Bug: tbl_svysummary() INDEX error

[ddsjoberg]

dat_dput <- 
  structure(
    list(
      region = c(3, 3, 3, 3, 3, 3, 3, 3, 7, 7), 
      year = c(1972, 1972, 1972, 1972, 1972, 1972, 1972, 1972, 1972, 1972), 
      weights = c(0.663196271930943, 0.917370028585327, 0.897412512251031, 1.06634082743438, 0.94432371066466, 0.526887241987567, 
                  0.526887241987567, 0.546578869586901, 0.283198307048893, 0.494322145606406)), 
    row.names = c(NA, 10L), class = "data.frame")
 

dat_svy <- dat_dput |> 
  srvyr::as_survey_design(
    ids = region, 
    strata = year, 
    weights = weights, 
    nest = TRUE)

cardx::ard_continuous(dat_svy, variables = weights)
#> {cards} data frame: 3 x 8
#>   variable   context stat_name stat_label  stat fmt_fn
#> 1  weights continuo…    median     Median 0.897      1
#> 2  weights continuo…       p25  25% Perc… 0.527      1
#> 3  weights continuo…       p75  75% Perc… 0.944      1
#> ℹ 2 more variables: warning, error
gtsummary::tbl_svysummary(dat_svy, include = weights)
#> Warning in svymean.survey.design2(reformulate2(variable), design = data, :
#> Sample size greater than population size: are weights correctly scaled?
#> Error in tapply(y, by, sum, na.rm = na.rm, default = 0L): 'INDEX' is of length zero

^{Created on 2024-12-03 with reprex v2.1.1}

[ddsjoberg]

Hi @larmarange , It looks like this error arises from survey::svytable(). Before I close this out, is there something I am missing? Should we be supporting this type of tabulation?

library(magrittr)

dplyr::tribble(
  ~region, ~year,          ~weights,
  3,        1972, 0.663196271930943,
  3,        1972, 0.917370028585327,
  3,        1972, 0.897412512251031,
  3,        1972,  1.06634082743438,
  3,        1972,  0.94432371066466,
  3,        1972, 0.526887241987567,
  3,        1972, 0.526887241987567,
  3,        1972, 0.546578869586901,
  7,        1972, 0.283198307048893,
  7,        1972, 0.494322145606406
) %>%
  survey::svydesign(
    data = .,
    ids = ~region,
    strata = ~year,
    weights = ~weights,
    nest = TRUE
  ) %>%
  survey::svytable(formula = ~weights, design = .)
#> Error in tapply(y, by, sum, na.rm = na.rm, default = 0L): 'INDEX' is of length zero

^{Created on 2024-12-27 with reprex v2.1.1}

[larmarange]

Yes, svytable() has a bug when applied to the weighting variable. The issue could eventually be raised to the survey package.

However, I do not see when generating such table. I do understand that you are interested by looking at the weighting variable. But in that case you will perform a tbl_summary() (unweighted stats). I do not see that would be the purpose of describing the weight variable using the weights themselves...

[ddsjoberg]

Thanks for the details @larmarange ! I'll close it out and if someone raises it with survey in the future that'll be sufficient.