Skip to contents

Effective Sample Size for a Survey Design

Source: R/analysis-effective-n.R
get_effective_n.Rd

Computes the effective sample size of a survey design using either the Kish (1965) weight-only approximation (method = "kish") or the full design-effect-based formula for a specified variable (method = "deff").

Usage

get_effective_n(
  design,
  x = NULL,
  group = NULL,
  method = c("kish", "deff"),
  na.rm = TRUE,
  decimals = NULL,
  min_cell_n = 30L,
  ...,
  .id = NULL,
  .if_missing_var = NULL
)

Arguments

design

A survey design object: survey_taylor, survey_replicate, survey_twophase, survey_nonprob, or a survey_collection.

x

<tidy-select> A single unquoted numeric variable name. Required when method = "deff"; ignored (with a message) when method = "kish". Default NULL.

group

<tidy-select> Optional grouping variable(s). Combined with any grouping set by group_by(). Default NULL.

method

Character(1). "kish" (default) or "deff". Controls the effective-N formula. Matched via match.arg().

na.rm

Logical. If TRUE (default), exclude observations with NA weights or group variables from the Kish computation; passed to get_means() for the DEFF computation.

decimals

Integer or NULL. Rounds n_eff and deff columns to this many decimal places. n is always integer and is never rounded. Default NULL.

min_cell_n

Integer. Minimum unweighted cell count before surveycore_warning_small_cell fires (Kish method only). Default 30L.

...

Unused. Reserved so that .id and .if_missing_var remain named-only when a survey_collection is passed.

.id

Character(1) or NULL. Column name identifying each survey in a survey_collection. Default NULL (uses the collection's stored @id).

.if_missing_var

"error", "skip", or NULL. Handling for surveys in a collection that lack x. Default NULL.

Value

A survey_effective_n tibble (also inheriting survey_result). Columns, in order:

  • [.id] — survey identifier column (when design is a collection).

  • [group_cols...] — group variable columns (when grouping is active).

  • n — integer. Unweighted count of observations.

  • n_eff — numeric. Effective sample size.

  • deff_kish — numeric. Weight-based design effect (n / n_eff). Present when method = "kish" only.

  • deff — numeric. Full design effect (Var_design / Var_SRS). Present when method = "deff" only.

Use meta(result)$method to retrieve the formula used. For DEFF, meta(result)$x is a named list with variable metadata.

Details

The Kish method (method = "kish") computes effective N from survey weights alone: n_eff = sum(w)^2 / sum(w^2). It captures only weight variation. For clustered designs with equal weights, deff_kish = 1.0 even when the true design effect is substantially greater due to clustering. Use method = "deff" to capture the full design effect for a specific analysis variable.

The DEFF method (method = "deff") computes effective N as n_eff = n / DEFF, where DEFF = Var_design / Var_SRS for variable x. It captures clustering, stratification, and weight variation jointly.

See also

Other analysis: clean(), get_anova(), get_corr(), get_covariance(), get_diffs(), get_freqs(), get_means(), get_pairwise(), get_quantiles(), get_ratios(), get_t_test(), get_totals(), get_variance(), meta()

Examples

d <- as_survey(nhanes_2017, ids = sdmvpsu, weights = wtint2yr,
               strata = sdmvstra, nest = TRUE)

# Kish effective N (weight-only approximation)
get_effective_n(d)
#> # A tibble: 1 × 3
#>       n n_eff deff_kish
#>   <int> <dbl>     <dbl>
#> 1  9254 3820.      2.42

# Full DEFF effective N for a specific variable
get_effective_n(d, ridageyr, method = "deff")
#> # A tibble: 1 × 3
#>       n n_eff  deff
#>   <int> <dbl> <dbl>
#> 1  9254 2425.  3.82