{"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,"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","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"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"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2610013184506083,"score_gpt":0.5385518179316583,"score_spread":0.27755049948105,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}