tabsurvey: Publication-Ready Analysis of Complex Survey Data

In R4VN: Health Data Analysis and Publication-Ready Reporting

Publication-Ready Analysis of Complex Survey Data

Description

Performs R4VN-style descriptive analysis, hypothesis testing, and effect estimation for complex survey data. The interface deliberately mirrors tab() while adding survey weights, strata, clusters, replicate weights, domain analysis, design-based standard errors, weighted and unweighted results, and optional population totals.

Usage

tabsurvey(
  data = NULL,
  vars = NULL,
  by = NULL,
  design = NULL,
  weight = NULL,
  strata = NULL,
  cluster = NULL,
  fpc = NULL,
  repweights = NULL,
  rep_type = NULL,
  weightscale = c("relative", "population"),
  nest = TRUE,
  subpop = NULL,
  result = c("weighted", "unweighted", "both"),
  bothstyle = c("columns", "rows"),
  statcols = c("separate", "compact"),
  rawn = TRUE,
  digit = 1,
  p_digit = 3,
  effect_digit = 2,
  level = 0.95,
  missing = c("ifany", "no", "always"),
  row = FALSE,
  col = TRUE,
  cell = FALSE,
  overall = c("first", "last", "none"),
  descriptive = TRUE,
  rvrow = NULL,
  rvcol = FALSE,
  test = TRUE,
  pvalue = TRUE,
  survey_test = c("F", "Chisq", "Wald", "adjWald"),
  or = FALSE,
  rr = FALSE,
  pr = FALSE,
  event = NULL,
  adjusted = NULL,
  multi = NULL,
  effect_ref = NULL,
  ci = TRUE,
  cimethod = c("logit", "likelihood", "beta", "mean", "asin", "xlogit"),
  quantile_method = c("mean", "beta", "xlogit", "asin", "score", "quantile"),
  se = FALSE,
  deff = FALSE,
  cv = FALSE,
  population = FALSE,
  lonely = NULL,
  bold_p = TRUE,
  p_bold = 0.05,
  test_note = TRUE,
  template = c("journal", "clean", "minimal"),
  append = NULL,
  file = NULL,
  raw = FALSE,
  name = FALSE,
  title = NULL,
  report = c("auto", "brief", "full", "custom"),
  interpretation = FALSE,
  show = TRUE
)

