MI (predictive mean matching)
Mass imputation by predictive mean matching: a regression y ~ x1 + x2 is fit on the non-probability sample, donors are chosen by closeness in either predicted-y (pmm_match_type = 1) or observed-y (pmm_match_type = 2) space, and the donor outcomes are aggregated over the probability sample.
nonprobsvy version: 0.3.0 | R: 4.6.0 | run: 2026-05-24 08:08:18 | commit: 4b2ba9a
| script | pmm_match_type | var_method | rep_type | num_boot | alpha | n_reps | bias | rmse | mc_se | mean_se | coverage | ci_width |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| mi_pmm | 1 | analytic | NA | NA | 0.05 | 500 | 0.011 | 0.084 | 0.004 | 0.054 | 0.794 | 0.213 |
| mi_pmm | 2 | analytic | NA | NA | 0.05 | 500 | 0.005 | 0.079 | 0.004 | 0.051 | 0.802 | 0.199 |
| mi_pmm_boot | 1 | bootstrap | subbootstrap | 50 | 0.05 | 100 | 0.006 | 0.080 | 0.008 | 0.052 | 0.780 | 0.202 |
| mi_pmm_boot | 2 | bootstrap | subbootstrap | 50 | 0.05 | 100 | 0.003 | 0.074 | 0.007 | 0.046 | 0.780 | 0.180 |
Notes
- DGP:
default. - Analytical SE typically under-states the variance for PMM for the same reason as NN imputation — donor selection variability is unmodelled.
- Sweeps
pmm_match_type ∈ {1, 2}atk = 5.