Compute the per-respondent survey weights

Returns the weight vector a scheme implies for a dataset – the same numbers the summary helpers use internally – so you can inspect them, attach them as a column, or feed them to another tool.

Usage

weight_vector(data = NULL, weights = NULL)

Arguments

data

A data frame. If omitted, the session default (use_dataset()) is used.

weights

A weighting scheme (any form accepted by set_weights()). If NULL (default), the session scheme is used.

Value

A numeric vector of length nrow(data), normalised to mean 1.

Details

This is the function behind the automatic weighting: each respondent's weight is target_share / sample_share for their category (raked across variables when there are several), normalised so the weighted base equals the unweighted one. Pair it with set_weights()'s reported design effect to judge how much precision the weighting costs.

See also

set_weights(), calc_percentage().

Other weighting: clear_weights_cache(), get_weights(), set_weights()

Examples

target <- c(variable = "demo_gender",
            "Male" = 0.49, "Female" = 0.50, "Non-binary" = 0.01)
w <- weight_vector(podracing_survey, target)

# The sample under-represents women, so their weights are above 1 and the
# weighted share moves towards the target. Categories the target does not
# mention keep a weight of 1, which is why their shares do not move.
podracing_survey %>%
  mutate(weight = w) %>%
  group_by(demo_gender) %>%
  summarise(
    unweighted_pct = round(n() / nrow(podracing_survey) * 100, 1),
    weighted_pct = round(sum(weight) / nrow(podracing_survey) * 100, 1)
  )
#> # A tibble: 5 × 3
#>   demo_gender            unweighted_pct weighted_pct
#>   <chr>                           <dbl>        <dbl>
#> 1 ""                                3.2          3.2
#> 2 "Female"                         39.9         47.5
#> 3 "Male"                           52.5         46.6
#> 4 "Non-binary"                      2.7          1  
#> 5 "Prefer not to answer"            1.7          1.7