Arguments

data	Optional data frame. Normally omitted when a stored surveyset() design is used.
vars	Variables to summarize, created with vars(). R4VN prefixes are supported: unprefixed or b2./b3. categorical variables, c. mean/SD, q. median/IQR, and f. full continuous summaries. Deferred selectors such as vars(.), wildcards, and exclusions are resolved against the survey data.
by	Optional outcome/grouping variable. An unprefixed variable is treated as categorical. Use by = c.outcome for a continuous outcome with mean-oriented inference or by = q.outcome for a continuous outcome with rank-oriented descriptive tests.
design	Survey design. May be an r4vn_survey object, a stored design name, or a design object from the survey package. If omitted, the active design created by surveyset() is used.
weight, strata, cluster, fpc	Direct design arguments for one-off analyses. These are alternatives to design= and have the same meaning as in surveyset().
repweights	Optional replicate weights for a one-off design.
rep_type	Replicate design type when repweights is used.
weightscale	"relative" or "population" for a one-off design. Stored designs retain the value declared in surveyset().
nest	Logical for a one-off multistage design.
subpop	Optional logical domain/subpopulation expression, for example subpop = age >= 60 & sex == "Female". Domain estimation preserves the original survey design rather than naively rebuilding it after row deletion.
result	Which analysis system to show: "weighted" (default), "unweighted", or "both". "both" applies to descriptive statistics, tests, effect estimates, and confidence intervals, not only to percentages.
bothstyle	When result = "both", "columns" puts weighted and unweighted results in parallel columns; "rows" stacks them using an Analysis column.
statcols	Presentation of descriptive statistics and effect estimates. "separate" (default) places sample n, estimate, confidence interval, SE, DEFF, CV, population N, model effect, model confidence interval, and p-value in separate publication-ready columns. "compact" keeps the older compact style in which estimates and confidence intervals are combined in one cell.
rawn	Include the actual unweighted sample n in descriptive cells. This is especially important beside weighted estimates and also keeps n visible for unweighted continuous summaries. The default is TRUE.
digit	Decimal places for descriptive estimates.
p_digit	Decimal places for p-values.
effect_digit	Decimal places for OR, PR, RR, and beta estimates.
level	Confidence level. The default is 0.95.
missing	"ifany", "no", or "always" for categorical missing-value rows.
row, col, cell	Percentage denominator for categorical variables when by is categorical. Exactly one should be TRUE. The default is column percentage, matching a conventional Table 1/Table 2 layout.
overall	Position of the overall descriptive column: "first" (default), "last", or "none".
descriptive	Logical. Include descriptive statistics.
rvrow	Optional categorical row reversal, matching tab(). Use TRUE to reverse every categorical variable, or identify selected variables with vars(...), c(...), or a character vector. Reversing display order does not silently change the regression reference.
rvcol	Logical. Reverse the displayed levels of a categorical by variable, matching tab(). Event selection still follows the original outcome order unless event= is supplied.
test	Logical. Include omnibus/group-comparison tests.
pvalue	Logical. Include coefficient-level p-values beside effect estimates.
survey_test	Statistic for categorical design-adjusted association tests passed to survey::svychisq(). The default "F" is the Rao-Scott second-order F correction. Other useful choices include "Chisq", "Wald", and "adjWald".
or	Logical. For a binary categorical outcome, estimate odds ratios using logistic regression.
rr	Logical. For a binary outcome, estimate risk/prevalence ratios with a log-link modified Poisson model. In cross-sectional surveys this is interpreted as a prevalence ratio.
pr	Logical. Estimate prevalence ratios with a log-link modified Poisson model. Weighted models use survey::svyglm() with quasipoisson(link="log"); unweighted models use Poisson regression with a sandwich/robust covariance estimate.
event	Event level for a binary categorical outcome. By default the last observed outcome level is the event.
adjusted	Optional adjustment set. Supply vars(...), a character vector, TRUE, or "ALL". A separate adjusted model is fitted for each focal predictor.
multi	Optional multivariable set. Supply vars(...), a character vector, TRUE, or "ALL". Each reported focal effect comes from a model containing the complete requested multivariable set; this is equivalent to reporting coefficients from the common model. Reference prefixes inside multi = vars(...) are respected even when they differ from the descriptive/crude reference.
effect_ref	Optional backward-compatible explicit reference mapping for crude and separately adjusted categorical effects, for example effect_ref = list(sex = "Male", smoking = "No"). A named character vector is also accepted. b2./b3. prefixes remain the preferred compact R4VN syntax. Multivariable references come from multi= when that specification supplies its own prefix.
ci	Logical. Show confidence intervals at the selected level where they are available.
cimethod	Confidence-interval method for weighted proportions: "logit" (default), "likelihood", "beta", "mean", "asin", or "xlogit". If a method cannot handle an observed proportion of exactly 0 or 1, R4VN falls back to a design-based Wald interval and constrains displayed limits to the interval from 0 to 1.
quantile_method	Interval method used by survey::svyquantile(). The default is "mean"; alternatives supported by the installed survey version include "beta", "xlogit", and "asin". "score" is for ordinary survey designs; "quantile" is for replicate-weight designs and is not appropriate for jackknife quantile SEs.
se	Logical. Add a separate standard-error column for descriptive estimates. When unweighted results are requested, their conventional SE is also reported where defined.
deff	Logical. Add a separate with-replacement design-effect column for weighted statistics where the underlying survey statistic supports it.
cv	Logical. Add a separate coefficient-of-variation/relative-SE column where defined for weighted and unweighted descriptive estimates.
population	Logical. Append estimated population N and its confidence interval for categorical cells. This requires a design declared with weightscale = "population". R4VN will not relabel normalized weights as population totals.
lonely	Optional lonely-PSU rule for this analysis. If omitted, the rule stored in the design is used.
bold_p	Logical. Bold p-values smaller than p_bold in HTML.
p_bold	Threshold used when bold_p = TRUE.
test_note	Logical. Add footnotes describing the tests used.
template	HTML style: "journal", "clean", or "minimal".
append	Optional previous R4VN table object to place before this table in the generated HTML page.
file	Optional HTML output path. A temporary file is used when omitted.
raw	Logical. Use raw variable names instead of variable labels.
name	Logical. When labels exist, append the raw variable name in square brackets.
title	Optional table title.
report	Reporting profile: "auto" (simple publication-ready survey output), "brief" (weighted descriptives only unless the user explicitly requests more), "full" (weighted and unweighted results stacked by rows with SE, DEFF, CV, tests, and model details where available), or "custom" (legacy defaults plus exactly the options requested by the user).
interpretation	Logical. Add a cautious deterministic interpretation table. The default is FALSE.
show	Logical. Open the generated HTML report in the Viewer/browser.

