2910028,0.02377501,0.006298621,0.0009157691,0.0004519772,0.0002846548,0.006395217],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8232513,"threshold_uncertainty_score":0.9347477,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2164498941","doi":"10.1002/sim.2580","title":"A comparison of the ability of different propensity score models to balance measured variables between treated and untreated subjects: a Monte Carlo study","year":2006,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":1134,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"","keywords":"Propensity score matching; Confounding; Observational study; Outcome (game theory); Statistics; Matching (statistics); Medicine; Average treatment effect; Variables; Econometrics; Mathematics","authors":[{"name":"Peter C. Austin","is_ca":true},{"name":"Paul Grootendorst","is_ca":true},{"name":"Geoffrey M. Anderson","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2043349314626578,"gpt":0.4063998310840797,"spread":0.2020648996214219,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.04845132,0.0008524176,0.001680212,0.001426422,0.0007175,0.001347724,0.001399815,0.001813798,0.001168846],"category_scores_gemma":[0.1220498,0.0007098514,0.002424733,0.001186239,0.001682484,0.001906955,0.001131405,0.001636814,0.0001568893],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001986346,"about_ca_system_score_gemma":0.00167404,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007454513,"about_ca_topic_score_gemma":0.003993871,"domain_scores_codex":[0.9815192,0.01597861,0.0005199485,0.0008710177,0.0007402697,0.0003708512],"domain_scores_gemma":[0.740361,0.2402902,0.005693611,0.009473369,0.003088922,0.001092955],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.006049864,0.001247361,0.04852712,0.0002950386,0.002514031,0.0002295064,0.0006461812,0.8653165,0.001168928,0.03875001,0.001361311,0.03389415],"study_design_scores_gemma":[0.0004035115,0.00111595,0.007384622,0.00008313108,0.0002840767,0.0001026954,0.000104505,0.9747498,0.0009780137,0.01404984,0.0006708329,0.00007308325],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8463221,0.001526185,0.147948,0.000655163,0.00006792765,0.0006625701,0.0003037346,0.0001557298,0.002358643],"genre_scores_gemma":[0.9491639,0.0005614044,0.04845314,0.0002611997,0.00003691712,0.0004652184,0.000405913,0.00005329924,0.0005989801],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9515487,"threshold_uncertainty_score":0.2562381,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2103371988","doi":"10.1002/sim.5705","title":"The performance of different propensity score methods for estimating marginal hazard ratios","year":2012,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":995,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Work & Health; Institute for Clinical Evaluative Sciences; Public Health Ontario; University of Toronto","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences; Heart and Stroke Foundation of Canada","keywords":"Propensity score matching; Covariate; Statistics; Odds ratio; Estimator; Matching (statistics); Medicine; Mathematics","authors":[{"name":"Peter C. Austin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.220207531187187,"gpt":0.4904911441960751,"spread":0.2702836130088881,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.08822343,0.002135994,0.00260438,0.006001812,0.001082461,0.00325052,0.003723077,0.001975625,0.004914273],"category_scores_gemma":[0.2533046,0.001278077,0.005449877,0.007219727,0.001711088,0.004071855,0.002929022,0.003688971,0.001151221],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001936039,"about_ca_system_score_gemma":0.003483006,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008023952,"about_ca_topic_score_gemma":0.006251296,"domain_scores_codex":[0.9445329,0.0427882,0.003535114,0.003798114,0.004837677,0.0005080369],"domain_scores_gemma":[0.8561339,0.1137137,0.00880006,0.01264468,0.008081779,0.0006259437],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001250341,0.0003003823,0.06174538,0.002009457,0.006442712,0.0002785465,0.001276182,0.2026966,0.001411559,0.1096856,0.01044673,0.6024566],"study_design_scores_gemma":[0.000765769,0.0005923841,0.02605526,0.001010339,0.001514827,0.000585917,0.0003417928,0.788509,0.003236327,0.1551221,0.02180606,0.0004602984],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01159006,0.002482056,0.982255,0.0005612064,0.0001496168,0.0007394147,0.0005586121,0.0005693061,0.001094631],"genre_scores_gemma":[0.149355,0.003587149,0.8403804,0.0004836649,0.0002539513,0.002448234,0.001760233,0.0005919473,0.001139475],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9117765,"threshold_uncertainty_score":0.4665757,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2404984931","doi":"10.1016/j.jclinepi.2016.04.014","title":"Specifying a target trial prevents immortal time bias and other self-inflicted injuries in observational analyses","year":2016,"lang":"en","type":"review","venue":"Journal of Clinical Epidemiology","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":974,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Jewish General Hospital; McGill University; McGill University Health Centre","funders":"National Center for Advancing Translational Sciences; National Cancer Institute","keywords":"Observational study; Emulation; Randomized controlled trial; Computer science; Medical physics; Research design; Simple (philosophy); Clinical trial; Medicine; Psychology; Statistics; Social psychology; Surgery; Mathematics","authors":[{"name":"Miguel A. Hernán","is_ca":false},{"name":"Brian C. Sauer","is_ca":false},{"name":"Sonia Hernández–Dı́az","is_ca":false},{"name":"Robert W. Platt","is_ca":true},{"name":"Ian Shrier","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8894193828691276,"gpt":0.6870352047744127,"spread":0.2023841780947149,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.05851218,0.000979829,0.0036489,0.001451765,0.0003373972,0.002247904,0.003214155,0.003595918,0.00424848],"category_scores_gemma":[0.1912318,0.0007543521,0.004541694,0.001803295,0.001870405,0.002940983,0.001306602,0.004668731,0.0006396406],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007071297,"about_ca_system_score_gemma":0.003477061,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008003105,"about_ca_topic_score_gemma":0.0009327474,"domain_scores_codex":[0.9531825,0.0390859,0.002961678,0.001660888,0.002832215,0.000276917],"domain_scores_gemma":[0.7363408,0.2502698,0.00575043,0.005638171,0.001718684,0.0002820024],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.001448012,0.0002092795,0.003008103,0.04736565,0.007548453,0.0001866949,0.0003279383,0.01182065,0.0002834218,0.1249196,0.01003843,0.7928438],"study_design_scores_gemma":[0.003796926,0.001073286,0.005403555,0.03127364,0.01788552,0.001429597,0.0001875918,0.03917804,0.001595954,0.8060647,0.09194345,0.0001676855],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"methods","genre_scores_codex":[0.006353679,0.5020225,0.4755419,0.008640566,0.001262983,0.000540177,0.0007292248,0.0004182613,0.004490711],"genre_scores_gemma":[0.2604527,0.398072,0.3217796,0.008448577,0.003832814,0.002842221,0.001144804,0.0003175633,0.003109784],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9414878,"threshold_uncertainty_score":0.3094457,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W2132557528","doi":"10.3386/w13039","title":"Regression Discontinuity Designs: A Guide to Practice","year":2007,"lang":"en","type":"report","venue":"National Bureau of Economic Research","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":888,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"Social Sciences and Humanities Research Council of Canada; National Science Foundation","keywords":"Regression discontinuity design; Discontinuity (linguistics); Regression; Computer science; Econometrics; Statistics; Mathematics","authors":[{"name":"Guido W. Imbens","is_ca":false},{"name":"Thomas Lemieux","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7507059545647676,"gpt":0.6941272470872978,"spread":0.05657870747746985,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09030104,0.002758994,0.004278574,0.005766373,0.001278387,0.00638596,0.007361208,0.007150094,0.02131987],"category_scores_gemma":[0.1731178,0.003018493,0.00337223,0.007005304,0.003942677,0.003725791,0.003717046,0.01079674,0.01103767],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002669733,"about_ca_system_score_gemma":0.008812854,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004150753,"about_ca_topic_score_gemma":0.005103311,"domain_scores_codex":[0.909315,0.07429093,0.005018477,0.002735717,0.008197663,0.0004423238],"domain_scores_gemma":[0.8346459,0.1426892,0.003602442,0.008811633,0.009305669,0.0009451432],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0002888841,0.0004259888,0.001481437,0.006675344,0.0007014074,0.0005027069,0.002331723,0.01660748,0.0009840616,0.3730246,0.1760256,0.4209507],"study_design_scores_gemma":[0.000566496,0.0003901676,0.0009278332,0.005714095,0.0002432385,0.0005597613,0.0005352041,0.02829406,0.001143884,0.5018719,0.4595134,0.0002400545],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0003219549,0.008900467,0.9714057,0.003726405,0.0008750232,0.00293971,0.001749521,0.001617827,0.008463297],"genre_scores_gemma":[0.003852276,0.007582771,0.9758968,0.001220044,0.0003550916,0.007732264,0.0006187738,0.0005801179,0.002161747],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.09030104,"threshold_uncertainty_score":0.4775633,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2127992265","doi":"10.1093/ije/dyp334","title":"Illustrating bias due to conditioning on a collider","year":2009,"lang":"en","type":"article","venue":"International Journal of Epidemiology","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":823,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University","funders":"National Institute of Environmental Health Sciences; National Institute of Allergy and Infectious Diseases; National Cancer Institute; National Institute on Alcohol Abuse and Alcoholism","keywords":"Collider; Conditioning; Selection bias; Outcome (game theory); Selection (genetic algorithm); Association (psychology); Confounding; Computer science; Physics; Statistics; Psychology; Mathematics; Artificial intelligence; Particle physics; Mathematical economics","authors":[{"name":"Stephen R. Cole","is_ca":false},{"name":"Robert W. Platt","is_ca":true},{"name":"Enrique F. Schisterman","is_ca":false},{"name":"Haitao Chu","is_ca":false},{"name":"Daniel Westreich","is_ca":false},{"name":"David J. Richardson","is_ca":false},{"name":"Charles Poole","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3891479896255752,"gpt":0.5215592792262753,"spread":0.1324112896007001,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0687522,0.0007451895,0.001555937,0.001048456,0.00148144,0.001836747,0.001928825,0.002811924,0.008119376],"category_scores_gemma":[0.2153856,0.0005230742,0.002186329,0.00183031,0.00647478,0.003449895,0.004588967,0.004704218,0.0004520124],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001826903,"about_ca_system_score_gemma":0.001625636,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004566786,"about_ca_topic_score_gemma":0.002448003,"domain_scores_codex":[0.9584944,0.03195968,0.001443738,0.002968943,0.003901902,0.001231382],"domain_scores_gemma":[0.6769437,0.283583,0.01191584,0.02202625,0.004650919,0.0008802562],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0008913076,0.0001448688,0.02647579,0.0003752189,0.0007034442,0.001477066,0.003017263,0.01670218,0.001208489,0.877115,0.005819594,0.06606987],"study_design_scores_gemma":[0.0003216781,0.0001860363,0.003700349,0.0001448901,0.0002520781,0.0007364415,0.0002268052,0.046608,0.001504973,0.9398205,0.006440674,0.00005756803],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.09242499,0.001432925,0.8686174,0.01838076,0.0004095846,0.0005151324,0.0005860284,0.0003649897,0.01726815],"genre_scores_gemma":[0.8258871,0.0007643208,0.1641562,0.005361356,0.0004028725,0.000581377,0.0002236326,0.00007693269,0.002546201],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9312478,"threshold_uncertainty_score":0.3636007,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4393132304","doi":"10.1016/j.envint.2024.108602","title":"A