srvyr brings parts of dplyr’s syntax to survey analysis, using the survey package.
srvyr focuses on calculating summary statistics from survey data,
such as the mean, total or quantile. It allows for the use of many dplyr
verbs, such as summarize, group_by, and
mutate, the convenience of pipe-able functions, rlang’s
style of non-standard evaluation and more consistent return types than
the survey package.
You can try it out:
install.packages("srvyr")
# or for development version
# remotes::install_github("gergness/srvyr")First, describe the variables that define the survey’s structure with
the function as_survey()with the bare column names of the
names that you would use in functions from the survey package like
survey::svydesign(), survey::svrepdesign() or
survey::twophase().
library(srvyr, warn.conflicts = FALSE)
data(api, package = "survey")
dstrata <- apistrat %>%
as_survey_design(strata = stype, weights = pw)Now many of the dplyr verbs are available.
mutate() adds or modifies a variable.dstrata <- dstrata %>%
mutate(api_diff = api00 - api99)summarise() calculates summary statistics such as mean,
total, quantile or ratio.dstrata %>%
summarise(api_diff = survey_mean(api_diff, vartype = "ci"))
#> # A tibble: 1 × 3
#> api_diff api_diff_low api_diff_upp
#> <dbl> <dbl> <dbl>
#> 1 32.9 28.8 37.0group_by() and then summarise() creates
summaries by groups.dstrata %>%
group_by(stype) %>%
summarise(api_diff = survey_mean(api_diff, vartype = "ci"))
#> # A tibble: 3 × 4
#> stype api_diff api_diff_low api_diff_upp
#> <fct> <dbl> <dbl> <dbl>
#> 1 E 38.6 33.1 44.0
#> 2 H 8.46 1.74 15.2
#> 3 M 26.4 20.4 32.4my_model <- survey::svyglm(api99 ~ stype, dstrata)
summary(my_model)
#>
#> Call:
#> svyglm(formula = api99 ~ stype, design = dstrata)
#>
#> Survey design:
#> Called via srvyr
#>
#> Coefficients:
#> Estimate Std. Error t value Pr(>|t|)
#> (Intercept) 635.87 13.34 47.669 <2e-16 ***
#> stypeH -18.51 20.68 -0.895 0.372
#> stypeM -25.67 21.42 -1.198 0.232
#> ---
#> Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
#>
#> (Dispersion parameter for gaussian family taken to be 16409.56)
#>
#> Number of Fisher Scoring iterations: 2Here are some free resources put together by the community about srvyr:
dplyr::across and rlang’s “curly curly”
{{}})
Still need help?
I think the best way to get help is to form a specific question and ask it in some place like posit’s community website (known for it’s friendly community) or stackoverflow.com (maybe not known for being quite as friendly, but probably has more people). If you think you’ve found a bug in srvyr’s code, please file an issue on GitHub, but note that I’m not a great resource for helping specific issue, both because I have limited capacity but also because I do not consider myself an expert in the statistical methods behind survey analysis.
Have something to add?
These resources were mostly found via vanity searches on twitter & github. If you know of anything I missed, or have written something yourself, please let me know in this GitHub issue!
minimal changes to my #r #dplyr script to incorporate survey weights, thanks to the amazing #srvyr and #survey packages. Thanks to @gregfreedman & @tslumley. Integrates soooo nicely into tidyverse
–Brian Guay (@BrianMGuay on Jun 16, 2021)
Spending my afternoon using
srvyrfor tidy analysis of weighted survey data in #rstats and it’s so elegant. Vignette here: https://CRAN.R-project.org/package=srvyr/vignettes/srvyr-vs-survey.html–Chris Skovron (@cskovron on Nov 20, 2018)
- Yay!
–Thomas Lumley, in the Biased and Inefficient blog
I do appreciate bug reports, suggestions and pull requests! I started this as a way to learn about R package development, and am still learning, so you’ll have to bear with me. Please review the Contributor Code of Conduct, as all participants are required to abide by its terms.
If you’re unfamiliar with contributing to an R package, I recommend the guides provided by Rstudio’s tidyverse team, such as Jim Hester’s blog post or Hadley Wickham’s R packages book.