Details

Dependency-light implementation. Beyond R4VN itself, tabsurvey() requires only the survey package for complex-survey estimation. Publication HTML is generated with base R; ggplot2, plotly, htmlwidgets, flextable, and similar presentation packages are not required. tabsurvey() is a table/inference function and does not create a plot, so it deliberately adds no plotting dependency. R4VN functions that do create plots should embed every requested plot directly in their Viewer/HTML report.

Weighted and unweighted are complete analysis modes. With result = "both", R4VN computes two parallel analyses. The unweighted side uses ordinary sample descriptions and conventional tests or regressions. The weighted side uses the declared survey design for descriptive estimates, standard errors, confidence intervals, Rao-Scott or design-based tests, and survey-weighted regression. This is intentionally more comprehensive than merely displaying a raw n beside a weighted percentage.

Default publication display. The default statcols = "separate" uses distinct columns for sample n, estimate, and confidence interval instead of combining them in one long cell. Optional SE, DEFF, CV, population totals, model effects, model confidence intervals, and model p-values are also separate columns. The default result = "weighted", rawn = TRUE shows the actual sample n together with the survey-weighted estimate. For categorical variables the weighted statistic is a percentage with a design-based confidence interval. For c. variables the weighted mean and weighted population SD are shown, with a design-based CI for the mean. For q. variables the weighted median and weighted IQR are shown, with a median CI when available.

Full summaries. A variable declared with f. produces separate mean (SD), median (IQR), and range rows so weighted and unweighted summaries can be compared without compressing incompatible statistics into one number.

Tests. For categorical predictor by categorical outcome, weighted inference uses survey::svychisq() and defaults to the second-order Rao-Scott F correction. Weighted continuous comparisons use design-based t/Wald tests for mean-oriented variables and survey::svyranktest() for median/rank-oriented variables.

Regression estimates. OR uses survey-weighted logistic regression. PR and RR use a log-link survey-weighted quasi-Poisson model. A continuous by = c.outcome or by = q.outcome automatically reports unstandardized beta coefficients; the q. prefix changes the descriptive/group test but beta remains a linear-regression coefficient, consistent with R4VN tab() conventions.

Reference categories. Categorical references follow vars() prefixes. For example b2.sex makes the second observed/displayed level the model reference. The same requested reference is used in weighted and unweighted models.

Domain analysis. Use subpop= instead of physically deleting observations and rebuilding a complex design. The survey domain/subset machinery keeps the design information needed for valid variance estimation.

Population totals. population = TRUE is intentionally blocked unless weightscale = "population". Weighted percentages, means, tests and regressions remain valid with normalized/relative survey weights, but their sum must not automatically be interpreted as the represented population.

Continuous outcomes. When by is continuous, predictor descriptions remain available and association tests/effect columns concern the continuous outcome. Categorical predictors are compared with t/ANOVA or rank tests as appropriate; numeric predictors are assessed by the slope test. The effect is an unstandardized beta coefficient with a confidence interval.

Replicate-weight designs. Replicate weights may be defined in surveyset() or directly in tabsurvey(). All statistics are then delegated to the corresponding survey replicate-design methods.

Reporting profiles. report = "auto" is the recommended default: it keeps the main table compact and weighted, automatically includes design-based tests when a by variable is present, and shows supporting design/test/effect tables in the Viewer. "brief" is deliberately descriptive. "full" adds the unweighted comparison plus SE, DEFF, and CV and stacks weighted/unweighted results by rows to avoid excessively wide tables. "custom" preserves the older option-by-option behavior. Interpretation is never automatic; set interpretation = TRUE.

Value

Invisibly returns an object of classes r4vn_tabsurvey, r4vn_tab, and list. Important components include:

• data: flat publication-ready table, compatible with tabexport();

• html, table_html, and file: rendered table;

• design: the R4VN survey design metadata;

• survey_design: the underlying survey design used after any domain restriction;

• metadata: resolved R4VN variable specifications;

• tests: long-form machine-friendly test results;

• effects: long-form machine-friendly OR/PR/RR/beta results;

• notes: table footnotes;

• tables: named end-user report tables including Main, Design, Tests, Effects, Precision, and Interpretation when available;

• diagnostics: survey-design and precision diagnostics;

• models: fitted survey/unweighted regression models used for reported effects;

• interpretation: optional deterministic interpretation table;

• subpop: domain expression, when used.

