surveycore (version 0.8.3)

as_survey_nonprob: Create a Calibrated / Non-Probability Survey Design

Description

Usage

as_survey_nonprob(data, weights, calibration = NULL)

Value

A survey_nonprob object.

Arguments

data

A data.frame containing the survey responses with pre-computed calibration weights. Must have at least one row and unique column names.

weights

<tidy-select> Calibration weight column (a single column, values strictly > 0). Typically produced by an external raking function (e.g., anesrake::anesrake()) or a surveywts calibration function.

calibration

Optional. The calibration provenance object returned by a surveywts calibration function (e.g., surveywts::rake()). Stored in @calibration for reproducibility. Supply NULL (the default) when calibration was performed externally and provenance metadata is not available. The object's structure is defined by surveywts and will be formally specified in Phase 2.5.

Phase 2.5 skeleton

This constructor is a skeleton. The resulting survey_nonprob object supports estimation via a model-assisted SRS variance assumption — the same as calling as_survey() with weights only. Full bootstrap re-calibration variance (which re-applies the raking procedure on each replicate) will be implemented in Phase 2.5 alongside the surveywts package.

When to use

Use as_survey_nonprob() instead of as_survey() when:

• Your data comes from a non-probability sample (online panel, quota sample, MTurk/Prolific, etc.)

• You have calibration or raking weights but no probability sampling design structure (no PSU IDs, strata, etc.)

• You want to explicitly record the provenance of your calibration weights for reproducibility

If your data comes from a probability sample with known design structure, use as_survey(), as_survey_replicate(), or as_survey_twophase() instead.

Variance estimation note

Standard errors from a survey_nonprob object assume simple random sampling within the calibrated weights. This is consistent with common applied practice for raked non-probability samples, but is technically a model-assisted approximation rather than design-based variance. See vignette("creating-survey-objects") for details and limitations.

Details

Creates a survey design object for non-probability samples and post-hoc calibrated designs (e.g., raked online panels, post-stratified samples). Accepts pre-computed calibration weights and optionally stores calibration provenance from surveywts output for reproducibility.

See Also

as_survey() for probability designs with Taylor variance, as_survey_replicate() for replicate-weight designs

Other constructors: as_survey(), as_survey_replicate(), as_survey_twophase(), survey_data(), survey_glm(), survey_glm_fit(), survey_nonprob(), survey_replicate(), survey_taylor(), survey_twophase()

Examples

# Minimal: pre-computed calibration weights from an external tool
df <- data.frame(
  y      = rnorm(200),
  age    = sample(c("18-34", "35-54", "55+"), 200, replace = TRUE),
  cal_wt = runif(200, 0.5, 2.5)
)
d <- as_survey_nonprob(df, weights = cal_wt)