svyder: Design Effect Ratio Diagnostics for Bayesian Survey Models

Computes parameter-level Design Effect Ratios (DER) for Bayesian hierarchical models fitted to complex survey data. The package implements a compute-classify-correct diagnostic framework:

Details

1. Compute: Calculate DER for each parameter using a sandwich variance estimator.

2. Classify: Assign estimable parameters by information source and flag those exceeding a threshold.

3. Correct: Apply selective Cholesky correction to flagged parameters only.

Core pipeline

• der_compute(): Compute DER values.

• der_classify(): Estimable-parameter classification.

• der_correct(): Selective Cholesky correction.

• der_diagnose(): All-in-one wrapper.

Analysis

• der_decompose(): Decompose DER into DEFF, B, R_k, kappa.

• der_sensitivity(): Threshold sensitivity analysis.

• der_theorem_check(): Verify theoretical predictions.

• der_compare(): Compare across clustering definitions.

Backend support

• brms: der_compute.brmsfit(), extract_draws.brmsfit()

• cmdstanr: der_compute.CmdStanMCMC(), extract_draws.CmdStanMCMC()

• rstanarm: der_compute.stanreg(), extract_draws.stanreg()

• survey: extract_design.survey.design2()

Bundled datasets

• nsece_demo: Synthetic NSECE-like survey data (binomial, J=51).

• sim_hlr: Balanced hierarchical linear regression (gaussian, J=10).

References

Lee, J., Williams, M. R., & Savitsky, T. D. (2026). Design Effect Ratios for Bayesian Survey Models: A Diagnostic Framework for Identifying Survey-Sensitive Parameters. Journal of Survey Statistics and Methodology. Submitted.

See also

Useful links:

• https://joonho112.github.io/svyder/

• https://github.com/joonho112/svyder

• Report bugs at https://github.com/joonho112/svyder/issues

Author

Maintainer: JoonHo Lee jlee296@ua.edu (ORCID)

Authors:

• JoonHo Lee jlee296@ua.edu (ORCID)