Recommended reporting

For a publication or survey report, describe the sampling design and source of the final analytic weight, identify strata and PSU variables, state any domain/subpopulation restriction, and report the actual sample n together with survey-weighted estimates and design-based confidence intervals. When a hypothesis test is reported, the survey-adjusted test should normally be treated as the inferential result for a complex probability sample.

When result = "both", the unweighted analysis is useful for data checking, transparency, and showing how weighting/design affects the result; it does not replace the design-based inference.

Common mistakes avoided by R4VN

• Do not interpret the sum of normalized/relative weights as a population size. Use weightscale = "population" only when the survey documentation supports an expansion-weight interpretation.

• Do not create a survey domain by deleting all observations outside the target subgroup and rebuilding the design. Prefer subpop = ....

• Do not assume one weight is correct for every variable in a public survey. When different analytic components require different weights, create multiple named designs with surveyset().

• Do not silently treat propensity-score IPTW, frequency weights, or analytic regression weights as sampling/design weights.

References

Lumley T. Complex Surveys: A Guide to Analysis Using R. Wiley; 2010.

Lumley T. Analysis of complex survey samples. Journal of Statistical Software. 2004;9(1):1-19.

See Also

surveyset, tab, vars, tabexport

Other R4VN survey: surveyset()

Other R4VN tables: tab(), tabexport(), tabforest(), tablong(), tabmeta(), tabmulti(), tabscale(), tabscore(), vars()

Examples

# Reproducible complex-survey data used throughout the examples.
set.seed(2026)
d <- expand.grid(
  person = 1:2, household = 1:5, psu = 1:6, strata = 1:4,
  KEEP.OUT.ATTRS = FALSE
)
n <- nrow(d)
d$sex <- factor(sample(c("Female", "Male"), n, TRUE),
                levels = c("Female", "Male"))
d$age <- pmin(85, pmax(18, round(rnorm(n, 46, 14))))
d$bmi <- round(rnorm(n, 23.5, 3.4), 1)
d$income <- round(exp(rnorm(n, log(8), .5)), 1)
d$smoking <- factor(sample(c("No", "Yes"), n, TRUE, c(.72, .28)),
                    levels = c("No", "Yes"))
d$education <- factor(
  sample(c("Primary", "Secondary", "College+"), n, TRUE),
  levels = c("Primary", "Secondary", "College+")
)
d$wt <- exp(.15 * (d$sex == "Male") + rnorm(n, 0, .3))
d$labwt <- d$wt * exp(rnorm(n, 0, .12))
d$popwt <- d$wt * 5000
d$fpc1 <- 30
d$fpc2 <- 100
lp <- -5 + .055 * d$age + .08 * (d$bmi - 23) +
      .45 * (d$sex == "Male") + .55 * (d$smoking == "Yes")
d$hypertension <- factor(rbinom(n, 1, plogis(lp)),
                         levels = 0:1, labels = c("No", "Yes"))
d$sbp <- 82 + .72 * d$age + .85 * d$bmi +
         5 * (d$sex == "Male") + rnorm(n, 0, 13)

# Declare the survey design once; later tabsurvey() calls can stay short.
usedf(d)
surveyset(weight = wt, strata = strata, cluster = psu, nest = TRUE)

# 1. Simplest weighted publication table. report="auto" is the default.
s1 <- tabsurvey(vars = vars(c.age, sex, c.bmi, smoking), show = FALSE)
s1$tables$Main
s1$tables$Design

# 2. R4VN continuous prefixes: c.=mean, q.=median, f.=full summary.
s2 <- tabsurvey(vars = vars(c.age, q.income, f.bmi, sex), show = FALSE)

# 3. Table by a binary outcome; design-based tests are automatic.
s3 <- tabsurvey(
  vars = vars(c.age, sex, c.bmi, smoking, education),
  by = hypertension, show = FALSE
)
s3$tables$Tests

# 4. Compare complete unweighted and weighted analyses side by side.
s4 <- tabsurvey(
  vars = vars(c.age, sex, q.income, c.bmi, smoking),
  by = hypertension, result = "both", bothstyle = "columns",
  show = FALSE
)

# 5. Full profile: both analyses stacked by rows plus SE, DEFF, and CV.
s5 <- tabsurvey(
  vars = vars(c.age, sex, c.bmi, smoking),
  by = hypertension, report = "full", show = FALSE
)
s5$tables$Precision

# 6. Brief profile: weighted descriptive summary only unless overridden.
s6 <- tabsurvey(
  vars = vars(c.age, sex, q.income, c.bmi),
  report = "brief", show = FALSE
)

