Skip to main content
Fig. 4 | BMC Medical Research Methodology

Fig. 4

From: Minimizing bias in massive multi-arm observational studies with BCAUS: balancing covariates automatically using supervision

Fig. 4

Comparison between BCAUS and baseline logistic regression and neural network propensity score models on the diabetes dataset. For every covariate, box plots show distributions of standardized differences, Δ between control and treatment cohorts for 133 intervention arms. Whiskers are at 5 and 95 percentiles. Red dashed lines show the threshold at 0.1 below which covariates are considered balanced. a Raw unadjusted data without IPW weighting. b Covariates weighted with IPWs from baseline logistic regression (LR) model. c Covariates weighted with IPWs from baseline neural network (NN) model. d Covariates weighted with IPWs from BCAUS

Back to article page