Skip to contents

Empirical likelihood for survey designs (NMAR)

Source: R/engines_el_impl_survey.R
el_survey.Rd

Internal method dispatched by el() when data is a survey.design.

Usage

# S3 method for class 'survey.design'
el(
  data,
  formula,
  auxiliary_means = NULL,
  standardize = TRUE,
  strata_augmentation = TRUE,
  trim_cap = Inf,
  control = list(),
  on_failure = c("return", "error"),
  variance_method = c("bootstrap", "none"),
  bootstrap_reps = 500,
  n_total = NULL,
  start = NULL,
  trace_level = 0,
  family = logit_family(),
  ...
)

Arguments

data

A survey.design created with survey::svydesign().

formula

Two-sided formula with an NA-valued outcome on the LHS; auxiliaries on the first RHS and, optionally, missingness predictors on the second RHS partition.

auxiliary_means

Named numeric vector of population means for auxiliary design columns. Names must match the materialized model.matrix columns on the first RHS (after formula expansion), including factor indicators and transformed terms. The intercept is always excluded.

standardize

Logical; standardize predictors.

strata_augmentation

Logical; when TRUE (default), augment the auxiliary design with stratum indicators and stratum shares when a strata structure is present in the survey design.

trim_cap

Numeric; cap for EL weights (Inf = no trimming).

control

List; solver control for nleqslv::nleqslv(control = ...).

on_failure

Character; "return" or "error" on solver failure.

variance_method

Character; "bootstrap" or "none".

bootstrap_reps

Integer; reps when variance_method = "bootstrap".

n_total

Optional analysis-scale population size N_pop; required for respondents-only designs.

start

Optional list of starting values passed to solver helpers.

trace_level

Integer 0-3 controlling estimator logging detail.

family

Missingness (response) model family specification (defaults to logit).

...

Passed to solver.

Value

c("nmar_result_el","nmar_result").

Details

Implements the empirical likelihood estimator with design weights. If n_total is supplied, it is treated as the analysis-scale population size N_pop used in the design-weighted QLS system. If n_total is not supplied, sum(weights(design)) is used as N_pop. Design weights are not rescaled internally; the EL equations use respondent weights and N_pop via \(T_0 = N_{\mathrm{pop}} - \sum d_i\) in the linkage equation. When respondents-only designs are used (no NA in the outcome), n_total must be provided; if auxiliaries are requested you must also provide population auxiliary means via auxiliary_means. Result weights are the unnormalized EL masses \(d_i / D_i(\theta)\) on this analysis scale; weights(result, scale = "population") sums to N_pop.