# 7. Row or cell percentages instead of the default column percentages.
s7_row <- tabsurvey(
  vars = vars(sex, smoking, education), by = hypertension,
  row = TRUE, col = FALSE, cell = FALSE, show = FALSE
)
s7_cell <- tabsurvey(
  vars = vars(sex, smoking, education), by = hypertension,
  row = FALSE, col = FALSE, cell = TRUE, show = FALSE
)

# 8. Crude survey-weighted odds ratios in the same publication table.
s8 <- tabsurvey(
  vars = vars(c.age, b2.sex, c.bmi, b2.smoking, education),
  by = hypertension, or = TRUE, event = "Yes", show = FALSE
)
s8$tables$Effects

# 9. Separately adjusted OR for every focal predictor.
s9 <- tabsurvey(
  vars = vars(c.age, b2.sex, c.bmi, b2.smoking),
  by = hypertension, or = TRUE, event = "Yes",
  adjusted = vars(c.age, b2.sex), show = FALSE
)

# 10. One common multivariable model containing all requested predictors.
s10 <- tabsurvey(
  vars = vars(c.age, b2.sex, c.bmi, b2.smoking, education),
  by = hypertension, or = TRUE, event = "Yes",
  multi = TRUE, show = FALSE
)
s10$models

# 11. Prevalence ratio via survey-weighted modified Poisson regression.
s11 <- tabsurvey(
  vars = vars(c.age, b2.sex, c.bmi, b2.smoking),
  by = hypertension, pr = TRUE, event = "Yes",
  multi = TRUE, show = FALSE
)

# 12. Explicit named reference levels; b2./b3. are also supported.
s12 <- tabsurvey(
  vars = vars(sex, smoking, education, c.age),
  by = hypertension, or = TRUE, event = "Yes",
  effect_ref = list(sex = "Male", smoking = "Yes",
                    education = "Secondary"),
  show = FALSE
)

# 13. Continuous outcome: unstandardized beta is reported automatically.
s13 <- tabsurvey(
  vars = vars(c.age, b2.sex, c.bmi, b2.smoking),
  by = c.sbp, multi = TRUE, result = "both", show = FALSE
)

# 14. q. continuous outcome requests rank-oriented group tests; effect is beta.
s14 <- tabsurvey(
  vars = vars(b2.sex, b2.smoking, education),
  by = q.sbp, result = "both", show = FALSE
)

# 15. Correct domain/subpopulation analysis; do not rebuild a reduced design.
s15 <- tabsurvey(
  vars = vars(c.age, sex, c.bmi, smoking),
  subpop = age >= 60 & sex == "Female", show = FALSE
)
s15$diagnostics$domain

# 16. Missing rows can be shown if present, always, or never.
d$smoking[1:4] <- NA
surveyset(d, name = "missing_demo", weight = wt, strata = strata,
          cluster = psu)
s16 <- tabsurvey(
  vars = vars(smoking, sex), design = "missing_demo",
  missing = "ifany", show = FALSE
)

# 17. Confidence level is fully dynamic, including the displayed CI label.
s17 <- tabsurvey(
  vars = vars(c.age, sex, c.bmi), by = hypertension,
  or = TRUE, event = "Yes", level = .90, show = FALSE
)
names(s17$data)  # contains "90% CI"

# 18. Request SE, design effect, and CV explicitly in a custom report.
s18 <- tabsurvey(
  vars = vars(c.age, sex, c.bmi, smoking),
  report = "custom", se = TRUE, deff = TRUE, cv = TRUE,
  show = FALSE
)
s18$tables$Precision

# 19. Population totals require declared expansion/population weights.
surveyset(d, name = "population", weight = popwt, strata = strata,
          cluster = psu, weightscale = "population", active = FALSE)
s19 <- tabsurvey(
  vars = vars(sex, education, hypertension), design = "population",
  population = TRUE, show = FALSE
)

# 20. One-off design: no prior surveyset() call is required.
s20 <- tabsurvey(
  d, vars = vars(c.age, sex, c.bmi, hypertension),
  weight = wt, strata = strata, cluster = psu, show = FALSE
)

# 21. Multiple named weight systems can coexist.
surveyset(d, name = "laboratory", weight = labwt,
          strata = strata, cluster = psu, active = FALSE)
s21 <- tabsurvey(
  vars = vars(c.age, sex, c.bmi), design = "laboratory", show = FALSE
)

