Beautiful tables in R: kable, kableExtra, gt, gtsummary and gtExtras

A good table is an argument made visible. It should let the reader identify the population, comparison, units, denominators and uncertainty without reverse-engineering the data frame that produced it.

This resource covers five related tools:

• knitr::kable() for small, portable tables;
• kableExtra for styling and structure around a kable table;
• gt for composing a display table from defined parts;
• gtsummary for analytical summaries and model results; and
• gtExtras for optional visual elements added to gt tables.

Select Run R above an example to execute it in the browser and render the resulting table directly beneath the code. The first use of a package can take longer because its WebAssembly build and dependencies must be loaded. The computation happens on your device. RStudio and Quarto remain the appropriate environment for saving the table into a complete report.

Begin with table-ready data

Formatting cannot repair an unclear analysis table. Construct one row per intended table row and one column per displayed quantity.

summary_data <- data.frame(
  group = c("Control", "Intervention"),
  n = c(120, 118),
  age_mean = c(61.24, 60.87),
  age_sd = c(8.91, 9.16),
  event_n = c(24, 15)
)

summary_data$event_percent <- 100 * summary_data$event_n / summary_data$n
summary_data

Keep numeric columns numeric until presentation. Early conversion to strings such as "20.0%" prevents correct sorting, colour scales and later calculations.

Choosing a table system

Need	Sensible starting point
A compact data frame in HTML, PDF or Word	knitr::kable()
Spanners, grouped rows and styling around kable()	kableExtra
A carefully composed display table	gt
Baseline characteristics or regression results	gtsummary
Sparklines, bars and other in-cell additions	gtExtras

Do not combine packages merely to make the pipeline longer. Choose the system whose data model matches the table.

knitr::kable(): a dependable foundation

knitr::kable(
  summary_data,
  digits = c(0, 0, 1, 1, 0, 1),
  col.names = c("Group", "N", "Mean age", "SD", "Events", "Events (%)"),
  caption = "Participant characteristics by study group",
  align = c("l", "r", "r", "r", "r", "r")
)

Useful arguments include:

• digits for numeric precision;
• col.names for reader-facing labels;
• align for column alignment;
• caption for the table title; and
• format when output must be forced to HTML, LaTeX, pipe or another supported form.

Right-align quantities so readers can compare magnitudes. Left-align prose. Use decimal places to represent meaningful precision, not the precision stored by the computer.

Extend kable() with kableExtra

library(kableExtra)

summary_data |>
  knitr::kable(
    format = "html",
    digits = 1,
    col.names = c("Group", "N", "Mean", "SD", "n", "%"),
    caption = "Participant characteristics by study group"
  ) |>
  add_header_above(c(" " = 2, "Age (years)" = 2, "Outcome" = 2)) |>
  kable_styling(
    bootstrap_options = c("striped", "hover"),
    full_width = FALSE,
    position = "left"
  ) |>
  column_spec(1, bold = TRUE)

The pipe reads as a sequence: make the table, add a grouped header, choose broad styling, then target a column. Keep the styling restrained; repeated borders, colour and bolding compete with the values.

Conditional formatting

summary_data |>
  transform(
    event_percent = cell_spec(
      round(event_percent, 1),
      color = ifelse(event_percent > 15, "#6e2538", "#214c57"),
      bold = TRUE
    )
  ) |>
  knitr::kable(format = "html", escape = FALSE) |>
  kable_styling(full_width = FALSE)

When escape = FALSE, HTML is allowed through. Use it only with content you control; untrusted text could contain unwanted markup.

HTML and PDF are different targets

Some kableExtra styling maps to both HTML and LaTeX, while other behaviour is format-specific. Test the actual output format early. A table that works in an HTML preview may overflow or need LaTeX packages in PDF.

The anatomy of a gt table

gt starts with data and builds a display table from components: header, column labels, body, stub, row groups, spanners, footnotes and source notes.

library(gt)

summary_data |>
  gt(rowname_col = "group") |>
  tab_header(
    title = "Participant characteristics",
    subtitle = "Summary by allocated group"
  ) |>
  tab_spanner(
    label = "Age (years)",
    columns = c(age_mean, age_sd)
  ) |>
  tab_spanner(
    label = "Primary event",
    columns = c(event_n, event_percent)
  ) |>
  cols_label(
    n = "N",
    age_mean = "Mean",
    age_sd = "SD",
    event_n = "n",
    event_percent = "%"
  ) |>
  fmt_number(columns = c(age_mean, age_sd, event_percent), decimals = 1) |>
  tab_source_note("Values are illustrative.")

