Confidence intervals for survey result objects — confint.survey

Computes confidence intervals for the parameters of a survey_result object using the t-distribution with per-parameter degrees of freedom stored in attr(object, ".survey_result")$df.

Usage

# S3 method for class 'survey_result'
confint(object, parm, level = 0.95, ...)

Arguments

object: A survey_result object from any supported get_*() function.
parm: A numeric, integer, or character vector selecting a subset of parameters, or missing (all parameters). Character parm is matched against names(coef(object)).
level: Confidence level, strictly between 0 and 1. Default 0.95.
...: Unused.

Value

A numeric matrix with two columns. Column names are "2.5 %" and "97.5 %" for level = 0.95 (matching stats::confint.default() format). Rownames are names(coef(object))[parm]. Dimensions: length(parm) × 2.

Details

For non-calibrated Taylor, replicate, and nonprob designs, df = Inf is stored, so qt(1 - (1 - level)/2, df = Inf) equals qnorm(...) — a normal approximation. For calibrated Taylor designs, per-cell finite df values are used. Either way the stored df is the one behind the result's ci_low and ci_high columns, so for get_means(), get_totals(), get_freqs(), get_ratios() and get_covariance(), confint(object, level = l) returns the same interval as those columns when the result was computed at conf_level = l. This holds on a filtered (domain) design too.

When SE(object)[i] is NA, the corresponding CI row is c(NA, NA).

Examples

d <- as_survey(
  nhanes_2017,
  ids = sdmvpsu,
  weights = wtint2yr,
  strata = sdmvstra,
  nest = TRUE
)
result <- get_means(d, ridageyr, variance = "se")
confint(result)
#>             2.5 %   97.5 %
#> ridageyr 37.39616 39.45179
confint(result, level = 0.90)
#>              5 %     95 %
#> ridageyr 37.5614 39.28654