tool to assess risk of bias in non-randomized follow-up studies of exposure effects (ROBINS-E)","year":2024,"lang":"en","type":"article","venue":"Environment International","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":806,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Cochrane; McMaster University; Ottawa Hospital; Impact","funders":"Medical Research Council; National Institute of Environmental Health Sciences; National Institute for Health and Care Research; National Institutes of Health; Cancer Research UK; CHDI Foundation; National Council for Air and Stream Improvement; NIHR Bristol Biomedical Research Centre; United Soybean Board","keywords":"Randomized controlled trial; Environmental health; Environmental science; Medicine; Internal medicine","authors":[{"name":"Julian P. T. Higgins","is_ca":false},{"name":"Rebecca L. Morgan","is_ca":true},{"name":"Andrew A. Rooney","is_ca":false},{"name":"Kyla W. Taylor","is_ca":false},{"name":"Kristina A. Thayer","is_ca":false},{"name":"Raquel A. Silva","is_ca":false},{"name":"Courtney Lemeris","is_ca":false},{"name":"Elie A. Akl","is_ca":false},{"name":"Thomas F. Bateson","is_ca":false},{"name":"Nancy D Berkman","is_ca":false},{"name":"Barbara Glenn","is_ca":false},{"name":"Asbjørn Hróbjartsson","is_ca":false},{"name":"Judy S. LaKind","is_ca":false},{"name":"Alexandra McAleenan","is_ca":false},{"name":"Joerg J Meerpohl","is_ca":false},{"name":"Rebecca Nachman","is_ca":false},{"name":"Julie Obbagy","is_ca":false},{"name":"Annette M. O’Connor","is_ca":false},{"name":"Elizabeth G. Radke","is_ca":false},{"name":"Jelena Savović","is_ca":false},{"name":"Holger J. Schünemann","is_ca":true},{"name":"Beverley Shea","is_ca":true},{"name":"Kate Tilling","is_ca":false},{"name":"Jos Verbeek","is_ca":false},{"name":"Meera Viswanathan","is_ca":false},{"name":"Jonathan A C Sterne","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1252290287888377,"gpt":0.4028231892404779,"spread":0.2775941604516402,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.634735,0.007479682,0.01038552,0.04406264,0.002422041,0.009822313,0.006726859,0.007588742,0.02182468],"category_scores_gemma":[0.8217387,0.004222886,0.03360912,0.02381319,0.006918431,0.01323023,0.01724186,0.007369274,0.002965883],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.008568889,"about_ca_system_score_gemma":0.01719823,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001862638,"about_ca_topic_score_gemma":0.003466036,"domain_scores_codex":[0.2149442,0.5748643,0.1536325,0.01017913,0.04471756,0.001662312],"domain_scores_gemma":[0.05851803,0.8561773,0.03621797,0.02051747,0.02772661,0.000842569],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.009332706,0.0004601839,0.03545513,0.1625951,0.03750682,0.0007583775,0.007722257,0.01521325,0.00156523,0.08792187,0.0741244,0.5673447],"study_design_scores_gemma":[0.01810513,0.005793126,0.05102591,0.1385178,0.04958469,0.004248572,0.006027467,0.09949417,0.01050386,0.2832738,0.3301835,0.003242164],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01138509,0.01362072,0.8329772,0.007021471,0.003049669,0.0903815,0.02046708,0.009005071,0.01209227],"genre_scores_gemma":[0.02838466,0.001325615,0.8711556,0.00110603,0.0002677029,0.09471441,0.001930522,0.0004575797,0.0006578413],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.365265,"threshold_uncertainty_score":0.4504368,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2142912667","doi":"10.1001/jama.297.3.278","title":"Analysis of Observational Studies in the Presence of Treatment Selection Bias","year":2007,"lang":"en","type":"article","venue":"JAMA","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":788,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Health Sciences Centre; St. Michael's Hospital; Sunnybrook Health Science Centre; Institute for Clinical Evaluative Sciences; University of Toronto; York University","funders":"National Institute on Aging; Institute for Clinical Evaluative Sciences; Canadian Institutes of Health Research; Dartmouth College","keywords":"Medicine; Propensity score matching; Relative risk; Observational study; Confidence interval; Cardiac catheterization; Myocardial infarction; Selection bias; Internal medicine; Proportional hazards model; Cohort study; Emergency medicine","authors":[{"name":"Thérèse A. Stukel","is_ca":true},{"name":"Elliott S. Fisher","is_ca":true},{"name":"David E. Wennberg","is_ca":true},{"name":"David A. Alter","is_ca":true},{"name":"Daniel J. Gottlieb","is_ca":true},{"name":"Marian J. Vermeulen","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5264523418573861,"gpt":0.503462976493724,"spread":0.02298936536366214,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4962091,0.001777566,0.006804421,0.007068696,0.001647803,0.005792558,0.003936998,0.005081857,0.002604682],"category_scores_gemma":[0.737581,0.001735512,0.009700562,0.01077152,0.004450665,0.003680573,0.004875804,0.004019257,0.000342213],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003071018,"about_ca_system_score_gemma":0.006512826,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002051437,"about_ca_topic_score_gemma":0.001501154,"domain_scores_codex":[0.2243666,0.6842287,0.0485758,0.01800681,0.02344178,0.001380366],"domain_scores_gemma":[0.116254,0.7428423,0.08607119,0.04569504,0.008356133,0.0007812799],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.006561192,0.0003635557,0.5523098,0.02826001,0.1817647,0.002386453,0.003417478,0.01462419,0.001291375,0.0526931,0.0104316,0.1458967],"study_design_scores_gemma":[0.01097617,0.004788965,0.3624138,0.02077281,0.126357,0.004365416,0.002311983,0.1309611,0.005232648,0.2755475,0.05533636,0.0009362007],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1311039,0.06596904,0.7570085,0.01149868,0.004984085,0.01662139,0.006409445,0.0007865221,0.005618521],"genre_scores_gemma":[0.7161672,0.008176116,0.2360319,0.006286675,0.002918694,0.02632141,0.002970088,0.0001721576,0.0009558853],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.5037909,"threshold_uncertainty_score":0.6212639,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2016403661","doi":"10.1002/bimj.200810488","title":"Some Methods of Propensity‐Score Matching had Superior Performance to Others: Results of an Empirical Investigation and Monte Carlo simulations","year":2009,"lang":"en","type":"article","venue":"Biometrical Journal","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":769,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"","keywords":"Propensity score matching; Statistics; Monte Carlo method; Matching (statistics); Standard deviation; Logit; Mathematics; Medicine","authors":[{"name":"Peter C. Austin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.374473599393049,"gpt":0.4837530024844751,"spread":0.109279403091426,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1601167,0.001600468,0.003490126,0.004028091,0.001334855,0.002938575,0.002063923,0.002888115,0.003480498],"category_scores_gemma":[0.4240412,0.0009736824,0.004367013,0.005659943,0.002588862,0.005545076,0.001741923,0.002224531,0.0006122217],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001919297,"about_ca_system_score_gemma":0.002767602,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01098205,"about_ca_topic_score_gemma":0.004551782,"domain_scores_codex":[0.9190413,0.06629736,0.004355052,0.004371841,0.004895106,0.001039371],"domain_scores_gemma":[0.2738614,0.6694188,0.01419554,0.02803061,0.01353532,0.0009583452],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00733176,0.001388351,0.1587101,0.002799296,0.009898307,0.0003732284,0.001866974,0.5436212,0.001168535,0.03923329,0.006658756,0.2269503],"study_design_scores_gemma":[0.0009120329,0.00172068,0.04012752,0.0004354954,0.001274669,0.0002823612,0.0003854761,0.9239079,0.001972218,0.02522084,0.003481693,0.0002790436],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6003972,0.00593877,0.3795111,0.002430291,0.0001976652,0.002140918,0.001652113,0.000975211,0.006756803],"genre_scores_gemma":[0.854463,0.00163781,0.1401811,0.0004439019,0.0001354878,0.0009740955,0.001044119,0.0002717242,0.000848673],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8398833,"threshold_uncertainty_score":0.8467882,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2013188809","doi":"10.1023/b:ejep.0000036568.02655.f8","title":"Loss to Follow-Up in Cohort Studies: How Much is Too Much?","year":2004,"lang":"en","type":"article","venue":"European Journal of Epidemiology","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":732,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Lunenfeld-Tanenbaum Research Institute; Mount Sinai Hospital; Institute for Work & Health; University of Toronto","funders":"","keywords":"Medicine; Confounding; Cohort; Logistic regression; Missing data; Cohort study; Odds; Statistics; Demography; Internal medicine; Mathematics","authors":[{"name":"Vicki L. Kristman","is_ca":true},{"name":"Michael Manno","is_ca":true},{"name":"Pierre Côté","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3441335696772531,"gpt":0.4711298412925512,"spread":0.1269962716152981,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4180975,0.001878546,0.009481296,0.003188269,0.003199247,0.005921212,0.009879319,0.01102037,0.004982052],"category_scores_gemma":[0.6889658,0.002675734,0.009446476,0.006276231,0.01173415,0.01827515,0.006952909,0.01842756,0.0005774865],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00301679,"about_ca_system_score_gemma":0.004877098,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008757475,"about_ca_topic_score_gemma":0.01173984,"domain_scores_codex":[0.6709807,0.2706053,0.02197904,0.01806793,0.01557573,0.002791283],"domain_scores_gemma":[0.1490204,0.7876668,0.02438668,0.03090697,0.005419747,0.002599583],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.007661877,0.0008631864,0.2849585,0.01834326,0.05167634,0.001834308,0.006618555,0.01397677,0.0003681829,0.1017633,0.03998075,0.4719552],"study_design_scores_gemma":[0.002526494,0.001590746,0.09972019,0.01240015,0.0388668,0.002686568,0.002142648,0.045602,0.000777773,0.7590706,0.03404327,0.000572673],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.0689307,0.2488172,0.4104423,0.2541144,0.01059893,0.0008532885,0.002921392,0.0006815958,0.002640347],"genre_scores_gemma":[0.7433625,0.05313065,0.09774308,0.0818928,0.01595168,0.00214357,0.001826032,0.0003945596,0.003555096],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.5819025,"threshold_uncertainty_score":0.7175894,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2891367027","doi":"10.1513/annalsats.201808-564ps","title":"Control of Confounding and Reporting of Results in Causal Inference Studies. Guidance for Authors from Editors of Respiratory, Sleep, and