Non-probability Samples

A survey design object for non-probability samples (e.g., online panels, quota samples, volunteer panels) with calibration weights (including raking and post-stratification) or inverse probability weighting (IPW) pseudo-weights. Create with as_survey_nonprob().

Usage

survey_nonprob(
  data = data.frame(),
  metadata = survey_metadata(),
  variables = list(),
  groups = character(0),
  call = NULL,
  calibration = NULL,
  reference_sample = NULL
)

Arguments

data

A data.frame containing the survey data. Prefer as_survey_nonprob() over calling this constructor directly.

metadata

A survey_metadata object. Created automatically by as_survey_nonprob().

variables

A named list of design specification (weights, probs_provided). Set automatically by as_survey_nonprob().

groups

Set by surveytidy's group_by(). Always character(0) in standalone surveycore use.

call

Language object capturing the construction call.

calibration

The calibration provenance object returned by a surveywts calibration function (e.g., surveywts::rake()), or NULL if calibration was performed externally. Stores the calibration targets, variables, and trimming parameters for reproducibility and future bootstrap re-calibration. Default NULL.

reference_sample

Optional survey_taylor object representing the probability-based reference sample used to estimate propensity scores or calibration targets. Stored for reproducibility. Default NULL.

Value

A survey_nonprob object.

Variance estimation

Two modes are available, selected by whether @variables$repweights is NULL:

SRS approximation (no replicate weights)

Standard errors treat the calibrated weights as fixed and assume simple random sampling. This understates calibration uncertainty and should only be used when replicate weights are unavailable.

Replicate variance (repweights supplied)

Bootstrap or jackknife replicate weights propagate calibration uncertainty into the variance estimate. Each replicate column must contain calibrated weights re-estimated on one replicate draw. This is the recommended approach.

See as_survey_nonprob() for the full parameter interface, including type, scale, rscales, and mse.

Non-probability samples

Unlike as_survey(), as_survey_replicate(), and as_survey_twophase(), this class does not assume a probability sampling design. When no replicate weights are supplied, standard errors rest on a model-assisted SRS assumption, which is consistent with common practice for calibrated non-probability samples (e.g., raked online panels). When replicate weights are supplied, bootstrap or jackknife variance is used instead. See vignette("creating-survey-objects") for guidance on choosing between these modes and the limitations of each.

Design variables (@variables)

weights

Character string naming the (calibrated) weight column.

repweights

Character vector of bootstrap replicate weight column names, or NULL when no replicate weights are present.

type

Replicate type. Only "bootstrap" is supported for non-probability samples ("JK1", "JK2", and "JKn" are not accepted); or NULL when no replicate weights are present.

scale

Numeric scale factor for the variance formula, or NULL.

rscales

Per-replicate scale factors, or NULL.

mse

Logical. TRUE for MSE form of variance, or NULL.

probs_provided

Always FALSE for calibrated designs.

Calibration provenance (@calibration)

When calibration is performed via surveywts, the returned calibration object is stored here. It contains the calibration targets, variables used, trimming cap, effective sample size before and after, and design effect. NULL when calibration was performed externally (e.g., via anesrake).

See also

as_survey_nonprob() to create a survey_nonprob object.

Other constructors: as_caldata(), as_survey(), as_survey_nonprob(), as_survey_replicate(), as_survey_twophase(), survey_glm(), survey_glm_fit(), survey_replicate(), survey_taylor(), survey_twophase()