Warning when using weights < 1 in tbl_svysummary about gtsummary HOT 5 CLOSED

Thank you @LucyMcGowan for the report and the simple reprex (much appreciated!).

The warning appears from using the deff=TRUE argument.

library(survey, warn.conflicts = FALSE)
#> Loading required package: grid
#> Loading required package: Matrix
#> Loading required package: survival

d <- mtcars[, 1, drop = FALSE]
d$weight <- runif(nrow(mtcars), 0, 0.5)
des <- svydesign(~1, weights = ~weight, data = d)
svymean(~mpg, des, na.rm = TRUE, deff = TRUE)
#> Warning in svymean.survey.design2(~mpg, des, na.rm = TRUE, deff = TRUE): Sample
#> size greater than population size: are weights correctly scaled?
#>        mean      SE DEff
#> mpg 20.3106  1.1444   NA

^{Created on 2023-11-28 with reprex v2.0.2}

@larmarange the line in question is coming from here:

I don't know why/if this is needed. Can you chime in, please?

from gtsummary.

Ah! I was trying to fiddle with it and somehow missed that option — estimating the Design Effect definitely doesn’t make sense from a propensity score weighting perspective but not sure how that is used internally

from gtsummary.

@LucyMcGowan The design effect is only used to estimate the proportion of missing data and not in the primary mean calculation. @larmarange can comment on these differences with more authority than I can!

As an aside: we're planning a rather large refactor, and if there are pain points or suggestions, please feel free to voice them here!

from gtsummary.

Hi, following #1486, we have added design effects to the list of statistics available in tbl_svysummary().

Due to the design of the functions, inherited from tbl_summary(), summarize_continuous_survey() computes deff only on demand, while summarize_categorical_survey() always computes all the possible statistics.

To be noted, even for a continuous variable, summarize_categorical_survey() is called to calculate the number of observations.

We probably need to update summarize_categorical_survey() to compute deff only when required.

from gtsummary.

Thanks @larmarange !

FYI @LucyMcGowan in the next version of the package, the messaging will be more clear where any warnings come from, for example it will say that this warning comes from estimating the statistic to measure the proportion of missing data.

from gtsummary.