Skip to contents

Set a survey weighting scheme for the session

Source: R/weights.R
set_weights.Rd

Defines target shares for one or more categorical variables (e.g. a known population split by gender or region). Once set, the summary helpers (calc_percentage(), calc_nps(), calc_summary(), crosstab()) weight their results automatically, so the weighted margins match your targets. Pass weights = FALSE to any of them to opt out for a single call.

Usage

set_weights(...)

Arguments

...

One scheme per variable. Two forms are accepted:

  • vector form (one argument per variable) carrying a variable entry and a share per category: c(variable = "demo_gender", Male = 0.49, Female = 0.50, "Non-binary" = 0.01);

  • named form: demo_gender = c(Male = 0.49, Female = 0.50), where the argument name is the column. Shares need not sum to 1 – they are normalised per variable (so raw percentages or counts work too).

Value

Invisibly, the parsed scheme (a named list of normalised targets).

Details

With a single variable the weights are exact post-stratification: target_share / sample_share for each category. With several variables the weights are found by raking (iterative proportional fitting) so every variable's weighted margin matches its targets. Weights are normalised to mean 1 (so the weighted base equals the sample size), and respondents whose weighting value is blank / a non-answer (see na_blank()) are left unadjusted. A category that appears in the data but is missing from your targets is an error – give every observed category a share. If a default dataset is set, this prints the Kish design effect and the effective sample size so you can see the precision cost.

See also

clear_weights(), get_weights(), weight_vector(), calc_percentage().

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

Examples

set_weights(c(variable = "demo_gender",
              "Male" = 0.49, "Female" = 0.50,
              "Non-binary" = 0.01))
#> Weighting scheme stored for: demo_gender. It applies once a dataset is in play.
calc_percentage(podracing_survey, satis_return)   # gains a wpct column
#> # A tibble: 5 × 4
#>   satis_return      n   pct  wpct
#>   <ord>         <int> <dbl> <dbl>
#> 1 Very unlikely    75     8     8
#> 2 Unlikely        162    16    16
#> 3 Not sure        238    24    24
#> 4 Likely          270    27    27
#> 5 Very likely     255    26    25
clear_weights()