External validation: the longitudinal MI tutorial (simulated CATS)¶
The R (and Stata) syntax published with "Multiple imputation for
longitudinal data: A tutorial" (Statistics in Medicine,
doi:10.1002/sim.10274) is longmi's external methodological oracle. The
actual CATS participant data are not publicly available; the authors
provide a simulated dataset designed to resemble CATS (CATS_dataL.csv)
together with the script that generated it (Simulation_of_data.R).
Role in longmi's validation strategy:
| Example | Purpose |
|---|---|
epil |
Real public longitudinal count data; main user tutorial |
| simulated CATS | Methodological parity with the published R tutorial |
| longmi synthetic fixture | Fast deterministic unit and CI tests |
The tutorial repository is treated as an external pinned checkout
(no explicit upstream license → nothing is copied or redistributed):
located via LONGMI_TUTORIAL_REPO, pinned by commit in
validation/external_repositories.toml,
verified by
examples/cats_tutorial_parity/external_repo.py.
Status: the external checkout and validation contract (pinning, required-file checks) are implemented. Cross-language statistical parity for the imputation distributions and pooled analyses — authors' R implementation vs longmi on the same data, missingness, imputation specification, substantive model, and pooling — remains in progress. Comparisons will be statistical (pooled estimates, standard errors, fractions of missing information, imputed margins), never draw-for-draw.