Critical Care Journals","year":2018,"lang":"en","type":"article","venue":"Annals of the American Thoracic Society","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":722,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University; McGill University; Queen's University","funders":"","keywords":"Confounding; Causal inference; Fallacy; Medicine; Statistics; Econometrics; Computer science; Mathematics; Linguistics","authors":[{"name":"David J. Lederer","is_ca":false},{"name":"Scott C. Bell","is_ca":false},{"name":"Richard D. Branson","is_ca":false},{"name":"James D. Chalmers","is_ca":false},{"name":"Rachel Marshall","is_ca":false},{"name":"David M. Maslove","is_ca":true},{"name":"David E. Ost","is_ca":false},{"name":"Naresh M. Punjabi","is_ca":false},{"name":"Michael Schätz","is_ca":false},{"name":"Alan R Smyth","is_ca":false},{"name":"Paul W. Stewart","is_ca":false},{"name":"Samy Suissa","is_ca":true},{"name":"Alex A. Adjei","is_ca":false},{"name":"Cezmi A. Akdiş","is_ca":false},{"name":"Élie Azoulay","is_ca":false},{"name":"Jan Bakker","is_ca":false},{"name":"Zuhair K. Ballas","is_ca":false},{"name":"Philip G. Bardin","is_ca":false},{"name":"Esther Barreiro","is_ca":false},{"name":"Rinaldo Bellomo","is_ca":false},{"name":"Jonathan A. Bernstein","is_ca":false},{"name":"Vito Brusasco","is_ca":false},{"name":"Timothy G. Buchman","is_ca":false},{"name":"Sudhansu Chokroverty","is_ca":false},{"name":"Nancy A. Collop","is_ca":false},{"name":"James D. Crapo","is_ca":false},{"name":"Dominic A. Fitzgerald","is_ca":false},{"name":"Lauren Hale","is_ca":false},{"name":"Nicholas Hart","is_ca":false},{"name":"Felix Herth","is_ca":false},{"name":"Theodore J. Iwashyna","is_ca":false},{"name":"Gísli Jenkins","is_ca":false},{"name":"Martin Kolb","is_ca":true},{"name":"Guy B. Marks","is_ca":false},{"name":"Peter J. Mazzone","is_ca":false},{"name":"J. Randall Moorman","is_ca":false},{"name":"Thomas M. Murphy","is_ca":false},{"name":"Terry L. Noah","is_ca":false},{"name":"Paul N. Reynolds","is_ca":false},{"name":"Dieter Riemann","is_ca":false},{"name":"Richard Russell","is_ca":false},{"name":"Aziz Sheikh","is_ca":false},{"name":"Giovanni Sotgiu","is_ca":false},{"name":"Erik R. Swenson","is_ca":false},{"name":"Rhonda D. Szczesniak","is_ca":false},{"name":"Ronald Szymusiak","is_ca":false},{"name":"Jean–Louis Teboul","is_ca":false},{"name":"Jean‐Louis Vincent","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4090745820242512,"gpt":0.5752142049379743,"spread":0.166139622913723,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.7191381,0.002934813,0.007901735,0.01531048,0.006452209,0.02012183,0.01367649,0.02037773,0.01152806],"category_scores_gemma":[0.9262456,0.004283652,0.004478055,0.0132328,0.02010208,0.02381686,0.01229245,0.02915555,0.01003895],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006360445,"about_ca_system_score_gemma":0.03562342,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002360648,"about_ca_topic_score_gemma":0.003558801,"domain_scores_codex":[0.2080951,0.6230816,0.09536444,0.008904447,0.06252616,0.00202835],"domain_scores_gemma":[0.02726836,0.8163003,0.05429543,0.03340573,0.06627011,0.002460033],"domain_codex":"methods","domain_gemma":"reporting","domain_candidate":"reporting","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0006573169,0.0001462871,0.001935114,0.01128752,0.0008118724,0.0003995989,0.009037915,0.0007392662,0.000607822,0.0508524,0.7071018,0.2164231],"study_design_scores_gemma":[0.0009084641,0.0004382441,0.002376979,0.02421363,0.000728053,0.001053785,0.00232526,0.005139656,0.00171391,0.2289218,0.7316939,0.0004865268],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"methods","genre_scores_codex":[0.0009950303,0.05586488,0.3329791,0.4936485,0.103129,0.004722694,0.001144186,0.00162557,0.005890992],"genre_scores_gemma":[0.02196413,0.04012019,0.6968084,0.1368341,0.08304755,0.01466808,0.0007092392,0.001783141,0.00406523],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.2808619,"threshold_uncertainty_score":0.3463528,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2024692342","doi":"10.1016/j.jeconom.2003.10.006","title":"How robust is the evidence on the effects of college quality? Evidence from matching","year":2004,"lang":"en","type":"article","venue":"Journal of Econometrics","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":690,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Propensity score matching; Matching (statistics); Regression; Quality (philosophy); Linear regression; Econometrics; Statistics; Average treatment effect; Regression analysis; Mathematics","authors":[{"name":"Dan A. Black","is_ca":false},{"name":"Jeffrey A. Smith","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4967390886913728,"gpt":0.404828367581793,"spread":0.09191072110957987,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.08900952,0.00109086,0.004936589,0.00437198,0.002025674,0.005759739,0.007254205,0.006589694,0.03221538],"category_scores_gemma":[0.4670679,0.0009970024,0.004685979,0.01139275,0.006261171,0.00597164,0.006140175,0.00388899,0.002467104],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001354335,"about_ca_system_score_gemma":0.001868799,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01596021,"about_ca_topic_score_gemma":0.005591518,"domain_scores_codex":[0.9018046,0.06829105,0.009099577,0.008938082,0.008588186,0.003278548],"domain_scores_gemma":[0.2588137,0.58089,0.06174454,0.08466306,0.01003137,0.003857365],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.02774613,0.003637977,0.5268528,0.006603045,0.04148316,0.0008603358,0.003610308,0.01320962,0.001287464,0.1089798,0.0161453,0.249584],"study_design_scores_gemma":[0.0118813,0.004747625,0.5432152,0.003000634,0.04937813,0.0008295741,0.004606572,0.0180531,0.009565986,0.3095365,0.04474343,0.0004419575],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8438649,0.04233491,0.03775464,0.02620451,0.001145146,0.0005830307,0.005830029,0.0002264087,0.04205636],"genre_scores_gemma":[0.9878555,0.003383377,0.003029934,0.002193948,0.0004942612,0.00007312796,0.0009640863,0.00004905185,0.001956781],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9109905,"threshold_uncertainty_score":0.470733,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2160704894","doi":"10.1093/aje/kwq224","title":"Statistical Criteria for Selecting the Optimal Number of Untreated Subjects Matched to Each Treated Subject When Using Many-to-One Matching on the Propensity Score","year":2010,"lang":"en","type":"article","venue":"American Journal of Epidemiology","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":667,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Clinical Evaluative Sciences","funders":"Canadian Institutes of Health Research; University of Toronto","keywords":"Propensity score matching; Matching (statistics); Statistics; Estimator; Mathematics; Average treatment effect; Mean squared error; Observational study; Medicine","authors":[{"name":"Peter C. Austin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.260989989819204,"gpt":0.4622396814012198,"spread":0.2012496915820159,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1482718,0.001310897,0.005615314,0.006165978,0.001677188,0.002954714,0.003664391,0.003901966,0.003036596],"category_scores_gemma":[0.3915726,0.001245255,0.002389806,0.00506803,0.003127254,0.002688161,0.003101914,0.00332539,0.0005988106],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001633417,"about_ca_system_score_gemma":0.003623835,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001245572,"about_ca_topic_score_gemma":0.001125724,"domain_scores_codex":[0.8817927,0.09461372,0.009956491,0.004485859,0.008235248,0.0009160957],"domain_scores_gemma":[0.6170748,0.3365382,0.01699555,0.01336199,0.01368485,0.002344655],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.008826949,0.001685319,0.06101311,0.002980879,0.003142511,0.0009659466,0.001078485,0.17738,0.004580033,0.1637856,0.01653469,0.5580266],"study_design_scores_gemma":[0.005095374,0.003955806,0.02921369,0.00190152,0.001675815,0.001119473,0.0005274167,0.6193871,0.008762997,0.3093519,0.01857612,0.0004327903],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.03433602,0.001586408,0.9554309,0.002180278,0.0002031991,0.003503233,0.0003865751,0.0003191236,0.002054253],"genre_scores_gemma":[0.1609482,0.0004902479,0.8323458,0.0009043633,0.0002135156,0.004062646,0.0004467453,0.0001009241,0.0004875907],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8517282,"threshold_uncertainty_score":0.7841458,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2051866975","doi":"10.1016/j.jtcvs.2007.07.021","title":"Propensity-score matching in the cardiovascular surgery literature from 2004 to 2006: A systematic review and suggestions for improvement","year":2007,"lang":"en","type":"review","venue":"Journal of Thoracic and Cardiovascular Surgery","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":598,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences","keywords":"Propensity score matching; Systematic review; Matching (statistics); Medicine; MEDLINE; General surgery; Surgery; Political science; Pathology","authors":[{"name":"Peter C. Austin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.19215338945083,"gpt":0.4069807759266218,"spread":0.2148273864757918,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.04690378,0.00159779,0.0129872,0.01088956,0.000633801,0.002926758,0.002755844,0.002086148,0.003196317],"category_scores_gemma":[0.1364134,0.001339023,0.01428071,0.01446768,0.001307793,0.004267544,0.002270395,0.001716677,0.0002397552],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003067232,"about_ca_system_score_gemma":0.01806114,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006703288,"about_ca_topic_score_gemma":0.02048842,"domain_scores_codex":[0.9665933,0.01665877,0.01052456,0.002634339,0.003213348,0.0003756809],"domain_scores_gemma":[0.8939812,0.08120663,0.01660139,0.002179668,0.00553621,0.0004948701],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0006999846,0.00005450568,0.006894065,0.7202207,0.07717006,0.0001230075,0.0002836117,0.000689202,0.0001786061,0.00126309,0.003135069,0.1892881],"study_design_scores_gemma":[0.002648001,0.000743149,0.02451222,0.4401195,0.4955629,0.0006783635,0.0006620529,0.001150823,0.0005959463,0.006676752,0.02649577,0.0001544627],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.001372962,0.9953467,0.00151867,0.0009595946,0.0001998291,0.000240783,0.000249198,0.00001243989,0.00009985153],"genre_scores_gemma":[0.03949598,0.9491074,0.008787117,0.001168058,0.0002918047,0.0006590014,0.0003850369,0.000009346149,0.00009620495],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9870128,"threshold_uncertainty_score":0.2480538,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2117317003","doi":"10.1002/pds.1357","title":"Immortal time bias in observational studies of drug effects","year":2007,"lang":"en","type":"article","venue":"Pharmacoepidemiology and Drug Safety","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":569,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true},"ca_institutions":"McGill University; McGill University Health Centre","funders":"Canadian Institutes of Health Research; AstraZeneca","keywords":"Medicine; Observational study; Pharmacoepidemiology; COPD; Cohort study; Drug; Disease; Pulmonary disease; Cohort; Internal medicine; Intensive care medicine; Pharmacology; Medical prescription","authors":[{"name":"Samy