DR (GLM outcome)
Doubly-robust estimator combining a logistic propensity score with a GLM outcome model. Both nuisance models are fit on x1 and x2.
nonprobsvy version: 0.3.0 | R: 4.6.0 | run: 2026-05-24 08:46:46 | commit: ec5beda
| var_method | num_boot | alpha | n_reps | mean_se | coverage | ci_width |
|---|---|---|---|---|---|---|
| analytic | NA | 0.05 | 500 | 0.126 | 0.994 | 0.492 |
| analytic | NA | 0.10 | 500 | 0.126 | 0.976 | 0.413 |
| bootstrap | 200 | 0.05 | 500 | 0.074 | 0.910 | 0.292 |
| bootstrap | 200 | 0.10 | 500 | 0.074 | 0.848 | 0.245 |
Notes
- DGP:
default. - Bias is small (DR’s claim to fame), but the analytical SE is conservative relative to the empirical sd of the point estimates — leading to over-coverage. Bootstrap variance is the recommended remedy.
bias_correction = TRUE(the joint-estimating-equations variant) is currently unavailable on the dev branch ofnonprobsvydue to an upstream bug; once fixed, adr_bias_corr.Rscript will be added here.