Format by meaning

financial <- data.frame(
  service = c("A", "B", "C"),
  cost = c(1203.4, 987.2, 1510.8),
  proportion = c(0.42, 0.37, 0.21),
  date = as.Date(c("2026-01-10", "2026-02-14", "2026-03-03"))
)

financial |>
  gt() |>
  fmt_currency(columns = cost, currency = "AUD") |>
  fmt_percent(columns = proportion, decimals = 0) |>
  fmt_date(columns = date, date_style = "d MMM y")

The stored proportions remain 0–1; fmt_percent() controls display. This is preferable to multiplying and pasting percent symbols into the underlying data.

Footnotes and missing values

summary_data |>
  gt() |>
  sub_missing(missing_text = "Not available") |>
  tab_footnote(
    footnote = "SD = standard deviation.",
    locations = cells_column_labels(columns = age_sd)
  )

Do not silently convert missing values to zero. “Not measured”, “not applicable” and “suppressed” may require different labels.

Targeted styling

summary_data |>
  gt() |>
  tab_style(
    style = cell_text(weight = "bold", color = "#6e2538"),
    locations = cells_body(columns = event_percent, rows = event_percent > 15)
  ) |>
  tab_options(
    table.font.names = c("Ubuntu", "Arial", "sans-serif"),
    table.border.top.color = "#6e2538",
    data_row.padding = px(7)
  )

Conditional styling should encode a meaningful rule that is stated or obvious. Avoid colouring cells merely because the package can.

gtsummary for descriptive Table 1

library(gtsummary)

trial |>
  dplyr::select(trt, age, grade, response) |>
  tbl_summary(
    by = trt,
    missing = "ifany",
    statistic = list(
      all_continuous() ~ "{mean} ({sd})",
      all_categorical() ~ "{n} ({p}%)"
    ),
    label = list(
      age ~ "Age (years)",
      grade ~ "Tumour grade",
      response ~ "Treatment response"
    )
  ) |>
  add_overall() |>
  bold_labels()

tbl_summary() chooses summaries according to variable type. Inspect those types and override defaults deliberately. A numeric variable containing category codes should usually be converted to a factor before summarising.

Adding comparisons

trial |>
  dplyr::select(trt, age, grade, response) |>
  tbl_summary(by = trt) |>
  add_p() |>
  add_n() |>
  bold_labels()

add_p() is not automatically desirable. Baseline significance testing in a randomised trial is generally uninformative because allocation generated the groups. In an observational table, p-values do not measure confounding or decide adjustment. Include comparisons only when they answer a stated question.

Regression tables

fit <- glm(response ~ age + grade + trt, data = trial, family = binomial)

fit |>
  tbl_regression(
    exponentiate = TRUE,
    label = list(age ~ "Age (years)", grade ~ "Tumour grade", trt ~ "Treatment")
  ) |>
  bold_labels()

With a logistic model and exponentiate = TRUE, coefficients appear as odds ratios. The display transformation does not change what the model estimates. State the reference categories and adjustment set.

Combine models

model_1 <- lm(marker ~ age + trt, data = trial)
model_2 <- glm(response ~ age + trt, data = trial, family = binomial)

tbl_merge(
  tbls = list(
    tbl_regression(model_1),
    tbl_regression(model_2, exponentiate = TRUE)
  ),
  tab_spanner = c("**Continuous outcome**", "**Binary outcome**")
)

Combine models only when side-by-side comparison helps. A dense wall of coefficients is not made clearer merely by placing it inside one table.

Additions with gtExtras

gtExtras extends gt with functions such as in-cell bars, point plots and sparklines. These can be useful for compact descriptive displays.

library(gt)
library(gtExtras)

summary_data |>
  gt() |>
  gt_plt_bar(
    column = event_percent,
    color = "#214c57",
    keep_column = TRUE
  ) |>
  cols_label(event_percent = "Events (%)")

An in-cell graphic should support the exact numeric value, not replace it. Consider whether the result still works in PDF, grayscale, screen readers and narrow layouts.

Exporting tables

table_object <- summary_data |>
  gt() |>
  fmt_number(columns = where(is.numeric), decimals = 1)

gtsave(table_object, "participant-table.html")
# gtsave(table_object, "participant-table.png")
# gtsave(table_object, "participant-table.docx")

Available export types and dependencies vary. Inspect the saved file rather than assuming it matches the Viewer pane.

Accessibility and editorial checks

Before publishing, ask:

• Does the caption state what population and comparison the table represents?
• Are counts accompanied by denominators where needed?
• Are units in labels rather than repeated in every cell?
• Are abbreviations expanded in footnotes?
• Are reference groups explicit?
• Can colour-coded meaning also be understood from text or symbols?
• Is rounding consistent with measurement precision?
• Are suppressed or missing values explained?
• Does the table remain usable on a phone and when printed?

Useful official references

• The gt reference
• gtsummary table-summary tutorial
• knitr::kable() and kableExtra