Suissa","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3359460404749271,"gpt":0.5112862260978849,"spread":0.1753401856229578,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3560762,0.001469097,0.002964022,0.004330137,0.001682092,0.00367028,0.003787249,0.00356492,0.002724841],"category_scores_gemma":[0.6126016,0.001276056,0.006300724,0.004620301,0.008384516,0.003993156,0.004224182,0.004460725,0.0005386668],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002490164,"about_ca_system_score_gemma":0.002745578,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00504805,"about_ca_topic_score_gemma":0.002692865,"domain_scores_codex":[0.5990047,0.3397064,0.02382316,0.01900837,0.01648568,0.001971628],"domain_scores_gemma":[0.2920925,0.5566009,0.05983131,0.08024321,0.0101796,0.001052557],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.003550082,0.0001847508,0.6575087,0.006006694,0.02239146,0.002032014,0.006722813,0.02773799,0.0009169948,0.1630981,0.005734733,0.1041157],"study_design_scores_gemma":[0.001661232,0.001889636,0.2983094,0.006491232,0.01272018,0.004798803,0.001687275,0.09815542,0.005799208,0.5242372,0.0435483,0.0007021329],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1732664,0.04072936,0.7547255,0.01143796,0.003863508,0.002899792,0.00344698,0.0007644944,0.00886607],"genre_scores_gemma":[0.9009963,0.004534064,0.08230861,0.004726306,0.001300382,0.002631056,0.001416328,0.0001913649,0.001895653],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.6439238,"threshold_uncertainty_score":0.7940727,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2102520493","doi":"10.1002/sim.2328","title":"A comparison of propensity score methods: a case‐study estimating the effectiveness of post‐AMI statin use","year":2005,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":565,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"","keywords":"Propensity score matching; Statin; Statistics; Medicine; Internal medicine; Computer science; Mathematics","authors":[{"name":"Peter C. Austin","is_ca":true},{"name":"Muhammad Mamdani","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3195657084475746,"gpt":0.5520056064708857,"spread":0.2324398980233111,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1602731,0.001386997,0.002962042,0.003939331,0.0009784806,0.002534657,0.002575693,0.004596793,0.004196114],"category_scores_gemma":[0.3171187,0.001156537,0.007926463,0.002899309,0.002142516,0.0027392,0.002530872,0.00206121,0.0004343371],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00171994,"about_ca_system_score_gemma":0.001738637,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002435663,"about_ca_topic_score_gemma":0.001369295,"domain_scores_codex":[0.7997295,0.183722,0.005987247,0.0035064,0.006288556,0.0007662192],"domain_scores_gemma":[0.5786208,0.381482,0.01477414,0.01982217,0.004505003,0.0007958965],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.12085,0.006861855,0.314453,0.006946105,0.07888829,0.002213478,0.002451003,0.06738662,0.00223261,0.07985225,0.005979885,0.311885],"study_design_scores_gemma":[0.04950507,0.05221142,0.1831572,0.003986435,0.06093164,0.006454949,0.003474533,0.5229113,0.005480115,0.09008022,0.0206837,0.001123513],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"methods","genre_scores_codex":[0.6975621,0.0377024,0.2486651,0.005035453,0.001027204,0.003872346,0.001404724,0.0001706066,0.004559962],"genre_scores_gemma":[0.9365861,0.004597637,0.05500231,0.0003704702,0.0002762365,0.002023075,0.0004936794,0.00004849537,0.0006019714],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.1602731,"threshold_uncertainty_score":0.8476155,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2063548620","doi":"10.1016/j.jclinepi.2004.10.016","title":"Propensity score methods gave similar results to traditional regression modeling in observational studies: a systematic review","year":2005,"lang":"en","type":"review","venue":"Journal of Clinical Epidemiology","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":497,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto; Institute for Clinical Evaluative Sciences","funders":"","keywords":"Propensity score matching; Observational study; Confounding; Medicine; Odds ratio; Statistics; Regression analysis; Hazard ratio; Confidence interval; Internal medicine; Mathematics","authors":[{"name":"Baiju R. Shah","is_ca":true},{"name":"Andreas Laupacis","is_ca":true},{"name":"Janet E. Hux","is_ca":true},{"name":"Peter C. Austin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9794651742196404,"gpt":0.7432106376404088,"spread":0.2362545365792316,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.07861588,0.0024957,0.01312587,0.00576441,0.0005051288,0.002708866,0.002088752,0.00242586,0.004821931],"category_scores_gemma":[0.2501971,0.001669266,0.02500507,0.008211569,0.001200598,0.003406444,0.001520957,0.002794789,0.0005729907],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001833312,"about_ca_system_score_gemma":0.004224126,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004279312,"about_ca_topic_score_gemma":0.007452198,"domain_scores_codex":[0.9458497,0.03603142,0.0102133,0.003295111,0.004303166,0.0003073386],"domain_scores_gemma":[0.7112432,0.2682002,0.00826977,0.005158632,0.00680604,0.0003221634],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"systematic_review","study_design_scores_codex":[0.003775469,0.00007819363,0.007380491,0.3529834,0.2248395,0.0002384956,0.000265107,0.002646494,0.0003074433,0.001997081,0.004173695,0.4013147],"study_design_scores_gemma":[0.005457046,0.001726401,0.01468176,0.06888321,0.8759802,0.0007148747,0.0001894954,0.002767567,0.0008909712,0.008738724,0.01981972,0.0001500403],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00363432,0.9768487,0.01692438,0.0005648874,0.0002680671,0.0005902228,0.0005716333,0.00008783855,0.0005100413],"genre_scores_gemma":[0.1061275,0.8625534,0.02733249,0.001498381,0.0004184176,0.000944937,0.0006005228,0.0001136795,0.0004105956],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9868741,"threshold_uncertainty_score":0.4157655,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2609185034","doi":"10.1186/s12874-017-0338-0","title":"Double-adjustment in propensity score matching analysis: choosing a threshold for considering residual imbalance","year":2017,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":491,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Hamilton Health Sciences; McMaster University; Population Health Research Institute","funders":"University of Oxford; Cancer Research UK","keywords":"Covariate; Propensity score matching; Confounding; Statistics; Residual; Matching (statistics); Logistic regression; Regression; Mathematics; Regression analysis; Econometrics; Medicine","authors":[{"name":"Tri‐Long Nguyen","is_ca":true},{"name":"Gary S. Collins","is_ca":false},{"name":"Jessica Spence","is_ca":true},{"name":"Jean‐Pierre Daurès","is_ca":false},{"name":"P.J. Devereaux","is_ca":true},{"name":"Paul Landais","is_ca":false},{"name":"Yannick Le Manach","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8896096151371321,"gpt":0.6436425630183503,"spread":0.2459670521187818,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.06941801,0.0009543284,0.001737593,0.001402466,0.0008438102,0.001565041,0.002064843,0.001619828,0.001819712],"category_scores_gemma":[0.1776288,0.0006040176,0.002167034,0.001739772,0.001880939,0.002116708,0.002513743,0.002156743,0.0002565408],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008855787,"about_ca_system_score_gemma":0.002236418,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001363129,"about_ca_topic_score_gemma":0.001299551,"domain_scores_codex":[0.9553469,0.03528474,0.002519446,0.003289063,0.002795828,0.0007640489],"domain_scores_gemma":[0.8937799,0.08483876,0.008100331,0.01004067,0.002545616,0.000694756],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.005074192,0.0006712202,0.1529049,0.002671024,0.004823809,0.001391796,0.00274634,0.1283654,0.01785288,0.1035447,0.007271467,0.5726824],"study_design_scores_gemma":[0.001056088,0.002276202,0.03577777,0.0006992397,0.001812976,0.001455852,0.0003686956,0.666053,0.02089773,0.256198,0.01313734,0.0002670366],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.03188892,0.0005892638,0.9656529,0.0005447534,0.0001145814,0.0003670945,0.0000696229,0.0003500978,0.0004227745],"genre_scores_gemma":[0.5114285,0.0002380622,0.4863264,0.0003842831,0.0001095765,0.0008301976,0.0001727486,0.0001652647,0.0003450406],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.930582,"threshold_uncertainty_score":0.3671219,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2119863846","doi":"10.1016/j.jclinepi.2014.12.018","title":"Interrupted time series analysis in drug utilization research is increasing: systematic review and recommendations","year":2015,"lang":"en","type":"review","venue":"Journal of Clinical Epidemiology","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":486,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University; St. Michael's Hospital; University of Toronto","funders":"","keywords":"Interrupted time series; Interrupted Time Series Analysis; Medicine; MEDLINE; Systematic review; Psychological intervention; Meta-analysis; Research design; Statistics; Pathology; Mathematics","authors":[{"name":"Racquel Jandoc","is_ca":true},{"name":"Andrea M. Burden","is_ca":true},{"name":"Muhammad Mamdani","is_ca":true},{"name":"Linda E. Lévesque","is_ca":true},{"name":"Suzanne M. Cadarette","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8623636112980363,"gpt":0.7162576385668367,"spread":0.1461059727311996,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.1157503,0.002004367,0.00934683,0.01825599,0.001223545,0.005792519,0.004780444,0.005057519,0.005805454],"category_scores_gemma":[0.2883806,0.001811084,0.00969514,0.01639019,0.001940181,0.008568156,0.002387903,0.004352239,0.0008347995],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.008411096,"about_ca_system_score_gemma":0.04338147,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0116168,"about_ca_topic_score_gemma":0.02432096,"domain_scores_codex":[0.9190809,0.04480172,0.02290744,0.002540292,0.009843199,0.0008264031],"domain_scores_gemma":[0.6296389,0.2886266,0.03901174,0.005362835,0.03572841,0.001631527],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0002211185,0.00005142313,0.001324682,0.7331922,0.006432075,0.0001236748,0.0004662008,0.0005423655,0.0001758694,0.00336227,0.01966814,0.23444],"study_design_scores_gemma":[0.0002387279,0.0001322356,0.001767259,0.8983541,0.01710274,0.0001770097,0.0004203587,0.000666094,0.0002045819,0.004168294,0.0766902,0.000078484],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0001786729,0.9889981,0.001686964,0.007148815,0.001123384,0.0004296611,0.0001784495,0.00002640475,0.0002295092],"genre_scores_gemma":[0.003649416,0.9805649,0.009963559,0.003369793,0.0006516038,0.001426986,0.000212442,0.00001366282,0.0001476632],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9906532,"threshold_uncertainty_score":0.6121535,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2158460222","doi":"10.1177/0272989x09341755","title":"The Relative Ability of Different Propensity Score Methods to Balance Measured Covariates Between Treated and Untreated Subjects in Observational Studies","year":2009,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":456,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"","keywords":"Propensity score matching; Covariate; Statistics; Observational study; Inverse probability weighting; Weighting; Medicine; Mathematics","authors":[{"name":"Peter