Convert a tbl_sample to a replicate-weight survey design

Creates a svyrep.design object from a tbl_sample by first converting to a survey::svydesign() object via as_svydesign(), then converting with survey::as.svrepdesign().

Usage

as_svrepdesign(x, ...)

# S3 method for class 'tbl_sample'
as_svrepdesign(
  x,
  ...,
  type = c("auto", "JK1", "JKn", "BRR", "bootstrap", "subbootstrap", "mrbbootstrap",
    "Fay")
)

Arguments

x: A tbl_sample object produced by execute().
...: Additional arguments passed to survey::as.svrepdesign().
type: Replicate method passed to survey::as.svrepdesign(). One of "auto", "JK1", "JKn", "BRR", "bootstrap", "subbootstrap", "mrbbootstrap", or "Fay".

Value

A svyrep.design object from the survey package.

Details

Replicate conversion supports single-phase designs. For unequal-probability designs (PPS or random-size Poisson), "subbootstrap" and "mrbbootstrap" are the supported replicate types. Other types emit a warning and may fail because inclusion probabilities vary within strata. For fixed-size PPS variance estimation, linearization via as_svydesign() is generally preferred. Two-phase designs should be exported with as_svydesign().

Bootstrap escape hatch for random-size Poisson at stage 1

Some designs cannot be expressed as a linearization-based survey::svydesign() object. Specifically, multi-stage designs with a random-size Poisson method (bernoulli or pps_poisson) at stage 1, and single-stage designs with cluster_by() and multiple rows per sampled cluster, are rejected by as_svydesign() for those methods.

For these cases as_svrepdesign(type = "subbootstrap") (or "mrbbootstrap") is the recommended path. The design is exported with a permissive specification (no finite-population correction at the Poisson stage, no pps argument), and the bootstrap resampler supplies the variance through replicate weights.

This is the package's bootstrap approximation for designs that exact Horvitz-Thompson linearization cannot express in survey::svydesign(). The subbootstrap and mrbbootstrap methods were developed for fixed-size PPS sampling (Antal and Tille 2011); their behavior on random-size Poisson designs, especially at multiple stages, has weaker theoretical backing and should be treated as an approximation. In particular, the resampling is fixed-size, so it does not capture the variance contribution of the random sample size and can materially understate the total variance of a Poisson-type design. When the exact Poisson linearization is available (single-stage designs), prefer as_svydesign().

Bounded cube, LPM2, and SCPS designs likewise have no native, design-specific replicate variance estimator in samplyr. as_svrepdesign(type = "subbootstrap") and "mrbbootstrap" export a generic PPS bootstrap approximation for them; they do not reproduce the original cube constraints or spatial selection algorithm within each replicate. Treat the resulting variance estimates as approximations, not as exact variance estimators for those designs.

Examples

sample <- sampling_design() |>
  stratify_by(region, alloc = "proportional") |>
  draw(n = 300) |>
  execute(bfa_eas, seed = 42)

rep_svy <- as_svrepdesign(sample, type = "auto")
survey::svymean(~households, rep_svy)
#>              mean     SE
#> households 71.668 3.6922