Core Survey Analysis Infrastructure.

surveycore

surveycore is the foundation of the surveyverse ecosystem — a modern, tidyverse-compatible replacement for the survey and srvyr packages in R.

It provides S7-based survey design objects with:

• A tidy-select interface (ids = c(psu, ssu), no formula syntax)
• Automatic preservation of haven-style variable labels and value labels
• Exact variance estimation (Taylor linearization, replicate weights, two-phase designs)
• Seamless conversion to and from survey::svydesign and srvyr::tbl_svy

For a side-by-side comparison with survey and srvyr, see vignette("surveycore-vs-survey").

Installation

# From CRAN:
install.packages("surveycore")

# Development version from GitHub:
# install.packages("pak")
pak::pak("JDenn0514/surveycore")

What surveycore provides

• S7 survey objects: survey_taylor, survey_replicate, survey_twophase, survey_nonprob
• Constructors: as_survey(), as_survey_replicate(), as_survey_twophase(), as_survey_nonprob()
• Metadata system: set_var_label(), set_val_labels(), extract_var_label(), extract_val_labels() — with automatic haven attribute import
• Analysis functions: get_freqs(), get_means(), get_totals(), get_corr(), get_quantiles(), get_ratios(), get_diffs()
• Regression: survey_glm() for survey-weighted GLMs with clean() for tidy coefficient tables
• Design utilities: update_design(), as_svydesign(), from_svydesign(), as_tbl_svy(), from_tbl_svy()

Who is this for?

surveycore is intended for:

• Survey researchers and methodologists who analyse complex probability samples and need design-consistent variance estimates (stratified, clustered, replicate-weight, and two-phase designs).
• Social scientists, epidemiologists, and public health researchers working with population surveys such as NHANES, ACS, GSS, or custom organizational surveys.
• R users who want a tidyverse-compatible interface for the survey analysis workflows currently served by survey and srvyr.

The software is designed to analyse rectangular survey microdata: one row per respondent, numeric or categorical outcome variables, and either explicit survey weights or a design specification (ids, strata, FPC). It supports:

• Data frames, tibbles, and data.table objects as input.
• Variables with haven-style variable labels and value labels (e.g. from .xpt or .sav files read with haven).
• Grouped analyses (via surveytidy::group_by()).

Each analysis function accepts specific types of outcome variables:

Basic usage

library(surveycore)

# ── Simple SRS design ──────────────────────────────────────────────────────────
set.seed(42)
df <- data.frame(
  psu = rep(1:10, each = 10),
  strata = rep(c("A", "B"), each = 50),
  weight = runif(100, 0.5, 2),
  income = rnorm(100, 50000, 10000),
  age = sample(18:80, 100, replace = TRUE)
)

d <- as_survey(df, ids = psu, weights = weight, strata = strata, nest = TRUE)
d
#> 
#> ── Survey Design ───────────────────────────────────────────────────────────────
#> <survey_taylor> (Taylor series linearization)
#> Sample size: 100
#> 
#> # A tibble: 100 × 5
#>      psu strata weight income   age
#>    <int> <chr>   <dbl>  <dbl> <int>
#>  1     1 A       1.87  53219.    42
#>  2     1 A       1.91  42162.    33
#>  3     1 A       0.929 65757.    71
#>  4     1 A       1.75  56429.    41
#>  5     1 A       1.46  50898.    50
#>  6     1 A       1.28  52766.    78
#>  7     1 A       1.60  56793.    55
#>  8     1 A       0.702 50898.    60
#>  9     1 A       1.49  20069.    58
#> 10     1 A       1.56  52849.    39
#> # ℹ 90 more rows

# ── Weighted mean and total ────────────────────────────────────────────────────
get_means(d, income)
#> # A tibble: 1 × 4
#>     mean ci_low ci_high     n
#>    <dbl>  <dbl>   <dbl> <int>
#> 1 50206. 47921.  52490.   100
get_totals(d, income)
#> # A tibble: 1 × 4
#>      total   ci_low  ci_high     n
#>      <dbl>    <dbl>    <dbl> <int>
#> 1 6460063. 5906356. 7013770.   100

Complex survey designs

# ── Replicate weights (BRR) ───────────────────────────────────────────────────
df_rep <- data.frame(
  y = rnorm(20),
  wt = runif(20, 1, 3),
  rep1 = runif(20, 0.5, 2),
  rep2 = runif(20, 0.5, 2),
  rep3 = runif(20, 0.5, 2),
  rep4 = runif(20, 0.5, 2)
)

d_rep <- as_survey_replicate(
  df_rep,
  weights = wt,
  repweights = starts_with("rep"),
  type = "BRR"
)
d_rep
#> 
#> ── Survey Design ───────────────────────────────────────────────────────────────
#> <survey_replicate> (BRR, 4 replicates)
#> Sample size: 20
#> 
#> # A tibble: 20 × 6
#>         y    wt  rep1  rep2  rep3  rep4
#>     <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#>  1 -2.00   2.30 1.09  0.849 0.705 1.71 
#>  2  0.334  2.84 0.619 1.37  0.766 1.90 
#>  3  1.17   1.73 1.74  1.76  1.28  1.75 
#>  4  2.06   2.71 0.609 0.698 1.72  0.691
#>  5 -1.38   1.60 0.672 1.84  0.673 1.47 
#>  6 -1.15   1.93 1.46  1.18  1.84  1.54 
#>  7 -0.706  1.29 0.981 1.84  1.36  0.548
#>  8 -1.05   2.62 0.783 0.873 0.720 1.88 
#>  9 -0.646  2.33 1.09  0.626 1.85  1.22 
#> 10 -0.185  1.12 1.79  0.573 0.880 0.900
#> # ℹ 10 more rows

Variable labels

surveycore preserves haven-style labels automatically when reading .xpt or .sav files. You can also set labels manually:

d2 <- set_var_label(d, income = "Annual household income (USD)")
d2 <- set_var_label(d2, age = "Respondent age in years")

extract_var_label(d2, income)
#>                          income 
#> "Annual household income (USD)"
extract_var_label(d2, age)
#>                       age 
#> "Respondent age in years"

Conversion to/from survey and srvyr

# To survey::svydesign
svy <- as_svydesign(d)
class(svy)
#> [1] "survey.design2" "survey.design"

# Back to surveycore
d_rt <- from_svydesign(svy)
d_rt

The surveyverse ecosystem

surveycore is the foundation of the surveyverse — a family of packages built around it:

• surveytidy — dplyr verbs (filter(), select(), mutate(), group_by()) that respect survey design structure, so grouped summaries and subsetting always propagate weights and strata correctly.
• surveywts — calibration and post-stratification for survey weights. Coming soon.

Development status

The package API is stable. The core classes, constructors, and analysis functions (get_freqs() through get_diffs()) are not expected to change in breaking ways. New analysis functions may be added in future releases. See NEWS.md for the full changelog.

Code of Conduct

Please note that the surveycore project is released with a Contributor Code of Conduct. By contributing to this project, you agree to abide by its terms.

License

GPL-3. Variance estimation code vendored from the survey package (Thomas Lumley, GPL-2/GPL-3) — see VENDORED.md for full attribution.

References

Lumley T (2004). “Analysis of Complex Survey Samples.” Journal of Statistical Software, 9(1), 1–19.

Lumley T (2010). Complex Surveys: A Guide to Analysis Using R. John Wiley and Sons.