C. Austin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4399813510196111,"gpt":0.5151726562683904,"spread":0.07519130524877926,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2641273,0.001509035,0.002079567,0.006519463,0.0007846547,0.004115808,0.002731375,0.002338548,0.002317864],"category_scores_gemma":[0.4796374,0.001280572,0.003375234,0.007904064,0.003611069,0.00412484,0.003571634,0.003065813,0.0004799682],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001883143,"about_ca_system_score_gemma":0.00228123,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00197236,"about_ca_topic_score_gemma":0.001758702,"domain_scores_codex":[0.7154419,0.2575209,0.009908682,0.006777897,0.009861588,0.0004890449],"domain_scores_gemma":[0.4445318,0.5105355,0.01908763,0.01962145,0.005527124,0.0006965204],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001602924,0.0003210986,0.112522,0.003486309,0.0111788,0.0002066252,0.00208377,0.07002088,0.001411484,0.1601281,0.004273449,0.6327646],"study_design_scores_gemma":[0.001439019,0.00169405,0.09036706,0.0027887,0.003965213,0.00100731,0.0006323428,0.2858962,0.005507547,0.5801725,0.02606719,0.0004627336],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02521487,0.009473683,0.9587231,0.00297565,0.0003037412,0.0007306787,0.0003058174,0.0002028409,0.002069565],"genre_scores_gemma":[0.2725471,0.006687344,0.71613,0.001183722,0.0005347327,0.001541493,0.0004669664,0.0001193583,0.0007892441],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.7358727,"threshold_uncertainty_score":0.9074621,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2065417422","doi":"10.1080/00273171.2011.540480","title":"A Tutorial and Case Study in Propensity Score Analysis: An Application to Estimating the Effect of In-Hospital Smoking Cessation Counseling on Mortality","year":2011,"lang":"en","type":"article","venue":"Multivariate Behavioral Research","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":421,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"Canadian Institutes of Health Research; U.S. Public Health Service; Ontario Ministry of Health and Long-Term Care","keywords":"Propensity score matching; Covariate; Medicine; Inverse probability weighting; Observational study; Matching (statistics); Statistics; Internal medicine; Mathematics","authors":[{"name":"Peter C. Austin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4995694276247035,"gpt":0.5544291490721748,"spread":0.05485972144747131,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02633445,0.00182313,0.001606363,0.002950819,0.001179121,0.002055506,0.002279116,0.003634437,0.01541281],"category_scores_gemma":[0.06575539,0.0009831543,0.002628596,0.003634386,0.002568983,0.003736755,0.003459948,0.005533811,0.003006242],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001305183,"about_ca_system_score_gemma":0.002172423,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003426624,"about_ca_topic_score_gemma":0.003482344,"domain_scores_codex":[0.9867932,0.01095522,0.00054773,0.0005663426,0.001008421,0.0001291683],"domain_scores_gemma":[0.9505283,0.04532723,0.0009128576,0.001705658,0.00113873,0.0003872047],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0001632273,0.0003867101,0.005960238,0.001001484,0.000357779,0.002391269,0.001085575,0.02915178,0.0007611637,0.5999829,0.07783631,0.2809216],"study_design_scores_gemma":[0.0001816938,0.0001956419,0.001610088,0.0004658271,0.0001233946,0.002856963,0.0002578875,0.1569151,0.0006518859,0.7326699,0.1039429,0.0001286528],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0008189544,0.0008730586,0.9917429,0.002588795,0.0002828768,0.0002193146,0.000184407,0.0002944279,0.002995292],"genre_scores_gemma":[0.01402616,0.002155231,0.9776562,0.0008737294,0.0006053913,0.001203583,0.0002135275,0.0001566484,0.003109529],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9736655,"threshold_uncertainty_score":0.1392716,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2153434255","doi":"10.1111/j.1524-4733.2009.00600.x","title":"Good Research Practices for Comparative Effectiveness Research: Defining, Reporting and Interpreting Nonrandomized Studies of Treatment Effects Using Secondary Data Sources: The ISPOR Good Research Practices for Retrospective Database Analysis Task Force Report—Part I","year":2009,"lang":"en","type":"article","venue":"Value in Health","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":410,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto; St. Michael's Hospital","funders":"","keywords":"Comparative effectiveness research; Observational study; Medicine; Outcomes research; Research design; MEDLINE; Management science; Medical education; Alternative medicine; Political science","authors":[{"name":"Marc L. Berger","is_ca":false},{"name":"Muhammad Mamdani","is_ca":true},{"name":"David C. Atkins","is_ca":false},{"name":"Michael L. Johnson","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.827067872307364,"gpt":0.6782163137175784,"spread":0.1488515585897856,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.8188614,0.004278625,0.009915556,0.02271955,0.006520698,0.01546916,0.01268458,0.01598727,0.005702613],"category_scores_gemma":[0.9128755,0.01065009,0.01017243,0.02234435,0.0180657,0.009132237,0.01001025,0.02305412,0.005009573],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006934171,"about_ca_system_score_gemma":0.04072893,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004846707,"about_ca_topic_score_gemma":0.007637077,"domain_scores_codex":[0.07467246,0.727362,0.1487023,0.005365074,0.0428963,0.001001934],"domain_scores_gemma":[0.03152167,0.71008,0.04355283,0.1343812,0.07832268,0.002141598],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.002207298,0.0007734415,0.01180095,0.06770996,0.007401831,0.0006186959,0.02734688,0.003605696,0.003449989,0.09682884,0.192542,0.5857146],"study_design_scores_gemma":[0.005405722,0.002235779,0.0328681,0.1902726,0.009962453,0.00371923,0.005144582,0.01272535,0.01977039,0.2866005,0.4298891,0.001406338],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.002608685,0.02845155,0.8551591,0.04164129,0.004527185,0.05206628,0.003922966,0.002363596,0.009259432],"genre_scores_gemma":[0.01172381,0.005306092,0.9413075,0.004026701,0.0008203619,0.03516576,0.0006526015,0.0004787897,0.0005182768],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.1811386,"threshold_uncertainty_score":0.2233762,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2133554893","doi":"10.1111/j.1467-985x.2012.01032.x","title":"Experimental Designs for Identifying Causal Mechanisms","year":2012,"lang":"en","type":"article","venue":"Journal of the Royal Statistical Society Series A (Statistics in Society)","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":407,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"York University; National Science Foundation","keywords":"Causality (physics); Imperfect; Causation; Computer science; Causal model; Causal inference; Process (computing); Criticism; Key (lock); Natural (archaeology); Black box; Data science; Cognitive psychology; Psychology; Management science; Artificial intelligence; Epistemology; Econometrics; Engineering; Mathematics; Computer security","authors":[{"name":"Kosuke Imai","is_ca":false},{"name":"Dustin Tingley","is_ca":false},{"name":"Teppei Yamamoto","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1689160987547828,"gpt":0.425026705623513,"spread":0.2561106068687302,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1541085,0.001837296,0.002525386,0.003004183,0.001797654,0.003417612,0.0035656,0.004529509,0.01541494],"category_scores_gemma":[0.3901162,0.001220796,0.002600028,0.002907312,0.009954951,0.005753833,0.003969822,0.00673982,0.001440474],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003276631,"about_ca_system_score_gemma":0.003578692,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005908413,"about_ca_topic_score_gemma":0.0005012686,"domain_scores_codex":[0.7528006,0.2206811,0.006338624,0.008151219,0.01113389,0.0008945023],"domain_scores_gemma":[0.3808837,0.5377774,0.02820754,0.04051205,0.01131061,0.001308648],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.000698287,0.0002222131,0.001556151,0.001579326,0.0004208351,0.00009496213,0.0007011333,0.007047759,0.0008679235,0.9384453,0.002684437,0.0456817],"study_design_scores_gemma":[0.001006394,0.001187538,0.001309478,0.0008108995,0.0002836515,0.0001020931,0.0002264296,0.02306491,0.001219303,0.9558483,0.01484455,0.00009643698],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.008232003,0.001155884,0.9729522,0.00338905,0.0006158042,0.004423254,0.0005665064,0.0002550079,0.008410315],"genre_scores_gemma":[0.1332855,0.001061874,0.8303804,0.001560359,0.0005128513,0.03159438,0.0002669505,0.00007365017,0.001263942],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8458915,"threshold_uncertainty_score":0.8150133,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1988221738","doi":"10.1080/00273171.2010.498292","title":"Assessing Mediational Models: Testing and Interval Estimation for Indirect Effects","year":2010,"lang":"en","type":"article","venue":"Multivariate Behavioral Research","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":398,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Markov chain Monte Carlo; Bootstrapping (finance); Computer science; Type I and type II errors; Bayesian probability; Statistics; Confidence interval; Statistical hypothesis testing; Econometrics; Null hypothesis; Prior probability; Mathematics","authors":[{"name":"Jeremy C. Biesanz","is_ca":true},{"name":"Carl F. Falk","is_ca":true},{"name":"Victoria Savalei","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6989123915737223,"gpt":0.608930368580515,"spread":0.08998202299320734,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2144728,0.00269559,0.003986045,0.008753337,0.002317026,0.004694523,0.009248227,0.003379619,0.01262069],"category_scores_gemma":[0.6872378,0.001502211,0.008435718,0.01140783,0.004947328,0.01111219,0.007221518,0.006815868,0.0009996388],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001793746,"about_ca_system_score_gemma":0.004242151,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00248955,"about_ca_topic_score_gemma":0.001215474,"domain_scores_codex":[0.695399,0.2520319,0.009691572,0.01772343,0.02345175,0.001702185],"domain_scores_gemma":[0.1468337,0.805248,0.01644277,0.02176544,0.008976197,0.0007338804],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.002991079,0.001081052,0.161306,0.006969944,0.01883306,0.0007750303,0.01192465,0.02884867,0.001220145,0.2246362,0.01252205,0.5288922],"study_design_scores_gemma":[0.00118063,0.005253247,0.09455208,0.004552107,0.007185369,0.00193225,0.008381294,0.3374296,0.005326481,0.5069368,0.02653182,0.0007384128],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.05021634,0.002017036,0.933268,0.00090172,0.0004018445,0.002786033,0.001746818,0.001196953,0.00746526],"genre_scores_gemma":[0.3299869,0.0008336384,0.6545684,0.0003319769,0.0001496084,0.0120715,0.001102552,0.0004195652,0.0005358619],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.2144728,"threshold_uncertainty_score":0.9686949,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4366992020","doi":"10.1001/jama.2023.4221","title":"Emulation of Randomized Clinical Trials With Nonrandomized Database Analyses","year":2023,"lang":"en","type":"article","venue":"JAMA","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":379,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University","funders":"National