# 22. Multistage clusters and finite-population corrections.
surveyset(
  d, name = "multistage", weight = wt, strata = strata,
  cluster = vars(psu, household), fpc = vars(fpc1, fpc2),
  nest = TRUE, active = FALSE
)
s22 <- tabsurvey(
  vars = vars(c.age, sex, c.bmi), design = "multistage", show = FALSE
)

# 23. Compact legacy cells and display-order controls.
s23 <- tabsurvey(
  vars = vars(sex, smoking), by = hypertension,
  statcols = "compact", rvrow = TRUE, rvcol = TRUE,
  report = "custom", show = FALSE
)

# 24. Interpretation is opt-in and remains separate from statistical output.
s24 <- tabsurvey(
  vars = vars(c.age, b2.sex, c.bmi, b2.smoking),
  by = hypertension, pr = TRUE, event = "Yes", multi = TRUE,
  report = "full", interpretation = TRUE, show = FALSE
)
s24$tables$Interpretation

# 25. Consistent result contract for custom reporting and downstream code.
names(s24$tables)
s24$descriptive
s24$tests
s24$effects
s24$diagnostics
s24$models
s24$interpretation

# 26. tabsurvey objects inherit from r4vn_tab and export with tabexport().
h <- tabexport(
  s3, s8, s11, s24, export = "html",
  file = tempfile("survey_report_"), quiet = TRUE
)
unlink(h$files)



# Additional syntax catalogue. These examples are intentionally not run by
# automatic checks, but are kept in ?tabsurvey for copy/paste use.

# 27. Risk ratio using the same modified-Poisson engine.
s27 <- tabsurvey(
  vars = vars(c.age, b2.sex, c.bmi, b2.smoking),
  by = hypertension, rr = TRUE, event = "Yes", multi = TRUE
)

# 28. Alternative CI methods for proportions and weighted quantiles.
s28_prop <- tabsurvey(
  vars = vars(sex, smoking, hypertension), cimethod = "beta"
)
s28_quantile <- tabsurvey(
  vars = vars(q.income, q.bmi), quantile_method = "beta"
)

# 29. Choose another design-adjusted categorical test.
s29 <- tabsurvey(
  vars = vars(sex, smoking, education), by = hypertension,
  survey_test = "Wald"
)

# 30. Display controls: no CI, no raw n, two decimals, Overall last.
s30 <- tabsurvey(
  vars = vars(c.age, sex, c.bmi, smoking), by = hypertension,
  ci = FALSE, rawn = FALSE, digit = 2, overall = "last"
)

# 31. Variable-name and HTML presentation controls.
s31_raw <- tabsurvey(
  vars = vars(c.age, sex, c.bmi), raw = TRUE,
  template = "clean", title = "Raw variable names"
)
s31_name <- tabsurvey(
  vars = vars(c.age, sex, c.bmi), name = TRUE,
  template = "minimal", title = "Labels plus names"
)

# 32. Inference/model-only table with descriptive cells suppressed.
s32 <- tabsurvey(
  vars = vars(c.age, b2.sex, c.bmi, b2.smoking),
  by = hypertension, or = TRUE, event = "Yes", multi = TRUE,
  descriptive = FALSE, report = "custom"
)

# 33. Explicitly suppress tests, coefficient p-values, and test notes.
s33 <- tabsurvey(
  vars = vars(c.age, b2.sex, c.bmi), by = hypertension,
  or = TRUE, event = "Yes", test = FALSE, pvalue = FALSE,
  test_note = FALSE, bold_p = FALSE, report = "custom"
)

# 34. Append two R4VN survey tables into one HTML page.
a34 <- tabsurvey(vars = vars(c.age, sex), show = FALSE)
f34 <- tempfile(fileext = ".html")
b34 <- tabsurvey(
  vars = vars(c.bmi, smoking), append = a34,
  file = f34, show = FALSE
)
unlink(f34)

# 35. A survey-package replicate design can be passed directly.
base35 <- survey::svydesign(
  ids = ~psu, strata = ~strata, weights = ~wt, data = d, nest = TRUE
)
rep35 <- survey::as.svrepdesign(base35, type = "bootstrap", replicates = 40)
s35 <- tabsurvey(
  vars = vars(c.age, sex, c.bmi, hypertension),
  design = rep35, report = "full"
)

# 36. Override the lonely-PSU rule for one analysis only.
s36 <- tabsurvey(
  vars = vars(c.age, sex, c.bmi), lonely = "average"
)

R4VN documentation built on Sept. 30, 2026, 5:13 p.m.