Allow cur_svy() to work inside mutate()?

[bschneidr]

I'm working on a pull request to try and implement the cur_svy_wts() function suggested in #136, and as a result I noticed that cur_svy() works inside summarize.tbl_svy() but not inside mutate.tbl_svy().

library(srvyr)

library(survey)
data(api)

# Create a survey design object
dstrata <- apistrat %>%
  as_survey_design(strata = stype, weights = pw)

# Able to access weights in summarize
dstrata %>%
  summarize(sum_of_wts = sum(1/cur_svy()$prob))
#> # A tibble: 1 x 1
#>   sum_of_wts
#>        <dbl>
#> 1      6194.

# But not in mutate
dstrata %>%
  mutate(wts = 1/cur_svy()$prob)
#> Error: Problem with `mutate()` column `wts`.
#> i `wts = 1/cur_svy()$prob`.
#> x Survey context not set

^{Created on 2021-12-19 by the reprex package (v2.0.0)}

Is there a design reason why cur_svy() isn't supported inside mutate()? If not, I'm happy to see if we can get it to work.

[gergness]

Yeah that makes sense. I just hadn’t thought of a situation where you’d want the survey context set in a mutate, maybe needs it for all dplyr verbs (filtering by weights too?)

[gergness]

Filtering by comparing to the mean in a single step is probably a more likely situation

[bschneidr]

Thanks, Greg. That's a good common example. Another more specialized reason for filtering on the weights is doing diagnostics of nonresponse adjustments. For example, you might want to look at cases where the adjusted weights are much different than the original weights. Or you might want to look at cases with especially large or small weights so that you can trim them to try to reduce variances.

Do you think it would make sense to add context for mutate() and filter()? I don't really see a downside other than hassle of implementation, but maybe I'm missing something. If you think it makes sense, I'm happy to put together a pull request for it.

[gergness]

Yep, makes sense! Thanks! I think it’s as simple as adding two lines like this, but I may be forgetting something
https://github.com/gergness/srvyr/blob/main/R/summarise.r#L6

[ray-p144]

I just ran into this today. I was trying to figure out how many missing values there were for the group with non-zero weights so I tried to use cur_svy() in filter(). It didn't work, so I tried using it in mutate() and that didn't work either.

I logged on to GitHub to see if there were any open issues on it just to find that you guys are already on top of it!

[gergness]

@bschneidr - did you get a chance to work on this? I'm hoping for some time to look at srvyr in the next week or two, would be nice to get this in the release. Happy to take a look myself if you haven't.

[bschneidr]

Yep, sorry for the delayed reply @gergness. I had to take some time off of open-source work the last few weeks but finally have time now to work on this.

[gergness]

closed via #139