Institute of Arthritis and Musculoskeletal and Skin Diseases; National Institute on Aging; National Heart, Lung, and Blood Institute; U.S. Food and Drug Administration","keywords":"Medicine; Emulation; Randomized controlled trial; Clinical trial; MEDLINE; Database; Internal medicine","authors":[{"name":"Shirley Wang","is_ca":false},{"name":"Sebastian Schneeweiß","is_ca":false},{"name":"Jessica M. Franklin","is_ca":false},{"name":"Rishi Desai","is_ca":false},{"name":"William B. Feldman","is_ca":false},{"name":"Elizabeth M. Garry","is_ca":false},{"name":"Robert J. Glynn","is_ca":false},{"name":"Kueiyu Joshua Lin","is_ca":false},{"name":"Julie M. Paik","is_ca":false},{"name":"Elisabetta Patorno","is_ca":true},{"name":"Samy Suissa","is_ca":true},{"name":"Elvira D’Andrea","is_ca":false},{"name":"Dureshahwar Jawaid","is_ca":false},{"name":"Hemin Lee","is_ca":false},{"name":"Ajinkya Pawar","is_ca":false},{"name":"Sushama Kattinakere Sreedhara","is_ca":false},{"name":"Helen Tesfaye","is_ca":false},{"name":"Lily G. Bessette","is_ca":false},{"name":"Luke E. Zabotka","is_ca":false},{"name":"Su Been Lee","is_ca":false},{"name":"Nileesa Gautam","is_ca":false},{"name":"Cassie York","is_ca":false},{"name":"Heidi Zakoul","is_ca":false},{"name":"John Concato","is_ca":false},{"name":"David Martin","is_ca":false},{"name":"Dianne Paraoan","is_ca":false},{"name":"Kenneth Quinto","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5650319051167182,"gpt":0.5950935289172615,"spread":0.03006162380054334,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.7791423,0.004099497,0.008361276,0.008248789,0.00180377,0.0102706,0.008159081,0.007070943,0.01240058],"category_scores_gemma":[0.8927141,0.003912398,0.0153033,0.007906293,0.01017738,0.01202398,0.008215318,0.008251142,0.002750016],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006197435,"about_ca_system_score_gemma":0.008162488,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007804646,"about_ca_topic_score_gemma":0.0006301482,"domain_scores_codex":[0.09806203,0.8091662,0.04910225,0.02036558,0.02221178,0.001092126],"domain_scores_gemma":[0.058611,0.7818273,0.0465089,0.09988521,0.01220073,0.0009669637],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.03294379,0.001999385,0.06882428,0.0290424,0.05573493,0.001347597,0.007038959,0.1295182,0.002444757,0.2644418,0.02921426,0.3774495],"study_design_scores_gemma":[0.01988716,0.01191476,0.02363903,0.01381653,0.01638463,0.001189458,0.001111378,0.4312463,0.01082955,0.3690735,0.09999783,0.0009099571],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0143689,0.00275161,0.9214587,0.003914464,0.002384698,0.04582738,0.00250582,0.001653215,0.005135189],"genre_scores_gemma":[0.1919209,0.0007741176,0.7033464,0.003168157,0.0006909238,0.09758791,0.001391744,0.0002656845,0.0008541279],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.2208577,"threshold_uncertainty_score":0.272357,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1975994236","doi":"10.1002/sim.3854","title":"The performance of different propensity-score methods for estimating differences in proportions (risk differences or absolute risk reductions) in observational studies","year":2010,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":359,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences; Heart and Stroke Foundation of Canada","keywords":"Propensity score matching; Covariate; Statistics; Confidence interval; Observational study; Estimator; Mean squared error; Relative risk; Matching (statistics); Inverse probability weighting; Mathematics; Econometrics; Medicine","authors":[{"name":"Peter C. Austin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4011751616883879,"gpt":0.5200551633034011,"spread":0.1188800016150132,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4396123,0.002058965,0.003944528,0.00764646,0.00156653,0.005256041,0.004350678,0.003243097,0.002818059],"category_scores_gemma":[0.7092298,0.001779044,0.01046743,0.01240259,0.003932175,0.005454705,0.004308516,0.004812075,0.0005815061],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003350895,"about_ca_system_score_gemma":0.004824952,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005815591,"about_ca_topic_score_gemma":0.003693299,"domain_scores_codex":[0.508193,0.413272,0.03620483,0.01625932,0.02482633,0.001244452],"domain_scores_gemma":[0.2754267,0.6212667,0.04339208,0.04115813,0.01791665,0.0008398208],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.004236432,0.0004164733,0.1859894,0.00841173,0.03437318,0.0004708153,0.003481008,0.1572051,0.001278455,0.09154662,0.009445312,0.5031455],"study_design_scores_gemma":[0.003740691,0.003445243,0.1456593,0.007521451,0.01458931,0.001705226,0.0009385151,0.5086161,0.007220493,0.2627421,0.04248298,0.001338544],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.03475623,0.01307819,0.9416749,0.001761582,0.0005445868,0.004067523,0.00152141,0.0004699106,0.002125594],"genre_scores_gemma":[0.3292992,0.006368985,0.6510851,0.001250981,0.000412748,0.008857602,0.001723367,0.0004349096,0.000567083],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.5603877,"threshold_uncertainty_score":0.6910578,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2050419204","doi":"10.1111/1468-2354.t01-1-00074","title":"2001 Lawrence R. Klein Lecture Estimating Distributions of Treatment Effects with an Application to the Returns to Schooling and Measurement of the Effects of Uncertainty on College Choice*","year":2003,"lang":"en","type":"article","venue":"International Economic Review","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":351,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Kellogg's (Canada)","funders":"","keywords":"Counterfactual conditional; LISREL; Veil of ignorance; Econometrics; Economics; Matching (statistics); Pairwise comparison; Probit model; Actuarial science; Mathematics; Statistics; Psychology; Microeconomics; Structural equation modeling; Counterfactual thinking; Social psychology","authors":[{"name":"Pedro Carneiro","is_ca":false},{"name":"Karsten T. Hansen","is_ca":true},{"name":"James J. Heckman","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.05282705166160347,"gpt":0.3680734952222458,"spread":0.3152464435606424,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01971663,0.001365084,0.002363477,0.004480042,0.0009198937,0.003733575,0.001649039,0.002226901,0.02682458],"category_scores_gemma":[0.07212189,0.002029049,0.002284384,0.006162832,0.003167674,0.005536459,0.003111224,0.004579217,0.005894071],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003303581,"about_ca_system_score_gemma":0.003541943,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01212076,"about_ca_topic_score_gemma":0.01036072,"domain_scores_codex":[0.9949515,0.003632797,0.0002025922,0.000493688,0.0006076383,0.0001118045],"domain_scores_gemma":[0.9569334,0.0375631,0.001122686,0.002442121,0.001685636,0.0002530143],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0001012787,0.00007746326,0.00386898,0.0002753095,0.000381639,0.000147548,0.0002249306,0.0583698,0.0002487883,0.5306035,0.06159702,0.3441037],"study_design_scores_gemma":[0.00006548693,0.00001990289,0.001265124,0.0002259763,0.00009643376,0.000108429,0.00007572913,0.0856832,0.0002429677,0.8756062,0.03656283,0.00004767203],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.002220365,0.007730817,0.9636132,0.0103395,0.000309101,0.00007123325,0.0005860317,0.0006096195,0.01452017],"genre_scores_gemma":[0.1101177,0.0267225,0.8239672,0.001517911,0.0009694239,0.0007414487,0.001309572,0.0008415853,0.03381277],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.02682458,"threshold_uncertainty_score":0.1042728,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2766207844","doi":"10.1136/bmj.j4587","title":"Handling time varying confounding in observational research","year":2017,"lang":"en","type":"review","venue":"BMJ","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":346,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University; University of British Columbia","funders":"","keywords":"Confounding; Observational study; Weighting; Marginal structural model; Statistics; Econometrics; Parametric statistics; Inverse probability weighting; Estimation; Causal inference; Medicine; Computer science; Mathematics; Propensity score matching; Engineering","authors":[{"name":"Mohammad Alì Mansournia","is_ca":false},{"name":"Mahyar Etminan","is_ca":true},{"name":"Goodarz Danaei","is_ca":false},{"name":"Jay S. Kaufman","is_ca":true},{"name":"Gary S. Collins","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8963818942939966,"gpt":0.6776851293144192,"spread":0.2186967649795775,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.06189795,0.001263774,0.003200913,0.004645024,0.0008936159,0.002978487,0.003829909,0.004973656,0.004718808],"category_scores_gemma":[0.140982,0.0007353167,0.003665261,0.007957282,0.005055876,0.004247671,0.003401465,0.005678431,0.001576729],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002296924,"about_ca_system_score_gemma":0.007384914,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003875302,"about_ca_topic_score_gemma":0.003507428,"domain_scores_codex":[0.9440285,0.044698,0.0027241,0.002061628,0.006108658,0.0003791717],"domain_scores_gemma":[0.8379079,0.14928,0.00485397,0.004130792,0.003475737,0.0003514969],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00008720256,0.00004828323,0.001219111,0.03616777,0.001493044,0.0001479795,0.0006770229,0.002206593,0.0001369245,0.1378189,0.01671446,0.8032827],"study_design_scores_gemma":[0.0001592294,0.0001646616,0.00357026,0.05180787,0.001356472,0.001258625,0.0004138802,0.002718173,0.0004498983,0.6016386,0.3363289,0.0001334605],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"review","genre_gemma":"methods","genre_scores_codex":[0.0004055576,0.9193934,0.06716642,0.009048861,0.001450988,0.0001495363,0.0001512938,0.00006745008,0.002166551],"genre_scores_gemma":[0.008762324,0.9374323,0.0459902,0.003815734,0.002783596,0.0004419082,0.0001531035,0.00004838847,0.0005724315],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9381021,"threshold_uncertainty_score":0.3273516,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1995875958","doi":"10.1257/aer.91.2.112","title":"Reconciling Conflicting Evidence on the Performance of Propensity-Score Matching Methods","year":2001,"lang":"en","type":"article","venue":"American Economic Review","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":336,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Western University","funders":"","keywords":"Propensity score matching; Economics; Matching (statistics); Econometrics; Actuarial science; Positive economics; Statistics; Mathematics","authors":[{"name":"Jeffrey A. Smith","is_ca":true},{"name":"Petra