Survey weighting typically links model-based population estimates with design-based inference through equivalent weights. A new arXiv preprint introduces a hierarchical Gaussian process approach to this problem, aimed at structured populations where simple weighting assumptions may break down.

For nonlinear models, the authors define equivalent weights locally, using sensitivity to derive weights that vary across the population. This local definition is a departure from global weighting schemes and is made tractable through the hierarchical Gaussian process structure.

The abstract is brief, so details of the estimator and its properties are not fully described here. Still, the work points toward a more flexible way to combine regression-based modeling with survey design considerations in complex, structured settings.