Todd","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4399859587587839,"gpt":0.4966393531473604,"spread":0.05665339438857642,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4428159,0.003357312,0.006386835,0.01512421,0.002088596,0.0104932,0.009567719,0.008925912,0.006394959],"category_scores_gemma":[0.8382776,0.002435838,0.004825614,0.02045833,0.01095486,0.009833198,0.01031356,0.00910789,0.001280458],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003267827,"about_ca_system_score_gemma":0.004948714,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003752323,"about_ca_topic_score_gemma":0.002710696,"domain_scores_codex":[0.4322239,0.5130273,0.01926942,0.01227574,0.02136605,0.001837561],"domain_scores_gemma":[0.06131589,0.8892987,0.01577824,0.02017162,0.0122795,0.001156005],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.01037891,0.0004007425,0.08546851,0.01437469,0.05885617,0.0008986702,0.003017033,0.02595783,0.0004858219,0.2370836,0.04815839,0.5149198],"study_design_scores_gemma":[0.003661867,0.0007867505,0.03177276,0.009339084,0.01963408,0.0007243162,0.001165382,0.04903422,0.001993782,0.8475447,0.03386707,0.0004759863],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"review","genre_gemma":"methods","genre_scores_codex":[0.09689811,0.3776979,0.308378,0.1923441,0.0063913,0.0005525378,0.00325384,0.0006908969,0.01379336],"genre_scores_gemma":[0.7870291,0.07057434,0.1038452,0.02391266,0.009456587,0.0006483958,0.002566781,0.0005172906,0.001449629],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.5571841,"threshold_uncertainty_score":0.6871072,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2809826847","doi":"10.1080/13645579.2018.1488449","title":"Natural experiment methodology for research: a review of how different methods can support real-world research","year":2018,"lang":"en","type":"review","venue":"International Journal of Social Research Methodology","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":335,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Waterloo","funders":"Canadian Institutes of Health Research; Public Health Agency of Canada","keywords":"Randomized experiment; Research design; Randomized controlled trial; Natural experiment; Natural (archaeology); Intervention (counseling); Management science; Psychological intervention; Computer science; Risk analysis (engineering); Psychology; Sociology; Engineering; Medicine; Social science; Mathematics; Statistics","authors":[{"name":"Scott T. Leatherdale","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.967881105314908,"gpt":0.8051552063693992,"spread":0.1627258989455088,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.09554453,0.002673871,0.00621125,0.01269302,0.001674418,0.007904281,0.00428605,0.005612397,0.008575454],"category_scores_gemma":[0.2080633,0.001597938,0.004308633,0.01449448,0.007147145,0.007275702,0.003595678,0.005624624,0.002509098],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.007129967,"about_ca_system_score_gemma":0.01458223,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004539828,"about_ca_topic_score_gemma":0.007237655,"domain_scores_codex":[0.8937596,0.07661869,0.01066572,0.003761841,0.01465119,0.0005429208],"domain_scores_gemma":[0.5735026,0.4041039,0.005710071,0.007117027,0.008977191,0.0005891862],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001447745,0.0001074702,0.0005393108,0.1844176,0.001316668,0.0001612996,0.001257004,0.001213593,0.0003931623,0.1192149,0.01996352,0.6712707],"study_design_scores_gemma":[0.0001690456,0.0003179654,0.001857353,0.2007617,0.001816112,0.0006345849,0.0006295406,0.0009102423,0.0006503645,0.1250054,0.6670151,0.0002324656],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0001144062,0.9691339,0.02144787,0.003547122,0.001051791,0.0004551255,0.0001367349,0.00007019749,0.004042895],"genre_scores_gemma":[0.003748277,0.9466132,0.04404032,0.002286194,0.0007310684,0.001915597,0.0001170062,0.0000725803,0.000475772],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9044555,"threshold_uncertainty_score":0.5052938,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W3047016911","doi":"10.1257/aer.20190687","title":"Methods Matter: p-Hacking and Publication Bias in Causal Analysis in Economics","year":2020,"lang":"en","type":"article","venue":"American Economic Review","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":333,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Ottawa","funders":"","keywords":"Credibility; Regression discontinuity design; Instrumental variable; Publication bias; Hacker; Causal inference; Omitted-variable bias; Economics; Identification (biology); Econometrics; Randomized experiment; Selection bias; Psychology; Statistics; Political science; Computer science; Law; MEDLINE; Mathematics; Computer security","authors":[{"name":"Abel Brodeur","is_ca":true},{"name":"Nikolai Cook","is_ca":true},{"name":"Anthony Heyes","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2339322466141202,"gpt":0.4709007945413937,"spread":0.2369685479272735,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2436466,0.001414048,0.002444718,0.00677963,0.002666393,0.009806413,0.003394207,0.00752794,0.0212011],"category_scores_gemma":[0.7354751,0.001063575,0.002293528,0.01687395,0.01720678,0.01204407,0.004310194,0.008946305,0.002696216],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003443468,"about_ca_system_score_gemma":0.00938127,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002262182,"about_ca_topic_score_gemma":0.001363832,"domain_scores_codex":[0.645991,0.2967421,0.01419951,0.01698731,0.02458312,0.001496922],"domain_scores_gemma":[0.1182152,0.8003928,0.034261,0.03518365,0.01077261,0.001174823],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.0007289485,0.0001028203,0.03233462,0.006073265,0.002954297,0.0006603128,0.002413908,0.004871654,0.0003270879,0.5415632,0.0831809,0.324789],"study_design_scores_gemma":[0.0002624742,0.0001417461,0.006315905,0.002573766,0.0005148085,0.0006691461,0.0004089249,0.009815304,0.0005533712,0.9341918,0.04446544,0.00008728425],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01426271,0.06523076,0.6506999,0.2136247,0.01208786,0.0008860338,0.001834952,0.001428352,0.0399447],"genre_scores_gemma":[0.6465842,0.03451182,0.2324712,0.05004157,0.01914088,0.003276563,0.0008280451,0.001132888,0.01201287],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7563534,"threshold_uncertainty_score":0.9327184,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2013920814","doi":"10.1002/sim.4200","title":"Comparing paired vs non‐paired statistical methods of analyses when making inferences about absolute risk reductions in propensity‐score matched samples","year":2011,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":325,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Clinical Evaluative Sciences; Public Health Ontario; University of Toronto","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences; Heart and Stroke Foundation of Canada","keywords":"Propensity score matching; Statistics; Confidence interval; Confounding; Sample size determination; Selection bias; Observational study; Statistical significance; Statistical inference; Matching (statistics); Causal inference; Medicine; Mathematics","authors":[{"name":"Peter C. Austin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6053372113795614,"gpt":0.5275261212026646,"spread":0.0778110901768968,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3921677,0.001958479,0.003827057,0.003772027,0.001552199,0.005874407,0.004651726,0.003755744,0.008487406],"category_scores_gemma":[0.7571114,0.001709421,0.007897763,0.004512876,0.006821116,0.006840642,0.004921835,0.008159158,0.001184618],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002002545,"about_ca_system_score_gemma":0.003287762,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001084618,"about_ca_topic_score_gemma":0.001071493,"domain_scores_codex":[0.3320736,0.6052327,0.01939249,0.01550575,0.02658703,0.001208378],"domain_scores_gemma":[0.1440299,0.7566448,0.0317837,0.05181341,0.01491351,0.0008146727],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.01301471,0.002176652,0.06633946,0.01309498,0.0349429,0.001193749,0.0100308,0.04449741,0.003884831,0.2143002,0.02376537,0.5727589],"study_design_scores_gemma":[0.006752149,0.01620483,0.07463906,0.00922901,0.01476885,0.002174336,0.003535578,0.265134,0.02951462,0.5006917,0.07604178,0.001314232],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02143456,0.002079766,0.9641409,0.001356372,0.001450131,0.004995408,0.0005793584,0.000549677,0.003413794],"genre_scores_gemma":[0.2332365,0.0009323264,0.7443196,0.001942047,0.0005003312,0.01689554,0.0005524331,0.0005322728,0.001088874],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.6078323,"threshold_uncertainty_score":0.7495655,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2035404753","doi":"10.1177/0962280215584401","title":"The performance of inverse probability of treatment weighting and full matching on the propensity score in the presence of model misspecification when estimating the effect of treatment on survival outcomes","year":2015,"lang":"en","type":"article","venue":"Statistical Methods in Medical Research","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":325,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"National Institute of Mental Health","keywords":"Propensity score matching; Observational study; Matching (statistics); Statistics; Inverse probability weighting; Hazard ratio; Inverse probability; Covariate; Average treatment effect; Selection bias; Confounding; Weighting; Marginal structural model; Econometrics; Selection (genetic algorithm); Mathematics; Computer science; Medicine; Confidence interval; Posterior probability; Bayesian probability; Artificial intelligence","authors":[{"name":"Peter C. Austin","is_ca":true},{"name":"Elizabeth A. Stuart","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5863130353224311,"gpt":0.5806095152849405,"spread":0.005703520037490528,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1233669,0.001446421,0.002521161,0.003142737,0.000909724,0.00253803,0.002101374,0.001963449,0.002198558],"category_scores_gemma":[0.3221789,0.001120473,0.003246184,0.002910829,0.002118835,0.003791014,0.004931708,0.002434645,0.0004148356],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001351644,"about_ca_system_score_gemma":0.002875415,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003744821,"about_ca_topic_score_gemma":0.002104199,"domain_scores_codex":[0.9080955,0.07778683,0.003532212,0.004155118,0.005721883,0.0007084448],"domain_scores_gemma":[0.7511629,0.2047143,0.01159833,0.02604081,0.005690562,0.0007931436],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.002757457,0.000300988,0.05036103,0.001392835,0.004375666,0.0002595816,0.00122268,0.2773727,0.002522402,0.05173722,0.001979167,0.6057182],"study_design_scores_gemma":[0.0005681652,0.001125598,0.0219243,0.0005419058,0.001210023,0.0005680492,0.0001984979,0.8412926,0.003994491,0.1232643,0.005049702,0.00026241],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.09423201,0.00231488,0.8993744,0.0005758584,0.0001676171,0.0006758491,0.0001994689,0.0003966085,0.002063386],"genre_scores_gemma":[0.5164662,0.001180669,0.4796246,0.0004002146,0.0001234205,0.0008669392,0.0004346655,0.000185459,0.0007177654],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8766332,"threshold_uncertainty_score":0.6524341,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3112964578","doi":"10.1161/circulationaha.120.051718","title":"Emulating Randomized Clinical Trials With Nonrandomized Real-World Evidence Studies","year":2020,"lang":"en","type":"article","venue":"Circulation","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":297,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"National Heart, Lung, and Blood Institute; McGill University; National Institute on Aging; Harvard University; Vanderbilt University","keywords":"Medicine; Randomized controlled trial; Confounding; Propensity score matching; Hazard ratio; Placebo; Clinical trial; Clinical endpoint; Research design; Proxy (statistics); Clinical study design; Confidence interval; Internal medicine; Statistics; Alternative medicine","authors":[{"name":"Jessica M. Franklin","is_ca":false},{"name":"Elisabetta Patorno","is_ca":false},{"name":"Rishi Desai","is_ca":false},{"name":"Robert J. Glynn","is_ca":false},{"name":"David Martin","is_ca":false},{"name":"Kenneth Quinto","is_ca":false},{"name":"Ajinkya Pawar","is_ca":false},{"name":"Lily G. Bessette","is_ca":false},{"name":"Hemin Lee","is_ca":false},{"name":"Elizabeth M. Garry","is_ca":false},{"name":"Nileesa Gautam","is_ca":false},{"name":"Sebastian Schneeweiß","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7113316494914241,"gpt":0.5902829400891217,"spread":0.1210487094023024,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.8319889,0.004713647,0.009840949,0.01134424,0.00161447,0.01288326,0.01191513,0.008400866,0.007398614],"category_scores_gemma":[0.9054288,0.005331047,0.01438662,0.008763396,0.01311773,0.01269084,0.01043721,0.008229705,0.001884533],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.007618624,"about_ca_system_score_gemma":0.0124819,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007780949,"about_ca_topic_score_gemma":0.0007830357,"domain_scores_codex":[0.05302263,0.8649918,0.0505246,0.01290105,0.0178799,0.000680089],"domain_scores_gemma":[0.03863227,0.8140991,0.04969933,0.08574565,0.01110209,0.0007216009],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0263824,0.00264973,0.04380776,0.06881231,0.07255123,0.001398027,0.006214097,0.1263407,0.00260229,0.2740687,0.01117268,0.3640002],"study_design_scores_gemma":[0.04003824,0.0173528,0.01260391,0.02905811,0.02515062,0.001235805,0.001113001,0.2204157,0.009435215,0.5654876,0.07704242,0.001066713],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0132592,0.006281731,0.9239872,0.004747826,0.002032474,0.04460464,0.001070765,0.0008312997,0.003184829],"genre_scores_gemma":[0.121334,0.001659505,0.8086676,0.002723033,0.0007665109,0.06374163,0.0006577414,0.0001472796,0.0003027306],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.1680111,"threshold_uncertainty_score":0.2071876,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2153012086","doi":"10.1111/j.1524-4733.2009.00602.x","title":"Good Research Practices for Comparative Effectiveness Research: Analytic Methods to Improve Causal Inference from Nonrandomized Studies of Treatment Effects Using Secondary Data Sources: The ISPOR Good Research Practices for Retrospective Database Analysis Task Force Report—Part III","year":2009,"lang":"en","type":"article","venue":"Value in Health","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":290,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Causal inference; Inference; Confounding; Statistical inference; Comparative effectiveness research; Marginal structural model; Medicine; Propensity score matching; Instrumental variable; Outcomes research; Research design; Computer science; Econometrics; Management science; Statistics; Machine learning; Alternative medicine; Artificial intelligence; Engineering; Mathematics","authors":[{"name":"Michael L. Johnson","is_ca":false},{"name":"William H. Crown","is_ca":false},{"name":"Bradley C. Martin","is_ca":false},{"name":"Colin R. Dormuth","is_ca":true},{"name":"Uwe Siebert","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8166517374954501,"gpt":0.7018213479349174,"spread":0.1148303895605327,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.6625069,0.004504994,0.008170646,0.01467366,0.004130589,0.008334371,0.008192461,0.009741425,0.00858853],"category_scores_gemma":[0.8581904,0.008448055,0.008323249,0.01392388,0.01018199,0.007947354,0.008001915,0.01889145,0.003886446],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004597995,"about_ca_system_score_gemma":0.01878429,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002488,"about_ca_topic_score_gemma":0.003812047,"domain_scores_codex":[0.1846051,0.6939537,0.06974253,0.006814035,0.04406402,0.0008206039],"domain_scores_gemma":[0.06870992,0.7520049,0.02284319,0.1075712,0.04724993,0.001620802],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.001540166,0.0006161772,0.006060954,0.04367522,0.00830146,0.0003621519,0.01087952,0.005616106,0.00192805,0.1346639,0.1100141,0.6763422],"study_design_scores_gemma":[0.005615057,0.001503268,0.01321623,0.07513554,0.01028098,0.002239007,0.001932959,0.03648195,0.01489575,0.6444806,0.1933158,0.0009027511],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0007430957,0.006339973,0.974567,0.005983197,0.0008310063,0.007809747,0.0008677439,0.0009664946,0.001891718],"genre_scores_gemma":[0.005548303,0.001353094,0.9806483,0.0007212819,0.0002112668,0.01094156,0.0001642357,0.0002427486,0.0001691521],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.3374931,"threshold_uncertainty_score":0.4161891,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2162163884","doi":"10.1136/jech.55.7.508","title":"A heuristic approach to the formulas for population attributable fraction","year":2001,"lang":"en","type":"article","venue":"Journal of Epidemiology & Community Health","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":289,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University","funders":"","keywords":"Confounding; Fraction (chemistry); Attributable risk; Medicine; Heuristic; Population; Representation (politics); Limit (mathematics); USable; Econometrics; Statistics; Public health; Distribution (mathematics); Estimation; Environmental health; Computer science; Artificial intelligence; Mathematics; Pathology","authors":[{"name":"James A. Hanley","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5047912795363017,"gpt":0.5411379398954912,"spread":0.03634666035918954,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01472217,0.001868814,0.001284956,0.005285968,0.001033217,0.00365482,0.003657368,0.002265318,0.01171525],"category_scores_gemma":[0.07911607,0.0008467226,0.001852844,0.004482408,0.00405439,0.006414939,0.003101367,0.005228224,0.003473888],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002706726,"about_ca_system_score_gemma":0.002398453,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002494287,"about_ca_topic_score_gemma":0.002086011,"domain_scores_codex":[0.9898545,0.006137764,0.0005822554,0.0008004172,0.002359806,0.0002654037],"domain_scores_gemma":[0.9753945,0.01956187,0.001019375,0.001926927,0.001967993,0.0001293936],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00002186692,0.00002148414,0.0004255655,0.0001506201,0.00003116122,0.000146425,0.0002334949,0.01462244,0.0001747805,0.9212099,0.008649228,0.05431314],"study_design_scores_gemma":[0.00002401789,0.00002925123,0.0001880479,0.0001481961,0.00003420764,0.0003476392,0.00007068007,0.06796742,0.0004013471,0.9081026,0.02265309,0.00003362842],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0005234293,0.0004488854,0.9942755,0.0009204717,0.0001412751,0.00005853299,0.00009050528,0.0001709207,0.00337043],"genre_scores_gemma":[0.03744952,0.001092968,0.9560311,0.0009052032,0.0004556846,0.0006041339,0.0001793741,0.0002446391,0.003037384],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.01472217,"threshold_uncertainty_score":0.07785922,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"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","authors":[{"name":"Amelia M. Haviland","is_ca":false},{"name":"Daniel S. Nagin","is_ca":false},{"name":"Paul R. Rosenbaum","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7505569186862165,"gpt":0.6029601374272204,"spread":0.1475967812589961,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1179349,0.001268932,0.002548949,0.006589707,0.0008403763,0.002551336,0.002011587,0.001505572,0.003739397],"category_scores_gemma":[0.2128375,0.0008797659,0.003086274,0.01133426,0.001311148,0.002266846,0.004435262,0.002303367,0.0006858302],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001131214,"about_ca_system_score_gemma":0.003480615,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003529795,"about_ca_topic_score_gemma":0.003267095,"domain_scores_codex":[0.8568908,0.1305165,0.003146546,0.0043555,0.004605399,0.0004852359],"domain_scores_gemma":[0.8617817,0.101223,0.01002018,0.02233024,0.003583957,0.001060933],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.002434263,0.0009505604,0.1538162,0.002414399,0.0184748,0.0005647708,0.001177857,0.06416535,0.001438641,0.09799723,0.01251014,0.6440558],"study_design_scores_gemma":[0.002754193,0.003098546,0.09804085,0.0009167027,0.007038114,0.001040497,0.0006788975,0.3723814,0.002636098,0.4827306,0.02831409,0.0003700011],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01344952,0.001289266,0.980451,0.0007379437,0.0002665622,0.00174847,0.0005220608,0.0003782155,0.001156898],"genre_scores_gemma":[0.2318855,0.00142328,0.760515,0.0006034064,0.0002904625,0.00324614,0.001129962,0.000133358,0.0007728262],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.1179349,"threshold_uncertainty_score":0.6237069,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W213628847","doi":"10.1016/j.cjca.2015.05.015","title":"A Review of Propensity-Score Methods and Their Use in Cardiovascular Research","year":2015,"lang":"en","type":"review","venue":"Canadian Journal of Cardiology","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":283,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":true,"about_ca":false},"ca_institutions":"Sunnybrook Hospital; University of Toronto; St. Michael's Hospital","funders":"Canadian Institutes of Health Research","keywords":"Propensity score matching; Observational study; Medicine; Covariate; Matching (statistics); Weighting; Inverse probability weighting; Selection bias; Strengths and weaknesses; Randomized controlled trial; Econometrics; Statistics; Internal medicine; Mathematics; Pathology; Psychology","authors":[{"name":"Saswata Deb","is_ca":false},{"name":"Peter C. Austin","is_ca":false},{"name":"Jack V. Tu","is_ca":false},{"name":"Dennis T. Ko","is_ca":false},{"name":"C. David Mazer","is_ca":true},{"name":"Alex Kiss","is_ca":true},{"name":"Stephen E. Fremes","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7264268470364439,"gpt":0.5542337611935539,"spread":0.1721930858428899,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.008752034,0.001250619,0.003510374,0.005378904,0.0003474445,0.00190234,0.001835207,0.002052446,0.00507448],"category_scores_gemma":[0.02539242,0.0007162314,0.002105138,0.007760947,0.00125897,0.002073026,0.001220998,0.002853369,0.00138734],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008763695,"about_ca_system_score_gemma":0.003756011,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002418691,"about_ca_topic_score_gemma":0.003569044,"domain_scores_codex":[0.9969323,0.001382455,0.0005887823,0.000348505,0.0006903361,0.00005765774],"domain_scores_gemma":[0.9753054,0.0217857,0.001078981,0.0004283926,0.001243384,0.000158166],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00008547947,0.00007068961,0.0006054663,0.05079843,0.0006831568,0.00009277593,0.00007020468,0.001595824,0.0002135981,0.01518375,0.01927458,0.9113261],"study_design_scores_gemma":[0.00022555,0.0004157383,0.006290864,0.05624467,0.003307434,0.002165121,0.0001941335,0.004248604,0.0008679571,0.09535995,0.8304283,0.0002517824],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00007776628,0.9956844,0.003088449,0.0006220864,0.0001494861,0.00001136357,0.00006132963,0.00001487664,0.0002902866],"genre_scores_gemma":[0.0008955181,0.9942311,0.003879597,0.0003488054,0.0004098759,0.00002442004,0.00005957116,0.000005870162,0.0001453283],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.991248,"threshold_uncertainty_score":0.04628575,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}